Phase transitions in the assembly and function of human miRISC

SUMMARY

miRISC is a multi-protein assembly that uses microRNAs (miRNAs) to identify mRNAs targeted for repression. Dozens of miRISC-associated proteins have been identified and interactions between many factors have been examined in detail. However, the physical nature of the complex remains unknown. Here, we show that two core protein components of human miRISC, Argonaute2 (Ago2) and TNRC6B, condense into phase-separated droplets in vitro and in live cells. Phase separation is promoted by multivalent interactions between the glycine/tryptophan (GW)-rich domain of TNRC6B and three evenly spaced tryptophan-binding pockets in the Ago2 PIWI domain. miRISC droplets formed in vitro recruit deadenylation factors and sequester target RNAs from the bulk solution. The condensation of miRISC is accompanied by accelerated deadenylation of target RNAs bound to Ago2. The combined results may explain how miRISC silences mRNAs of varying size and structure, and provide experimental evidence that protein-mediated phase separation can facilitate an RNA processing reaction.

In Brief

Phase separation of miRISC proteins leads to target mRNA sequestration and facilitates accelerated deadenylation.

INTRODUCTION

MicroRNAs (miRNAs) are post-transcriptional regulators of gene expression that are integral to diverse physiological process in animals, including epithelial regeneration (Chivukula et al., 2014), cardiac function (Xin et al., 2013), brain morphogenesis (Giraldez et al., 2005), and the progression of cancer (Lujambio and Lowe, 2012). To date, more than 500 unique human miRNA sequences have been reported (Fromm et al., 2015). The most evolutionarily conserved of these can be grouped into 87 miRNA families, which collectively regulate more than half of the mRNAs in the human transcriptome (Friedman et al., 2009).

To exert their regulatory function, miRNAs associate with Argonaute proteins to form the core of the miRNA-induced silencing complex (miRISC) (Liu et al., 2004; Meister et al., 2004). miRISC uses the sequence information in the miRNA as a guide to bind complementary sequences in mRNAs targeted for silencing (Bartel, 2009).

Once bound to its target mRNA, miRISC induces silencing by a complex mechanism that remains an active area of inquiry. Silencing includes elements of both translational repression and mRNA decay (Bagga et al., 2005; Bazzini et al., 2012; Djuranovic et al., 2012; Guo et al., 2010). There is no known upper limit for the size of target mRNA miRISC can repress and, although most miRNA binding-sites are found in the 3' untranslated regions of targets, miRISC appears to be able to silence from any position on an mRNA to which it can stably bind (Agarwal et al., 2015).

Insights into the structure of miRISC have come through pairwise studies of miRISC components (Boland et al., 2013; Chen et al., 2014; Christie et al., 2013; Elkayam et al., 2017; Jinek et al., 2010; Kozlov et al., 2010; Mathys et al., 2014; Petit et al., 2012; Schafer et al., 2014; Schirle and MacRae, 2012; Till et al., 2007; Wolf et al., 2014). Through these efforts a detailed model for the molecular architecture of miRISC has emerged (Jonas and Izaurralde, 2015). The central component in this model is the scaffold protein GW182 (TNRC6 in mammals) (Ding et al., 2005; Liu et al., 2005; Meister et al., 2005; Rehwinkel et al., 2005). GW182/TNRC6 proteins contain large intrinsically disordered regions (IDRs) that bridge interactions between Argonaute and downstream effectors, such as the deadenylase complexes PAN2–PAN3 (Braun et al., 2011; Chen et al., 2009) and CCR4–NOT (Behm-Ansmant et al., 2006; Chekulaeva et al., 2011; Fabian et al., 2011), which in turn interacts with the translational repressor and decapping activator DDX6 (Chen et al., 2014; Mathys et al., 2014).

In contrast to many globular miRISC components, which are characterized in detail, the miRISC scaffold proteins themselves are less well understood. Fragments of GW182/TNRC6 have been studied in isolation or in complex with globular binding partners (Chen et al., 2014; Christie et al., 2013; Elkayam et al., 2017; Jinek et al., 2010; Mathys et al., 2014; Pfaff et al., 2013), but because of its disordered nature, the intact protein has proven recalcitrant to structural and biophysical analysis. Consequently, the physical nature of the assembled miRISC remains unexplored.

Here, we use purified samples of human Argonaute2 (Ago2) and TNRC6B to examine the assembly, activity and physical properties of reconstituted human miRISC. We demonstrate that the PIWI domain of Ago2 possesses a tryptophan (trp) binding region, composed of three trp-binding pockets (two identified previously (Elkayam et al., 2017; Schirle and MacRae, 2012)), which allows Ago2 to make diverse interactions with the unstructured GW-rich ABD of TNRC6B. Interactions between Ago2 and TNRC6B promote the assembly of massive complexes that condense into phase-separated droplets in vitro. We further show that TNRC6B and Ago2 assemble into similar molecular condensates within live mammalian cells. In vitro reconstitution demonstrates that Ago2 remains functionally active in the separated phase, allowing droplets to sequester and concentrate target RNAs in a sequence dependent fashion. When exposed to mammalian cell lysate, miRISC droplets also recruit components of the CCR4-NOT complex as well as deadenylase activity, which can be accelerated by more than two orders of magnitude upon phase separation. These observations suggest a model in which miRISC uses molecular condensation to sequester miRNA targets and concentrate them with factors that mediate mRNA decay. The use of phase separating systems to control RNA localization and stability may be a feature common to many aspects of RNA metabolism (Berry et al., 2015; Feric et al., 2016; Fromm et al., 2014; Lin et al., 2015; Molliex et al., 2015; Strzelecka et al., 2010; Vourekas et al., 2016; Ying et al., 2017).

RESULTS AND CONCLUSIONS

Discovery of a third tryptophan-binding pocket in the Argonaute trp-binding region

To investigate the assembly and structure of human miRISC, we first sought to better understand how the two core protein components, Ago2 and TNRC6B, interact with each other. TNRC6B (and other GW182 proteins) contains an N-terminal Argonaute binding domain (ABD) with multiple regions that interact with Argonaute (Baillat and Shiekhattar, 2009; El-Shami et al., 2007; Elkayam et al., 2017; Eulalio et al., 2009; Lian et al., 2009; Pfaff et al., 2013; Takimoto et al., 2009; Till et al., 2007). The ABD is unstructured and rich in gly and trp residues, which are recognized through two trp-binding pockets on the surface of Argonaute (Jannot et al., 2016; Kuzuoglu-Ozturk et al., 2016; Pfaff et al., 2013; Schirle and MacRae, 2012). Most recently, the crystal structure of human Ago1 bound to a peptide (termed the “Ago-hook”) derived from human TNRC6A demonstrated that Argonaute uses these two pockets to bind tandem trp residues that are connected by a flexible linker (Elkayam et al., 2017).

After our initial report of two trp-binding pockets in Ago2 (Schirle and MacRae, 2012), we conducted a biochemical analysis of interactions between Ago2 and the ABD of TNRC6B (described below). Over the course of this work, we discovered that the two known trp-binding pockets are insufficient to explain how Ago2 recognizes some trp-residues in the ABD. We therefore searched for additional trp-binding pockets by determining new structures of Ago2 in the presence of free tryptophan (Table S1). This approach reconfirmed trp-binding pocket-1 and pocket-2, and identified a neighboring trp-binding pocket-3, which was masked by a lattice contact in the original Ago2 crystals (Figure 1A and 1B). All three pockets are clustered together in a small region near the bottom of the Ago2 PIWI domain.

Figure 1. Structure of the trp-binding region in Ago2.

(A) Cartoon representation of Ago2 illustrates the position of the trp-binding region. (B) Close up view of the trp-binding region. Fo-Fc tryptophan omit map, contoured at 2.5 σ, (green mesh) shows well-ordered indole side chains of three bound tryptophan molecules. (C) Surface representation of the region illustrates the three trp-binding pockets. Curved lines indicate approximate distances between adjacent pockets. See also Figure S1 and Table S1.

Measuring along the surface of Ago2, each trp molecule is separated from the other two by a distance of about ~25 Å, which corresponds to a flexible linker length of 10–15 amino acid residues (Figure 1C). This spacing of tryptophan residues is prevalent in GW-rich regions of diverse Argonaute-binding proteins (Figures S1). Notably, the trp-binding pockets are arranged in an equilateral triangle on the surface of Ago2 (Figure 1C). Thus, appropriately spaced trp residues could potentially interact with any two trp-binding pockets. Additionally, close examination reveals that all three pockets only contact the indole side chains of bound trp molecules, leaving the main chain atoms free to rotate about the C±-C² bond. This observation indicates that an individual pocket may be able to bind a trp-residue without reference to the polarity of the GW polypeptide chain. Thus, the structure of the trp-binding region suggests a model in which Ago2 is capable of diverse modes of engaging GW-rich proteins.

Diverse interaction modes between the trp-binding region of Ago2 and the GW-rich ABD of TNRC6B

We examined the model for Argonaute-GW recognition by monitoring the ability of Ago2 proteins to bind variants of the ABD from TNRC6B (Figure 2A). FLAG-tagged Ago2 was immobilized on an anti-FLAG agarose support and incubated with purified samples of the TNRC6B-ABD (residues 437-1056) fused to maltose binding protein (MBP). Proteins retained on the resin after washing were visualized by SDS PAGE. This assay showed a clear interaction between Ago2 and the TNRC6B-ABD (Figure 2B). The interaction was greatly reduced in an all-alanine (AllA) mutant ABD, in which all 36 tryptophans were mutated to alanine. Adding back two tryptophan residues found in “motif I” (W623 and W634), a region in TNRC6 known to interact with Argonaute (Baillat and Shiekhattar, 2009; Elkayam et al., 2017; Pfaff et al., 2013; Takimoto et al., 2009), to the AllA protein was sufficient to restore binding. Importantly, neither trp residue alone was sufficient to restore binding, indicating both that trp residues in motif I are recognized by Ago2.

Figure 2. Dissection of interactions between Ago2 and TNRC6B.

