Mapping RNA–protein interactions with subcellular resolution using colocalization CLIP

Colocalization CLIP reveals HuR target interactions at subcellular resolution

To establish a universal method for delineating subcellular RBP–RNA interactions, we integrated CLIP with proximity labeling, using the engineered ascorbate peroxidase APEX2 (Lam et al. 2015; Hung et al. 2016). APEX labeling and similar proximity-based strategies (Qin et al. 2021) have been previously used to define subcellular proteomes (Rhee et al. 2013; Hung et al. 2014, 2017; Loh et al. 2016; Markmiller et al. 2018), transcriptomes (Kaewsapsak et al. 2017; Fazal et al. 2019; Padrón et al. 2019), and protein-occupied RNAs (Benhalevy et al. 2018). We hypothesized that combining a proximity approach with CLIP would allow us to simultaneously identify interactions of individual RBPs with their RNA targets at the subcellular level (Fig. 1A,B). We began by engineering cells to express APEX2 at predetermined locations such as the cytoplasm, nucleus, or within SGs using localization domains. For cytoplasmic localization, we used the nuclear export signal (NES) from cAMP-dependent protein kinase inhibitor (PKI) (Wen et al. 1995). For nuclear localization, we used the nuclear localization signal (NLS) from the SV40 large T antigen (Kalderon et al. 1984). Last, for SG targeting, we created a fusion protein with G3BP1 (Tourrière et al. 2023). After initiating the APEX reaction, we evaluated biotin labeling using fluorescent streptavidin and identified a streptavidin-reactive signal primarily confined to the labeling locations, which was stress-dependent in the case of G3BP1–APEX (Fig. 1C; Supplemental Fig. S1A).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 1.

Colocalization CLIP (coCLIP) for subcellular resolution of RBP–RNA interactions. (A) An APEX2 enzyme is localized to a specific compartment, depicted here as either nuclear, cytoplasmic, or within stress granules as a G3BP1 fusion. The proximity labeling reaction using biotin phenol and hydrogen peroxide is performed, followed by UV cross-linking. (B) After cell lysis, the sample undergoes antibody-based RBP–CLIP. A second streptavidin pulldown following RBP IP enables the capture of local RBP targets. (C) Immunofluorescence for streptavidin reactive signal with and without labeling reagents for arsenite-stressed Huh7.5 cells with nuclear, cytoplasmic, or G3BP1 localized APEX2. Images were captured at 20× magnification. Scale bar, 10 µm. (D) Example autoradiograms for input HuR CLIP (left) or cytoplasmic HuR coCLIP (right) from two replicates. The size of HuR is indicated with an arrow and excised HuR:RNA complexes are indicated with boxes. (OD) RNAse overdigestion, (-Lab) no label. (E) HuR peak counts (left) and proportions (right) for mRNAs within input CLIP or from nuclear, cytoplasmic, or G3BP1 coCLIP in Huh7.5 cells. (F) HuR peak counts (left) and proportions (right) as in E for ncRNA peaks. See also Supplemental Figures S1–S3.

We next focused on HuR as a candidate RBP to assess the viability of our coCLIP approach. HuR is a ubiquitously expressed member of the ELAV family of RBPs (Abdelmohsen and Gorospe 2010). HuR shuttles between the nucleus and cytoplasm and binds to AU-rich motifs in 3′ untranslated regions (3′ UTRs) of mRNAs to affect stability (Abdelmohsen and Gorospe 2010). In the nucleus, HuR also interacts with introns to modulate the alternative splicing of specific mRNAs (Lebedeva et al. 2011). Typically, >90% of HuR is located in the nucleus; however, under stress conditions, HuR relocates to the cytoplasm and accumulates in SGs (David et al. 2007), as illustrated in Supplemental Figure S1B. Given HuR's ability to interact with different transcript features depending on its location—intronic in the nucleus and 3′ UTR in the cytoplasm—it emerged as a suitable candidate RBP for defining subcellular RBP–RNA interactions with coCLIP.

