A global screening identifies chromatin-enriched RNA-binding proteins and the transcriptional regulatory activity of QKI5 during monocytic differentiation

Background Cellular RNA-binding proteins (RBPs) have multiple roles in post-transcriptional control, and some are shown to bind DNA. However, the global localization and the general chromatin-binding ability of RBPs are not well-characterized and remain undefined in hematopoietic cells. Results We first provide a full view of RBPs’ distribution pattern in the nucleus and screen for chromatin-enriched RBPs (Che-RBPs) in different human cells. Subsequently, by generating ChIP-seq, CLIP-seq, and RNA-seq datasets and conducting combined analysis, the transcriptional regulatory potentials of certain hematopoietic Che-RBPs are predicted. From this analysis, quaking (QKI5) emerges as a potential transcriptional activator during monocytic differentiation. QKI5 is over-represented in gene promoter regions, independent of RNA or transcription factors. Furthermore, DNA-bound QKI5 activates the transcription of several critical monocytic differentiation-associated genes, including CXCL2, IL16, and PTPN6. Finally, we show that the differentiation-promoting activity of QKI5 is largely dependent on CXCL2, irrespective of its RNA-binding capacity. Conclusions Our study indicates that Che-RBPs are versatile factors that orchestrate gene expression in different cellular contexts, and identifies QKI5, a classic RBP regulating RNA processing, as a novel transcriptional activator during monocytic differentiation. Supplementary Information The online version contains supplementary material available at 10.1186/s13059-021-02508-7.

Ren et al. perform a systematic proteomics screen to identify RNA binding proteins that are associated with the chromatin in three different cell lines. Among the candidate list they decide to focus on seven proteins that have been linked to haematopoiesis. For these proteins they perform genome-wide studies on RNA and DNA binding. From these experiments they find that QKI5 is binding to promoter regions. There it acts independent of its role as RNA binding protein and controls critical monocytic differentiation-associated genes. The overall topic of the interplay between Chromatin and RNA biology is very relevant and the finding are interesting. However, in the current state the quality of the genome-wide datasets and analysis are difficult to judge and will require some additional experiments and analyses before publication. For more details see comments below. Comments: 1) The authors perform CLIP experiments for several RBPs. However, it is difficult to evaluate the quality of these experiments. This is of particular interest for QKI5 CLIP since the authors report binding patterns that contradict previous studies. Are the antibodies used for Immununoprecipitation (IP) specific to QKI5? The authors should show that for the different IPs the protein of interest is pulled down and no contaminations. For QKI5 it would be required to perform the CLIP IP in a QKI5 knockdown/knockout background to show that the purified QKI5-RNA complexes are lost.
Also, it would be required to report some statistics in the methods section: How many reads remained per sample after mapping. Has PCR duplicate removel been performed? How many replicates have been performed (At least 2-3 would be required!)? How many binding sites are there for the different proteins? Are they reproducible between the replicates? 2) Considerations from comment 1 also hold true for the Chip-seq experiments. For example, in Figure 5C it looks like the read coverage for the QKI5 Chip is very low, with 5 or 10 reads forming a peak. Have these experiments been performed in replicates? This would be required and should be shown for some examples, at least as supp. Figure. 3) The question if RBP binding leads to changes in transcript abundance upon knockdown of the RBP is very interesting. However, I am not convinced by the current analysis. To make it more convincing it would be required to make use of the dataset as a whole. Since the authors have RBP binding data for seven proteins as well as knockdown data for the same proteins they should compare all by all. Meaning calculating the odds ratio of all RBP binding sets versus all knockdown effects. This would be a nice internal control to show that the observed effects are specific.

Point-by-point responses to the reviewers' comments:
Referee 1: Comments: This paper starts with identification of chromatin-associated RNA-binding proteins and then focuses on a representative RBP-QKI5 to reveal its functional role during monocytic differentiation. The authors show that QKI5 binds to the promoters of many differentiation-associated genes, in a manner independent of the RNA-binding activity of QKI5. Depletion or overexpression of QKI5 led to impaired differentiation. These observations are potentially interesting. However, the experimental data fall short to support their current conclusions. A major revision should be made in order to publish this work.
Our response: We thank this reviewer for the comments and advices. We have carefully revised our manuscript according to the reviewer's suggestion, in principle, we have (i) added more information and description in both results and methods section; (ii) provided detailed explanation to each relevant question, for instance, we have performed more comparative analyses between RBPs' ChIP-seq and CLIP-seq results and tried to make a conclusive speculation of the regulatory pattern of these Che-RBPs as well as more intensive analyses of QKI5 ChIP-seq data along with other data to provide more supporting evidences for the potential transcriptional activation ability of QKI5; (iii) included some essential quality control experiments were to support the main findings.