(A) Linear schematic of TNRC6B domain structure. MBP fusion of the ABD used in pull down assays, and sequence context of motif I (W623 and W634) are indicated. (B) Coomassie stained SDS gel showing pull down of ABD variants by wild type (WT) Ago2. AllA is an ABD variant in which all trp residues have been replaced with alanine. Ago2, wild type ABD variants, and AllA ABD variants are indicated by single, double, and triple asterix, respectively. (C) Schematic of trp-binding pockets in Ago2 with residues mutated to disable individual pockets indicated in blue (K660S, P590G, and R688S mutations in pockets 1, 2 and 3, respectively). (D) Pull down of ABD variants by Ago2 with individual trp-binding pockets inactivated. (E) Pull down of ABD by Ago2 with combinations of trp-binding pockets inactivated. (F) Schematic of ABD construct with sequence context motif II (W875, W896, and W910) indicated. (G) Pull down of W875, W896, and W910 AllA-ABD variants by WT Ago2, and (H) by trp-binding pocket Ago2 mutants. See also Figure S2.

We next generated Ago2 mutants in which individual trp-binding pockets were disabled (Figure 2C and Figure S2). All trp-binding pocket mutants retained the ability to bind the wild-type ABD, demonstrating that no single pocket is required (Figure 2D). Disabling combinations of pockets revealed that any single pocket is sufficient to maintain observable, albeit reduced, pull-down of wild type ABD (Figure 2E). Disruption of pocket-1 severely reduced interactions with the ABD variant containing only the two trp residues from motif I (W623 and W634). In contrast, mutation of pocket-2 had no effect. This is the observation that led us to search for additional pockets (described above), as the requirement of two trp residues indicated the involvement of two trp-binding pockets. Subsequent mutation of pocket-3 disrupted interactions with trp residues in motif I (Figure 2D). We therefore conclude that, unlike the Ago-hook motif (Elkayam et al., 2017), motif I is specifically recognized via trp-binding pockets 1 and 3.

Another region of TNRC6B known to interact with Argonaute, termed “motif II”, contains three tryptophan residues (W875, W896, and W910) (Elkayam et al., 2017; Takimoto et al., 2009; Till et al., 2007) (Figure 2F). Adding any adjacent pair of trp residues from this region to the AllA-ABD increased interactions with Ago2, and adding all three led to pull down levels comparable to the wild type ABD (Figure 2G). Interactions with motif II are most dependent on trp-binding pockets 1 and 2, while mutation of pocket-3 had a modest effect (Figure 2H).

The combined results indicate that all three pockets in the trp-binding region of Ago2 contribute to interactions with TNRC6B. Specific trp residues in TNRC6B have pocket preferences, suggesting that the three trp-binding pockets are not perfectly redundant with each other. However, each pocket is able to interact with multiple different trp residues in TNRC6B. These observations support a model in which interactions between Ago2 and TNRC6B are multivalent and structurally pliable.

Ago2 drives TNRC6B into phase-separated droplets

The nature of the trp-binding region in Ago2, capable of diverse binding modes, suggested that the structures formed with TNRC6B may be heterogeneous and complex. Indeed, under physiological salt concentrations (100 mM KOAc, 20 mM NaCl), solutions containing both Ago2 and the TNRC6B-ABD quickly became opaque, indicative of the formation of massive particles (Figure 3A and S3). In contrast, solutions containing Ago2 and the AllA-ABD remained clear. Concentrated solutions of the ABD alone also became turbid at temperatures below 15°C (Figure 3B and 3C).

Figure 3. Ago2 drives phase separation of the TNRC6B-ABD.

(A) Initial observation of ABD-Ago2 phase separation. Solutions containing the TNRC6B ABD become turbid upon introduction of Ago2. Solutions of the AllA-ABD mutant remain clear. (B) Turbidity, measured by absorbance of 480 nm light, of solutions containing Ago2 (20 µM), ABD (40 µM), or ABD + Ago2 (40µM and 20µM, respectively) plotted as a function of temperature. (C) Turbidity versus temperature for ABD samples (10 µM) with various concentrations of Ago2. (D) Light microscopy images of mixtures of ABD (20 µM) and Ago2 taken at room temperature (~23 °C). Scale bar, 10 µm. (E) Time-lapse images showing fusion of three adjacent ABD-Ago2 droplets (left), and confocal microscopy images showing fusion of ABD-Ago2 droplets containing Alexa-488 labeled ABD (right). (F) Fluorescence microscopy images from a FRAP experiment in which an entire ABD-Ago2 droplet was bleached. 10% of Ago2 molecules were labeled with TMR, and ~10% ABD molecules were labeled with Alexa Fluor 488. (G) FRAP recovery curves for three ABD-Ago2 droplets with error bars indicating SEM. All droplets were formed in 100 mM KOAc and 20 nM NaCl. See also Figures S3, and Movies S1 and S2.

Viewing samples under a light microscope revealed that turbidity correlated with the formation of microscopic structures (Figure 3D). These structures were usually round and thus appeared to be phase-separated droplets. Droplet size, ranging from <0.2 µm to >10 µm in diameter, was dependent on the input amount of Ago2, indicating that phase separation of the ABD is driven or stabilized by interactions with Ago2 (Figure 3D). Indeed, trp-binding pocket-1 or pocket-3 Ago2 mutants generated 15–20% fewer droplets than equivalent concentrations of wild-type Ago2, and mutation of pocket-2 reduced droplet numbers and size by 35–50%. Droplet size and number were reduced by 80–90% by the pocket-2/pocket-3 double mutation (Figure S3).

ABD-Ago2 structures fuse upon direct contact (Figure 3E and Movie S1) and disassemble via formation of internal vacuoles upon introduction of concentrated NaCl (Movie S2). These behaviors are indicative of dynamic, liquid-like properties (Banerjee et al., 2017; Brangwynne et al., 2009). To further examine their material properties we formed ABD-Ago2 droplets containing Alexa-488 labeled TNRC6B-ABD and Ago2 labeled with tetramethylrhodamine (TMR) (Figure 3E). Fluorescence recovery after photobleaching (FRAP) experiments indicate that about 40% of both Ago2 and the ABD molecules are mobile and exchange with the bulk solvent (Figure 3F and 3G). Incomplete recovery of FRAP signal reveals that ABD-Ago2 droplets may not be simple liquids, but instead also exhibit partially solid-like properties. Recovery times for the ABD (t_1/2 = 110 sec) were less that half of that of Ago2 (t_1/2 = 245 sec), indicating that Ago2 diffuses more slowly into the separated phase than the ABD in our experimental conditions.

Ago2 also drove the formation of droplets when mixed with full-length TNRC6B (Figure 4A). Droplets initially appeared as small spheres, which coalesced into larger clusters over time (Figure 4B). Despite their amorphous appearance, FRAP experiments revealed that ~22% of the Ago2 molecules can exchange with the bulk solvent (t_1/2 = 80 sec). In contrast TNRC6B remained largely immobile in the separated phase, with only ~1% fluorescence recovery 10 minutes after photobleaching (Figure 4C and 4D).

Figure 4. Phase separation of miRISC.

(A) Fluorescence microscopy images showing Ago2 (TMR labeled) promotes phase separation of full length TNRC6B (Alexa Fluor 488 labeled) in vitro. (B) Spherical miRISC droplets formed in vitro coalesced into larger clusters over time. (C) Representative fluorescence microscopy images from miRISC FRAP experiments. (D) FRAP recovery curves for three miRISC droplets with error bars indicating SEM. (E) Live cell images showing fusion of two GFP-TNRC6B cytoplasmic foci in HEK 293 cells over time. (F) Representative live cell images of GFP-TNRC6B cytoplasmic foci FRAP experiments. (G) FRAP recovery curves for three GFP-TNRC6B cytoplasmic foci with error bars indicating SEM. (H) Representative live cell images of GFP-TNRC6B/mCherry-Ago2 cytoplasmic foci FRAP experiments. (I) FRAP recovery curves for four GFP-TNRC6B/mCherry-Ago2 cytoplasmic foci with error bars indicating SEM. See also Figure S4, and Movies S3, S5 and S6.

The combined observations demonstrate that TNRC6B and its intrinsically disordered GW-rich ABD are prone to phase separation. Phase separation is driven or stabilized by interactions with Ago2, resulting in microscopic structures with viscoelastic properties.

Phase separation of TNRC6B in living cells

Fritzler and Chan first reported that human GW182 (often called TNRC6A) concentrates in cytoplasmic structures 15 years ago (Eystathioy et al., 2002). These structures, termed GW bodies (GWBs, sometimes termed P-bodies or GW/P-bodies), vary in size from 0.2 to 1 µm, are dynamic within the cytoplasm, and dissipate and reform over the course of the cell cycle (Eystathioy et al., 2002). Adjacent GWBs were noted to fuse upon contact and return to a spherical shape (Yang et al., 2004), indicative of liquid-like behavior (Brangwynne, 2013). We therefore examined the hypothesis that TNRC6B undergoes phase separation in cells in a manner similar to the purified protein in vitro.

We first generated a 293 HEK cell line that stably expresses GFP-tagged TNRC6B. As seen previously (Yang et al., 2004), cells contained GFP-labeled, dynamic foci 0.2 to 1 µm in size (Figure S4 and Movie S3). In some cases, adjacent foci fused upon contact, while in other cases adjacent foci clustered without immediately fusing (Figure 4E, Movie S3 and Movie S4). FRAP experiments indicate that ~45% of the TNRC6B molecules in foci exchanged with the cytoplasm, with an average recovery half time of 90 seconds (Figure 4F and 4G and Movie S5). This finding shows that the cellular environment maintains condensates of full-length TNRC6B in a more liquid-like state than our in vitro conditions. Transfecting cells with a plasmid encoding mCherry-labeled Ago2 revealed that, as seen previously (Jakymiw et al., 2005; Liu et al., 2005), Ago2 accumulated in foci with TNRC6B (Figure 4H and Movie S6). FRAP experiments indicate that ~45% and ~40% of TNCR6B and Ago2 molecules in foci exchanged with the bulk cytoplasm, respectively (Figure 4H and 4I and Movie S6). We conclude that cellular conditions are conducive to phase separation of TNRC6B and Ago2, and that upon phase separation these miRISC components are incorporated into a cellular material with viscoelastic properties resembling those formed by the purified proteins in vitro.