After confirming that UV cross-linking does not induce spurious APEX2 activity or nonspecific biotinylation (Supplemental Fig. S2A), we initiated APEX labeling in cells, followed immediately by UV cross-linking (see Materials and Methods). We carried out antibody-based CLIP for HuR in arsenite-stressed and unstressed conditions (Fig. 1D; Supplemental Fig. S2B) and performed streptavidin enrichment of eluates from HuR pulldowns (Fig. 1D; Supplemental Fig. S2C). We observed no significant differences in HuR–RNA signal intensities from input HuR CLIP across samples (Supplemental Fig. S2B). However, we noted that HuR coCLIP in the nucleus gave a more robust signal than coCLIP from the cytoplasm or SGs (Supplemental Fig. S2C), reflecting the distribution of HuR at steady state.

We performed coCLIP and created sequencing libraries from NLS, NES, or G3BP1-tethered APEX2 under mock or sodium arsenite conditions, all with associated input and enriched libraries, for 58 sequenced libraries in total (Supplemental Table S1). To facilitate robust and consistent data analysis, similar to previous efforts for CLIP in nonmodel organisms (Rozen-Gagnon et al. 2021), we further created a custom single command-line analysis pipeline, termed “CLIPpityCLIP,” to process raw sequencing data into peaks for subsequent analysis and visualization (Supplemental Fig. S3; see Materials and Methods). We first compared coCLIP data to input HuR CLIP in both mock and stressed conditions. We retained peaks if they appeared in at least half of the replicates and had reads above the median count across all samples. In the mock condition, input HuR CLIP yielded ∼10⁴ HuR peaks on mRNAs with a characteristic preference for intronic (60%) and 3′ UTR (30%) binding as has been previously reported (Fig. 1E; Lebedeva et al. 2011; Mukherjee et al. 2011). Upon arsenite stress, we observed closer to 20,000 peaks, with a similar distribution in mRNA-binding preferences as in the mock condition (Fig. 1E). In coCLIP data sets, nuclear coCLIP showed an enrichment for introns (75%), similar to the input HuR CLIP, whereas cytoplasmic and G3BP1 coCLIPs revealed a preference for 3′-UTR binding (75%) (Fig. 1E). With stress, the HuR binding preferences were generally conserved across HuR coCLIPs, with a 14% increase in the 3′-UTR binding proportion for G3BP1 coCLIP. Between coCLIP data sets, we observed ∼10-fold more HuR target peaks from nuclear coCLIP relative to cytoplasmic and G3BP1 coCLIP in unstressed cells, likely reflecting the steady state distribution of HuR (Fig. 1E). Moreover, with arsenite stress, we observed a threefold and fivefold increase in the number of 3′-UTR peaks in cytoplasmic and G3BP1 coCLIPs, respectively (Fig. 1E).

A similar binding correlation was observed among noncoding HuR targets, where input HuR CLIP and nuclear coCLIP resembled each other in both conditions. In contrast, cytoplasmic and G3BP1 coCLIP yielded 10-fold fewer peaks in unstressed cells compared to nuclear coCLIP (Fig. 1F), with the number of peaks increasing by fivefold and 2.5-fold, respectively, with arsenite stress. We observed a unique enrichment of transposable elements (TEs) in cytoplasm and near G3BP1 upon arsenite treatment, consistent with previously observed accumulation of LINE/SINE elements such as ALUs in SGs (Fig. 1F; Moon and Namkoong 2023). Taken together, these results suggest that coCLIP can parse dynamic subpopulations of HuR within different cellular compartments under mock and stress conditions in situ.

