Dynamic recruitment of single RNAs to processing bodies depends on RNA functionality

SUMMARY

Cellular RNAs often colocalize with cytoplasmic, membrane-less ribonucleoprotein (RNP) granules enriched for RNA processing enzymes, termed processing bodies (PBs). Here, we track the dynamic localization of individual miRNAs, mRNAs and long non-coding RNAs (lncRNAs) to PBs using intracellular single-molecule fluorescence microscopy. We find that unused miRNAs stably bind to PBs, whereas functional miRNAs, repressed mRNAs and lncRNAs both transiently and stably localize within either the core or periphery of PBs, albeit to different extents. Consequently, translation potential and 3` versus 5` placement of miRNA target sites significantly impact PB-localization dynamics of mRNAs. Using computational modeling and supporting experimental approaches we show that partitioning into the PB phase attenuates mRNA silencing, suggesting that physiological mRNA turnover occurs predominantly outside PBs. Our data support a PB role instead in sequestering unused miRNAs for surveillance and provides a framework for investigating the dynamic assembly of RNP granules by phase separation at single-molecule resolution.

eTOC BLURB

Single cellular RNAs stably or transiently associate with the core or periphery of processing bodies (PBs). The kinetics, localization patterns and enrichment of RNAs at PBs depend on RNA type and function. While PBs may not be designated sites of mRNA degradation, they contribute towards miRNA surveillance.

Graphical Abstract

INTRODUCTION

Sub-cellular, membrane-free granules have emerged as critical components of normal biology and pathophysiology (Banani et al., 2017; Shin and Brangwynne, 2017), owing to their key role in spatial regulation of gene expression (Martin and Ephrussi, 2009; Spector, 2006). Processing bodies (PBs) are one such class of ribonucleoprotein (RNP) granules that persist during cellular homeostasis and are enriched for RNA processing and degradation enzymes (Eulalio et al., 2007a; Parker and Sheth, 2007). These granules are observed in almost all eukaryotes, ranging from yeast to mammals, and have been implicated in multiple biological processes, including oogenesis, progression through early development, and mediation of neuroplasticity (Buchan, 2014).

More specifically, mammalian PBs have been functionally associated with storage, translational repression and/or degradation of mRNAs (Buchan, 2014; Hubstenberger et al., 2017; Liu et al., 2005; Schutz et al., 2017), as a result of which PBs are predominantly composed of translationally repressed messenger RNAs (mRNAs), mRNA-regulating miRNAs and, to a lesser extent, regulatory long non-coding RNAs (lncRNAs). Such a large RNP complex is hypothesized to assemble via RNA dependent phase separation (Banani et al., 2017), wherein multiple translationally repressed RNPs are concentrated within dense foci through strong multivalent interactions and individual or oligomeric RNPs loosely interact with these dense regions to create dynamic shells (Cougot et al., 2012; Van Treeck and Parker, 2018). Consequently, PBs, as whole granules, display a wide array of dynamic behaviors (Aizer et al., 2008), but the intra- and peri-granular RNP dynamics and RNP recruitment – processes that govern the maintenance, maturation and putative gene regulatory functions of PBs – are largely unknown. While mRNP-PB colocalization and mRNA regulation have been shown to be tightly correlated (Buchan, 2014; Parker and Sheth, 2007), the question of whether mRNPs are degraded at microscopically visible, and thus relatively large (> 250 nm), PBs also remains unresolved.

Here, we dissect the fundamental principles governing the dynamic localization of functionally distinct classes of RNPs at phase separated PBs and unravel the functional consequence of RNA-PB colocalization. To this end, we developed methodologies to simultaneously observe single RNA molecules (miRNAs, mRNAs or lncRNAs) and individual PB foci inside both living and fixed human cells. We demonstrate that a majority of miRNAs and repressed mRNAs are stably anchored within PBs, whereas translationally active mRNAs and lncRNAs associate with PBs only transiently, suggesting a strong correlation between PB-localization and RNA class. While, miRNAs and mRNAs localized at core or shells of PBs, lncRNAs were predominantly found at PB-shells. Furthermore, we find that unused (target-less) miRNAs are enriched at PB and that the 3` versus 5`terminal positioning of cis-regulatory miRNA response elements (MREs) dictates the PB localization patterns and dynamics of mRNAs. Finally, in silico modeling and experimental validation through hyperosmotic-stress induced phase separation suggest that the stochastic collision of mRNAs with freely diffusing, sub-microscopic PBs leads to more efficient mRNA regulation than their recruitment to microscopic PBs. Taken together, our observations reveal the nanoscale principles that govern the compositional complexity of mesoscale RNP granules, and a novel suggested function for PBs in accumulating target-less miRNAs for miRNA surveillance.

RESULTS

Super-resolved single-molecule fluorescence microscopy probes RNA-PB interactions

To dissect the localization dynamics of RNAs at and near PBs, we created a U2-OS cell line that stably expressed GFP tagged Dcp1a, an mRNA decapping co-activator and PB marker (Aizer et al., 2008; Hubstenberger et al., 2017). We selected a clone (hereon termed UGD) with similar number and composition (based on colocalization with other PB markers) of Dcp1a foci as endogenously found in U2-OS cells (Figure S1A-D). Next, mature regulatory miRNAs, whose size (~22 nt per strand) precludes endogenous labeling strategies (Pitchiaya et al., 2014), were chemically synthesized with a fluorescent Cy5 dye at the 3`end of one of their two complementary strands, typically the guide strand. Since transfection results in the sequestration of RNA within subcellular vesicles (Cardarelli et al., 2016), we chose to deliver these miRNAs via microinjection (Figure 1A-C), which enables controlled delivery (Figure S1E-G) of physiologically relevant miRNA molecules per cell (~10–20,000 copies, i.e. 1/10^th the total number of miRNAs per cell) and defines a clear starting point for our assays by instantaneously exposing RNAs to the cellular milieu (Pitchiaya et al., 2012; Pitchiaya et al., 2013; Pitchiaya et al., 2017). We confirmed that fluorophore labeling and microinjection did not affect the gene-repressive function (Figure S1H-K) of let-7 miRNA (l7/l7* and l7-Cy5/l7*) (Pitchiaya et al., 2012), alter the sub-cellular abundance and behavior of PBs (Figure S1J-K), or induce stress granule (SG) formation (Figure S1L-O).

Figure 1. A super-resolution imaging tool for probing RNA-granule dynamics and stoichiometry.

(A) Schematic of iSHiRLoC assay for probing miRNA-PB dynamics and colocalizations. (B and C) Representative pseudo-colored and contrast-adjusted images from live-cell imaging (B) and fixed cell imaging (C) assays of UGD cells expressing GFP-labeled PBs (green) that were microinjected with l7-Cy5/l7* miRNA (red) and imaged 2 h post injection. Scale bar, 10 μm. (D) Representative single-particle trajectories of PBs (green) and l7-Cy5/l7* miRNA (red) from yellow and magenta boxes in B, representing diffusing miRNAs in PBs and in the cytoplasm (Cyt) respectively. Scale bar, 1 μm. Dotted green circle represents PB outline in the first frame of the movie. Distribution of l7-Cy5/l7* miRNA diffusion constants in PB and Cyt are also depicted. Green area on the plot depicts the range of PB diffusion constants (n = 3, 15 cells). (E) Zoomed-in view of orange and violet boxes in C, from fixed UGD cells. Scale bar, 2 μm. Step-wise photobleaching trajectories PB- and Cyt-localized l7-Cy5/l7* is also shown. (F) Distribution of l7-Cy5/l7* miRNA stoichiometry as monomeric (Mono, 1 photobleaching step) or multimeric (Multi, ≥ 2 photobleaching steps) complexes in PB and Cyt within fixed UGD cells (n = 3, 15 cells). See also Figure S1.