Major Concerns: 1. The authors concluded that the role of QKI on chromatin is independent of its RNAbinding activity, based on the assays using an RRM mutant of QKI. And RNase treatment failed to affect QKI mutant's chromatin association as shown in Figure S3h. Yet, this conclusion seems to be contradictory to the chromatin fractionation assays. As shown in Fig.2b, RNase treatment led to the decreased association of QKI on chromatin, which appeared to be consistent among different cell types. In addition to QKI5, the binding of ADAR and ELAVL1 to chromatin seemed to partially rely on RNA. The author needs to provide explanation for this discrepancy.
Our Response: Thanks for the comments. In Fig. 2b  2. The authors performed ChIP-seq and CLIP-seq for several RBPs. But the data mining is far from enough. Here are my questions and suggestions.
(1) What kinds of genes are bound by these RBPs at the DNA and/or RNA levels? Do they tend to bind house-keeping genes or cell-type specific genes?
Our response: According to the reviewer's suggestion, we first analyzed the distribution of each hChe-RBP in different gene types, which indicated that hChe-RBPs exhibited dissimilar binding patterns in different types of genes either at DNA or RNA level. As shown in Fig. 2c, ChIP-seq analyses revealed that each hChe-RBP possessed a unique distribution pattern in genome, yet all presenting a trend to bind protein-coding genes, while PTBP3 showed binding preference to lncRNA genes (Supported file 1-2: a, left panel, see "Methods" for detail). Additionally, CLIP-seq results indicated that hChe-RBPs preferred to bind to proteincoding RNA transcripts, among which NUDT21 also showed a tendency to bind small RNA transcripts. The enrichment level of each hChe-RBP, evaluated by fold-change between IP and Input samples, also differed among different gene types at DNA or RNA levels (Supported file 1-2: b). For example, QKI5 tended to accumulate on genomic regions of both lncRNA and small RNA but only showed distinct enrichment on RNA transcripts of small RNA. NUDT21 exhibited high binding-intensity on RNA transcripts of small RNA but showed no gene-type specific binding-tendency on genome. Subsequently, besides distribution pattern analysis of gene types, we also analyzed different types of protein-coding genes from hChe-RBPs' ChIP-seq and CLIP-seq results. According to the reviewer's advice, we divided protein-coding genes into house-keeping (HK) genes and cell-type specific (SP) genes in THP-1 cells, acquired from the human protein atlas database (https://www.proteinatlas.org/). For each hCheRBP, we calculated the ratio of HK and SP genes in the RBP-bound genes at DNA or RNA level and took the ratio of HK or SP genes of THP-1 cells in overall coding-genes as the reference. The binding tendency was determined by comparing the proportion of HK or SP genes in each hChe-RBP's bound genes with the reference ratio. For ChIP-bound genes: hChe-RBPs possessed different binding tendencies towards HK genes while showed no preference to SP genes in THP-1 cells, as ADAR, KHSRP, ELAVL1 and SETD1A preferred to bind HK genes. Of note, SETD1A, a known methyltransferase which methylates histone H3 lysine 4 (H3K4) (Wysocka J,et. al., Genes Dev, 2003,17 (7):896-911), exhibited the highest percentage (38.7%) of HK gene binding. For CLIP-bound genes: all these RBPs showed different degree of preference for HK genes. However, ELAVL1 also showed tendency to bind SP genes in THP-1 cells suggesting its specific regulatory role in cell-type specific pathways (Supported file 1-2: c).
(2) What's the relationship between ChIP-seq and CLIP-seq for each RBP. Besides correlation analysis at the gene level, the authors should provide more sequencing tracks or peak-level analysis. And how is the association of these two datasets related to their sensitivity to RNase treatment?
