Comprehensive Identification of RNA-Binding Domains in Human Cells

Abstract

Mammalian cells harbor more than a thousand RNA-binding proteins (RBPs), with half of these employing unknown modes of RNA binding. We developed RBDmap to determine the RNA-binding sites of native RBPs on a proteome-wide scale. We identified 1,174 binding sites within 529 HeLa cell RBPs, discovering numerous RNA-binding domains (RBDs). Catalytic centers or protein-protein interaction domains are in close relationship with RNA-binding sites, invoking possible effector roles of RNA in the control of protein function. Nearly half of the RNA-binding sites map to intrinsically disordered regions, uncovering unstructured domains as prevalent partners in protein-RNA interactions. RNA-binding sites represent hot spots for defined posttranslational modifications such as lysine acetylation and tyrosine phosphorylation, suggesting metabolic and signal-dependent regulation of RBP function. RBDs display a high degree of evolutionary conservation and incidence of Mendelian mutations, suggestive of important functional roles. RBDmap thus yields profound insights into native protein-RNA interactions in living cells.

Graphical abstract

Figure 1

In Vivo Identification of RBDs by RBDmap (A) Schematic representation of the RBDmap workflow. (B) LysC- and ArgC-mediated proteolysis was monitored without trypsin treatment. The protease digestion under RBDmap conditions or in buffers typically used in MS studies (optimal) were compared to in silico digestions defining 0% miscleavage. The missed cleavages were calculated and plotted. (C) Distribution of MS-identified LysC/ArgC fragments based on their number of amino acids. (D) Silver staining shows the protein pattern of purified RBPs prior to and after LysC treatment (crosslinking: CL). (E) Scatter plot comparing the peptide intensity ratios between RNA-bound and released fractions. The peptides enriched in the RNA-bound fraction at 1% (RBDpep) and 10% FDR (candidate RBDpep) are shown in red and salmon, respectively (Pearson correlation coefficient: r). (F) Peptide intensity ratios between LysC and ArgC experiments computed from three biological replicates. The dots represent released peptides (blue), RBDpeps (red), candidate RBDpeps (salmon), and background peptides (gray). (G) Venn diagram comparing the proteins within the RBDmap data set and the HeLa, HEK293, and Huh-7 RNA interactomes. (H) Comparison of the peptide intensity ratios from three biological replicates between UV-irradiated and non-irradiated inputs (x axis) and between RNA-bound and released fractions (y axis) (color code as above). (I) Number of proteins harboring recognizable or unknown RBDs in the HeLa mRNA interactome (left) and in RBDmap dataset (right). See also Table S1 and Figure S1.

Figure 2

Identification of Well-Established RBDs by RBDmap (A) Amino acid enrichment within RBDpeps (salmon) over released (blue) proteolytic fragments (^∗, 10% FDR and ^∗∗, 1% FDR). (B) Amino acid enrichment within RNA-binding protein surfaces (≤4.3 Å to the RNA) over distant regions (>4.3 Å from the RNA) extracted from protein-RNA co-structures. (C) Bar plot showing the odds ratio of the most enriched known RBDs. (D) Distribution of RBDpeps and released fragments in a classical RBP. The x axis represents the protein sequence from N to C terminus, and the y axis shows the RNA-bound/released peptide intensity ratios. The protein domains are shown in boxes under the x axis (LysC: L and ArgC: A). (E) Schematic representation of RBDpeps mapping within or outside of classical RBDs (left). The idealized outcome of a perfect correlation between RBDpeps and classical RBDs (top right) and random distribution are shown (bottom right). (F) Computed ratio of peptides mapping within known RBDs versus outside RBDs, regarding their peptide RNA-bound to released ratios. The horizontal line represents the baseline for uncorrelated data (i.e., the proportion of peptides mapping to classical RBD in the whole validation set in absence of enrichment; see E bottom). See also Table S2 and Figure S2.