We then combined a super-registration fluorescence microscopy-based tool (Grunwald and Singer, 2010) that measures intermolecular distances of spectrally distinct fluorescent molecules with intracellular single molecule, high-resolution localization and counting (iSHiRLoC)(Pitchiaya et al., 2012; Pitchiaya et al., 2017; Pitchiaya et al., 2013). Consequently, we were able to visualize miRNA-PB interactions in living cells and precisely quantify miRNA stoichiometry within PBs in fixed cells (Methods, Figure 1A-C and Supplementary movie 1). At a spatial accuracy of 30 nm and temporal resolution of 50 ms, we can visualize large (> 400 kDa) miRNPs, such as miRISC:mRNP complexes, in living cells and all miRNPs, irrespective of RNP size, in fixed cells (Figure S1P) (Pitchiaya et al., 2012; Pitchiaya et al., 2017; Pitchiaya et al., 2013). Using this new tool, we found that the tumor suppressive let-7 miRNA (l7-Cy5/l7*) diffused ~100–1,000-fold slower at PBs compared to in the cytosol (Figure 1G), supporting the notion that miRNAs physically dock to form higher order complexes at PBs and consistent with previous ensemble observations of miRNA accumulation at PBs (Liu et al., 2005; Pillai et al., 2005). However, we additionally observed that PB-localized miRNAs distributed between (at least) two populations of diffusion coefficients or molecular weights. Complementarily, fixed cell analysis showed that cytoplasmic l7-Cy5/l7* miRNA were predominantly monomeric, wherein a significant minority of monomeric (~40%) and a predominant fraction of multimeric (~60%) RNA complexes (Figure 1H) were observed at PBs. Moreover, the PB dynamics and localization extents of l7-Cy5/l7* in GFP-Dcp1a expressing HeLa cells were almost identical to those in UGD cells (Figure S1Q-R), underscoring the generality of our observations across cellular systems. Our data suggest that miRNPs of diverse sizes, and perhaps composition, localize to PBs via potentially distinct mechanisms, with the possibility to yield distinct regulatory outcomes.

miRNAs stably or transiently localize at the core or periphery of PBs

We next sought to understand whether the observed diverse miRNP diffusion and assembly states at PBs are based on the type of miRNA-PB interaction. To this end, we first inspected individual trajectories of PB-localized l7-Cy5/l7* in live cells to discover diversities in the kinetics and modalities of miRNA-PB interactions. We identified five distinct types RNA-PB interactions, each of which could be classified by a unique combination of diffusion coefficient (D), photobleaching corrected dwell time (T) and percentage of an RNA track colocalizing with a PB (P) (Figure 2A, S2A and Supplementary movie 2): 1) RNAs stably anchoring at PBs (D = 0.0001 – 0.1 μm²/s, T ≥ 15 s, P = 100%, Supplementary movie 2); 2) RNAs displaying significant dynamics within PBs (D = 0.001 – 0.1 μm²/s, T ≥15 s, P = 100%, Supplementary movie 2); 3) RNAs entering PBs from the cytosol (D = 0.0001 – 0.01 μm²/s, T = 7.9 ± 0.7 s, P = 52 – 89%, Supplementary movie 2); 4) RNAs transiently probing PBs (D = 0.0001 – 1 μm²/s, T = 0.9 ± 0.1 s, P = 3 – 72%, Supplementary movie 2); and 5) RNAs exiting a PB into the cytosol (D = 0.0001 – 1 μm²/s, T = 0.8 ± 0.1 s, P = 7 – 83%, Supplementary movie 2). The first three and latter two interaction types depict what we refer to as stable and transient RNA-PB localizations, respectively. These data suggest that the diffusion rate and dwell times of miRNPs defines the type of interaction with PBs. Next, we quantified the relative localization of PB-resident proteins or a few control proteins with respect to GFP-Dcp1a (Figure S2B). Using this intra-granular localization atlas as a template, we spatially mapped the localization of miRNPs with reference to PB boundaries and found that miRNAs localized near the core or the periphery/shell of PBs in fixed cells (Figure 2B). We then performed ratiometric quantification of core- or shell-localized immunofluorescence (IF) signal at PBs and the adjacent cytosol (Figure S2C), which yields similar information as the average percentage of IF signal within PBs per cell but also accounts for any heterogeneities between PBs within the same cell, and created a small compendia of proteins that were either enriched ( > 1) or depleted ( < 1) from PBs (Figure S2C). Combining this new quantification tool with single-molecule counting, we discovered that miRNAs were either clustered (enriched within PBs compared to the adjacent cytosol) or dispersed at PBs (Figure 2C). As a control, we also probed dl7-Cy5/dl7*, a control DNA oligonucleotide of the same sequence as let-7 miRNA, but incompetent for RNA silencing. In contrast to l7-Cy5/l7*, and as expected, we found that dl7-Cy5/dl7* neither localized to nor was enriched at/near PBs (Figure 2D). Taken together, these findings unravel a potentially tight relationship between miRNP composition and type of miRNP-PB interaction, and the requirement for small double-stranded (ds) oligonucleotides to assemble into large RNPs to stably interact with PBs.

Figure 2. miRNAs show diverse spatiotemporal localization patterns at PB core and periphery.

(A) Schematic and representative time-lapsed images of PBs (green) and l7-Cy5/l7* miRNAs (red) in live UGD cells. Scale bar, 1 μm. Embedded numbers in green/red overlay images (far-left and far right) represent time in seconds. Dotted green circles in red panels have been included to aid in the identification of PB boundaries. White arrow points to an individual RNA particle. Stable RNA-PB association patterns (static, dynamic and recruited) are represented in orange whereas transient ones (probe and escape) are represented in blue. n_PB = number of track localizations within PBs, n_Cyt = number of track localizations in the cytosol. (B) Schematic and representative images of PBs (green) and l7-Cy5/l7* (red) representing the localization of miRNAs within shells or cores of PBs in fixed UGD cells. Scale bar, 2 μm. Dotted green and red circles represent boundaries of PBs and miRNAs respectively. Relative localization (RL) values of l7-Cy5/l7* for these representative colocalizations are embedded in the green panels. (C) Schematic and representative images of PBs (green) and l7-Cy5/l7* (red) representing the enrichment of miRNAs in PBs within fixed UGD cells. Dotted yellow and red circles represent PB-miRNA colocalization and cytoplasmic miRNAs respectively. Enrichment of l7-Cy5/l7* per PB (EI) for these representative colocalizations are embedded in the green panels. Images are scaled as in B. (D) Scatter plot representing the % of RNA or DNA molecules that colocalize with PBs per fixed UGD cell (top). Each dot represents a cell. Scatter plot of enrichment of molecules per PB (below) is also shown. Each dot represents an individual PB in fixed UGD cells. n = 3, > 15 cells, ***p ≤ 0.0001 by two-tailed, unpaired Student’s t-test. Grey dotted line depicts an EI of one, which demarcates PB-enriched (> 1) from PB-depleted (< 1) factors. See also Figure S2.

mRNA-targeting and target-free miRNAs are both enriched at PBs but display distinct PB localization dynamics

Based on our observations that a functionally repressive l7-Cy5/l7* miRNA dynamically localized to PBs via diverse modes (Figure 2), we hypothesized that the regulatory potential of miRNAs impacts their PB localization. To test this hypothesis, we compared the PB-localization of functional l7-Cy5/l7* with l7/l7*-Cy5, let-7 miRNA Cy5-labeled on the passenger instead of the guide strand, where the passenger strand has very few endogenous targets and is at least 8-fold less stable than the guide strand, and with ml7-Cy5/ml7*, a seed-sequence mutated let-7 miRNA variant that cannot bind endogenous let-7 targets and is at least 4-fold less stable than let-7 miRNA (Figure 3A) (Pitchiaya et al., 2017). Strikingly, the fractional extents of PB localization and enrichment were significant and similar for l7-Cy5/l7*, l7/l7*-Cy5 and ml7-Cy5/l7* (Figure 3B-C). Similar trends (Figure S3A-C) were observed for all other small dsRNAs, namely an oncogenic miRNA miR-21 (m21-Cy5/m21*), an artificial miRNA cxcr4 (cx-Cy5/cx*) and scrambled control dsRNA (Scr-Cy5/Scr*). Considering that each of these dsRNAs have distinct regulatory potential and intracellular stability (Pitchiaya et al., 2017), our data strongly suggest that miRNA functionality is not necessary for PB localization. However, ml7-Cy5/ml7*, l7/l7*-Cy5, cx-Cy5/cx* and Scr-Cy5/Scr* rarely displayed any transient interactions (Figure 3D and Figure S3D), but instead exhibited monophasic dwell time distributions, residing in PBs for ≥ 15 s (Figure 3D and S3), significantly different from the PB-dynamics of l7-Cy5/l7* and m21-Cy5/m21*. These observations suggest that transient PB interactions of a miRNA are correlated with its ability to target mRNAs, whereas unused (target-less) miRNAs are more stably recruited to PBs. Further corroborating this notion, we found that, upon co-microinjecting its cognate (RL-ml7–2x) mRNA, the mRNA-targeting ml7-Cy5/ml7* exhibited a substantial 5-fold increase in the fraction of transient interactions, resulting in a biphasic dwell time distribution with T_fast = 0.7 s and T_slow = 13.2 s (Figure 3E-F and S3C-D). Taken together, our results are consistent with PBs stably capturing target-less, non-coding miRNAs for surveillance, and suggest that instead transient PB interactions are dominant for functional miRNAs engaging mRNA targets.

Figure 3. miRNA functionality influences miRNA-PB interaction kinetics.

