Dicer-dependent and -independent Argonaute2 Protein Interaction Networks in Mammalian Cells^*

Lysate Preparation and Immunoprecipitation

SILAC labeled cells were lysed separately in Lysis Buffer (150 mm KCl, 25 mm Tris-HCl, pH 7.5, 2 mm EDTA, 1 mm NaF, 5% glycerol, 0,5% Nonidet P-40, 0,5 mm dithiothreitol, 1× Complete Protease Inhibitor mixture (Roche). For immunoprecipitation 3–4 mg total lysate protein were incubated with 50 μl M2 FLAG agarose beads (Sigma) for 4 h at 4 °C with rotation. For RNA dependence samples 100 μg/ml RNase A was added to the lysate for the last 20 min of the incubation. Beads were washed three times with IP Wash Buffer (300 mm KCl, 50 mm Tris-HCl pH 7.5, 5 mm MgCl₂, 0.1% Nonidet P-40, 5% glycerol) and twice with Elution Buffer (150 mm NaCl, 25 mm Tris-HCl pH 7.5, 5 mm MgCl₂, 5% glycerol). Corresponding beads for the SILAC experiments were combined directly after washing and bound proteins were eluted with 500 μg/ml 3x FLAG peptide (N-MDYKDHDGDYKDHDIDYKDDDDK-C) in Elution Buffer for 90 min at 4 °C with 800 rpm.

Protein Digestion

Eluates were separated by one dimensional gel electrophoresis on 4–12% NuPAGE gels (Invitrogen, Carlsbad, CA) and visualized by staining with the NOVEX Colloidal Blue Stain Kit (Invitrogen). Lanes were cut into 8 slices and proteins were in-gel digested with trypsin (Promega, Madison, WI) (40) using iodacetamide as alkylation reagent. Peptides were concentrated and desalted using C₁₈ StageTips (41, 42).

LC-MS/MS Analysis

Peptides were separated on line to the mass spectrometer by using an easy nano-LC system (Proxeon Biosystems, Dreieich, Germany). Four microliter samples were loaded with a constant flow of 700 nl/min onto a 15-cm fused silica emitter with an inner diameter of 75 μm (Proxeon Biosystems) packed in house with RP ReproSil-Pur C18-AQ 3 μm resin (Dr. Maisch). Peptides were eluted with a segmented gradient of 5–60% solvent B over 105 min with a constant flow of 250 nl/min. The nano-LC system was coupled to a mass spectrometer (LTQ-Orbitrap; Thermo Fisher Scientific) via a nanoscale LC interface (Proxeon Biosystems). The spray voltage was set between 2.0 and 2.2 kV, and the temperature of the heated capillary was set to 200 °C.

Survey full-scan MS spectra (m/z = 300–2000) were acquired in the Orbitrap with a resolution of 60,000 at the theoretical m/z = 400 after accumulation of 1,000,000 ions in the Orbitrap. The five most intense ions from the preview survey scan delivered by the Orbitrap were sequenced by collision induced dissociation (collision energy 35%) in the LTQ after accumulation of 5000 ions concurrently to full scan acquisition in the Orbitrap. Maximal filling times were 1000 ms for the full scans and 150 ms for the MS/MS. Precursor ion charge state screening was enabled, and all unassigned charge states as well as singly charged peptides were rejected. The dynamic exclusion list was restricted to a maximum of 500 entries with a maximum retention period of 90 s and a relative mass window of 10 ppm. Orbitrap measurements were performed with the lock mass option enabled for survey scans to improve mass accuracy (polydimethylcyclosiloxane (PCM) ions at m/z 445.120025) (43).

The raw files were processed using the MaxQuant computational proteomics platform (44) version 1.1.1.27. The fragmentation spectra were searched against the IPI mouse database (version 3.68, 56,729 entries) supplemented with frequently observed contaminants, using the Andromeda search engine (45) with the initial precursor and fragment mass tolerances set to 7 ppm and 0.5Da, respectively, and with up to two missed cleavages allowed. Trypsin allowing for cleavage N-terminal to proline was chosen as enzyme specificity. Carbamidomethylated cysteins were set as fixed, oxidation of methionine, and N-terminal acetylation as variable modifications. Maximum false discovery rates (FDRs)—calculated by employing a reverse database strategy—were set to 0.01 both on peptide and protein levels and thus not dependent on the peptide score. Minimum required peptide length was six amino acids. Corresponding forward and reverse experiments were analyzed together and specified as “forward” and “reverse” in the experimental design.

Raw MS data, unfiltered protein groups tables and peptides tables can be downloaded from https://proteomecommons.org/tranche using the following HASH key:

a6LrT+dbaF4jmWZcpYPJuvIzNTFmI0VhkHfqhh4SwQ8l68MSFv5dnfBw35h8RNo99yQZUIAhTvWHpL+xXo7dW25eigAAAAAAAABXfA==.