Ago2-TNRC6B droplets sequester miRNA targets and recruit miRISC components

We also examined the biochemical properties of reconstituted Ago2-TNRC6B droplets. We noticed that larger droplets sink over time and thus can be isolated from the bulk solvent by centrifugation (Figure 5A). Running supernatant and pelleted fractions on an SDS polyacrylamide gel confirmed that droplets contain both Ago2 and TNRC6B (Figure 5B). Under conditions that promote formation of droplets large enough to pellet ([TNRC6B] ≥ 300 nM), highly purified TNRC6B phase separated efficiently when moved into our standard reaction buffer, with nearly all protein appearing in the pellet fraction (Figure S5). However, when using partially purified TNRC6B a substantial fraction of both TNRC6B and Ago2 remained in supernatant fraction (Figure 5B). Under these conditions the droplets specifically recruited full length TNRC6B and Ago2.

Figure 5. Biochemical characterization of Ago2-TNRC6B droplets.

(A) Ago2-TNRC6B droplets can be separated from the bulk solvent by centrifugation. Cartoon schematic of procedure (left), and images of droplets in input and supernatant fractions (right). (B) Droplets recruit full-length TNRC6B, Ago2, and miRNA target RNAs. TNRC6B (~1 mM, partially purified) was mixed with Ago2 (0.5 µM) loaded with either let-7 or miR122, and a ³²P-labeled let-7 target RNA (8xlet7 target, ~3 nM). After centrifugation, supernatant and pellet fractions were analyzed by Coomassie stained SDS PAGE (right, top panel) and phosphorimaging of a denaturing gel (right, bottom panel). (C) Ago2 remains active in the separated phase. TNRC6B (~ 1 µM) was mixed with Ago2-miR122 (250 nM) and a ³²P-labeled target RNA (~0.5 µM) with perfect complementarity to miR122 in the absence of divalent cations. After centrifugation MgCl₂ (3 mM) was added to the separated phase. Target RNA was extracted and analyzed by denaturing PAGE and phosphorimaging (right panel). (D) TNRC6B-Ago2 droplets recruit other miRISC components. TNRC6B (40 nM) was mixed with Ago2 (40 nM) and soluble lysate from HEK 293 cells (OD₂₆₀ ~3). Input, supernatant, and pellet fractions were analyzed by Western blot (right panel). See also Figure S5.

Ago2-TNRC6B droplets also selectively sequester miRNA targets from the bulk solution. When mixed with droplets containing Ago2-let7 (Ago2 loaded with the miRNA let-7), a target RNA with eight seed matched let-7 binding sites (8xlet7) appeared almost exclusively in the pellet fraction (Figure 5B). In contrast, the 8xlet7 target RNA remained in the supernatant when added to droplets formed with Ago2-miR122.

Ago2 can catalyze the endonucleolytic cleavage (termed “slicing”) of target RNAs with extensive complementarity to its small RNA guide (Liu et al., 2004; Meister et al., 2004). To examine the integrity of Ago2 in the separated phase, we measured slicing activity. Ago2-TNRC6B droplets were generated in the absence of divalent cations, which are required for slicing (Martinez and Tuschl, 2004). The phase separation reaction included Ago2-miR122 and a ³²P-labeled target RNA with perfect complementarity to miR122. Droplets were isolated by centrifugation and MgCl₂ was added to the pelleted material. After a brief incubation, RNA was extracted and applied to a denaturing polyacrylamide gel, which revealed a distinct band corresponding to the cleaved target RNA (Figure 5C). Thus, Ago2 remains catalytically active in the separated phase.

Finally, we tested if Ago2-TNRC6B droplets can incorporate other protein components of miRISC. Ago2 and TNRC6B were mixed in the presence of soluble lysate from HEK 293 cells and the resulting droplets were isolated by centrifugation. Analyzing by Western blot revealed that subunits of the CCR4-NOT deadenylase complex (CNOT7 and CNOT9) co-pelleted with TNRC6B (Figure 5D). Consistent with previous reports showing TNRC6/GW182 directly binds CCR4-NOT (Chen et al., 2014; Mathys et al., 2014), Ago2 was not required for this interaction. Actin, which is not a component of miRISC, remained in the supernatant fraction.

The combined results indicate that phase separation by Ago2-TNRC6B is a biochemically discriminating process, in terms of both protein and RNA components. Phase separation thereby provides a means for removing target RNAs from the bulk solvent and concentrating them with factors involved in mRNA degradation and repression.

Phase separation accelerates miRNA target deadenylation

After observing Ago2 slicing activity in miRISC droplets (Figure 5C) we wondered if deadenylases associated with miRISC might also function within the separated phase. To test this idea, we employed a 5' cap-radiolabeled 8xlet7 target RNA, which contains a defined 114 nt poly(A) tail and eight sites complementary to the seed region of the miRNA let-7 (Fukaya and Tomari, 2012). The target RNA was stable in soluble cell lysate (Figure 6A and 6B). As expected, adding Ago2-let7 induced poly(A)-dependent 3' end shortening (Figures 6B and S6). The majority of the target RNA remained in the supernatant fraction after centrifugation, indicating deadenylation occurred in miRISC assemblies too small to pellet. Therefore, to promote formation of heavy miRISC droplets, partially purified TNRC6B was added to the reaction mixture prior to initiation of deadenylation (via addition of Mg²⁺). The pellet fraction of the +TNRC6B condition contained ~97% of the total target RNA, the majority of which was deadenylated (Figure 6B). We conclude that phase-separated miRISC is capable of target RNA deadenylation.

Figure 6. Efficient target deadenylation in miRISC droplets.

(A) Schematic of experiment. (B) Deadenylation of a target RNA by miRISC. Ago2 (40 nM, final concentration), loaded with either let7 or miR122 (negative control), was mixed with a ³²P-5'-cap-labeled target RNA harboring binding sites for let-7 and a 114 nt. poly(A) tail in the presence of soluble lysate from HEK 293 cells (OD₂₆₀ ~3), with and without additional TNRC6B (~300 nM, partially purified). After a 15-minute incubation, supernatant and pellet fractions were isolated by centrifugation and target RNA was extracted and analyzed by denaturing PAGE and phosphorimaging. (C) Deadenylation timecourse. Reactions containing Ago2-let7 (40 nM) and soluble HEK 293 lysate (OD₂₆₀ ~3), with and without exogenous TNRC6B (~300 nM, final concentration) were fractionated at various times, and analyzed by denaturing gel. (D) Estimation of deadenylation rates. Target RNA bands in (C) were quantified and fraction of total intact RNA (A₁₁₄) for +/− TNRC6B conditions was plotted as a function of incubation time. Data were fit with a first order decay, yielding A₁₁₄ half-lives of 40 and 4 minutes for plus and minus exogenous TNRC6B, respectively. Plotted data are the average of three independent experiments with SEM indicated as error bars. See also Figure S6.

We noticed that target deadenylation rates increased ~10 fold with the introduction of exogenous TNRC6B (Figure 6C). This observation raised the possibility that deadenylation was stimulated by condensation of miRISC. Alternatively, accelerated deadenylation may have only been a consequence of increasing TNRC6B concentration and unrelated to phase separation. Indeed, the small amount (~3%) of target RNA in the supernatant fraction was deadenylated at rate closely matching the rate in the pellet fraction, suggesting either imperfect pelleting of miRISC droplets or that deadenylation rates in aqueous and separated phases were equally stimulated by addition of TNRC6B. To distinguish between these possibilities we sought a controlled experimental regime for assessing the effects of phase separation on miRNA target deadenylation.

Molecular crowding agents, such as polyethylene glycol (PEG), can reduce the critical concentration required for protein phase separation (Annunziata et al., 2002). Indeed, we found that adding PEG 8000 (5% weight/volume, final concentration) to solutions containing low concentrations of Ago2 and TNRC6B, promoted the formation of large miRISC droplets (Figure 7A and S7). Thus, introduction of PEG provides a controlled and innocuous method of inducing miRISC condensation without altering the stoichiometry of miRISC components in the reaction mixture.

Figure 7. Inducing miRISC phase separation accelerates target deadenylation.

(A) PEG 8000 promotes TNRC6B-Ago2 phase separation. Images of droplets formed from Alexa-488 labeled TNRC6B (~20 nM) and Ago2 (200 nM) in the presence and absence of 5% (w/v) PEG 8000. (B) Effects of PEG 8000 on target deadenylation reactions. 8xlet7 deadenylation reactions containing combinations of Ago2 (20 nM), HEK 293 lysate (OD₂₆₀ ~1.5), and exogenous TNRC6B (~30 nM) were treated with 5% (w/v) PEG 8000, separated into supernatant and pellet fractions, and analyzed by denaturing PAGE. (C) Effect of PEG 8000 on deadenylation rates. Reactions containing Ago2-let7 (20 nM), HEK 293 lysate (OD₂₆₀ ~1.5), and exogenous TNRC6B (~30 nM) were treated with 5 % (w/v) PEG 8000, separated into supernatant and pellet fractions, and analyzed by denaturing PAGE at various times. (D) PEG 8000 accelerates deadenylation. Bands corresponding to target RNA species in (C) were quantified and fraction of intact target (A₁₁₄) plotted as a function of time. Plotted data are the average of three independent experiments with SEM indicated as error bars. See also Figure S7.

We next determined the impact of PEG-induced miRISC phase separation on target deadenylation. Adding PEG to a deadenylation reaction containing HEK 293 lysate and 20 nM Ago2-let7 caused about half of the target RNA to move to the pellet fraction after centrifugation (Figure 7B). Including purified TNRC6B (~20 nM, final added concentration) and PEG in the reaction shifted almost all target RNA into the pellet. In both cases, pelleted fractions contained substantially more deadenylated target RNA than the supernatants. Time course experiments demonstrated that target deadenylation rates increased >100 fold with PEG-induced miRISC droplet formation (Figures 7C and 7D).