(A) Schematic of miRNAs used. P, lines and dots represent 5` phosphate, Watson-crick base pairing and wobble pairing respectively. (B) Scatter plot representing the % of RNA or DNA molecules that colocalize with PBs per fixed UGD cell. Each dot represents a cell. (C) Scatter plot of EI for different constructs. Each dot represents an individual PB in fixed UGD cells. Grey dotted line depicts an EI of one, which demarcates PB-enriched (> 1) from PB-depleted (< 1) factors. (D) Relative distribution of stable and transient interactions per live UGD cell for different miRNAs. (E) Comparison of fast and slow miRNA-PB interaction kinetics in live UGD cells. (F) Relative distribution of stable and transient interactions per live UGD cell for ml7-Cy5/ml7* RNAs co-injected with a seed mismatched (RL-l7–2x) or seed matched (RLml7–2x) mRNA target. (G) Comparison of fast and slow ml7-Cy5/ml7*-PB interaction kinetics in the presence of a seed mismatched (RL-l7–2x) or seed matched (RL-ml7–2x) mRNA target in live UGD cells. n = 3, 15 cells per sample, NS = not significant, **p ≤ 0.001 or ***p ≤ 0.0001 by two-tailed, unpaired Student’s t-test. See also Figure S3.

miRNA-targeted mRNAs localize to PBs depending on 3` versus 5` terminal positioning of MREs

Next, we probed whether miRNAs and their cognate mRNA targets displayed similar dynamics and localization patterns at PBs. mRNAs were endogenously expressed and tagged via a modified version of the widely used MS2-MCP labeling system (Fusco et al., 2003), wherein a total of up to ~1,000 Halo-MCP bound MS2-RNA molecules were visualized per living cell upon covalently coupling the Halo tag with the cell-permeable fluorescent dye JF646 (Figure 4A-B and Supplementary movie 3) (Grimm et al., 2015). mRNAs in fixed cells were instead visualized by standard single-molecule fluorescence in situ hybridization (smFISH, Figure 4A and 4C) (Raj et al., 2008). We created an MS2-MCP tagged construct bearing the firefly luciferase (FL) coding sequence (CDS) and an artificial 3`untranslated region (3`UTR) bearing six tandem miRNA response elements (MREs) for the tumor suppressive let-7 miRNA (l7–6x). Upon performing live and fixed cell imaging respectively, we found that that the mobility and assembly of FL-l7–6x-MS2 mRNA was similar to its cognate l7-Cy5/l7* miRNA (Figure 1 and 4A-F), strongly supporting the notion that a miRISC-mRNP complex interacts with PBs. As a control, we created an MS2-tagged FL gene with ml7–6x, a 3`UTR composed of six tandem mutant MREs ml7/ml7* that are not targeted by endogenous let-7 (Figure 4G). Considering that MRE-containing mRNAs are repressed, irrespective of whether the MREs are in the 3` or 5` UTR of the mRNA (Lytle et al., 2007), we created additional control constructs with either l7–6x or ml7–6x in the 5`UTR of the MS2-tagged FL gene, termed l7–6x-FL-MS2 and ml7–6x-FL-MS2 respectively (Figure 4G). As expected, ensemble activity assays showed that all MS2-tagged constructs were translated and regulated much like their untagged counterpart (Figure S4A-B). FL-ml7–6x-MS2 and ml7–6x-FL-MS2 were expressed to much higher extents (Figure 4H) than the let-7-MRE containing FL-l7–6x-MS2 and l7–6x-FL-MS2, which both were similarly repressed by let-7 miRNA (Figure 4H and S4C), thus corroborating prior reports that MREs embedded in either the 3` or 5`UTR are functional. However, the fractional extents of localization and enrichment of l7–6x-FL-MS2 at PBs were similar to those of the non-targeted FL-ml7–6x-MS2 and ml7–6x-FL-MS2, and significantly (at least 5-fold) lower than those of FL-l7–6x-MS2 (Figure 4I-J). Still, l7–6x-FL-MS2, FL-ml7–6x-MS2 and ml7–6x-FL-MS2, much like FL-l7–6x-MS2, interacted transiently with PBs and displayed biphasic interaction kinetics (Figure 4K-L and S4D). While the “fast” phase for l7–6x-FL-MS2, FL-ml7–6x-MS2 and ml7–6x-FL-MS2 (spanning ~0.5, 0.7 and 0.6 s, respectively) was similar to that of FL-l7–6x-MS2 (0.9 s), the “slow” phase for these constructs was ~3-fold faster than that of FL-l7–6x-MS2 (4.2 s, 3.7 and 2.5 s, respectively, compared to 15 s, Figure 4K), indicating a significant difference in behavior upon targeting the 3` versus 5`UTR. Similarly, a minority of l7–6x-FL-MS2, FL-ml7–6x-MS2 and ml7–6x-FL-MS2 particles did not photobleach and resided in PBs for the entire duration of acquisition (~ 15 s), with the number of such occurrences ~3-fold lower than for FL-l7–6x-MS2 (Figure S4E). Not only do these observations strongly support the notion that miRNAs and their cognate mRNA targets display generally similar PB localization kinetics and patterns, consistent with the hypothesis that they interact, but they uniquely demonstrate that 3`UTR versus 5`UTR positioning of MREs distinctly impacts PB colocalization in that only 3`UTR targeting leads to the most stable PB interactions. We posit that distinct aspects of translation are blocked when miRNAs engage the 3`UTR versus 5`UTR, resulting in compositionally distinct mRNPs that differentially recruit them to PBs.

Figure 4. mRNAs localize to PBs depending on 3` versus 5` terminal positioning of MREs and translation potential.

(A) Schematic of assay for probing mRNA-PB dynamics and colocalizations. (B and C) Representative pseudo-colored and contrast-adjusted images from live-cell imaging (B) and fixed cell imaging (C) assays of UGD cells expressing GFP-labeled PBs (green) and MCP tagged FL-l7–6x-MS2 mRNAs (red). Scale bar, 10 μm. (D) Representative single-particle trajectories of PBs (green) and FL-l7–6x-MS2 mRNAs (red) from yellow and magenta boxes in B, representing diffusing mRNAs in PBs and in the cytoplasm (Cyt) respectively. Scale bar, 1 μm. Dotted green circle represents PB outline in the first frame of the movie. Distribution of FL-l7–6x-MS2 mRNAs diffusion constants in PB and Cyt are also depicted. Green area on the plot depicts the range of PB diffusion constants (n = 3, 20 cells). (E) Zoomed-in view of orange and violet boxes in C, from fixed UGD cells. Scale bar, 2 μm. Intensity measurements of PB- and Cyt-localized FL-l7–6x-MS2 mRNAs is also shown. (F) Distribution of FL-l7–6x-MS2 mRNAs stoichiometry as monomeric (Mono, 1 photobleaching step) or multimeric (Multi, ≥ 2 photobleaching steps) complexes in PB and Cyt within fixed UGD cells (n = 3, 20 cells). (G and M) Schematic of different mRNA constructs with various 3` or 5` UTRs. Color-coded symbols for each transcript is shown and will be used to depict these respective transcripts from hereon. (H and N) Luciferase reporter assays represented as the ratio of luminescence form a firefly luciferase (FL) reporter gene and a renilla luciferase (RL) normalization gene in UGD cells. Data were normalized to the FL sample. Mean and s.e.m are represented (n = 12 replicates, ***p < 0.0001 based on two-tailed, unpaired Student’s t-test). (I and O) Scatter plot representing the % of mRNA molecules that colocalize with PBs per fixed UGD cell. Each dot represents a fixed UGD cell. (J and P) Scatter plot of EI for different mRNA constructs. Each dot represents a PB in fixed UGD cells. Grey dotted line depicts an EI of one, which demarcates PB-enriched (> 1) from PB-depleted (< 1) factors. (K and Q) Relative distribution of stable and transient interactions per live UGD cell for different mRNAs. (L and R) Comparison of fast and slow mRNA-PB interaction kinetics in live UGD cells. Black line depicts acquisition window (15 s). Green-black line depicts the mean magnitude of FL-l7–6x-MS2 for the respective observable. n = 3, ≥ 15 cells per sample, NS = not significant, * p ≤ 0.01, **p ≤ 0.001 or ***p ≤ 0.0001 by two-tailed, unpaired Student’s t-test. See also Figure S4.