HuR binding preferences remain consistent with coCLIP during arsenite stress

Next, we aimed to contrast coCLIP with standard CLIP derived from nuclear or cytoplasmic cell fractions. Cellular fractionation is widely used to provide insight into the function and composition of cellular organelles and is particularly useful when studying RBPs (Sanford et al. 2008; Brugiolo et al. 2017), such as HuR (Doller et al. 2008), that undergo nucleocytoplasmic trafficking. Following a commonly used protocol (Suzuki et al. 2010; Nabbi and Riabowol 2015), we conducted HuR CLIP from nuclear and cytoplasmic fractions with cells in both arsenite-induced stress and unstressed states (Supplemental Fig. S4A,B). As cellular stressors such as arsenite are known to alter nucleocytoplasmic transport (Zhang et al. 2018), we reasoned that this perturbation would allow us to observe alterations in HuR's binding preferences across RNA features and types. Importantly, we were unable to observe appreciable diffusion of labeled HuR from its nominal nuclear localization within the timescales of APEX labeling and UV cross-linking (Supplemental Fig. S4C,D). We generally observed twofold to 10-fold fewer peaks from coCLIP compared to fractionation CLIP (Fig. 2A). This discrepancy is reflective of a fundamental characteristic of coCLIP, which, involving a dual purification process, focuses on a specific subset of total interactions and is thus expected to yield a lower overall peak count compared to the broader scope captured by singular fractionation steps. Mirroring the coCLIP findings, the fractionation CLIP of unstressed cells revealed typical enrichment in mRNAs for intronic HuR binding events in the nucleus (75%) and primarily 3′-UTR binding events in the cytoplasm (60%) (Fig. 2A). Although coCLIP under stressed conditions maintained intronic preference in the nucleus and 3′-UTR preference in the cytoplasm, arsenite stress led to a balanced proportion of intronic and 3′-UTR binding in fractionation CLIP (Fig. 2A). For noncoding HuR targets, fractionation CLIP and coCLIP presented similar proportional distributions under normal conditions (Fig. 2B). In contrast, we observed a specific increase or TEs, snRNAs, and snoRNAs in cytoplasmic coCLIP under stress conditions. Overall, these findings indicate that arsenite can impact the results derived from fractionation methods, leading to changes in observed HuR binding patterns while emphasizing coCLIP's consistent performance under these conditions.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 2.

Consistency of HuR binding preferences with coCLIP under arsenite stress. (A) HuR peak proportions under mock and arsenite stress conditions for nuclear or cytoplasmic CLIP derived from classical fractionation or in situ labeling with coCLIP. (B) HuR peak proportions for ncRNAs as in A. (C) Venn diagrams depicting the overlap of peaks between fraction CLIP and coCLIP during mock and arsenite stress conditions for each compartment. (D–G) Scatterplots comparing nuclear to cytoplasmic ratios of HuR mRNA binding in coCLIP and fractionation CLIP during mock (D) and arsenite (E) stress conditions. Side histograms depict the relative density of each feature and are scaled for comparison within each feature and across mock and arsenite stress. (F,G) Similar scatterplots as in D,E for HuR binding on ncRNAs during mock (F) or arsenite (G) stress conditions. See also Supplemental Figure S4.

We next sought to directly compare nuclear and cytoplasmic enrichment for target HuR peaks in fractionation CLIP and coCLIP. We filtered for peaks that were present across either all NLS or NES coCLIP data sets or across all nuclear or cytoplasmic fractionation CLIP data sets for direct comparison and observed a high degree of overlap between coCLIP peaks and their fractionation counterparts (Fig. 2C). After calculating fold changes, we identified clear, annotation-dependent enrichment of nuclear introns and cytoplasmic 3′-UTR targets under mock conditions (Fig. 2D); this pattern was preserved exclusively in coCLIP during arsenite stress (Fig. 2E). For noncoding HuR targets, we noticed similar distributions between nuclear and cytoplasmic locations under both conditions; however, coCLIP demonstrated a pronounced nuclear enrichment for lncRNAs, snRNAs, and TEs in comparison to fractionation CLIP (Fig. 2F,G). We observed snoRNA enrichment in nuclear fractionation CLIP, but not in nuclear coCLIP, regardless of stress. Exploring the lack of snoRNA enrichment in nuclear coCLIP, we observed comparable snoRNA read frequencies across nuclear NLS and cytoplasmic NES coCLIP compared to cytoplasmic fractionation (Supplemental Fig. S4E). This observation suggests that the NLS-APEX fusion may not capture the entire spectrum of nuclear HuR interactions such as those of nucleolar-localized snoRNAs. Overall, these results suggest that although fractionation can produce anticipated outcomes under normal conditions, introducing arsenite stress may generate fractionation artifacts. By performing labeling in situ, coCLIP appears to circumvent these distortions.