All further analysis was done in a script-based manner employing R (http://www.r-project.org). Protein groups were further filtered requiring at least two unique peptides per protein identification, and 2 ratio counts (quantification events) in the forward as well as in the reverse experiment. For all analysis log2 transformed normalized ratios (as computed by MaxQuant) were used.

Median plus 1.2 standard deviations and median minus 1.2 standard deviations was used as significance threshold in the forward and reverse experiment, respectively. Proteins were defined as interactors if they passed the threshold either in the wt versus control, or the knockout versus control experiment in the forward and reverse experiment. For all further analysis, only proteins passing these criteria were considered. For hierarchical clustering only proteins that showed a ratio in four out of six experiments (wt versus control, ko versus control, wt versus ko; each forward and reverse) were considered.

The separate experiments were combined using the Uniprot identifier, and proteins were clustered employing an Euclidian distance matrix.

Generation of MEFs Stably Expressing Tagged Ago2

To identify Dicer-dependent and -independent Ago2 interactors, we established Dicer-deficient or Dicer wt MEF cell lines that stably express FH-tagged Ago2 (Fig. 1). We confirmed the stability of the Dicer knock out during culturing by PCR (Fig. 1A) and tested FH-Ago2 expression levels by Western blotting using antibodies against the HA-tag (Fig. 1B). As expected and also observed for endogenous Ago proteins (data not shown), FH-Ago2 levels are slightly lower in the Dicer-deficient cells compared with Dicer wt cells. This is probably due to active destabilization of unloaded Ago proteins. The GFP-only expressing cell lines serve as controls for all experiments. To minimize background binding during our biochemical Ago2 complex purification, we asked whether FH-Ago2 can be eluted with the FLAG-peptide (Fig. 1C). Indeed, an excess of FLAG-peptide efficiently removed bound FH-Ago2 from the anti-FLAG antibody matrix. As a further quality control, we analyzed miRNA expression in Dicer-deficient MEFs that express FH-Ago2 (Fig. 1D). Northern blotting against miR-19b showed that miRNAs were neither present in the lysates of Dicer-deficient MEFs nor in anti-FLAG immunoprecipitates from these cells. Therefore, the cell lines are suitable for the analysis of miRNA-dependent and -independent Ago2 interaction partners.

To further increase the robustness and information content of our quantitative mass spectrometry data, we performed forward and reverse (label swapping) experiments. For the forward experiment, we labeled the FH-Ago2-expressing cell line with heavy amino acids and the corresponding GFP-expressing cell line with light amino acids (Fig. 1E, left panel). For the reverse experiment, the labels of the cell lines were swapped (right panel). To prevent potential heavy to light exchange of specific transiently interacting partners during the purification procedure (46, 47), the FH-Ago2 complexes were immunoprecipitated with the FLAG antibody from total lysates for each SILAC state separately and combined during FLAG peptide elution. Eluates were separated by SDS-PAGE and cut into eight slices, proteins were in-gel digested and peptides were analyzed by high resolution LC-MS/MS on an LTQ Orbitrap instrument.

Specific interactors by definition show a high ratio in forward and a low ratio in reverse experiments, whereas unspecific background binders show a ratio close to 1 in both experiments. For data visualization, we plot the logarithmized normalized ratios of the forward and reverse experiments against each other. Every dot represents an identified and quantified protein. Each of the datasets contained on average 1214 identified proteins and 672 of these fulfilled our strict criteria for quantification (see Experimental procedures). Background binders constitute the vast majority of quantified proteins show no ratio change between SILAC pairs and are therefore clustered around zero (Fig. 1F, gray circle). Specific Ago2 interactors show high ratios in the forward and low ratios in the reverse experiment and are outliers from the background distribution in the lower right quadrant (green circle).

Identification of Dicer-independent Ago2 Interactors

To compare miRNA-free (unloaded) and miRNA-containing (loaded) Ago2 complex compositions, we isolated FH-Ago2 from Dicer wt and Dicer-depleted cell lines. GFP-transfected cells were used to identify unspecific binders (Fig. 2). Proteins interacting with Ago2 independently of miRNAs show significant ratios above the cut-off both in Dicer +/+ and Dicer −/− cell lines when measured against the GFP control (Fig. 2A, left and middle panel; specific binders appear in the lower right quadrant). To directly quantify the interaction as a function of the presence or absence of Dicer and miRNAs, we additionally precipitated FH-Ago2 from Dicer +/+ MEFs (heavy label) and Dicer −/− MEFs (light label). Labels were swapped for the reverse experiment. In this experiment, proteins binding independently of miRNAs appear together with the background binders clustering around zero (right panel).