mRNA-PB interactions depend on translation potential

Given that translationally unrepressed mutant FL-ml7–6x-MS2 and translationally repressed FL-l7–6x-MS2 mRNAs displayed distinct PB-dynamics and localization patterns (Figure 4), we hypothesized that the translation potential of an mRNA inversely correlates with PB localization. To test this hypothesis, we compared the PB localization dynamics of the let-7 regulated FL-l7–6x-MS2 mRNA (Figure 4) with those of FL-MS2 (lacking the regulatory 3`UTR), FL-l7–2x-cx-4x-MS2 (carrying a 3`UTR with two tandem MREs targetable by endogenous let-7 and four MREs for a non-endogenous cxcr4 miRNA) and FL-CX-6x-MS2 (carrying a 3`UTR with six tandem MREs for cxcr4 miRNA) (Figure 4M). Notably, protein expression of FL-MS2 and FL-cx-6x-MS2 was significantly higher (~2.7 fold) than FL-l7–2x-cx-4x-MS2, which in turn was higher (~2.2-fold) than FL-l7–6x-MS2 (Figure 4N). Consistent with our hypothesis, the fractional extents of localization and enrichment of FL-MS2 and FL-cx-6x-MS2 were significantly (at least 2.8 fold or 5-fold) lower than those of FL-l7–2x-cx-4x-MS2 or FL-l7–6x-MS2 (Figure 4G-H). Additionally, the interaction modalities and “slow” phase kinetics of FL-MS2 and FL-cx-6x-MS2 were distinct from FL-l7–2x-cx-4x-MS2 and FL-l7–6x-MS2, with the former set of constructs displaying at least ~2.5-fold more transient interactions and ~3-fold shorter dwell times at PBs (Figure 4I-J) compared to the latter set. A significant minority of FL-MS2 and FL-cx-6x-MS2 particles resided in PBs for the entire duration of acquisition (~15 s), potentially representing mRNAs that are currently translation inactive, but the number of such occurrences was ~2.9-fold and ~4.5-fold lower than those for FL-l7–2x-CX-4x-MS2 and FL-l7–6x-MS2 (Figure S4F), respectively. These observations strongly support the notion that actively translating mRNAs rarely localize to PBs, and conversely that the propensity for PB-localization increases with extent of mRNA repression.

miRNA-targeted mRNA turnover predominantly occurs outside of PBs

We find that almost all visible PBs colocalize with miRNA or mRNA molecules, irrespective of relative RNA enrichment (Figure 2–5, 5A), and a single PB associates with at least 3 labeled RNA molecules within our timeframe of imaging (Figure 5B). Considering this frequent encounter of mi/mRNAs and PBs, that miRNA-mediated translational repression would eventually lead to RNA decay (Djuranovic et al., 2012) and that PBs are enriched for mRNA degradation enzymes (Hubstenberger et al., 2017; Parker and Sheth, 2007), we sought to test whether PBs are designated sites of RNA decay responsible for the bulk of cellular mRNA turnover. While fluorescence microscopy can visualize large PBs (> 50 nm), it does not capture smaller functional complexes of RNA decay enzymes. We therefore kinetically modeled (Mourao et al., 2014) the mRNA degradation activity of microscopically visible and invisible PBs computationally (Figure 5C). We specifically tested miRNA-mediated mRNA decay, largely due its cellular prevalence and prior reports on miRNA programmed mRNA localization to PBs; however, our method is extendable to other decay processes. We devised a basic set of reactions, each with predefined rates, whereby the interaction of miRISC with mRNPs activates PB-mediated mRNA degradation. Upon computing the copy number of each of these molecular species as they diffused across the lattice through time, we found that mRNA degradation was most efficient when there was a large number of small, mobile PBs (Figure 5C). That is, while degradation is possible within large, microscopically visible PBs, the process is most efficient when degradation factors, perhaps individual molecules, are unconstrained in the cell, thus presenting a large surface area for capturing repressed mRNAs.

Figure 5. A majority of microscopically visible PBs associate with mRNAs, but mRNAs are more effectively degraded with a larger number of smaller, microscopically-invisible PBs.

(A) Scatter plot representing the % PBs that colocalize with RNAs, per fixed UGD cell (n = 3, ≥ 15 cells per sample). (B) Frequency distribution of the number of times an individual PB encounters an RNA in live UGD cells (n = 3, 155 cells, 2102 PBs). Dotted line represents the average number of RNA encounters per PB after correcting for photobleaching. (C) Schematic (left) of in silico kinetic modeling of RNA-PB interactions and RNA decay. Changes in the abundance of mRNA over the timescale of the simulation is also depicted (right). Im (highlighted text) represents simulations in which PBs were immobile, whereas PBs were mobile in all other conditions. (D) Experimental validation of simulations using microinjection-based miRNA activity assay. Left, representative images of U2-OS cells treated with isotonic or hypertonic (300 mM Na⁺) medium and co-injected with CB-Dextran, GFP mRNA,mCh mRNA with MREs for cxcr4 (cx/cx*) miRNA and either a scrambled, control siRNA (Scr/Scr*) or cx/cx*. Images were acquired 4 h after injection. Right, scatter plot representing the ratio of mCh : GFP intensity at various injection and treatment conditions. Each dot represents a U2-OS cell (n = 3, 60 cells for each sample). See also Figure S5.

To test our in silico predictions experimentally, we resorted to modulating PB number and size via hyperosmotic stress, a method that has been proven to increase PB number in yeast (Huch and Nissan, 2017). We confirmed that hyperosmotic treatment of UGD cells results in a high number of immobile GFP-Dcp1a foci (Figure S5A-D), which form due to local association of previously mobile, microscopically-unresolved GFP-Dcp1a proteins, an aspect that is efficiently recapitulated by our in silico kinetic modeling approach (Figure 5C, “Im”). Microinjection-based miRNA activity assays (Figure S1E) in U2-OS cells suggested that, as predicted, miRNA-mediated gene repression is alleviated when PBs are aggregated upon subjection of cells to hyperosmotic stress (Figure 5D). Taken together, our data predict that mRNA degradation is primarily mediated by degradation enzymes rendered more efficient by freely diffusing in the cytosol, relegating PBs to degrading only a small fraction of repressed mRNAs.

lncRNA-PB interactions are distinct from those of regulatory miRNAs and repressed mRNAs

Having discovered the importance of translation versus translational repression in mRNA-PB colocalization behavior, we hypothesized that lncRNAs that sparsely interact with the translational machinery must localize to PBs via mechanisms distinct from those involving miRNAs and mRNAs. To address this hypothesis, we chose as model the nucleo-cytoplasmic lncRNA THOR (Figure 6A) that binds PB-enriched IGF2BP1 protein (Hubstenberger et al., 2017). We confirmed that THOR-MS2 still mediates the oncogenic phenotype of the unmodified lncRNA (Hosono et al., 2017) as evident by it promoting cell growth and stimulating oncogene expression (Figure S6A-C). We then performed live cell imaging assays (Supplementary movie 4) and found that that THOR-MS2 molecules, on an average, diffused faster than miRNAs or mRNAs that we imaged, but distributed between at least two populations of diffusion constants at PBs, much like the other RNAs (Figure S6D). Fixed cell imaging showed that the stoichiometry of THOR at PBs was marginally higher than that found in the cytosol (Figure 6B and S6E). While the fractional extent of RNA-PB colocalization did not significantly differ between mRNAs on the one hand and lncRNAs on the other (Figure 6C-D), we found significant differences in the localization patterns and interaction kinetics between mi/mRNAs and THOR-MS2 lncRNAs (Figure 6). In particular, THOR-MS2 frequently localized to the shell of PBs, whereas l7-Cy5/l7* miRNA or FL-l7–6x-MS2 mRNA, ~2.5–5-fold more PB-enriched than THOR-MS2, predominantly localized near PB cores (Figure 6E). We also observed that a THOR version lacking IGF2BP1 binding sites (THOR-Δbs-MS2) only rarely localized to or interacted with PBs (Figure 6C-D), indicating that THOR-PB interactions are mediated by IGF2BP1. Moreover, THOR-MS2 displayed ~2–3-fold more transient PB interactions than FL-l7–6x-MS2 mRNA (Figure 6F-G). Although the dwell time distributions were bi-phasic for both (Figure S5C and S6F), l7-Cy5/l7* miRNA or FL-l7–6x-MS2 mRNA (T_fast = 0.6 s and T_slow ≥ 15 s) resided at PBs for a significantly longer time than THOR-MS2 (T_fast = 0.6 s and T_slow = 2.9 s, Figure 6B and S6B). We further found that oncogenic lncRNA ARlnc1 (Zhang et al., 2018), known to bind PB-enriched HuR (Hubstenberger et al., 2017), displayed similar PB-localization kinetics and patterns as THOR (Figure 6 and S6F); whereas oncogenic LINC00941 (L941) (Shukla et al., 2017), a lncRNA that lacks consensus binding motifs for PB-enriched proteins (Hubstenberger et al., 2017), only rarely localized to PBs and displayed mono-phasic interaction kinetics, much like THOR-Δbs-MS2 (Figure 6 and S6F). Together, these data support our hypothesis that regulatory miRNAs and miRNA-regulated mRNAs are stably captured by PBs; by contrast, regulatory, non-translating lncRNAs that bind PB-localizing protein factors only transiently associate with PBs. These specific, yet transient lncRNA-PB interactions are often missed in ensemble assays that largely rely on the enrichment of stable, high-affinity interactions, likely leading to the relative dearth of lncRNAs observed in the transcriptome of PB cores (Hubstenberger et al., 2017).

Figure 6. lncRNAs transiently interact with PB peripheries.

