High-Resolution Mapping of RNA-Binding Regions in the Nuclear Proteome of Embryonic Stem Cells

Abstract

Interactions between noncoding RNAs and chromatin proteins play important roles in gene regulation, but the molecular details of most of these interactions are unknown. Using protein-RNA photocrosslinking and mass spectrometry on embryonic stem cell nuclei, we identified and mapped, at peptide resolution, the RNA-binding regions in ∼800 known and previously unknown RNA-binding proteins, many of which are transcriptional regulators and chromatin modifiers. In addition to known RNA-binding motifs, we detected several protein domains previously unknown to function in RNA recognition, as well as non-annotated and/or disordered regions, suggesting that many functional protein-RNA contacts remain unexplored. We identified RNA-binding regions in several chromatin regulators, including TET2, and validated their ability to bind RNA. Thus, proteomic identification of RNA-binding regions (RBR-ID) is a powerful tool to map protein-RNA interactions and will allow rational design of mutants to dissect their function at a mechanistic level.

Figure 1. Development and Optimization of RBR-ID

(A) Mouse ESCs were pulsed with 4SU or not treated (1 and 2) and irradiated with different UV wavelengths (3). We isolated nuclei (4) and digested the crosslinked extracts with protease and RNase, producing a mixture of crosslinked and uncrosslinked peptides (5). Covalent crosslinks to RNA alters the peptide mass, and the mass spectrum of the corresponding uncrosslinked peptide decreases in intensity (6; red peaks). (B) Example averaged spectra from comparable retention time windows from untreated (left) and 4SU-treated ESCs (right). UV (312 nm) crosslinking caused decreased intensity of the highlighted spectrum in the 4SU sample. (C) Quantification of the extracted chromatogram for the control and HNRNPC peptides highlighted in (B). Bars indicate the average of the peak intensities normalized to the untreated sample (no 4SU) in six replicates + SEM. (D) Volcano plots showing log-fold changes in peptide intensities on the x axis and p values on the y axis for ±UV (254 nm) and ±4SU (312 and 365 nm). Peptides overlapping annotated RRM domains are in blue. The RNA-binding peptide from HNRNPC is highlighted in red. (E) Number of proteins with consistently (p < 0.05) depleted peptides and annotated as RBPs in the GO database (black) or not (gray). (F) Percentage of known RBPs according to the indicated studies and databases that were identified using different UV wavelengths for the crosslink. See also Figure S1.

Figure 2. Protein-Level Analyses of Proteins Identified by RBR-ID

(A) Scatterplot showing log-converted and normalized average intensities for peptides from biological replicates 1–3 and the additional replicates 4 and 5. (B) Top ten enriched GO terms (biological process and molecular function) for primary RBR-ID protein hits. p values are plotted on the x axis, and terms with false discovery rate (FDR) < 10% are shown in red. (C) Overlap of RBR-ID protein hits and all known RBPs, both experimentally identified and annotated in databases. p value is from the hypergeometric distribution. (D) Top ten enriched GO terms as in (B) but only for the RBR-ID protein hits not found in the set of already known RBPs. (E and F) Top ten non-redundant protein domains enriched in the primary RBR-ID protein hits (E) or only in the unknown RBP set (F). The black section of the stacked bar plots indicates the number of proteins containing the domain and found in the primary RBR-ID candidate list; the gray section indicates the number of proteins not in the RBR-ID list but detected by MS in the ESC nuclear extract. (G) Tukey boxplot for the distribution of maximum RBR-ID scores per protein comparing known RBPs and unknown putative RBPs from (C). See also Tables S1, S2, S3, S4, S5, S6, and S7 and Figure S2.

Figure 3. Mapping of the RBR for MS2-CP with RBR-ID In Vitro

(A) Recombinant MS2 coat protein and in vitro-transcribed stem-loop RNA with or without incorporated 4SU were allowed to form a complex, then crosslinked, digested, and analyzed by mass spectrometry. (B) Pull-down of MS2-SL RNA with or without 4SU and crosslinked to MS2-CP with different UV wavelengths. MS2-CP was detected via its fusion tag, GST. (C) Extracted ion chromatogram showing the elution profile of an RBR-overlapping peptide (red) and a peptide from an MS2-CP region that does not bind RNA (black) from MS2-CP crosslinked to natural (top) or 4SU-containing (bottom) MS2-SL RNA using 312 nm UV. (D) Quantification of peak intensities for the two peptides shown in (C) for three biological replicates each acquired in duplicates. Bars show average intensity + SEM. (E) Averaged and smoothed residue-level RBR-ID scores plotted along the primary sequence of MS2-CP. Regions with no peptide coverage are shown as gaps. Data are from three biological replicates each acquired in duplicates. Position of the known RBR and the uridine-interacting glu 63 residue are shown. See also Table S8 and Figure S3.