Among the Dicer-independent Ago2 binders (Table I; see supplemental Table S1 for the complete mass spec data set), we found a number of proteins that have been implicated in Ago2 function before, indicating high specificity in our analyses. For example, Hsp90 is involved in loading small RNAs into RISC (48, 49) and it regulates Ago2 localization (50). The co-chaperones FKBP5 and PTGES have not been described in Ago2 complexes before and but our results suggest that they may function in a similar manner. Another example is the putative DExD box helicase MOV10, which has been identified as RISC component before (17, 18, 34). Other proteins in this group are the mRNA binding proteins IGF2BP1–3, PUM2, DHX30, HNRNPL, a highly specific set of ribosomal proteins (RPS 19, 18, 14, 5, and 3a) and the ARE binding protein DHX36/RHAU. We also find proteins involved in mRNA decay such as the decapping enhancer EDC4, Upf1 and DDX6/Rck (51–53). As mentioned above, GW proteins directly interact with Ago proteins via a specific Ago-interaction domain (12) and consistently, we find TNRC6A and TNRC6B as Dicer-independent Ago2 binders. Finally, the novel Ago2 interacting protein STRAP/Unrip (54) found here has been shown to interact with poly-A binding proteins and are involved translational repression in Drosophila (55). We speculate that this protein is involved in miRNA-guided gene silencing as well.

Identification of Dicer-dependent Ago2 Interactors

We next analyzed proteins that bind to Ago2 preferentially in the presence of Dicer and miRNAs (Fig. 2B). Together with Dicer itself, this group of proteins is located in the lower right quadrant in the experiments using Dicer +/+ cells versus control (Fig. 2B, left panel) and in the Dicer +/+ versus Dicer −/− experiment (right panel). In the Dicer −/− versus control experiment, these proteins do not appear as specific binders (middle panel).

As expected, we find the Dicer cofactor TARBP2 (TRBP) (56, 57) in this group (Table I). Among the identified factors is the RNA helicase A/DHX9, which has been implicated in siRNA-loading (58). Proteins such as YBX1, Gemin4, Gemin5, HNRNPC, HNRNPUL1, the ARE binding protein ELAVL1/HuR, the poly-A binding proteins PABPC1 and 4, the mitochondrial protein Matrin3 and the Fragile X mental retardation protein paralog FXR2 have been found in Ago complex purifications before. In addition, we identified the PABPC1-binding protein LARP1 (59), the mRNA binding proteins FAM120A/Ossa and CSDA, which have not been implicated in Ago2 function before. In contrast to the GW protein family members TNRC6A and B, TNRC6C interacts with Ago2 only in the presence of Dicer, suggesting that it requires Ago2 to be loaded onto miRNA target mRNAs. Of note, it is also conceivable that proteins found in this group associate with Ago2 indirectly via interaction with Dicer.

Proteins that Interact with Ago2 Preferentially in the Absence of Dicer

Proteins that preferentially interact with unloaded Ago2 are found in the lower right quadrant of the plot generated from Dicer-deficient cells (Fig. 2C, middle panel).

In this group, we find the RNA binding protein RoRNP (60), the zinc finger protein ZNF521 (61) and the vesicle coat protein clathrin (CLTC) (Table I). Of note, Ago3 shows an increased association with Ago2 in the absence of Dicer and miRNAs. HERC5, a HECT-type E3 protein ligase that mediates conjugation of ISG15 to target proteins in human (62) is also in this group, raising the possibility that it post-translationally modifies unloaded Ago2.

Ago2 Associates with mRNPs in the Absence of Dicer and Mature miRNAs

Because miRNAs guide Ago proteins to specific mRNAs for gene silencing, RNA-binding proteins are associated with Ago proteins and many of these interactions are bridged by mRNAs (17–19, 34). However, a requirement for miRNA for these interactions have generally not been investigated. To our surprise, we noticed that several mRNA binding proteins bind to Ago2 independently of Dicer and mature miRNAs. Some of these proteins associate with Ago2 in an RNA-dependent manner as has been reported for IGF2BP1 and three for example (17) suggesting that Ago2 may associate with mRNAs even in the absence of mature miRNAs. To test this hypothesis, we immunoprecipitated Ago2 complexes from the F/H-Ago2 expressing Dicer wt and Dicer-depleted MEFs and isolated the bound RNAs (Fig. 3A). GFP expressing cell lines or RNase treatment served as control. The isolated RNAs were analyzed on an agarose gel and Ago2 levels were controlled by Western blotting. A significant amount of longer RNA is bound to Ago2 in the Dicer-depleted cells supporting our hypothesis that Ago2 stably associates with mRNAs in the absence of miRNAs (Fig. 3A).