(A) Schematic of different lncRNA constructs bound by their respective interacting protein partner. (B) Representative pseudocolored and contrast-adjusted images of fixed a UGD cell expressing GFP-Dcp1a (green) and stained for THOR-MS2 via smFISH (red). Dotted line, cell and nuclear outline. Scale bar, 10 μm. (C) Scatter plot representing the percentage of lncRNA molecules per cell that colocalize with PBs. Each dot is a cell. (D) Scatter plot for the enrichment of lncRNAs at PBs. Each dot is a PB. Grey dotted line depicts an EI of one, which demarcates PB-enriched (> 1) from PB-depleted (< 1) factors. (E) Representative pseudoclored and contrast-adjusted regions of fixed UGD cells with GFP-Dcp1a (green), stained for FL-l7–6x-MS2 mRNA or THOR-MS2 lncRNA via smFISH (red). Green and red dotted circles represent boundaries of PBs and THOR-MS2 respectively. Scale bar, 2 μm. Relative localization value is represented within the image. (F) Representative pseudocolored and contrast-adjusted images of a live UGD cells expressing GFP-Dcp1a (green) and THOR-MS2 (red). Dotted line, cell and nuclear outline. Scale bar, 10 μm. (G) Relative distribution of stable and transient interactions per live UGD cell for. (H) Comparison of fast and slow interaction kinetics in in live UGD cells. Green-black line depicts the mean magnitude of FL-l7–6x-MS2 for the respective observable. n = 3, ≥ 15 cells per sample, NS = not significant, * p ≤ 0.01, **p ≤ 0.001 or ***p ≤ 0.0001 by two-tailed, unpaired Student’s t-test. See also Figure S6.

DISCUSSION

Previous reports have provided exquisite static snapshots of RNA and protein colocalization with PBs (Cougot et al., 2012; Horvathova et al., 2017; Kedersha and Anderson, 2007; Liu et al., 2005), but could not assess the dynamics of the underlying recruitment processes. Others have provided valuable information regarding the dynamics of PB movement and the bulk exchange of proteins or mRNAs between PBs and the cytosol, but could not extract mechanistic information about the recruitment of biomolecules to PBs (Aizer et al., 2008; Aizer et al., 2014; Kedersha et al., 2008; Leung et al., 2006). Using single-molecule live-cell imaging we here uniquely demonstrate that miRNAs, mRNAs and lncRNAs dynamically localize to PB either stably or transiently (Figures 1 and 2). Having dissected the molecular anatomy of PBs (Figure S2), we find that stable anchoring at PBs is concordant with snapshots that visually portray RNA accumulation within PB “cores”, whereas more mobile localizations and transient interactions are more likely to depict the localization of RNAs in PB “shells”. In agreement with our data on mi/m/lncRNA-PB interactions during cellular homeostasis, recent reports (Moon et al., 2019; Wilbertz et al., 2019) have complementarily shown that mRNAs associate both stably and transiently with both stress granules (SGs) and PBs during the integrated stress response. The dwell times annotated as stable (~250 s) or transient (~10 s) in these reports are akin to particles in our datasets that dwell at PBs for the entire duration of acquisition (> 15 s) and for ~3–5 s, respectively. We have found an additional, highly dynamic interaction mode that lasts ~ 1 s, which potentially represents a relatively rapid PB-probing step. Based on the dwell times of THOR-Δbs-MS2 and L941-MS2 (~0.1–0.3 s, Figure S6F and Table S2), which seldom localize to PBs, it is unlikely that the dynamic interaction mode (~1 s) is an artifact of coincidental interaction/co-localization of RNAs with PBs. Upon RNP remodeling, these rapid encounters may transition into longer spans of granule probing or stable docking of RNAs to granules.

Elucidation of the PB-core transcriptome (Hubstenberger et al., 2017) has suggested that certain miRNAs, lncRNAs and repressed mRNAs are enriched in PBs, yet it is unclear whether the principles governing PB enrichment for these major classes of transcripts are similar or different. Strikingly, we found that miRNAs, mRNAs and lncRNAs have distinct PB localization signatures, which appear correlated with the distinct functionalities of these transcripts and the diversity in the types of RNPs they form (Figure 3). Based on our data, we propose a model that assigns PB localization patterns to specific RNA forms and functionalities (Figure 7). Stably anchored and PB-enriched miRNAs are predominantly dysfunctional – they do not have many mRNA targets and localize to PBs in their unbound or miRISC-bound (single-stranded or double-stranded) forms (Figure 4). Functional miRNAs, more likely to reside in RISC-mRNA complexes, display this behavior only in their minority and, when anchored, preferably localize within PB cores. These data are consistent with prior reports that both strands of both target-less and target-containing siRNAs localize to PBs (Jakymiw et al., 2005). We posit that, by contrast, transient associations at PB peripheries represent miRISC-mRNA complexes that do not yet have bound an important recruiting protein, such as GW182 or LAMP1 (Moon et al., 2019; Wilbertz et al., 2019), that is required for PB association. Conversely, highly translatable mRNAs that are not associated with miRNAs, while transiently associating peripherally, are not enriched at PBs. Based on recent reports (Moon et al., 2019; Wilbertz et al., 2019) and our data (Figures 2,3) we predict that non-translating mRNAs and translationally repressed mRNAs bearing MREs in their 3`UTR stably associate with PB cores, while only the latter are enriched at PBs. Furthermore, we find that miRNA-repressed mRNAs with MREs in the 5`UTR (Figure S5A-B) are not enriched at and only transiently associate with PBs, probably also due to the lack of a PB-recruitment factor bound to these RNPs (Figure 5). Prior reports have demonstrated that MREs in the 5’ UTR cause translational repression downstream of translation initiation sites (Lytle et al., 2007), potentially resulting in polysome bound non-translating mRNAs, which consequently cannot enter ribosome excluded PB cores (Parker and Sheth, 2007). By contrast, MREs in the 3’UTR typically result in inhibition of translation initiation, leading to non-translating mRNAs that are also free of ribosomes, which can then enter PB cores. Taken together, our data suggest that different modes of miRNA-mediated mRNA repression favor different types of PB localization.

Figure 7. Resulting model for the dynamic recruitment of specific RNAs to PBs.

RNAs dynamically associate with PB core or shell based on functionality. Target-free miRNAs, mRNA-targeting miRNAs and miRNA-targeted mRNAs with 3`UTR MREs are stably enriched within either cores or shells of PBs. The presence of a PB recruitment factor (PB-RF) may influence the dynamics and enrichment extent of miRNA-targeted mRNAs at PBs. lncRNAs transiently-yet-specifically associate with PB shells when the lncRNA binding protein (lncRNA-BP) is a PB enriched factor or is a PB-RF. Other lncRNAs, translating mRNAs and miRNA-targeted mRNAs with 5`UTR MREs transiently associate with PB shells, or are excluded from PBs. A majority of nuclease mediated RNA degradation occurs outside of PBs.

THOR, ARlnc1 and LNC00941 are recently discovered, oncogenic lncRNAs with distinct protein interactomes and functions. First, THOR is a highly conserved testis-specific lncRNA that is up-regulated in a broad range of human cancers and found to work in concert with IGF2BP1 (Hosono et al., 2017), a PB-enriched protein (Hubstenberger et al., 2017) that stabilizes transcripts via CRD (coding region instability determinant)-mediated mRNA stabilization (Weidensdorfer et al., 2009). Second, ARlnc1 is a lineage-specific lncRNA that collaborates with the PB-enriched protein HuR to enhance the stability of transcripts bound via an RNA-RNA interaction (Zhang et al., 2018). Third, LINC00941 (L941) is a lncRNA that is highly expressed in lung cancer (Shukla et al., 2017), that does not have consensus sequences for binding PB-resident proteins (data not shown). Based on our data, we propose that THOR, ARlnc1 and LINC00941 all assemble into slowly diffusing (D = 0.0001 – 0.1 μm²/s, Figure S6D) RNPs, which we posit correlate with their functions (Hosono et al., 2017). The frequent, transient associations of THOR and ARlnc1 with PBs may be linked to the regulatory role of these lncRNAs, wherein one can envision: 1) the lncRNAs depositing regulated mRNAs for storage at PBs; or 2) the lncRNAs instead selecting PB-stored mRNAs for reintroduction into the translating cytoplasmic pool. Of note, we rarely observed any stable anchoring or significant enrichment of THOR, ARlnc1 or LINC00941 at PBs, which suggests that the mere inability of an RNA to be translated is not a sole prerequisite for stable PB association and enrichment. Moreover, the mere ability of THOR and ARlnc1 to bind RNA-stabilizing proteins (IGF2BP1 and HuR respectively) may preclude stable, long lasting interactions with PBs that are enriched for RNA destabilizing factors. However, the relative contribution of stabilizing/destabilizing RBPs on PB-recruitment of mRNAs is yet to be determined and will clearly identify the molecular driving forces of RNP recruitment into phase separated granules and their subsequent regulation at these sites. Finally, our data also supports the notion that ncRNA-PB interactions are dependent on the size, as reported for SGs (Khong et al., 2017), and class of the regulatory ncRNA (Figure 6).