HuR exhibits unique binding preferences in stress granules via coCLIP

Many compartments, particularly membrane-less organelles such as SGs (Anderson and Kedersha 2008; Protter and Parker 2016) are often transient, condition-dependent, and challenging to purify (Khong et al. 2017). coCLIP could offer a means to probe specific RBP–RNA interactions within these structures, allowing investigations even in their unstressed states. Previous APEX-based studies of SG proteomes (Markmiller et al. 2018) and transcriptomes (Padrón et al. 2019) serve as enabling precedents, laying a solid foundation for the deployment of coCLIP to probe specific RBP–RNA interactions and to discern the particular HuR targets residing within SGs with greater precision.

We performed HuR coCLIP using APEX tagged to the SG core protein G3BP1 (Tourrière et al. 2023) to study HuR–RNA interactions specific to SGs. For clarity, we describe these interactions as occurring in proximity to G3BP1, particularly in mock conditions in which SGs are not microscopically visible. Our analysis of peak enrichment highlighted a notable 3′-UTR enrichment for NES and G3BP1 adjacent coCLIP (Fig. 1E). Interestingly, despite both NES and G3BP1 coCLIP showing similar cytoplasmic localizations under microscopic examination, distinct target sets emerged between them (Fig. 3A,B). This discrepancy underscores the ability of our coCLIP method to sensitively and specifically unravel the complexities of RNA–RBP interactions within specific cellular locales.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 3.

HuR exhibits unique binding preferences within stress granules. (A,B) Venn diagrams depict the overlap of peaks for all coCLIP data sets from either mock (A) or arsenite (B) stress conditions. (C,D) Scatterplots contrasting enriched coCLIP to input CLIP for mRNA-binding events as log₂ fold changes, comparing G3BP1 coCLIP versus nuclear NLS coCLIP under mock (C) and arsenite (D) stress conditions. Side histograms depict the relative density of each feature and are scaled for comparison within each feature and across mock and arsenite stress. (E,F) Scatterplots as in C,D comparing G3BP1 coCLIP versus cytoplasmic NES coCLIP under mock (E) and arsenite (F) stress conditions.

Next, we analyzed the enrichment of HuR peaks between NLS, NES, and G3BP1 coCLIP, comparing them to input HuR CLIP under mock and arsenite-induced stress conditions (Fig. 3C–F). We observed clear intronic enrichment for nuclear HuR and 3′-UTR enrichment for HuR near G3BP1 (Fig. 3C) that was enhanced upon arenite treatment (Fig. 3D). HuR in both the cytoplasm and near G3BP1 exhibited heightened 3′-UTR enrichments relative to input HuR CLIP (Fig. 3F). However, the HuR specifically proximal to G3BP1 displayed a more pronounced 3′-UTR peak enrichment, which was absent for cytoplasmic HuR coCLIP (Fig. 3E). Moreover, compared to its nuclear or cytoplasmic counterparts, the G3BP1 proximal HuR revealed elevated TE peak binding (Supplemental Fig. S5) without similar enrichments in other noncoding RNAs (e.g., snoRNA, snRNA, rRNAs, and lncRNAs). These results suggest that HuR exhibits localization-specific target enrichment and binding preferences.

HuR demonstrates localization and condition-specific binding preferences across genomic loci and motifs

To further characterize HuR binding preferences across distinct cellular compartments, we conducted a metagene analysis on six genomic loci: the transcription start and end sites, translation start and stop sites, and splicing donor/acceptor sites (Fig. 4A). For nuclear HuR coCLIP, we noticed HuR binding enrichment near 5′ and 3′ splice sites with reduced signal in the 3′ UTR, a trend consistent between both mock and stress conditions. Cytoplasmic HuR coCLIP showed more robust enrichment for peaks in 3′ UTR compared to the nucleus. Intriguingly, in mock conditions, cytoplasmic HuR also demonstrated binding proximal to splice sites, a feature that persisted upon stress, albeit at reduced levels. HuR labeled in SGs distinctly favored 3′ UTRs under both conditions, corroborating our observations in the peak enrichment analysis (Fig. 3A).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 4.