Because Ago2 may interact with mRNAs in the absence of miRNAs, it was not clear from our SILAC data which interactions were mediated by mRNAs. To analyze mRNA-bridged Ago2 interactions, we established RNase treatment conditions under which the mRNA is completely degraded but the miRNAs are not affected (supplemental Fig. S1). We performed a SILAC experiment in which Ago2-containing mRNPs were isolated and the immunoprecipitate from the light labeled cells was treated with RNase A whereas the immunoprecipitate from the heavy labeled cells was untreated. For reverse experiments, labels were swapped. We obtained 681 quantified proteins in the Dicer wt and 520 in the Dicer-depleted cell lines (supplemental Table S2). We exclusively considered the proteins that were identified as specific interactors in our previous experiment (Table I) and visualized their ratios in plots (Fig. 3B), a heat map (Fig. 3C) as well as a table (Table II). In the ratio plots, RNA-dependent interactors appear in the lower right quadrant and interactors that are not affected by RNase treatment appear together with the background binders in the center of the ratio plots. In the heat map, the red color indicates values above, blue below and gray around zero. Red-blue pairs for forward and reverse experiments are characteristic for an mRNA-dependent interactor.

As hypothesized, a high number of proteins associate with Ago2 in a RNA-dependent manner even in the absence of miRNAs (Table II). In this group, we find many RNA binding proteins including IGF2BP1–3, DHX36, DHX30, HNRNPL, and the ribosomal protein RPS14. Strikingly, genetic data in C. elegans demonstrated that PRS14 modulates mature let-7 function (63). The SILAC data therefore confirms that Ago proteins associate with mRNAs in the absence of mature miRNAs.

Additionally, we find RNA-dependent interactors that are only present when miRNAs are present (Table II). Among them are UPF1, FAM120A, YBX1, CSDA, ELAVL1/HuR, Matrin3, HNRNPC and LARP1. Interestingly, the poly-A binding proteins 1 and 4 also appear to require miRNAs for RNA-dependent Ago2 association.

A set of proteins associates with Ago2 in an mRNA-independent manner. This includes the TNRC6 proteins and Dicer. Other miRNA-independent protein-protein interactors of Ago2 are the HSP90 alpha and beta proteins with their cochaperones PTGES and FKBP5. Among the interactors preferentially binding in the absence of Dicer and miRNAs only Ago3, CLTC and ZNF521 were identified in the datasets and they show a direct binding behavior.

Taken together, our mass spectrometry approach revealed that Ago2 associates with larger RNA species even in the absence of small RNAs. Furthermore, several mRNA-binding proteins are specific to miRNA-free and miRNA-containing Ago2-mRNA complexes.

Validation of Ago2 Interactions

To further verify our mass spectrometry results, we performed Western blot analyses on a set of identified Ago2 interactors (Fig. 4A). Ago2 complexes were isolated from FH-Ago2-expressing cells (Fig. 4A, lanes 1–4). To test RNA requirements of the interactors, a set of samples was treated with RNase A (Fig. 4A, lanes 3 and 4). Immunoprecipitates from the GFP-expressing control cell lines served as background control (Fig. 4A, lanes 5 and 6). YBX1 and UPF1 are both miRNA- and mRNA-dependent interactors. We see a strong signal in the Dicer wt sample (Fig. 4A, lane 1) and a reduced signal in the Dicer −/− cells (Fig. 4A, lane 2) suggesting a miRNA-dependent interaction. This binding is bridged by mRNAs as apparent from the disappearance of the signal after RNase A treatment (lanes 3 and 4). ZNF521 is a protein that directly interacts with Ago2 in the absence of miRNAs and Dicer. As expected, we see signals in the samples from the Dicer −/− cells only (lane 2) and the signal intensity is not affected by the RNase A treatment (lane 4).

To functionally validate our interaction data, we used luciferase-based miRNA reporters (Figs. 4B and 4C). The 3′ UTR of Hmga2, a well-characterized let-7a target (64), was fused to firefly luciferase and transfected into HeLa cells in which ZNF521, CSDA and EDC4 were depleted by RNAi (Figs. 4B and 4C). In addition, we employed a reporter containing the Hmga2 3′ UTR with mutated let-7a target sites and normalized the data against each other (19) (Fig. 4C). As expected, Ago2 knock down led to increased luciferase activity. Knock down of EDC4, which has been implicated in miRNA function in Drosophila (52), resulted in specific luciferase up-regulation as well, suggesting that EDC4 is indeed involved in silencing of the Hmga2 reporter construct. Similar results were obtained for the mRNP component CSDA. ZNF521, however, is not involved in miRNA-guided gene silencing. Since ZNF521 is a putative transcription factor, it is possible that it cooperates with Ago2 in nuclear Ago functions.

In summary, our validation experiments show that novel Ago2 interactors discovered by proteomics can play roles in miRNA-guided gene silencing, highlighting the specificity of our Ago2 interaction network.