More broadly, our molecular observations of colocalizations of varying dynamics are consistent with phase transition principles that recently have been recognized to govern the assembly of large membrane-free granules (Protter and Parker, 2016; Shin and Brangwynne, 2017). Static, core-localized and enriched RNPs may serve as nucleating factors for large PBs, whereas dynamic, shell-localized and dispersed colocalizations may occur when the interfaces of the RNP, PB and surrounding cytoplasm are similar, as in a Neumann’s triangle observed in Cajal bodies attached to B-snurposomes (Shin and Brangwynne, 2017). Transient colocalizations may represent cases where the smaller RNP and PB come in close proximity, but the interfacial surface tension is too high for the two to fuse, presumably due to the absence of an appropriate PB-recruitment factor on the RNP.

Although there is general agreement on the phase-separation assembly principles of PBs and other RNA granules, the functions of these granules are still a topic of intense debate. Some reports have suggested that PBs may have stress dependent RNA decay or storage roles (Aizer et al., 2014), whereas others have suggested that PBs are sites of RNA storage, but not decay (Eulalio et al., 2007b; Horvathova et al., 2017; Stalder and Muhlemann, 2009; Tutucci et al., 2017). Notably, all previous studies have examined only microscopically visible PBs. Our computational simulations, which considered PBs both large and small, together with subsequent experiments using hyperosmotic stress to induce PB aggregation, suggest that microscopically visible PBs cannot account for the bulk of cellular mRNA decay (Figure 6). Our data instead suggest that fundamental principles of physical chemistry hold true for mRNA regulation processes within the complex cellular environment, in that the entropic gain from the larger degree of freedom and surface area of freely diffusing decay components dominates, an aspect that warrants additional lines of investigation. In addition to storing repressed mRNAs, our work unveils an additional housekeeping role for PBs in storing or possibly degrading unused miRNAs for their surveillance. Super-resolved fluorescence microscopy thus is shown to provide a powerful approach for mechanistically probing the dynamic assembly of RNP granules by phase separation at single-molecule resolution.

STAR⋆ METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Nils G. Walter (nwalter@umich.edu) or cocorresponding author Sethuramasundaram Pitchiaya (sethu@umich.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

HeLa (CCl-2, ATCC) and U2-OS (HTB-96, ATCC) cells were propagated in DMEM (GIBCO, #11995) and McCoy’s 5A (GIBCO, # 16600) basal media respectively supplemented with 10% FBS (GIBCO, # 16000). HeLa or U2-OS cells stably expressing GFP- Dcp1a (UGD) was created by transfecting U2-OS cells with pEGFP-Dcp1a and selecting for stable clones by G418 selection. UGD cells were grown in the abovementioned medium supplemented with 100 μg/mL G418 (Thermo-Fisher, # 10131027). All medium typically contained 1x Penicillin-Streptomycin (GIBCO, #15140). U2-OS cells stably expressing GFP-G3BP and RFP-Dcp1a (UGG-RD, gift from Nancy Kedersha) were propagated in McCoy’s 5A (GIBCO, # 16600) basal media supplemented with 10% FBS (GIBCO, # 16000). Phenol-red free McCoy’s 5A (GE-Amersham, # SH3027001) supplemented with 1% FBS was used for seeding and cells for imaging experiments. For hyperosmotic shock, cells were treated with the above media supplemented with 10 × PBS such that the final sodium concentration was 300 mM. Plasmid transfections for MS2-MCP imaging and cell growth assays were achieved using Fugene HD (Promega, # E2311). Cotransfection of plasmids with oligos was achieved using lipofectamine 2000 (Thermo-Fisher, # 11668027). For inducing stress granules (SGs), growth media of UGG-RD cells were supplemented with 0.5mM Sodium Arsenite (NaAsO₂) for 1 h. All cell lines were subjected to biweekly mycoplasma contamination and, HeLa and U2-OS cells were genotyped.

METHODS

DNA, RNA and LNA oligonucleotides

All DNA and RNA oligonucleotides used for iSHiRLoC experiments and reverse transcription, followed by quantitative polymerase chain reaction (RT-qPCR) were obtained from IDT. Oligonucleotides contained a 5` Phosphate (P) and, in the case of fluorophore labeled oligonucleotides, a Cy5 dye at the 3`end. Dyes were attached after oligonucleotide synthesis to a 3`amino group on a C6 carbon linker and were HPLC purified by the vendor. Guide and passenger strands were heat-annealed in a 1:1.1 ratio to achieve 10 μM stock solutions and were frozen until further use. Negative control siRNA (Scr/Scr*) was purchased as ready-to-use duplex samples from Ambion respectively. Six tandem let-7 (l7–6x) miRNA response elements (MREs) or mutant l7–6x (ml7–6x) MREs were purchased as gene blocks from IDT. AntimiR LNA oligos were purchased from Exiqon. Oligonucleotide and gene block sequences are listed in Table S1.

Plasmids

pEGFP-Dcp1a was constructed by ligating PCR amplified (using Pfu ultra polymerase, Agilent, # 600380) EGFP ORF (from pEGFP-C1, Clontech) into pmRFP1-hDcp1a (gift from Nancy Kedersha, Brigham Women’s hospital) within the AgeI and XhoI restriction enzyme (RE) sites. This replaces mRFP1 with EGFP in the plasmid. pEF6-mCh and pEF6-mCh-cx-6x construction was previously described previously (Pitchiaya et al., 2017). pEF6-mCh-l7–6x plasmid was constructed by ligating l7–6x gene block within NotI and XbaI sites of pEF6-mCh plasmid. Plasmids pRL-TK-let7-A, pRL-TK-let7-B, pRL-TK-cxcr4–6x, phage-ubc-nls-ha-2xmcp-HALO (a gift from Phil Sharp, Addgene plasmid # 11324, #11325, # 11308 and # 64540) and pmiR-GLO (pmG, Promega, # E1330) were purchased. pmG-MS2, encoding the firefly luciferase (FL) gene followed by 24 MS2 stem loops (FL-MS2), was created in two steps. First, the coding sequence (CDS) of IF2 was PCR amplified and ligated into the SbfI and NotI RE site of pmG, to create pmG-IF2. MS2 stem loops from pSL-MS2_24x (a gift from Robert Singer, Addgene plasmid # 31865) were then cloned into the EcoRI (introduced by above PCR)-NotI restriction enzyme sites pmG-IF2, to generate pmG-MS2. Clones containing the MS2 stem loops were created in SURE2 bacterial cells (Stratagene) to minimize recombination of the MS2 repeats with the bacterial genome. pmG-l7–6x-MS2 and pmG-ml7–6x-MS2 encoding FL-l7–6x-MS2 and FL-ml7–6x-MS2 respectively were constructed by ligating the l7–6x or ml7–6x gene blocks within the XhoI RE site in pmG-MS2. l7–6x-pmG and ml7–6x-pmG encoding l7–6x-FL and ml7–6x-FL respectively were created by ligating the synthesized I7–6x or ml7–6x fragment within the Esp3I and BbsI in pmG, between the human phosphoglycerate kinase promoter and FL CDS. l7–6x-pmG-MS2 and ml7–6x-pmG-MS2 encoding l7–6x-FL-MS2 and ml7–6x-FL-MS2 respectively were created using 24x MS2 stem loops from pMG-MS2 into EcoRI and NotI sites of l7–6x-pmG and ml7–6x-pmG. pmG-I7–2x-cx-4x was constructed by ligating the synthesized I7–2x-cx-4x fragment within the XhoI and EcoRI in pmG. pmG-I7–2x-cx-4x-MS2 was constructed by ligating 24x MS2 stem loops from pMG-MS2 into EcoRI and NotI sites of pmG-I7–2x-cx-4x. pmG-cx-6x-MS2 constructed by ligating 24x MS2 stem loops from pMG-MS2 into XhoI and NotI sites of pmG- cx-6x (Pitchiaya et al, 2017). pLenti6-THOR and pLenti6-RHOT (antisense of THOR) were constructed as described (Hosono et al., 2017). plenti6-THOR-MS2 was constructed by ligating 24x MS2 stem loops from pmG-MS2 into EcoRI and NotI sites of pLenti6-THOR. pCDH-ARlnc1-MS2 was constructed by ligating 24x MS2 stem loops from pMG-MS2 into XhoI and NotI sites of pCDH-ARlnc1 (Zhang et al. 2018). pCDH-LINC00941 was constructed by cloning LINC00941 into BstBt and BamHI sites of pCDH. pCDH- LINC00941-MS2 was then then constructed by ligating 24x MS2 stem loops from pMG-MS2 into XhoI and NotI sites of pCDH- LINC00941.