Genomic position and motif preferences for subcellular HuR. (A) Metagene profiles for HuR binding for input CLIP or coCLIP under mock or arsenite conditions centered on indicated mRNA features: transcription starts, start codons, 5′ and 3′ splice sites, 3′-UTR starts, and 3′-UTR ends. The number of peaks for each metagene profile is shown in the upper left corner. (B) Correlation matrix for HuR peak binding for all CLIP experiments by type. (Frac CLIP) Fractionation CLIP. (C) HuR binding motif ranks for all CLIP and coCLIP experiments ordered by PAR-CLIP rankings from Mukherjee et al. (2011). (D) Motif density within HuR peaks (top) and cumulative distribution for frequency of the UUUUU motif (bottom) in input CLIP or coCLIP in mock or stress conditions. (E) Motif density within HuR peaks (top) and cumulative distribution for frequency of the AAAAA motif (bottom) as in D. Indicated P-values calculated by Kolmogorov–Smirnov test.

Depending on localization, differential binding preferences for HuR were also observable at the motif level. We first calculated 5-mer motif densities around peaks (Supplemental Table S2). We used 34 target motifs previously characterized for HuR through PAR-CLIP (Mukherjee et al. 2011), and we determined the ranked order of motif enrichments for our data sets. Although we observed strong sample correlations between our samples as expected (R² > 0.8) (Fig. 4B), we discerned intriguing differences between samples. Notably, the distinctions between nuclear and cytoplasmic fractionation CLIPs diminished under the influence of arsenite treatment. Conversely, the differences between nuclear, cytoplasmic, and SG coCLIPs became more pronounced under stress.

The differences observed in the correlation matrix were more apparent when we investigated our samples at the individual motif level (Fig. 4C). Generally, we observed that the rank order of the enriched motifs from our experiments matched well with what has been reported in the field (Lebedeva et al. 2011; Mukherjee et al. 2011; Nostrand et al. 2016), especially for uridine-rich motifs. Unexpectedly, we detected a pronounced enrichment of adenine-rich motifs compared to published data. Whereas A-rich motifs such as “WAAAA,” “AAAUW,” “WAAAU,” “UAAAW,” and “WUAAA”—previously identified as HuR targets—showed elevated enrichment levels, the “AAAAA” motif showed strong enrichment exclusively in the CoCLIP samples.

Analysis of motif density within peaks revealed a strong representation of the “UUUUU” motif across coCLIP samples and conditions (Fig. 4D), with the most significant binding to the “UUUUU” motif observed in HuR near G3BP1 under mock conditions. In addition, the cumulative distribution of “UUUUU” appearances within peaks showed that roughly 50% of peaks containing “UUUUU” have four or fewer motifs (including overlaps) under all samples and conditions and exhibited no statistically significant differences.

In contrast, adenine-rich motifs were enriched within HuR peaks in the cytoplasm and SGs under stress conditions (Fig. 4E). The cumulative distribution of “AAAAA” appearances within cytoplasmic and near G3BP1 peaks displayed statistically significant shifts between mock and stress conditions (Kolmogorov–Smirnov test, P < 0.01). Although 50% of peaks in cytoplasmic HuR contained at most one “AAAAA” motif, the number of “AAAAA” motifs significantly increased upon stress. HuR near G3BP1 showed a similar shift in which 50% of peaks in the mock condition contained two or fewer “AAAAA,” whereas, upon stress, 50% contained at least eight “AAAAA” motifs. These findings suggest that although HuR binds to its canonical AU-rich elements such as “UUUUU” and “AAAAA,” this binding may not be equivalent. Overall, our data suggest that HuR binding to adenine-rich motifs is location- and condition-specific. Specifically, the sequence motif context is distinct from uridine-rich motifs, as HuR-bound adenine-rich motifs are considerably longer than uridine-rich motifs upon stress.