These results demonstrate that miRISC condensates are competent for target deadenylation. Moreover, inducing miRISC condensation by two distinct routes (increasing TNRC6B concentration, or introducing a molecular crowding agent) correlated with increased target deadenylation rates. Creation of heavy miRISC droplets is the only known variable common to both experiments. We therefore conclude that condensation of miRISC via phase separation enables the deadenylation of miRNA targets.

Discussion

In this study, we examined the assembly principles and biochemical activities of human miRISC. Previous studies identified discrete regions in GW-rich proteins that interact with Argonaute with high, moderate and low efficacy (Chekulaeva et al., 2009; El-Shami et al., 2007; Eulalio et al., 2009; Lian et al., 2009; Pfaff et al., 2013; Takimoto et al., 2009; Till et al., 2007; Yao et al., 2011). The finding that each trp-binding pocket in Ago2 has preferences for specific trp residues in TNRC6B (Figure 2) may explain why some GW-rich regions contribute more than others to stable interactions with Argonaute. We further suggest that the structure of the trp-binding region, capable of diverse binding modes (Figure 1), may explain how moderate and low affinity interactions with GW-rich proteins can be established in the absence of obvious or well-conserved consensus sequences.

A variety of cellular structures containing GW182 or other GW-rich proteins have been observed in animals, plants and fungi. In animal cells, TNRC6/GW182 has been found in GW/P-bodies (Eystathioy et al., 2003; Liu et al., 2005; Sen and Blau, 2005), cytoplasmic structures that contain non-translating mRNAs and arise as a consequence of post-transcriptional silencing (Chu and Rana, 2006; Eulalio et al., 2007). GW/P-body-like structures in mammalian neurons contain ribosomes and are actively transported along dendrites, with release of Ago2 upon neuronal activation, suggesting a role in mRNP localization (Cougot et al., 2008). Inhibiting nuclear export of TNRC6A in HeLa cells leads to formation of nuclear foci enriched in Ago2, but lacking general mRNA decay factors (Nishi et al., 2013). In Arabidopsis, GW-rich proteins NRPE1 and KTF1 colocalize with AGO4 in nuclear bodies associated with small RNA-directed DNA methylation (El-Shami et al., 2007; He et al., 2009; Huang et al., 2009; Li et al., 2008). Finally, the S. pombe protein Tas3 contains an Argonaute-binding GW-rich region (Till et al., 2007) and accumulates in nuclear foci associated with heterochromatin in fission yeast (Motamedi et al., 2004; Petrie et al., 2005). These observations, taken with the propensity of the GW-rich TNRC6B-ABD to phase separate (Figure 3), leads to the hypothesis that GW-protein phase transitions may contribute to diverse cellular bodies with distinct biological functions.

Previous pairwise interaction studies have provided a detailed model for the core structure of miRISC (Boland et al., 2013; Bridge et al., 2017; Chen et al., 2014; Christie et al., 2013; Elkayam et al., 2017; Jinek et al., 2010; Jonas et al., 2014; Jonas and Izaurralde, 2015; Kozlov et al., 2010; Mathys et al., 2014; Petit et al., 2012; Schafer et al., 2014; Schirle and MacRae, 2012; Till et al., 2007; Wolf et al., 2014). Our data reveal that human miRISC can also assemble into multimeric structures that manifest as a phase-separated viscoelastic material (Figures 3, 4 and Figure S7). The extent to which miRISC phase separation contributes to repression of miRNA targets remains an open question. If functional, miRISC condensates must be able to form on submicroscopic scales, as dissipation of GW/P-bodies did not impact miRNA-mediated repression (Chu and Rana, 2006; Eulalio et al., 2007). Consistent with this notion, submicroscopic P-body-related mRNP assemblies have been identified in yeast that are visibly P-body-deficient (Rao and Parker, 2017). Our in vitro reconstitution data indicate that phase separation can increase miRISC functionality both in terms of sequestering miRNA-targets and accelerating their deadenylation (Figure 5, Figure 6 and Figure 7). We therefore put forth the hypothesis that the ability to form higher order complexes via molecular condensation may allow miRISC to organize miRNA-targets within the cytoplasm and thereby modulate rates of mRNA translation and decay. The prevalence of IDRs and modular/multivalent structures in diverse RNA-binding proteins (Calabretta and Richard, 2015; Lin et al., 2015; Lunde et al., 2007; Strzelecka et al., 2010) suggest that this may be a theme common to many aspects of RNA biology.

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, Ian J. MacRae (macrae@scripps.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Bacterial strains and plasmids

Bacteria used for cloning were OminMAX™ chemically competent E. coli (ThermoFisher). Bacteria used for expression of protein were BL21 (DE3) chemically competent E. coli. Bacteria used for production of bacmid were DH10Bac™ chemically competent E. coli (ThermoFisher). Plasmids are listed in Key Resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit polyclonal anti-FLAG tag	Cell Signaling	Cat# 2368S
Rabbit polyclonal anti-human Ago2	Cell Signaling	Cat# 2897S
Mouse monoclonal anti-CNOT7	Abnova	Cat# H00029883
Rabbit polyclonal anti-RQCD1 (CNOT9)	Proteintech	Cat# 22503-1-AP
Mouse monoclonal anti-pan-actin (ACTN05 (C4))	Pierce	Cat# MA511869
Chemicals, Peptides, and Recombinant Proteins
Alexa Fluor™ 488 C5 Maleimide	ThermoFisher	Cat# A10254
HaloTag® TMR Ligand	Promega	Cat# G8251
cOmplete™ Mini EDTA-free Protease Inhibitor tablets	Roche	Cat# 4693159001
Murine RNase Inhibitor	New England Biolabs	Cat# M0314S
Ribonucleic acid, transfer from baker’s yeast	Sigma	Cat# 9014-25-9
Ni-NTA Agarose	Qiagen	Cat# 30210
StrepTactin Sepharose HP	GE Healthcare Life Sciences	Cat# 28935599
High Capacity NeutrAvidin™ Agarose	Pierce	Cat# 29202
Q-Sepharose Fast Flow	GE Healthcare Life Sciences	Cat# 17051001
Critical Commercial Assays
HaloTag® Protein Purification System	Promega	Cat# G6280
NEBuilder® HiFi DNA Assembly Cloning Kit	New England BioLabs	Cat# E5520S
Deposited Data
Crystal Structure of Human Argonaute2 Bound to Three Tryptophans	This paper	PDB ID: 6CBD
Experimental Models: Cell Lines
HEK 293 Flp-In™	Invitrogen	Cat# R75007
HEK 293 Flp-In™ TRex™	Invitrogen	Cat# R78007
HEK 293 Flp-In™ TRex™ GFP-TNRC6B-FL	This paper	N/A
Spodoptera frugiperda (Sf9)	Expression Systems, LLC	Cat# 94-001S
Experimental Models: Organisms/Strains
Oligonucleotides
p-let-7a guide RNA: p-UGAGGUAGUAGGUUGUAUAGU (p=5′ phosphate)	Integrated DNA Technologies	N/A
let-7a capture RNA: Biotin-mAmUmAmGmAmCmUmGmCmGmAmCmAmAmUmA mGmCmCmUmAmCmCmUmCmCmGmAmAmCmG (m=2′-O-methyl)	Integrated DNA Technologies	N/A
let-7a competitor DNA: Biotin-CGTTCGGAGGTAGGCTATTGTCGCAGTCTAT	Integrated DNA Technologies	N/A
p-miR-122 guide RNA: p-UGGAGUGUGACAAUGGUGUUUG (p=5′ phosphate)	Integrated DNA Technologies	N/A
miR-122 capture RNA: Biotin-mUmCmUmCmGmUmCmUmAmAmCmCmAmUmGmC mCmAmAmCmAmCmUmCmCmAmAmCmUmCmU (m=2′-O-methyl)	Integrated DNA Technologies	N/A
miR-122 competitor DNA: AGAGTTGGAGTGTTGGCATGGTTAGACGAGA	Integrated DNA Technologies	N/A
Recombinant DNA
pmycTNRC6B	(Meister et al., 2005)	Addgene #10978
pFastBac HT A	ThermoFisher	Cat# 10584027
pFastBac His-Flag-TEV TNRC6 Cterm-Strep-Strep	This paper	N/A
pSV272-TNRC6B-ABD	This paper	N/A
pcDNA5 Flag-hAgo2 WT	(De et al., 2013)	N/A
pcDNA5 Flag-hAgo2 K660S	This paper	N/A
pcDNA5 Flag-hAgo2 P590G	This paper	N/A
pcDNA5 Flag-hAgo2 R688S	This paper	N/A
pcDNA5 Flag-hAgo2 K660S;P590G	This paper	N/A
pcDNA5 Flag-hAgo2 K660S;R688S	This paper	N/A
pcDNA5 Flag-hAgo2 P590G;R688S	This paper	N/A
pFastBac His-Flag-TEV hAgo2 3×D;2×A WT	(Schirle and MacRae, 2012)	N/A
pFastBac His-Flag-TEV hAgo2 3×D;2×A P590G	This paper	N/A
pFastBac His-Flag-TEV hAgo2 3×D;2×A R688S	This paper	N/A
pFastBac His-Flag-TEV hAgo2 3×D;2×A P590G;R688S	This paper	N/A
pUC57-mini-let-7-A114	Gift from Yukihide Tomari	N/A
All-A TNRC6-ABD	Synthesized by Genewiz, Inc.	N/A
pcDNA5 FRT/TO GFP-TNRC6B-FL pcDNA5 FRT/TO mCherry Ago2	This paper This paper	N/A N/A
Software and Algorithms
Zen Black	Zeiss	https://www.zeiss.com/microscopy/us/downloads/zen.html#service-packs
Imaris 9	Bitplane, Oxford Instruments	http://www.bitplane.com/learning/imaris-9-launch-july-2017
Prism 6	Graphpad	https://www.graphpad.com/scientific-software/prism/
ImageJ	National Institutes of Health, USA	https://imagej.nih.gov/ij/
Coot	(Emsley et al., 2010)	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
Phenix	(Adams et al., 2010)	http://phenix-online.org
CCP4	(Winn et al., 2011)	http://www.ccp4.ac.uk
PyMOL	Schrödinger, LLC	https://pymol.org/2/
ImageQuant TL	GE Healthcare Life Sciences	https://www.gelifesciences.com/shop/protein-analysis/molecular-imaging-for-proteins/imaging-software/imagequant-tl-8-1-p-00110?current=29000737