Figure 3

RBDmap Identifies RNA-Binding Regions with High Accuracy (A) Schematic representation of proximal and non-proximal peptides (left). The proteins within protein-RNA co-structures were digested in silico with LysC or ArgC and predicted fragments aligned with the RBDpep supersets. The left bars represent the proportion of proximal and non-proximal LysC/ArgC fragments in the complete structure superset (random probability). The right bars show the % of aligned RBDpeps that are RNA proximal or non-proximal (^∗∗∗p < 0.001). (B) Schematic representation of the X-link peptide coverage analysis. (C) x axis represents the relative position of the RRM (from 0 to 1) and their upstream (−1 to 0) and downstream (1 to 2) regions. The ratio of the X-link over released peptides at each position of the RRM and surrounding regions using the LysC data set was plotted (top). The secondary structure prediction for each position of the RRM and flanking regions is shown (bottom). (D) The ratio of X-link over released peptides was plotted in a representative RRM-RNA structural model (PDB 2FY1) using a heatmap color code. (E) As in (C), but for the DEAD-box domain. (F) As in (D), but using the PDB 2J0S as a DEAD-box helicase model. (G) As in (C), but for the KH domain. (H) As in (D), but using the PDB 4B8T as a model for a KH domain. (I) As in (C), but for the CSD. (J) As in (D), but with the PDB 3TS2 as a model for a CSD. See also Table S2 and Figure S3.

Figure 4

Globular RBDs Discovered by RBDmap (A) Odds ratios for the most highly enriched RBDs. (B) RBDpep and released peptides mapping to TXN as in Figure 2D (top). The ratio of the X-link over released peptide coverage at each position of the TXN fold as in Figure 3C is shown (middle). The secondary structure prediction for each position of the TXN fold and flanking regions is shown (bottom). (C) Crystal structure of human TXN (PDB 3M9J), K3 and K8 are highlighted, and the identified RBDpep is shown in red. (D) Schematic representation of the protocol for measurement of RNA-binding using eGFP fusion proteins. (E) Relative total (input) or RNA-bound (eluate) green fluorescence signal from cells expressing different eGFP fusion proteins (^∗∗∗p < 0.01, t test, and n = 9). (F) As in (B), but for PDZ domain. (G) Ratio of X-link over released peptides plotted as a heatmap in a PDZ homology model. (H) As in (B), but for DZF domain. (I) As in (G), but using a DZF homology model. (J) Autoradiography of FLAG-HA tagged proteins after PNK assay. (K) Western blotting using an antibody against the HA tag. The polypeptides of the expected molecular masses are indicated by asterisks. See also Tables S2, S3, and S5 and Figure S4.

Figure 5

Disordered Protein Regions as RBDs (A) Number of RBDpeps mapping to globular and disordered domains. (B) Number of proteins mapped by at least one RBDpep solely in globular domains, in globular and disordered domains, or only in disordered motifs. (C) Amino acid enrichment between globular (violet) and disordered (pink) RBDs (^∗, 10% FDR and ^∗∗, 1% FDR). (D) Multiple sequence alignment of short, disordered RBDpeps with clustal omega. The sequence logos were extracted from aligned disordered fragments. (E) Examples of alignment of K-rich protein motifs. (F) Disordered RNA-binding motifs from FUS and MECP2 expressed as eGFP fusion. (G) Relative total (input) or RNA-bound (eluate) green fluorescence signal from cells expressing FUS_449–518-eGFP, MECP2_267–316-eGFP, or unfused eGFP as a negative control (^∗∗p < 0.01, t test, and n = 6). See also Figure S5.

Figure 6

Features of Known and Previously Unknown RBDs (A) Dots show the mean isoelectric point of all LysC and ArgC fragments (the bars represent SEM) (^∗∗∗p < 0.01 and not statistically significant: n.s.). (B) Density plot comparing mRNA abundances of known RBPs and previously unknown globular and disordered RBPs. (C) Dots show the mean of the mRNA abundance of the protein groups described in (B) (^∗p < 0.05 and not statistically significant: n.s.). (D) Bar plot showing the conservation of RBDpeps and released fragments using Homo sapiens as reference (^∗p < 0.05 and ^∗∗p < 0.01). (E) Odds ratios for the most enriched PTMs in RBDpeps versus released fragments. (F) Sequence logos of conserved amino acids around posttranslational modifications. A position weight matrix is computed from all 12-mer sequences around the modified residue (10% FDR amino acids are shown). (G) Bar plot showing the odds ratio of Mendelian mutations occurring in RNA-bound over released fragments. (H) RBDmap of FUS. The position of the disease-associated mutations is represented as red or blue colored circles if mapping within or outside an RBDpep, respectively. See also Table S4.