Gene level analysis of subcellular HuR targeting reveals distinct gene set enrichments

Expanding our analysis beyond peak-level enrichment, we honed in on the interaction between HuR and RNA at the gene level. Our primary objective was to eliminate the possibility that the distinct HuR binding patterns we observed in specific locations were artifacts of transcriptomic alterations among different APEX-cell lines or between mock and arsenite treatment conditions. To this end, we conducted RNA-seq and differential gene expression analysis on the cell lines used for CoCLIP experiments, finding no statistically significant disparities between cell lines or our two conditions (Supplemental Fig. S6). Further, comparisons between stressed G3BP1–APEX CoCLIP and mock coCLIPs from any compartment revealed significant changes to HuR target selection, but no RNA-seq level changes (Fig. 5A). These data suggest that the observed changes in HuR binding across conditions are occurring because of the relocalization of HuR and/or target transcriptomes, and not because of changes in transcript levels. Moreover, HuR target genes exclusively enriched in the proximity of G3BP1 showed longer overall transcript lengths than targets enriched in the nucleus or cytoplasm (Fig. 5B), consistent with previous observations (Khong et al. 2017).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 5.

Gene level features of subcellular HuR targeting from coCLIP. (A) Stress versus mock comparisons of G3BP1 coCLIP data with respective mock conditions compared to matched gene level RNA-seq data. (B) Transcript length boxplots for input or coCLIP-derived HuR targets. Outliers are omitted for clarity. (C) Gene ontology analysis of HuR binding sites from input CLIP or compartment-specific coCLIP under mock or arsenite stress conditions. Individual gene ontology terms are shown in the right. Dot size represents statistical significance after FDR correction. Manually curated grouped GO terms are shown on the left. See also Supplemental Figure S6.

We next performed gene ontology enrichment analysis on the HuR target genes identified for different cellular compartments (Fig. 5C) to uncover potential associations attributed to HuR targets within different cellular environments. All target gene ontologies identified in input mock HuR CLIP were identified in at least one coCLIP data set, suggesting a level of conserved targeting. However, unique clusters of target genes became apparent under stress and cellular localization. For instance, HuR interaction with genes associated with mRNA nuclear export was exclusively observed in cytoplasmic HuR under mock conditions and was subsequently lost with arsenite stress induction. Genes associated with SG assembly were prominently enriched in G3BP1 CoCLIP under stress (Fig. 5C). These results shed light on the dynamics of HuR targeting and hint at a structured network where distinct subcellular RNA targets are functionally interconnected.

De novo HuR binding events and HuR flow across compartments

Our gene set enrichment analysis prompted us to investigate how the HuR translocation from one compartment to another affects its target gene repertoire. As our analysis is far from exhaustive by focusing on three compartments, we also wanted to account for HuR targeting events that appeared to be “de novo” upon stress. As the observed changes in HuR binding upon arsenite stress induction were not due to alterations in transcript abundance (Fig. 5A; Supplemental Fig. S6), these targeting events could be due to HuR:RNA translocation from other compartments as well as new binding events from HuR not bound to RNA. By dividing the identified HuR target genes into subgroups exclusive to each compartment and condition, we tracked target genes shared between mock and stress conditions in the same compartment or genes that were in one compartment in the mock condition but appeared in a different compartment upon stress. To our surprise, almost all (748/763 or 98%) HuR target genes uniquely enriched in nuclear coCLIP under stress conditions were de novo targets not associated with HuR in any tested compartment in the mock condition (Fig. 6A).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

FIGURE 6.

De novo HuR binding events and HuR flow across compartments. (A) Graphical illustration of common and unique HuR targeting events from nuclear, cytoplasmic, and SG coCLIP under mock and arsenite stress conditions. Arrows indicate the flow of binding events starting from the mock condition. De novo events are defined as those that appear only in stressed conditions. (B) De novo HuR binding event counts (left) and proportions (right) for nuclear, cytoplasmic, and SG coCLIP data sets. (C–G) Gene browser tracks of normalized HuR binding for input CLIP and coCLIP data on the (C) CCND1 3′ UTR, (D) FOS 3′ UTR, (E) SLK 3′ UTR, (F) VAPA 3′ UTR, and (G) PCBP1 3′ UTR. Highlighted regions indicate significant peaks of interest. Length and normalized peak height scales are indicated at the top and at the bottom right, respectively. (H) DDX17 intronic and 3′-UTR HuR binding. Sequence-level PhyloP scores and multisequence alignments for peaks of interest are shown. (I) G3BP1 3′-UTR HuR binding is shown with sequence-level PhyloP scores and multisequence alignments for indicated peaks.