Bacterial media and growth conditions

All bacterial cultures were grown in Luria-Bertani (LB) medium. For protein expression, BL21 (DE3) cultures were grown at 37°C to an OD₆₀₀ of 0.8 and protein production was induced with 1 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG), followed by growth either overnight at 18°C (MBP fusion proteins used in pull down experiments) or for 4–5 hours at room temperature (ABD proteins used in phase separation experiments). For cloning, transformed cultures were grown overnight at 37°C, except for the pUC57-mini-let-7-A₁₁₄ construct, which was grown for a maximum of 10 hours at 37°C (to avoid loss of poly-A encoding region). When applicable, media was supplemented with one or more of the following at the indicated concentrations: ampicillin (100 µg/mL), kanamycin (40 µg/mL), tetracycline (5 µg/mL), gentamycin (7 µg/mL), 5-Bromo-4-Chloro-3-Indolyl β-D-Galactopyranoside (X-gal, 20 µg/mL in dimethylformamide), and/or IPTG (1mM).

Insect cell media and growth conditions

Sf9 cells were grown in Lonza Insect XPRESS™ medium supplemented with penicillin (100 units/mL), streptomycin (100 µg/mL), L-glutamine (2.92 mg/mL) at 27 °C in suspension.

Mammalian cell media and growth conditions

HEK 293 FlpIn™ and FlpIn™ TRex™ cells were grown in Gibco Dulbecco’s Modified Eagle Medium (DMEM) with high glucose and sodium pyruvate supplemented with penicillin (100 units/mL), streptomycin (100 µg/mL), L-glutamine (2.92 mg/mL) and 10% fetal bovine serum. Cells were cultured at 37°C with 5% CO₂ in an adherent monolayer. For passage, cells were dissociated with 0.05% Trypsin EDTA. Prior to microscopy, cells were plated on dishes with tissue culture-treated coverslip bottoms.

METHOD DETAILS

Purification of Ago2 (bound to assorted cellular small RNAs)

Ago2 protein was expressed in Sf9 cells using a baculovirus expression system (Schirle et al., 2014). Sf9 cells were infected at 27°C for ~60 hours and then harvested by centrifugation. Cell pellets were resuspended in Lysis Buffer (300mM NaCl, 0.5 mM TCEP, 50 mM Tris, pH 8) and lysed in a single pass through a M-110P lab homogenizer (Microfluidics, Westwood, MA). Lysate was centrifuged to pellet insoluble material and Ago2 was purified from clarified lysate by nickel-affinity purification using agarose Ni-NTA resin (Qiagen). The resin was washed three times with Nickel Wash Buffer (300 mM NaCl, 25 mM imidazole, 0.5 mM TCEP, 50 mM Tris, pH 8). Bulk cellular RNAs were then degraded by on-resin treatment with RNase A. After several additional washes, protein was eluted in Nickel Elution Buffer (300 mM NaCl, 300 mM imidazole, 0.5 mM TCEP, 50 mM Tris, pH 8). The N-terminal FLAG-His₆ tags were removed using TEV protease during dialysis against Hi-Trap Dialysis Buffer (300 mM NaCl, 15 mM imidazole, 0.5 mM TCEP, 50 mM Tris, pH 8). The dialyzed protein was then passed through a Hi-Trap Chelating column (GE Healthcare Life Sciences) and the unbound material was collected and further purified by size exclusion chromatography using a Superdex 200 column (GE Healthcare Life Sciences) equilibrated in Superdex Buffer (100mM NaCl, 0.5mM TCEP, 50mM Tris pH 8). Purified Ago2 was concentrated and buffer was exchanged twice with Crystal Buffer (100mM NaCl, 0.5 mM TCEP, and 10 mM Tris pH 8). Concentrated samples were flash frozen in liquid nitrogen and stored at −80 °C.

Purification of Ago2 (bound to a defined guide RNA)

Protein expression and early purification steps were performed as for samples bound to cellular small RNAs (above). However, Micrococcal Nuclease was used instead of RNase A to degrade bulk co-purifying RNAs. After elution from the Ni-NTA resin, ethylene glycol tetraacetic acid (EGTA), pH 8.0, was added to the eluate to a final concentration of 5 mM to chelate any remaining calcium ions. The synthetic guide RNA (single stranded, 21-nucleotides in length, with a 5'-phosphate) was then added to the nickel purified Ago2, followed by the addition of TEV protease for cleavage of the purification tags and dialysis against Hi-Trap Dialysis Buffer overnight at 4°C. The dialyzed protein was then passed through a Hi-Trap Chelating column (GE Healthcare Life Sciences) and the unbound material was collected. Ago2 molecules loaded with the desired guide RNA were then isolated from those bound to co-purifying cellular small RNAs using a modified Arpón method (Flores-Jasso et al., 2013). Briefly, a biotinylated capture oligonucleotide was immobilized on High Capacity NeutrAvidin resin. Dialyzed Ago2 was supplemented with 0.01% CHAPS and 2mM magnesium acetate before being passed over the capture resin by gravity flow. The resin was then washed with ~200 column volumes Wash A (0.01% CHAPS, 2 mM magnesium acetate, 100 mM potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8), ~350 column volumes Wash B (0.01% CHAPS, 2 mM magnesium acetate, 2M potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8), and ~50 column volumes Wash C (0.01% CHAPS, 2 mM magnesium acetate, 1M potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8). Captured protein was eluted in Wash C using a biotinylated competitor DNA complementary to the capture oligonucleotide. Excess competitor DNA was removed by incubation with High Capacity NeutrAvidin followed by anion exchange using Q-Sepharose Fast Flow resin. Loaded Ago2 proteins were further purified by size exclusion chromatography using a Superdex 75 16/60 column (GE Healthcare Life Sciences) equilibrated in 1 M NaCl, 50 mM Tris pH 8, 0.5 mM TCEP. Purified Ago2 samples were dialyzed against either Crystal Buffer (for crystallography) or 50 mM HEPES (pH 7.5), 0.1 M potassium acetate, and 0.5 mM TCEP (for biochemical assays). Samples were concentrated, flash frozen and stored at −80 °C.

Crystallization of Ago2 bound to Tryptophan

Ago2-guide-target complexes were formed by mixing Ago2-guide complexes with target RNAs at a 1:1.2 molar ratio. Crystals were grown by hanging drop vapor diffusion at 20°C. Drops contained a 1:1 ratio of Ago2-guide-target (1.5 mg/ml) to reservoir solution (14% PEG 3350, 0.1 M Tris pH 8.2, 12% Isopropanol, 10 mM MgCl₂, L-tryptophan to saturation). Crystals were harvested with nylon loops, moved to a cryoprotectant containing reservoir solution supplemented with 25% ethylene glycol, and flash frozen in liquid N₂. Data were collected under cryogenic conditions remotely at beam line 12-2 at the Stanford Synchrotron Radiation Lightsource (SSRL). Data were processed using XDS and Scala (Gonzalez, 2010; Kabsch, 2010; Winn et al., 2011).

Structure Refinement

The Ago2-guide-target structure (PDB ID 4W5O) served as a starting model for refinement. The model was built using Coot (Emsley et al., 2010) and subjected to XYZ coordinate and B-factor refinement using PHENIX (Adams et al., 2010). Tryptophan molecules were identified by manual inspection of F_obs-F_calc difference maps. Model building and refinement continued iteratively until all interpretable electron density was modeled. The R_free set was identical to that used in refinement of the original 4W5O structure. Structure figures were generated with PyMOL (Schrödinger, LLC).

Northern Blotting

Mutant Ago2 proteins were overexpressed in HEK 293 Flp-In cells by transient transfection with Lipofectamine 2000 (Invitrogen). Cells were grown for 48 hours after transfection and then lysed in Triton-X Lysis Buffer (50 mM Tris pH 8, 150 mM NaCl, 0.5% Triton-X, 0.5 mM TCEP) using a glass dounce homogenizer with a low clearance pestle. Insoluble material was pelleted by centrifugation and the supernatant bound to Anti-FLAG M2 Affinity Gel (Sigma) for 1.5 hours at 4°C. Resin was washed five times with FLAG Wash Buffer (50mM Tris pH 8, 150 mM NaCl, 0.5 mM TCEP) and then protein and bound RNAs eluted in Formamide Loading Buffer (95% Formamide, 0.025% Bromophenol Blue, 0.025% Xylene Cyanol, 5mM EDTA). Bound RNAs were resolved by 15% denaturing poly-acrylamide gel electrophoresis. RNA was transferred to Hybond nylon membrane (GE Lifesciences) using a Trans-Blot SD semi-dry transfer apparatus (BioRad) and chemically cross-linked using 1-Ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride (EDC). Membranes were incubated with a radiolabeled LNA probe (Exiqon) against miR-16, washed, and microRNA levels visualized by phosphorimaging (Typhoon FLA 7000, GE Lifesciences).

Construction of TNRC6B-ABD constructs

Regions of TNRC6B cDNA encoding amino acids 437-1056 (obtained from pmyc-TNRC6B construct) were inserted into the pSV272 (His₆-MBP fusion protein encoding vector) multiple cloning site. DNA encoding the ABD all-Ala mutant, in which all 36 tryptophans were mutated to alanine, was synthesized by Genewiz, Inc. and similarly inserted into psV272. Specific tryptophan codons were added back to the AllA-ABD construct using the Quickchange mutagenesis method (Stratagene).

