Robust transcriptome-wide discovery of RNA-binding protein binding sites with enhanced CLIP (eCLIP)

Abstract

As RNA-binding proteins (RBPs) play essential roles in cellular physiology by interacting with target RNA molecules, binding site identification by UV crosslinking and immunoprecipitation (CLIP) of ribonucleoprotein complexes is critical to understanding RBP function. However, current CLIP protocols are technically demanding and yield low-complexity libraries with high experimental failure rates. We have developed an enhanced CLIP (eCLIP) protocol that decreases requisite amplification by ∼1,000-fold, decreasing discarded PCR duplicate reads by ∼60% while maintaining single-nucleotide binding resolution. By simplifying the generation of paired IgG and size-matched input controls, eCLIP improves specificity in the discovery of authentic binding sites. We generated 102 eCLIP experiments for 73 diverse RBPs in HepG2 and K562 cells (available at https://www.encodeproject.org), demonstrating that eCLIP enables large-scale and robust profiling, with amplification and sample requirements similar to those of ChIP-seq. eCLIP enables integrative analysis of diverse RBPs to reveal factor-specific profiles, common artifacts for CLIP and RNA-centric perspectives on RBP activity.

Figure 1. Improved identification of RNA binding protein (RBP) targets by enhanced CrossLinking and ImmunoPrecipitation followed by high-throughput sequencing (eCLIP-seq)

(a) RBP-RNA interactions are stabilized with UV crosslinking, followed by limited RNase I digestion, immunoprecipitation of RBP-RNA complexes with a specific antibody of interest, and stringent washes. After dephosphorylation of RNA fragments, an “inline barcoded” RNA adapter is ligated to the 3′ end. After protein gel electrophoresis and nitrocellulose membrane transfer, a region 75 kDa (~220 nt of RNA) above the protein size is excised and proteinase K treated to isolate RNA. RNA is further prepared into paired-end high-throughput sequencing libraries, where read 1 begins with the inline barcode and read 2 begins with a random-mer sequence (added during the 3′ DNA adapter ligation) followed by sequence corresponding to the 5′ end of the original RNA fragment (which often marks reverse transcriptase termination at the crosslink site (red X)). (b) Bars indicate the number of reads remaining after processing steps. PCR duplicate reads that map to the same genomic position and have the same random-mer as previously considered reads are discarded, leaving only “Usable reads”. (c) Varying numbers of uniquely mapped reads were randomly sampled from RBFOX2 iCLIP and eCLIP experiments and PCR duplicate removal was performed. Points indicate the mean of 100 downsampling experiments (for all, s.e.m. is less than 0.1% of mean value). (d) RBFOX2 read density in reads per million usable (RPM). Shown are iCLIP, two biological replicates for eCLIP with paired size-matched input (SMInput) and IgG-only controls. CLIPper-identified clusters indicated as boxes below, with dark colored boxes indicating binding sites enriched above SMInput.

Figure 2. Improved CLIP signal-to-noise and reproducibility by normalization with paired Size-Matched Input (SMInput)

(a) Enrichment for SLBP clusters relative to SMInput was determined for all 23,034 CLIPper-identified clusters. Histograms show the number of clusters with indicated fold-enrichment, with histone-overlapping clusters in pink. (b) The subset of 2,821 SLBP eCLIP clusters with either pre-normalized (CLIPper) or SMInput normalized p-value ≤ 10⁻⁵ were selected and ranked by (left) pre-normalized CLIPper p-value or (right) by SMInput normalization; histograms indicate the number of histone-overlapping binding sites in each bin. (center) For 271 clusters overlapping histone RNAs, pink lines indicate the change in rank, with significance determined by Kolmogorov-Smirnov test. (c) Histogram indicates the number of RBFOX2 eCLIP clusters with indicated fold-enrichment in eCLIP relative to SMInput, with clusters overlapping introns flanking RBFOX2-dependent cassette exons indicated in green. (d) SLBP clusters were identified in Replicate 1, and for each cluster the fold-enrichment was determined for both Replicate 1 and Replicate 2 eCLIP. Histone-overlapping points are indicated in pink, with significantly enriched peaks indicated in blue. Attached histograms show the number of significantly-enriched peaks with specified fold-enrichment in Replicate 1 (top) and Replicate 2 (right). (e) Graphs indicate Irreproducible Discovery Rate (IDR) analysis performed on eCLIP fold-enrichment for (top) RBFOX2 biological replicates and RBFOX2 compared to IgG-only eCLIP, and (bottom) SLBP biological replicates.

Figure 3. Scalable RBP target identification with eCLIP

(a) Pie charts indicate experimental success results for 209 eCLIP experiments (each including two biological replicates plus an SMInput control) for which successful immunoprecipitation was performed in K562 (top) or HepG2 (bottom) cell lines, with colors indicating the amount of amplification required to obtain 100 fmoles of library (eCT). 102 experiments pass validation metrics and are deposited at http://www.encodeproject.org (Supplementary Table 2). (b) Each point represents a successful experiment in (a), with the x-axis indicating the eCT of the best replicate (denoted as replicate 1) and the y-axis indicating the increase in eCT between replicate 1 and replicate 2 (indicating decreased efficiency in the second replicate). Seven IgG-only eCLIP experiments are indicated by black lines, covering all 75 kDa intervals from 25 to 250 kDa. (c) The fraction of usable (non-PCR duplicated) reads out of all uniquely mapped reads is shown for eCLIP, public iCLIP experiments (12 performed for the ENCODE consortium as well as 115 published iCLIP datasets) and 152 published CLIP datasets (including PAR-CLIP and HITS-CLIP), shown as points with underlaid kernel density smoothened histogram.

Figure 4. eCLIP enables RNA-centric identification of protein binding to abundant noncoding RNAs

(a) Bars indicate the distribution of RBFOX2 clusters enriched (eCLIP/SMInput read density > 1) in RBFOX2 eCLIP relative to input (light bar) as compared to those depleted (dark bar). (b) Points indicate the percent of CLIPper-identified clusters identified within given regions that are enriched when compared against the paired SMInput for 102 eCLIP experiments (in biological duplicate) in K562 and HepG2 cells. (c) Bars indicate fold-enrichment of the most enriched peak overlapping lincRNA MALAT1 in each of 204 K562 and HepG2 datasets. Labels indicate biological replicates of RBPs with specific localization patterns. (d) Read density tracks along lincRNA MALAT1 for Replicate 1 of subset of datasets labeled in (c), with others shown in Supplementary Figure 15g.