mRNA synthesis

pRL-TK-cx6x, pRL-TK-let7-A and pRL-TK-let7-B were linearized with NotI to generate RL-cx6x RL-l7–2x and RL-ml7–2x mRNAs respectively. pEF6-mCh-cx6x and pEF6-mCh-l7–6x were linearized with XbaI to generate mCh-cx6x and mCh-l7–6x mRNA respectively. The pCFE-GFP plasmid (Thermo Scientific) was directly used in the in vitro transcription reactions to generate the GFP mRNA. The linearized plasmids were extracted with phenol and chloroform and subsequently ethanol precipitated. In vitro transcriptions were performed using the MegaScript T7 kit (Thermo-Fisher, # AM1334) according to manufacturer’s protocol. Transcription reactions were then DNase treated (turbo DNase supplied with kit) and the respective RNAs were purified by sequential gel-filtration chromatography (Nap-5 followed by Nap-10, GE healthcare, # 17085301 and #17085401 respectively) and ethanol precipitation. The RNAs were 5’capped (ScriptCap™ m7G Capping System, CELLSCRIPT, # C-SCCE0625) and polyadenylated (A-Plus™ Poly(A) Polymerase Tailing Kit, CELLSCRIPT, # C-PAP5104H) and were further purified by sequential gel-filtration chromatography and ethanol precipitation. The length of the polyA tails was estimated based on electrophoretic mobility on a 1.2% formaldehyde agarose gel.

Luciferase reporter assays

100 μL of 10, 000 −20, 000 cells were seeded per well of a 96 well plate in antibiotics-free medium. Transfection conditions and luminescence readouts are as described previously (Pitchiaya et al., 2012; Pitchiaya et al., 2013; Pitchiaya et al., 2017). Briefly, cells were transfected with 60 ng of the indicated plasmid, 10 nM of the indicated dsRNA, and when appropriate 30 nM anti-ctrl or anti-l7 antimiRs, 0.4 μL of Lipofectamine 2000 (Invitrogen) and 50 μL of OptiMEM (GIBCO). 6 h after transfection the growth medium was replaced with fresh medium. 24 h after transfection, medium was replaced with phenol red-free McCoy’s 5A. Dual luciferase assays were performed using the Dual-Glo luciferase assay reagents (Promega, # E2920) as per the manufacturer’s protocol and luminescence was detected using a Genios Pro (Tecan) plate reader.

RT-qPCR

Cells were harvested and total RNA from cells were isolated using QIAzol Lysis reagent (Qiagen) and the miRNeasy kit (Qiagen) with DNase digestion according to the manufacturer’s instructions. cDNA was synthesized using Superscript III (Invitrogen) and random primers (Invitrogen). Relative RNA levels determined by qRT-PCR were measured on an Applied Biosystems 7900HT Real-Time PCR System, using Power SYBR Green MasterMix (Applied Biosystems). Expression was quantified by 2^ΔCt method, wherein Myc expression was first normalized to that of GAPDH and then this normalized expression was further normalized to Mock treatment.

Cell growth assays

100 μL of 10, 000 −20, 000 cells were seeded per well of a 96 well plate in antibiotics-free medium and were transfected every 24 h with the appropriate plasmid construct using Fugene HD (Promega, # E2311). Cell growth and viability was measured as an end point measurement for each time point using the Cell-titer GLO assay (Promega, # G7570) based on manufacturer’s instructions.

Microinjection

Cells grown on DeltaT dishes (Bioptechs, # 0420042105C) were microinjected as described (Pitchiaya et al., 2012; Pitchiaya et al., 2013, Pitchiaya et al., 2017). Briefly, injection solutions contained the appropriate miRNA at 1 μM concentration, 1x PBS and 0.5 mg/mL of 10 kDa cascade blue conjugated dextran (CB-Dex, Thermo-Fisher, # D1976). For microinjection based titration assays solution with 0 – 0.1 μM, 1x PBS and 0.1 mg/mL of 500 kDa cascade blue conjugated dextran (FITC-Dex, Thermo-Fisher, # D7136). For microinjection based miRNA activity assay, mRNAs were added at a stoichiometric amount based on the number of miRNA binding sites, for instance, 0.16 μM of RL-cx6x mRNA, bearing 6 cxcr4 binding sites, was added along with 1 μM cxcr4 miRNA. Solutions were filtered through a 0.45 μm Ultrafree-MC filter (Millipore, # UFC30HV00) and then centrifuged at 16,000 × g for 15min at 4 °C immediately before injection. The solution was loaded into a femtotip (Eppendorf, # E5242952008). Injections were performed using a Femtojet pump (Eppendorf) and an Injectman (Eppendorf) mounted to the microscope. Microinjections were performed at 100 hPa injection pressure for 0.5 s with 20 hPa compensation pressure. This pressure translates to a volume of 0.02 pL and 10,000–20,000 miRNA molecules.

Single-molecule fluorescence in situ hybridization

smFISH was performed as described (Hosono et al., 2017). Briefly, cells were grown on 8-well chambered coverglasses (Thermo-Fisher, # 155383PK), formaldehyde fixed and permeablized overnight at 4  ° C using 70% ethanol. Cells were rehydrated in a solution containing 10% formamide and 2 × SSC for 5 min and then treated with 100 nM fluorescence in situ hybridization probes (LGC-Biosearch) for 16 h in 2 × SSC containing 10% dextran sulfate, 2 mM vanadyl-ribonucleoside complex, 0.02% RNAse-free BSA, 1 μg μl−1 E. coli transfer RNA and 10% formamide at 37 °C. After hybridization, cells were washed twice for 30 min at 37 °C using a wash buffer (10% formamide in 2 × SSC). Cells were then mounted in solution containing 10 mM Tris/HCl pH 7.5, 2 × SSC, 2 mM trolox, 50 μM protocatechiuc acid and 50 nM protocatechuate dehydrogenase. Mounts were overlaid with mineral oil and samples were imaged immediately. Sequences of Q670 labeled probes against the FL gene are listed in Table S1 and probes against THOR and ARlnc1 were previously described (Hosono et al., 2017 and Zhang et al., 2018).

Immunofluorescence

Cells were grown on 8-well chambered coverglasses (Thermo-Fisher, # 155383PK), formaldehyde fixed and permeablized using 0.5% Triton-X100 (Sigma, T8787–100ML) in 1x PBS at room temperature (RT) for 10 min. Cells were then treated with blocking buffer containing 5% normal goat serum (Jackson Immunoresearch, 005–000-121), 0.1% Tween-20 (Sigma, P9416–50ML) in 1x PBS at RT for 1 h. Primary antibodies (pA) were diluted in blocking buffer to appropriate concentrations and cells were treated with pA at RT for 1 h. Following three washes with the blocking buffer for 5 min each cells were treated with secondary antibodies (sA) diluted in blocking buffer to appropriate concentrations. Following two washes with the blocking buffer and two washes with 1x PBS for 5 min each, cells were mounted in solution containing 10 mM Tris/HCl pH 7.5, 2 × SSC, 2 mM trolox, 50 μM protocatechiuc acid and 50 nM protocatechuate dehydrogenase. Mounts were overlaid with mineral oil and samples were imaged immediately.

Microscopy

Highly inclined laminated optical sheet (HILO) imaging was performed as described (Pitchiaya et al., 2012; Pitchiaya et al., 2013, Pitchiaya et al., 2017) using a cell-TIRF system based on an Olympus IX81 microscope equipped with a 60× 1.49 NA oilimmersion objective (Olympus), as well as 405 nm (Coherent ©, 100 mW at source, ~65 μW for imaging CB-Dex), 488 nm (Coherent ©, 100 mW at source, ~1.2 mW for imaging GFP), 561 nm (Coherent ©, 100 mW at source, ~50 μW for imaging mCh) and 640 nm (Coherent ©, 100 mW at source, 13.5 mW for imaging Cy5) solid-state lasers. Quad-band filter cubes consisting of z405/488/532/640rpc or z405/488/561/640rpc dichroic filters (Chroma) and z405/488/532/640m or z405/488/561/640m emission filters (Chroma) were used to filter fluorescence of the appropriate fluorophores from incident light. Emission from individual fluorophores was detected sequentially on an EMCCD camera (Andor IXon Ultra) for fixed cell imaging. For multicolour live-cell imaging, the emitted light was split onto two different EMCCDs using a single beamsplitter within a filter adapter (TuCam, Andor). Emission filters were placed just prior to each camera to minimize fluorescence bleed-through. For simultaneous detection of GFP and Cy5, a filter set with a 585dxcru dichroic that splits fluorescence into et525/50m and et705/100m emission filters respectively was placed in the Tucam adapter. For live cell imaging of MS2-MCP constructs, UGD cells on Delta T dishes were treated with 100 nM JF646- Halo ligand (a kind gift from Luke Lavis) for 30 min in growth medium without phenol red (Grimm et al., 2015). After the treatment, cells were washed three times in media and placed back in the incubator for 30 min, prior to imaging.