Expression and Purification of MBP-TNRC6B ABD fusion

MBP-TNRC6B ABD fusion proteins were expressed in E. coli strain BL21(DE3). Cultures were grown to an OD₆₀₀ of 0.8 and protein production was induced with 1 mM IPTG, followed by growth either overnight at 18°C (MBP fusion proteins used in pull down experiments) or for 4–5 hours at room temperature (ABD proteins used in phase separation experiments). Cells were collected by centrifugation, resuspended in Nickel Lysis Buffer (300 mM NaCl, 0.5 mM TCEP, 1 mM PMSF, 50 mM Tris, pH 8), and lysed by one passage through a Microfluidizer processor (Microfluidics). Lysate was clarified by high-speed centrifugation and then applied to Ni-NTA resin for 1 hour at 4°C. Resin was washed three times with Nickel Wash Buffer (300 mM NaCl, 25 mM imidazole, 0.5 mM TCEP, 1 mM PMSF, 50 mM Tris, pH 8), and protein eluted in Nickel Elution Buffer (300 mM NaCl, 300 mM imidazole, 0.5 mM TCEP, 50 mM Tris, pH 8). Excess imidazole was removed by overnight dialysis into Dialysis Buffer (300 mM NaCl, 15 mM imidazole, 0.5 mM TCEP, 0.5 mM PMSF, 50 mM Tris, pH 8). MBP-fusion proteins (used in pull down experiments) were then concentrated, glycerol was added to 20% (v/v), and the resulting samples were flash frozen in liquid N₂ and stored at −80°C.

Expression and Purification of TNRC6B-ABD

TNRC6B-ABD protein used in phase separation experiments was also expressed as an MBP fusion protein and underwent initial purification as above, but following elution from nickel resin, the MBP tag was removed by Tobacco Etch Virus (TEV) Protease cleavage overnight concurrent with dialysis. Protein was then passed through a Hi-Trap Chelating Column (GE Healthcare Life Sciences). The flow through solution, containing the ABD, was collected and concentrated to 1–2 ml. Concentrated samples were then heated to 65°C for 10 minutes to denature globular proteins and precipitated material was removed by centrifugation. The soluble fraction was then applied to a Superdex 200 16/60 gel filtration column (GE Healthcare Life Sciences), equilibrated in 0.1 M NaCl, 1 mM PMSF, 0.5 mM TCEP, 50 mM Tris, pH 8, and fractions containing full-length protein were pooled and concentrated to ~10 mg/ml (or until cloudy). Final samples were flash frozen in liquid N₂ and multiple freeze-thaw cycles avoided.

Expression and Purification of Full-Length TNRC6B

Full-length TNRC6B from the pDEST-myc TNRC6B construct was cloned into the pFastBac HTA backbone alongside an N-terminal His-FLAG tag and a C-terminal Strep-tag using the NEB HiFi DNA Assembly Cloning Kit (New England BioLabs). The vector was used to produce baculovirus using the Bac-to-Bac system (Thermo Fisher Scientific). It was observed that production of full-length TNRC6B decreased upon multiple amplification cycles of the virus, and therefore, early passages of virus were used for protein expression. Sf9 cells were infected for ~60 hours and harvested by centrifugation. Cells were resuspended in Detergent Lysis Buffer (300mM NaCl, 0.5% Triton-X, 5% glycerol, 0.5 mM TCEP, protease inhibitors, 50 mM NaH₂PO₄ pH 8) and homogenized using a glass dounce homogenizer with a low clearance pestle. Insoluble material was removed by centrifugation and soluble lysate applied to Ni-NTA for 30 minutes at 4°C. Resin was then washed three times with High Salt Nickel Wash Buffer supplemented with CHAPS detergent (1 M NaCl, 25 mM imidazole, 0.5 mM TCEP, 0.05% CHAPS, 50 mM Tris, pH 8) and eluted in High Salt Nickel Elution Buffer supplemented with CHAPS detergent (50 mM Tris pH 8, 1 M NaCl, 300 mM imidazole, 0.5 mM TCEP, 0.1% CHAPS, protease inhibitors). The eluate was then applied to 0.5 mL packed StrepTactin resin (Qiagen) and incubated for 1.5 hours at 4°C. Full-length protein was eluted using 5 mM desthiobiotin in StrepTactin Elution Buffer (50 mM Tris pH 8, 1 M NaCl, 0.5 mM TCEP, 0.1% CHAPS).

TNRC6B and TNRC6B-ABD Labeling

Samples of both TNRC6B and TNRC6B-ABD that were used for labeling were eluted and stored as above, but at pH 7.4 (to avoid decomposition of the maleimide). Prior to labeling, purified TNRC6B-ABD was incubated with a 10-fold molar excess (over protein concentration) of fresh TCEP for 5 minutes. Samples were then incubated with 2-fold molar excess AlexaFluor 488 Maleimide in for 2 hours at room temperature in the dark. Excess maleimide was quenched using excess β-mercaptoethanol followed by passage through to two sequential illustra MicroSpin G-25 columns (GE Healthcare Life Sciences) equilibrated in 0.1 M NaCl, 0.5 mM TCEP, 50 mM Tris, pH 8. Labeled protein was flash frozen in liquid N₂ and stored at −80°C.

TNRC6B-FL was labeled similarly to the ABD, but excess maleimide was removed by overnight dialysis (because full length TNRC6B bound the illustra MicroSpin G-25 column).

HaloTag Ago2 Labeling

DNA encoding a HaloTag (Promega Corporation) was inserted after the TEV protease cleavage site in pFastBac His-Flag-TEV hAgo2 3×D;2×A WT (Schirle and MacRae, 2012), such that cleavage of the expressed protein with TEV protease removes N-terminal the FLAG-His₆ tag and leaves the Halo-Ago2 fusion intact. Baculovirus were generated using the Bac-to-Bac system (Thermo Fisher Scientific). Halo-Ago2 protein was expressed in Sf9 cells and purified as described above. Purified Halo-Ago2 was labeled with the HaloTag TMR ligand, following the manufacturer’s protocol (Promega).

ABD Pull-down Assays

MBP-TNRC6B ABD fusion proteins were expressed in E. coli strain BL21(DE3), and purified as described above. FLAG-tagged Ago2 was expressed in Sf9 cells and purified by FLAG-IP: 50 ml cell pellets (~100 ×10⁶ infected cells) were resuspensed in Detergent Lysis Buffer (50mM NaH₂PO ₄ pH8, 300mM NaCl, 5% glycerol, 0.5% Triton-X, 0.5mM TCEP) and lysed by 15 strokes with a dounce homogenizer. Cleared lysate was applied to 100µL Anti-FLAG M2 Affinity Gel (Sigma) and bound for 1.5 hours at 4°C. Resin was washed three times with Flag Wash Buffer (0.15 M NaCl, 2 mM MgCl₂, 0.5 mM TCEP, 20 mM Tris, pH 8.0) then treated with RNase A for 1 hour at 4°C. Resin was then washed three times with Flag Wash Buffer, three times with High Salt Wash Buffer (50mM Tris pH8, 1M NaCl, 0.5 mM TCEP), and three times again with Flag Wash Buffer. Resin was dried and resuspended in 1mL Flag Wash Buffer. Relative quantities of immobilized Flag-Ago2 were determined by SDS-PAGE (data not shown).

Equal inputs of immobilized Flag-Ago2 were incubated with MBP-TNRC6B ABD at an approximately 1:2 Ago2:ABD molar ratio for 1 hour at 4°C. Resin was then washed three times with Detergent Lysis Buffer followed by two washes with Flag Wash Buffer. Beads were dried and bound proteins eluted by boiling in SDS Sample Buffer (2% SDS, 10 mM EDTA, 10% glycerol, 0.1% bromophenol blue, 50 mM Tris, pH 7.4). Pull-downs were replicated at least twice for each experimental condition. Gels shown in-text were all performed on the same day.

Generation of GFP-TNRC6B-FL HEK 293 inducible stable cell line and GW-body formation

Inducible, stably-expressing GFP-TNRC6B-FL HEK 293 cells were made using the Flp-In expression system (Invitrogen). Briefly, the coding sequence for GFP and TNRC6B-FL was cloned into pcDNA5/FRT/TO using the New England Biolabs HiFi DNA Assembly kit to generate a fusion protein construct. This construct was co-transfected with pOG44 into Flp-In TREx cells (Invitrogen) using Lipofectamine 2000. Recombined transformants were selected by treatment with 200 µg/ml hygromycin B over the course of 3–4 weeks. Cells were grown in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum, 100 unit/ml penicillin, 100 µg/ml streptomycin and 2 mM glutamine in a humidified 37°C incubator with a 5% CO₂ atmosphere.

Expression of GFP-TNRC6B-FL was induced by addition of 1 µg/ml tetracycline. Cells were visualized 24 hours post-induction. To visualize Ago2, the coding sequence for mCherry and Ago2 was cloned into pcDNA5/FRT/TO using the NEB HiFi DNA Assembly kit to generate a fusion protein construct. This construct was transfected into the GFP-TNRC6B-FL cell lines using Lipofectamine 2000 concurrent with tetracycline induction. Transfected cells were visualized 24 hours post-transfection.

Microscopy

Initial observations and characterization of ABD-Ago2 droplets were carried out on an Axiovert 40 CFL light microscope (Zeiss) fit with a Power Shot G9 camera (eBay; free carrying case included). Confocal images and FRAP experiments were conducted on a using a Zeiss LSM 780 laser scanning confocal microscope, equipped with a 63× oil immersion objective (Plan-Apochromat 63×/1.4 Oil DIC M27) using the Zen Black software to collect and process data. Excitation/emission wavelengths were 488 nm/503-549 nm for Alexa Fluor 488-labeled TNRC6B samples, and 561 nm/602-632 nm for TMR-labeled Ago2 samples. Cellular microscopy was conducted using the same microscope fitted with a LiveCell stagetop incubation system (Pathology Devices, Inc.). Experiments were performed at 37°C, 5% CO₂, and 40% humidity. Excitation/emission wavelengths were 488 nm/503–549 nm for GFP-TNRC6B visualization and 561 nm/602–632 nm for mCherry-labeled Ago2. Videos were generated using Imaris software (Bitplane).