Our response: As the reviewer suggested, in order to explore the interaction between hChe-RBPs' DNA and RNA binding, we analyzed the relative positions between ChIPseq and CLIP-seq peaks for each hChe-RBP: For each hChe-RBP, the overlapping rate between ChIP and CLIP peaks was very low, with only 0.14% ChIP peaks overlapped with CLIP peaks in average (Supported file 1-3: a). CLIP peaks only appeared concurrently in the range of 5kb~10kb or beyond 10kb of ChIP peaks (~17% and 80% in average, respectively), corresponding to the low overlapping rate at gene level presented by Jaccard index (Supported file 1-3: b). We speculated that the interaction of hChe-RBPs to chromatin might not be mediated by newly-transcribed RNAs, nor by neighbor nascent transcripts because the distances between these RBPs' DNA and RNA binding sites are mostly more than 5kb. Their interaction might be mediated by ncRNAs like LncRNAs which could function at long distance or by the high-ordered structures of chromatin. (3) ChIP-seq and CLIP-seq of these RBPs showed quite different binding patterns and had only small fractions of overlapped genes. The correlation index for ChIP-seq and CLIP-seq is also low for each RBP analyzed. Could it result from mRNA contamination in CLIP-seq? Do these RBPs possess both DNA-and RNA-binding capacities? Do they have RNA-dependent and independent functions?

Combining with RNase treated sub-cellular fractionation results in THP-1 cells (Supported
Our response: We thank the reviewer for this concern. Actually, with the high specificity of CLIP experiment, the probability of mRNA contamination is quite low. According to CLIPantibody quality control (Supported file 1-5: a) and reproductive test of CLIP-seq data (Supported file 1-5: b, right panel), the experiment process and the quality of data were standardized. To further address the reviewer's concern, we have compared the known RNA motifs of classical RBPs as QKI5, ADAR, KHSRP, PTBP3 and ELAVL1 with our CLIP-seq data, which showed high consistency between those identified motifs generated from our data and the published ones, indicating that the CLIP signals in our data possessed high specificity and accuracy (Supported file 1-5: c). As showing below (Supported file 1-5: c), 5 RBPs' (QKI5, ADAR, KHSRP, PTBP3, ELAVL1) CLIP-seq motifs were consistent with those previous studies (Eggington JM. et al. Nat Commun. 2011, 2, 319;Galarneau A. et al., Nat Struct Mol Biol, 2005,12(8), 691-8;Van Nostrand EL. et al., Nature, 2020, 583(7818), 711-719;Meisner, N.-C., et al., ChemBioChem, 2004, 5: 1432-1447.
According to our data, besides RBP's classical functions in RNA regulation, they might have transcriptional regulation potential. Among them, QKI5, KHSRP and SETD1A tended to regulate gene transcription by binding to target genes (Supported file 1-6). In addition, SETD1A is a known methyltransferase possessing DNA binding domain (Wysocka J, et. al., Genes Dev, 2003,17 (7):896-911); KHSRP and ADAR have also been reported to have DNA binding domains (Davis-Smyth T. et.al., J Biol Chem. 1996, 271(49):31679-87;Schwartz T, et al. Science. 1999, 284(5421):1841; as for QKI5, although recent studies have shown that it could be recruited to chromatin by transcription factors (Zhou X, et.al., J Clin Invest. 2020,130(5):2220-2236Shin S, et.al., Nat Commun. 2021,12 (1) (4) What about meta-plot pattern at the gene body of QKI5 ChIP targets compare to the input? The enrichment of QKI5 ChIP-seq signals at the promoter is not sufficient to indicate a regulatory role on gene transcription. What about Pol II ChIP-seq or H3K4me3 ChIP-seq signals on commonly-activated genes after QKI5 knockdown?
Our response: QKI5 preferred to enrich on gene promoters than gene body regions as shown below (Supported file 1-7: a). However, the enrichment on promoters was still low, which was probably due to the weak interaction between QKI5 and chromatin. As the reviewer mentioned, the enrichment on promoters was insufficient to indicate a transcriptional activation ability. To address this concern, we firstly analyzed the histone modifications around QKI5 ChIP peaks and found that the active histone markers as H3K4me3/H3K27ac tended to enrich on QKI5-located promoters indicating its transcriptional activation function. At the same time, H3K27me3, a repressive histone maker was relatively low at QKI5-located promoters (Supported file 1-7: b). Moreover, we performed H3K4me3 and Pol II ChIP-seq in QKI5 knock-downed (QKI5 KD) THP-1 and control cells, respectively. Our results showed that both H3K4me3 and Pol II signals decreased on QKI5 commonlyactivated genes in QKI5 KD group, providing further evidence on QKI5's transcriptional activation function (Supported file 1-7: c). In addition, loss of QKI5 only brought minor decreases of H3K4me3 and Pol II ChIP signals. This is probably because that QKI5 is not a general transcription factor or a histone modification enzyme so that the deficiency of QKI5 would not lead to broad or significant impact on the enrichment of H3K4me3 and Pol II, i.e., the chromatin environment.