Similarly, ∼80% of unique HuR targets proximal to G3BP1 (481/619) or NES-APEX (331/418) during stress were de novo targets. Furthermore, nuclear HuR targets uniquely enriched in mock conditions were also targets in the cytoplasm and near G3BP1 under stress conditions (80 and 130 genes, respectively). In contrast, only a small number of cytoplasmic and G3BP1 proximal HuR targets under mock conditions were observed in different compartments under stress. Within each compartment, 20% of nuclear targets were consistent between mock and stress conditions, whereas <5% of the targets in the cytoplasm and near G3BP1 were shared across conditions.

The de novo target genes for each compartment under stress conditions showed peak distributions (Fig. 6B) that were characteristic of each compartment (Figs. 2A and 3A,B). Notably, the enhanced TE peak binding observed in HuR proximal to G3BP1 under stress was absent among the de novo target genes. These results imply that HuR association with TEs is distributed evenly across the cellular compartments we have tested under mock conditions.

To further illustrate these variations in HuR binding, we examined individual gene tracks across different compartments and stress conditions. For example, HuR showed minimal binding on mRNA for the cell cycle regulator cyclin D1 under mock conditions (Fig. 6C). However, we observed strong HuR binding in the 3′ UTR of CCND1 mRNA in the input HuR CLIP upon stress induction. Through coCLIP, we discerned that this HuR binding predominantly occurred in the cytoplasm and within SGs, with no detectable binding in the nucleus. Likewise, the proto-oncogene Fos demonstrated no interaction with HuR across all compartments under mock conditions; however, under stress conditions, HuR binding to the 3′ UTR of the FOS transcript was evident, specifically in the nucleus and SGs (Fig. 6D). We also observed similar HuR binding patterns on the PCBP1 transcript, which encodes a multifunctional RBP implicated in alternative splicing, immune regulation, and cancer (Fig. 6G), where increased association with HuR in the 3′ UTR under stress conditions was primarily due to increased HuR binding in SG.

Furthermore, coCLIP data were sensitive enough to discern instances in which HuR binds to identical locations on a transcript during both mock and stress conditions but in different cellular compartments. For instance, the mRNA for the protein kinase SLK exhibited a robust association with HuR in the 3′ UTR of its gene in the cytoplasm under mock conditions (Fig. 6E). However, with arsenite-induced stress, although the HuR binding site remained constant, the interaction location transitioned from the cytoplasm to SG. Conversely, the VAPA transcript, encoding the endoplasmic reticulum membrane component VAP-A, demonstrated pronounced 3′-UTR binding with HuR in SG under mock conditions (Fig. 6F). Following arsenite treatment, this same binding interaction predominantly occurs in the cytoplasm. Our data also show intricate alterations in HuR binding that occur across different loci on a transcript, varying between subcellular compartments and conditions. For example, the DDX17 mRNA displays robust intronic HuR binding in the nucleus under mock conditions. However, upon the induction of stress, this intronic binding is reduced, but there is a corresponding peak increase in the 3′ UTR of the gene in the cytoplasm and near G3BP1 (Fig. 6H).

Last, coCLIP data displayed a poly(A) motif enrichment that was specific to arsenite-induced stress, most notably in the cytoplasmic and G3BP1 coCLIP. For example, although the aforementioned intronic peak in DDX17 mRNA contained a U-rich sequence, a 3′-UTR peak that was unique to HuR near G3BP1 upon stress contained a highly conserved A-rich sequence (Fig. 6H). We observed a similar change in motif preference in the 3′ UTR of the G3BP1 transcript, with a similar well-conserved A-rich stretch in the 3′-UTR peak in G3BP1 coCLIP (Fig. 6I). Collectively, our findings demonstrate that coCLIP is a robust and sensitive innovative method for studying RNA–RBP interactions with subcellular resolution.