Fluorescence Recovery after Photobleaching

Whole ABD-Ago2, TNRC6B-Ago2 droplets, and GW-bodies were photobleached with 488 nm and/or 561 nm laser lines. Time-lapse images were acquired at excitation wavelengths 488 nm and 561 nm. Images were processed and fluorescence intensities measured using Zen Black (Zeiss). For in vitro recombinant FRAP, fluorescence of a neighboring unbleached droplet was used as a reference to correct for uneven illumination after subtraction of background intensity. All in vitro FRAP experiments were conducted ≤3 hours after droplet preparation. Percent recovery and half lives did not notably change over this time period. A reference could not be used for GW-body FRAP due to movement of the bodies over time. At each time point, the corrected fluorescence intensity of the recovering droplet was divided by its corrected original (pre-bleached) fluorescence intensity. The normalized intensities were plotted as a function of time and fit to a single exponential growth curve to determine the half time and the percentage of bleached fluorescence recovery using GraphPad Prism 6 (GraphPad Software). Replicates were reported for three individual bodies of roughly equivalent age (for recombinant ABD and TNRC6B-Ago2 droplets), all collected on the same day. For GW-bodies, four bodies selected from four different cells, collected on the same day, were reported.

Cell Lysate Preparation

Cell lysate for sequestration and deadenylation assays was obtained from HEK 293 Flp-In cells (Invitrogen). Briefly, cells were scraped into ice-cold PBS and washed twice by centrifugation and resuspension in PBS. Washed cell pellets were incubated with 2× volume Hypotonic Lysis Buffer A (50 mM KCl, 0.5 mM TCEP, protease inhibitor tablet (Roche), 25 mM HEPES, pH 7.5) for 20 minutes on ice. Cells were then homogenized with ~30 strokes of a glass dounce homogenizer with a low clearance pestle. Lysate was re-salted with addition of 1/9 volume Buffer B (25 mM HEPES pH 7.5, 1 M potassium acetate, 0.5 mM TCEP, protease inhibitor tablet). Cell debris was pelleted and the concentration of the supernatant solution was determined by measuring OD₂₆₀. Lysate was flash-frozen and stored at −80°C.

Target RNA Generation

Target RNAs were transcribed by T7 RNA polymerase using puc57 mini-let-7-A₁₁₄, generously provided by the Tomari lab, as a template. A₁₁₄ transcripts were generated from plasmid digested using NsiI, while A₁₁₄-N₄₀ transcripts with an internalized poly-A tail were transcribed from plasmid digested with HindIII, as described previously (Fukaya and Tomari, 2012). Transcripts (subsequently referred to as 8xlet-7-A₁₁₄ or 8xlet-7-A₁₁₄-N₄₀ target RNAs, respectively) were 5' cap-labeled with α³²P-GTP using a Vaccinia Capping System (New England BioLabs) and gel-purified for sequestration and deadenylation experiments.

Recombinant Protein Sequestration Experiments (in the absence of lysate)

Sequestration experiments in the absence of lysate were performed using partially-purified TNRC6B-FL (Nickel purification only, see Figure S5), although experiments with purified samples behaved analogously (data not shown). 200 nM human Ago2 loaded with either miR-122 or let-7a guide RNA was incubated with tRNA (0.02mg/mL final) and approximately ~1 nM 8xlet-7 cap-labeled target RNA. Phase separation was then initiated by addition of TNRC6B-FL (300nM final concentration). Samples without TNRC6-FL added were supplemented with an equal volume of High Salt Nickel Elution Buffer to maintain equal concentrations of salt. Final concentrations of buffer components in reactions were approximately as follows: 0.01% CHAPS, 0.5mM Desthiobiotin, 42.5 mM HEPES pH 7.5, 30mM Imidazole pH 8, ~30mM NaCl, 85mM Potassium Acetate, 5mM 0.5mM TCEP, Tris pH 8. Phase separation was allowed to proceed for 15 minutes at room temperature. The samples were then centrifuged at 14,000×g for 2 minutes and the supernatant carefully removed. Protein in both the supernatant and pellet fractions was then removed by Proteinase K digestion in Proteinase K Buffer (2mg/mL Proteinase K, 200 mM Tris pH 8, 25 mM EDTA, pH 8, 300mM NaCl, 2% w/v sodium dodecyl sulfate, 20µg glycogen per reaction) for 30 minutes at 65°C followed by phenol-chloroform extraction and RNA precipitation. The resulting samples were resolved by 8% denaturing polyacrylamide gel electrophoresis (PAGE).

Sequestration of protein components was assessed using identical concentrations of protein components as those used for target RNA experiments, but without target RNA and tRNA. Phase separation was again allowed to proceed for 15 minutes at room temperature followed by centrifugation and separation of supernatant and pellet. Sample Buffer was added to the supernatant and the pellet was resuspended in 1× SDS Sample Buffer. Fractions were resolved by SDS-PAGE and detected by staining with Denville Blue™ Protein Stain (Denville).

miRISC Component Sequestration Experiments (in presence of lysate)

Sequestration of miRISC components from HEK 293 lysate, prepared as indicated above, were performed using fully-purified TNRC6B-FL samples, although experiments with partially-purified samples yielded similar results (data not shown). 40 nM human Ago2 (RNase A treated) was incubated with cell lysate (final OD₂₆₀ ~3.5) and phase separation was initiated with the addition of ~200 nM TNRC6B-FL. The final concentration of buffer components in the reaction were approximately as follows: 45mM HEPES pH 7.5, 100mM NaCl, 90mM Potassium Acetate, 0.5mM TCEP, 5mM Tris pH8. Phase separation was allowed to proceed for 15 minutes at room temperature. Samples were then spun and the supernatant solution was carefully removed. SDS Sample Buffer was added to the supernatant and the pellets were resuspended in 1× SDS Sample Buffer. Proteins were resolved by SDS-PAGE using a 4–20% precast gel (BioRad) and then transferred to nitrocellulose using a BioRad TransBlot Turbo system (BioRad). Membranes were blocked in 5% w/v milk in TBST (50mM Tris pH8, 150mM NaCl, 0.1% TWEEN) and then incubated with antibodies at 1:1000 dilution in 5% milk overnight at 4°C. Membranes were washed three times in TBST, then incubated with 1:10,000 secondary antibody dilution in 5% milk. Membranes were again washed three times in TBST and then incubated with Amersham ECL Prime reagent (GE Lifesciences). Protein bands were detected by film. Membranes were stripped using Restore™ Western Blot Stripping Buffer (Thermofisher) for sequential detections. Similar results as reported were obtained from replicates using different protein and lysate preparations.

Deadenylation Experiments (phase separation induced by addition of exogenous TNRC6B and reducing concentration of NaCl)

40 nM Ago2 loaded with let-7a (or miR-122, in case of negative control) was incubated with ~3 nM cap-labeled 8xlet-7 target RNA with 0.02 mg/mL tRNA (Sigma), 0.5 µL Murine RNase Inhibitor (New England Biolabs), in the presence of HEK cell lysate, prepared as described above, at a final OD₂₆₀ of ~3. Phase separation was initiated by addition of ~300 nM TNRC6B-FL (partially purified). The final concentration of buffer components in the reaction were approximately as follows: 1.5mM EDTA, 48mM HEPES pH 7.5, 12mM Imidazole pH 8, 12mM NaCl, 96mM Potassium Acetate, 0.5mM TCEP, 2mM Tris pH8. Phase separation was allowed to proceed for 10 minutes. Deadenylation was initiated by addition of MgCl₂ to a final concentration of 3 mM. Time points were taken by spinning the lysate for 2 minutes at 14,000 × g and careful removal of the supernatant. Both pellet and supernatant reactions were quenched in Proteinase K Buffer and incubation at 65°C for 30 minutes. RNA was extracted by phenol-chloroform followed by ethanol precipitation. RNA pellets were resuspended in Formamide Loading Buffer and resolved by 8% denaturing PAGE. RNAs were visualized by phosphorimaging. Due to variances in the partial purification of TNRC6B-FL, three technical replicates, wherein the same protein purification (with the same number of freeze/thaw cycles) and RNA and lysate preparation were used but the experiment was performed on separate days, were quantified and reported in text. However, comparable results were obtained from biological replicates using different protein, RNA, and lysate preparations.

Deadenylation Experiments (phase separation induced by PEG)

20 nM Ago2 (final) loaded with let-7a was mixed with cap-labeled 8xlet-7 target RNA, HEK cell lysate (as prepared above, final OD₂₆₀ ~1.5), and ~20 nM TNRC6B-FL in CHAPS Reaction Buffer (final concentration of buffer components approximately 0.076% CHAPS, 0.5mM Desthiobiotin, 1.3 mM EDTA, 2.5mM HEPES pH 7.5, 180mM NaCl, 15mM Potassium Acetate, 0.5 mM TCEP, 40mM Tris pH 8) supplemented with 0.02 mg/mL tRNA (Sigma) and 0.5 µL Murine RNase Inhibitor (New England Biolabs). Including CHAPS was important for inhibiting phase separation of purified TNRC6B in the reaction buffer prior to addition of PEG (below). Samples were incubated for 2 minutes and then pre-spun to remove any non-specific precipitate at 14,000×g for 2 minutes. The supernatant was removed for subsequent reactions. Phase separation was induced by addition of an equal volume of PEG Induction Buffer (10% PEG 8000, 0.1% CHAPS, 2 mM EDTA, 200 mM NaCl, 0.5mM TCEP, 50 mM Tris pH 8) or, in the case of non-phase separated samples, supplemented with an equal volume of CHAPS Reaction Buffer (0.1% CHAPS, 2 mM EDTA, 200 mM NaCl, 0.5mM TCEP, 50 mM Tris pH 8) and allowed to proceed for 10 minutes. The final buffer component concentrations were approximately as follows, ± PEG8000 at a final concentration of 5%: 0.25mM Desthiobiotin, 0.09% CHAPS, 1.65mM EDTA, 1.25mM HEPES pH 7.5, 195mM NaCl, 7.5mM Potassium Acetate, 0.5mM TCEP, 45mM Tris pH 8. Deadenylation was initiated by addition of MgCl₂ to a final concentration of 3 mM. Time points were taken by spinning the lysate for 2 minutes at 14,000 × g and careful removal of the supernatant. Both pellet and supernatant reactions were quenched in Proteinase K Buffer and incubation at 65°C for 30 minutes. RNA was extracted by phenol-chloroform followed by ethanol precipitation. RNA pellets were resuspended in Formamide Loading Buffer and resolved by 8% denaturing PAGE. RNAs were visualized by phosphorimaging. The reported rates were obtained from three technical replicates using the same buffer conditions, protein purifications, and RNA and lysate preparations but performed on separate days. However, acceleration of deadenylation in the presence of PEG-induced phase separation was observed across multiple biological replicates using slightly different buffer conditions, protein purifications, and RNA and lysate preparations.