Figure 4. RBR-ID Maps the Sites of Protein-RNA Interactions In Vivo

(A) The surface rendering of the MS2 coat protein in complex with its cognate RNA (PDB: 1ZDI; Valegârd et al., 1997) was color coded according to the residue-level RBR-ID score from the experiment shown in Figure 3. (B) Schematic representation of the U1 snRNP particle (Kondo et al., 2015; Pomeranz Krummel et al., 2009). Subunits found in the primary list of RBR-ID candidates are in dark red; proteins in the extended list are in light red. (C and D) Zoomed-in regions of the crystal structure of U1 snRNP (PDB: 4PJO; Kondo et al., 2015) showing protein surfaces color coded according to their RBR-ID score and interacting RNAs for two regions of U1-70K (C) and SmD2 (D). (E and F) Schematic representation of the mammalian RNA pol II (E) and RNA pol I (F) complex according to Wild and Cramer (2012). Color coding is same as in (B). Subunits detected in the nuclear proteome, but not identified by RBR-ID, are in gray, undetectable subunits in white. The mammalian homolog for yeast RNA pol I subunit A14 (dashed circle) is unknown. See also Figure S4.

Figure 5. Known and Unknown RNA-Binding Regions in the ESC Proteome

(A) All detected peptides were sorted according to their RBR-ID score (UV312 ± 4SU) or a control score (no UV ± 4SU). The frequency of peptides overlapping the RRM domain (left) or a control, non-RNA binding domain (IPR027417, right) in these ranked lists is shown. (B) Categories of Interpro annotations for all peptides detected (left) or peptides in the primary list from RBR-ID (right). (C–E) Enrichment of selected domains in the top-tier RBR-ID peptides compared to the full list of detected peptides. Classical (C) and non-classical (D) RNA-binding domains are shown as well as enriched domains not previously reported to bind RNA (E). (F) Tukey boxplot of the isoelectric point for the indicated sets of peptides. p value is from a Student’s t test. Tot, all detected peptides in the nuclear proteome; uRBRs, peptides in the primary candidate lists that did not overlap known RBDs; RRM, all detected peptides overlapping with the RRM domain. (G) Percentage of peptides overlapping with disordered regions from IUPred (Dosztányi et al., 2005; Oates et al., 2013). Values are shown for all detected peptides (tot), all top-tier RBR-ID peptides not mapping to a known RNA-binding domain (uRBRs), and all peptides overlapping RRM domains. p value is from a chi-square test. See also Figure S5.

Figure 6. Validation of RBRs in L1TD1 and TET2

(A) Primary sequence and known domains for L1TD1 (top); smoothed residue-level RBR-ID score plotted along the primary sequence (middle); and scheme of epitope-tagged WT and RBR-deleted (ΔRBR) constructs used for validation (bottom). (B) PAR-CLIP of transiently expressed WT and ΔRBR L1TD1 in HEK293 cells. Autoradiography for ³²P-labeled RNA (top) and control western blot (bottom). (C) PAR-CLIP for WT L1TD1 with and without treatment with RNase A (top) and control western blot (bottom). (D) Primary sequence and known domains for TET2 (top); smoothed residue-level RBR-ID score plotted along the primary sequence (middle); and scheme of epitope-tagged catalytic domain fragment (CD) and RBR-deleted (CD_ΔRBR) constructs used for validation (bottom). (E) PAR-CLIP of transiently expressed TET2-CD and TET2-CD_ΔRBR in HEK293 cells. Autoradiography for ³²P-labeled RNA (top) and control western blot (bottom). (F) PAR-CLIP for TET2 CD with and without treatment with RNase A (top) and control western blot (bottom). See also Table S8 and Figure S6.