Image analysis

The two cameras used for simultaneous acquisition of GFP and Cy5 fluorescence in live cells were first registered as described (Churchman et al., 2005). Registration was achieved by imaging 0.1 μm tetraspeck beads (Thermo-Fisher, # T7279), whose emission is similar to both GFP and Cy5, before or after imaging of live cells. The registration matrix was then applied to GFP and Cy5 images for accurate tracking of PBs and RNAs respectively. Single particle tracking was performed as described (Pitchiaya et al., 2012; Pitchiaya et al., 2013) with some minor modifications. Briefly, particle tracking analysis was performed in Imaris (Bitplane) using tracks that spanned at least four video frames and all tracks were fit to a Brownian diffusion model to extract diffusion coefficients. PB boundaries were detected using a local contrast/threshold approach in Image J and Imaris. An RNA particle was identified as colocalizing with a PB when the centroid of the RNA is at or within the boundary of a PB. The use of finite observation windows to measure the dwell times introduces a systematic bias in the observed dwell times. This was corrected for by measuring the aggregate time for Cy5 photobleaching (T_phb) and subtracting its reciprocal this from the reciprocal of the observed dwell time (T_obs) along with the reciprocal of the observation window (T_w), as described by T_actual = 1 / ((1/ T_obs) - (1/ T_phb) - (1/ T_w)), as described (Rueda et al., 2004). Dwell times of all transcripts are summarized in Table S2. Percentage of track colocalizing with PBs (track %) was calculated as n_PB / (n_PB + n_Cyt), where n_PB = number of track localizations within PBs, n_Cyt = number of track localizations in the cytosol and depicted in Figure 2. This measure, in addition to visual inspection of individual tracks were used to objectively define trajectory “phenotypes” as stable or transient.

Step-wise photobleaching analysis of fluorophore labeled miRNAs and intensity analysis of smFISH particles in fixed cells were done using custom written Lab-view codes and ImageJ macros that can be shared upon request, as described (Pitchiaya et al., 2012; Hosono et al., 2017). To overcome statistical biases of co-incidental colocalizations introduced merely by RNA abundance, we calculated the accumulation of RNA within PBs via an enrichment index (EI) – a ratio of the number of RNA molecules in PB to those outside of PBs (Figure 2 and S2). An E.I. of > 1 suggests that the RNA accumulates at PB, whereas the opposite is true if the E.I. is ≤ 1. We also calculated the percentage of RNA or protein signal within PBs per cell by calculating the ratio between the cumulative abundance of signal within PBs divided by the total signal within the cell. Mean abundance / cell of all transcripts are provided in Table S3. Relative localization (RL) of RNAs within PBs was calculated as d_CR / (d_RB + d_CB), where d_CR = distance of RNA centroid from PB centroid, d_RB = distance of RNA centroid from PB boundary, d_CB = distance of PB centroid from PB boundary and depicted in Figure S2. The centroid and boundary of PBs were obtained via a modest variation of the local/adaptive-threshold method previously described (Simonson et al., 2010).

mCh and GFP signal from microinjection based miRNA activity assay were extracted and analyzed as described (Pitchiaya et al., 2012; Pitchiaya et al., 2017). Briefly, mCh and GFP intensity threshold were set (Huang threshold in image J) to automatically identify cell boundary. Background intensity, outside of cell boundary, was subtracted from mCh and GFP signal to extract the corrected intensity, whose ratio was calculated on a per cell basis.

In silico kinetic modeling

The fundamental theory and basic methodology of modeling, including the lattice gas automata algorithm are as described (Mourao et al., 2014). Our simulation platform allows for the specification of a variable number of elementary reactions. Unless otherwise stated, the results presented here were obtained using two different reactions:

$$miRISC\text{+}mRNP\underset{{\text{k}}_{-1}}{\overset{{\text{k}}_1}{\rightleftarrows }}\text{miRISC/mRNP}$$ (1)

$$miRISC/mRNP+\text{PB}\underset{{\text{k}}_{-2}}{\overset{{\text{k}}_2}{\rightleftarrows }}\text{miRISC/mRNP/PB}\stackrel{{\text{k}}_3}{\to }PB+\text{miRISC}$$ (2)

The reaction in (2) represents a catalytic event. The rate coefficients k_i are modeled as reaction probabilities. For example, in (1) k₁ is modeled by the probability that a miRISC and an mRNP molecule will react to form complex miRISC/mRNP, given that they have collided. Unless otherwise stated, the probability of a forward reaction (on the basis of the rate coefficients k₁ and k₂) is set to 1 and the probability of a reverse reaction (on the basis of the rate coefficients k_-1 and k_-2) is set to 0.1. The probability of a catalytic reaction (on the basis of the rate coefficient k₃) is set to 0.1. Note that the forward reaction rates (e.g., k₁) may remain constant over time, in agreement with the law of mass action, or decay over time for diffusion-limited reactions, when the time required for any two reactants to interact increases with the level of obstruction to diffusion. In the latter case, it can be shown that log(k₁) decays linearly at long times in a logarithmic time scale, as described (Mourao et al., 2014).

Each simulation begins with all particles randomly placed on a 2D lattice of size 200×200 lattice points with cyclic boundary conditions. Particles can be initialized with different sizes, provided that they are square, i.e., each initial particle can only occupy x2 positions, × being at least 1. Our platform allows for the creation of initial aggregates of a particular number and size. With the restriction mentioned above, we modulate the number and size of P-body particles within an aggregate with the assumption that all P-body particles within an aggregate have the same size. Each aggregate of P-bodies is created in two main steps. In the first step occurs, we insert the first molecule of the aggregate in the lattice. This first molecule is placed in a random position in the lattice. In the second step, we randomly select an adjacent neighborhood of a random P-body in the existing aggregate as a destination for the new P-body. The addition of P-bodies to an aggregate follows the reaction:

$$PB+N_{1}PB\to \left(N_1+1\right)\text{PB}$$ (3)

where N₁ corresponds to the number of P-bodies in the existing aggregate. This is done iteratively until the pre-determined aggregated size is achieved. Every particle is randomly initialized with a given orientation and direction of rotation. There are six possible orientations, corresponding to the coordinate number of a triangular lattice. The direction of rotation is always clockwise (CW) or counter-clockwise (CCW). Note that, although the particle’s movement is independent of its orientation, reactant particles will only associate if their orientations are complementary.

QUANTIFICATION AND STATISTICAL ANALYSIS

Graphpad-Prizm and Origin were used for statistical analysis and plotting. For pairwise comparisons, p-values were calculated based on non-parametric unpaired t-tests with Kolmogorov-Smirnov test. For comparisons involving more than 2 samples, one-way-ANOVA tests were used with Geisser-Greenhouse correction.

DATA AND SOFTWARE AVAILABILITY

Raw image files pertaining to figure panels in the main text and supplementary information can be found in http://dx.doi.org/10.17632/65t29ys57x.1.

Supplementary Material

2

Movie S1. miRNA-PB dynamics in living UGD cells (Related to Figure 1). l7-Cy5/l7* miRNAs were simultaneously tracked with GFP-Dcp1a containing PBs in living UGD cells. Videos are played at the data acquisition rate 20 frames per second (fps). The entire field-of-view (FoV) is 68.096 μm.

3

Movie S2. Diverse modes of miRNA-PB interactions in living UGD cells (Related to Figure 2). l7-Cy5/l7* miRNAs were simultaneously tracked with PBs in living UGD cells and representative videos of static RNA within a PB (S2A), dynamic RNA within a PB (S2B), cytosolic RNA moving into a PB (S2C), RNA transiently probing a PB (S2D) and RNA exiting a PB (S2E) are shown. l7-Cy5/l7* miRNAs were simultaneously tracked with PBs in living UGD cells. Videos are played at the data acquisition rate 20 fps. FoV scaled as in Figure 2.

4

Movie S3. mRNA-PB dynamics in living UGD cells (Related to Figure 4). FL-l7-6xMS2 mRNAs were simultaneously tracked with GFP-Dcp1a containing PBs in living UGD cells. Videos are played at the data acquisition rate 20 fps. The entire field-of-view (FoV) is 68.096 μm.

5

Movie S4. lncRNA-PB dynamics in living UGD cells (Related to Figure 6). THORMS2 lncRNAs were simultaneously tracked with GFP-Dcp1a containing PBs in living UGD cells. Videos are played at the data acquisition rate 20 fps. The entire field-of-view (FoV) is 68.096 μm.

HIGHLIGHTS.

• RNAs exhibit diverse spatiotemporal localization patterns at PB core and periphery

• Extent of stable or transient RNA-PB interactions depends on RNA functionality

• Positioning of cis-regulatory miRNA target sites influences PB interaction kinetics

• PBs contribute to miRNA surveillance, but less to mRNA decay