QUANTIFICATION AND STATISTICAL ANALYSIS

Microscopy

For in vitro recombinant droplet imaging, replicates were obtained from different droplets of roughly the same age, all collected on the same day for each experimental condition (i.e. ABD vs. TNRC6B-FL droplets). N=3 for all in vitro reported FRAP data, where N is the number of droplets. For in cell GW-body imaging, replicates were obtained from different cells from the same dish, all collected on the same day for each experimental condition. N=4 for the GW-body FRAP data, where N is the number of GW-bodies, each from distinct cells. Images were processed and fluorescence intensities measured using Zen Black (Zeiss). Quantified data, mobile fractions, and half-lives were determined by plotting recoveries and fitting to a single exponential growth curve using Prism (Graphpad).

Deadenylation experiments

Deadenylation was quantified by determining the amount of intact 8xlet-7-A₁₁₄ RNA and shortened RNA (8xlet-7-A_n) present in each time point in order to determine the “fraction A₁₁₄ remaining.” This approach was chosen in order to include all deadenylated intermediates in rate determination. RNA was quantified by phosphorimaging using ImageQuant (GE Healthcare Life Sciences). N=3 for each deadenylation experiment, where N represents number of experimental replications. Rates were determined by plotting fraction intact remaining and fitting to a single exponential decay using Prism (Graphpad).

Supplementary Material

1. Movie S1. Related to Figure 3. Growth and fusion of ABD-Ago2 droplets.

High concentrations of TNRC6B–ABD (15 µM) and Ago2 (30 µM) were mixed and droplet fusion was visualized by fluorescence of Alexa488 labeled ABD. Movie plays at 10x real time (indicated in minutes and seconds displayed at bottom right).

Figure_S_3. Figure S3. Related to Figure 3. Ago2-ABD Droplets.

(A) Reversibility of ABD-Ago2 droplets by temperature. ABD (40 µM) and Ago2 (20 µM) were mixed and turbidity (as a read-out for droplet formation) was assessed by absorbance of 480 nm light as temperature was varied. Turbidity was high room temperature, decreased with increasing temperature, and then increased again upon re-cooling the same sample. (B) Representative fluorescence microscopy images of Ago2-ABD droplets (containing 10% Alexa Flour 488 labeled TNRC6B–ABD) generated using wild-type Ago2 or trp-binding pocket mutants. Ago2 and ABD protein concentrations (equimolar) indicated on the left. Scale bar, 2 µm. (C) Histograms representing the number of Ago-ABD droplets of varying radii settled in a fixed area (14000 µm2) after 20 minutes. Droplets with radii < 0.1 µm were not counted.

Figure_S_4. Figure S4. Related to Figure 4. Controls for GFP-TNRC6B in living cells.

(A) Confocal images of HEK 293 cells transfected with plasmids encoding either GFP, which distributes relatively evenly throughout the cell (left), or the GFP-TNRC6B fusion, with concentrates in cytoplasmic foci (right). (B) Anti-GFP Western blot of GFP-TNRC6B stable cells (plus and minus tetracycline induction) shows the full length GFP-TNRC6B fusion protein is intact in the living cells. Asterisk indicates a cross reacting band that serves as a loading control.

Figure_S_5. Figure S5. Related to Figure 5. Purification and fractionation of full length TNRC6B.

(A) Schematic of purification strategy. A combination of N-terminal (His6) and C-terminal (Strep-tag) tags were used to purify full length TNRC6B from degradation products and other protein contaminants. (B) Coomassie stained SDS polyacrylamide gel of TNRC6B samples at various stages of purification. (C) Fractionation of miRISC samples by centrifugation. (left) input samples of purified Ago2 and TNRC6B. (middle) supernatant solutions after centrifugation of indicated samples. (right) pelleted material after centrifugation of indicated samples.

Figure_S_6. Figure S6. Related to Figures 6 and 7. Target shortening is dependent on presence of poly(A) at 3’ end.

To verify that shortening of 8xlet7-A₁₁₄ transcript was mediated by a poly(A)-specific 3’–5’ exonuclease, phase separation/deadenylation experiments were performed using a target with an internalized poly(A) tail (8xlet7-A₁₁₄-N₄₀), is refractory to deadenylation (Fukaya and Tomari 2011). Denaturing gel of RNAs found in supernatant and pellet fractions shown for samples incubated in the presence and absence of Mg²⁺ (necessary for deadenylation) and/or 5% PEG 8000 (to induce miRISC phase separation). Bands corresponding to 8xlet7-A₁₁₄-N₄₀, 8xlet7-A₁₁₄, and 8xlet7-A₀ RNAs are indicated

Figure_S_7. Figure S7. Related to Figure 7. Analysis of PEG-induced miRISC droplets.

(A) PEG-induced TNRC6B droplets recruit Ago2. Representative fluorescence microscopy images of TNRC6B droplets formed in the absence and presence of Ago2. Scale bar, 5 µM. (B) Representative fluorescence microscopy images from miRISC FRAP experiments. Scale bar, 2 µm. (C) FRAP recovery curves for three miRISC droplets with error bars indicating SEM. Ago2 fluorescence recovery approached 54% with a t_1/2 of 200 sec. TNRC6B fluorescence recovery approached 10% with a t_1/2 of 300 sec. (D) PEG-induced droplets can be recovered by centrifugation. (left panel) input Ago2 and TNRC6B samples. (middle) supernatant solutions after centrifugation without 5% PEG 8K (+PEG supernatant not shown because PEG prevented proteins from entering the gel). (right) pellet fractions +/− 5% PEG 8K.

2. Movie S2. Related to Figure 3. Reversibility of ABD-Ago2 droplets by introduction of NaCl.

1 µl of 1 M NaCl was added to a 10 µl volume containing ABD-Ago2 droplets (visualized by Alexa488 labeled ABD). Droplets dissolve as they are apparently struck by a wave of high [NaCl] from the top right corner of the image. We do not know the concentration of NaCl at the wave front but, assuming complete mixing, the final salt concentration in the reaction was 90 mM KOAc, 160 mM NaCl. Time in minutes and seconds displayed at bottom right.

3. Movie S3. Related to Figure 4. Fusion of adjacent GFP-TNRC6b–FL foci in live HEK293 cells.

A single HEK293 cell is shown. White arrow indicates region near fusing droplets. Time (in minutes and seconds) displayed at bottom right.

4. Movie S4. Related to Figure 4. Fusion and non-fusion of adjacent GFP-TNRC6b–FL foci in live HEK293 cells.

“Fusing” arrow points to two droplets that immediately fuse upon contact. “Non-fusing” arrows point two clusters of droplets that appear to be physically tethered but do not immediately fuse (note some droplets move out of focus in some images). Time (in minutes and seconds) displayed at bottom right.

5. Movie S5. Related to Figure 4. Fluorescence recovery after photobleaching of GFP-TNRC6b–FL foci in live HEK293 cells.

Arrow points to GWB prior to photo-bleach. Time post-bleach (in minutes and seconds) displayed at bottom right.

6. Movie S6. Related to Figure 4. Fluorescence recovery after photobleaching of GFP-TNRC6b–FL and mCherry-Ago2 harboring foci in live HEK293 cells.

Arrow points to GWB prior to photo-bleach. Time post-bleach (in minutes and seconds) displayed at bottom right.

Figure_S_1. Figure S1. Related to Figure 1. Trp spacing in GW-rich domains.

(A) Schematic of TNRC6B primary sequence. Vertical lines indicate positions of trp residues. Ago-binding motif II, motif II and the Ago-hook are indicated. Black horizontal line indicates region illustrated in panel (B). (B) Regularly spaced trp residues within a GW-rich region in TNRC6B. Positions of trp residues in the primary sequence indicated as peaks. The mean gap length (in number of amino acid residues) between trp residues (± standard deviation), indicated on right. Positions of other residues of similar abundance in the same region are shown below for comparison. (B) Regular spacing of trp residues within GW-rich regions of diverse GW proteins.

Figure_S_2. Figure S2. Related to Figure 2. Quality control of Ago2 mutants.

(A) Trp-binding pocket mutants bind miRNAs. FLAG-tagged Ago2 mutants were ectopically expressed in HEK 293 cells. Material isolated by FLAG-IP was subject to Northern blotting and probed for miR-16 (endogenously expressed in HEK 293 cells). All mutants retained the ability to bind miR-16 at levels near wild-type Ago2. (B) Trp-binding pocket mutants behave as wild-type Ago2 on size exclusion chromatography. Prior to phase separation experiments Ago2 mutants were purified by size exclusion chromatography. Elution profiles were indistinguishable from wild-type Ago2, indicating the mutants were not globally misfolded or aggregated. (*) The R688S mutant was purified on a different size exclusion column, giving rise to a retention volume slightly different from the other preparations. We were unable to identify an Ago2 triple mutant, in which all three trp-binding pockets were disabled, that met these criteria.

HIGHLIGHTS.

• Ago2 has three pockets that bind Trp residues in the miRISC scaffold protein TNRC6B

• Ago2-TNRC6B interactions promote phase separation of miRISC

• miRISC droplets sequester miRNA target RNAs and recruit deadenylases

• miRISC condensation correlates with accelerated target deadenylation