3. The authors claimed that QKI promotes monocyte differentiation. Based on Figure S3a, the expression of QKI5 is increased on day 8 but recovered to the normal expression level on day 11 of differentiation in HSPCs and THP-1 cells. However, QKI5 KD or OE affected monocyte differentiation on D13 and D19 in HSPCs. The authors should check the phenotypes at D8 instead of later time points when QKI5's expression is decreased.
Our response: Thanks for the suggestion. We had checked the influence of QKI5 on early time point during HSC monocytic differentiation (day 8), and QKI5 also showed monocyticdifferentiation promoting effect at the indicated time point. As shown below, QKI5 and its mutant could promote monocytic differentiation, while knock-down of QKI5 impeded the same process (Supported file 1-8).
4. Based on ChIP-seq analysis, QKI binds to >1000 gene promoters. But the authors only focused on a small number of targets that are related to differentiation. How about other target genes?
5. EMSA detected weak supershifted signals between QKI and dsDNA probes. This weak DNA-binding affinity of QKI hardly explains its chromatin-binding. More control DNA probes with different sequences (such as the common motif of QKI5 ChIP targets) should be tested. In addition, the authors should compare the dsDNA-binding activity of QKI to its abilities to bind ssDNA, DNA/RNA hybrid, and RNA.
Our response: We thank the reviewer for suggesting this important control experiment. According to the reviewer's suggestion, we have generated ssDNA, DNA/RNA hybrid, and RNA probes with the same sequence of previously used CXCL2 promoter probe as well as QKI5 ChIP motif probe (two motif sequence tandemly connected). The result of EMSA assay using these probes showed that QKI5 had the strongest interaction with RNA probes, which was in accordance with its RBP identity. QKI5 also showed weak interaction with CXCL2 promoter probe (dsDNA) which was consistent with our previous finding. And QKI5 might interact with DNA/RNA hybrid probe reflected in the slight tailing emerged in the indicated lane (Supported file 1-10: a). In summary, QKI5 binds RNA with high intensity while shows low binding intensity with dsDNA, which is possibly because of that none of known DNAbinding domain existed in QKI5 protein.
However, there might be uncharacterized DNA binding structures, or the weak DNA binding feature of some known structures in QKI5 might not be recognized so far, which resulted the weak bands bound to QKI5 protein in the EMSA analysis. In addition, QKI5 could also interact with the ChIP motif probe, further demonstrating that QKI5 could bind DNA by recognizing specific sequence (Supported file 1-10: b).
Minor Concerns: 1. In the first part, the authors mentioned an observation that RNase treatment had different effects on RBP's chromatin associations in different cell lines, which indicates a context-dependent regulation. Yet it's a pity that the authors didn't extend this part or discuss the potential mechanism.
Our response: Thanks for the comments. Please refer to our response to Major concerns 2/Section 2 and 3 for more details as well as the discussion part in the revised manuscript from line 420 to 432.
2. The authors characterized QKI in-depth based on an odds ratio (OR) calculation. But QKI5's OR is a slightly higher than 1 (Fig. 2g). Please explain the logic to choose QKI.
Our response: We apologize for the unclear description of why choose QKI5. The logic or the filtering process to pick up QKI5 is: First, the overlapping rate of QKI5's ChIP/CLIP signals is relatively low suggesting it might possess unique regulatory function at the transcriptional level (Fig.2g, ADAR/QKI5/PTBP3/NUDT21/ELAVL1 possessed low ChIP/CLIP-gene cooccupancy rate). Next, histone markers labeling transcriptional active status enrich on QKI5located promoter regions that indicates its transcriptional activation role (Fig.2i, the promoter regions of ADAR/QKI5/KHSRP/SETD1A were enriched of active histone markers like H3K4me3 and H3K27ac. The candidates after the two-step filtration are only ADAR and QKI5). Finally, according to Fig.2j, the transcriptional regulatory potential of each RBP calculated by OR shows that QKI5 might regulate target-genes' expression at transcriptional level (Fig.2j, QKI5/KHSRP/SETD1A showed significant transcriptional regulatory potential presented by OR ratio. The only candidate after the final filtration is QKI5). In addition, the target genes regulated by QKI5 gathered in hematopoietic related pathways the most among other hChe-RBPs with transcriptional regulatory potential (Fig. 2k), thus making QKI5 a potentially functional hChe-RBP in hematopoiesis.
3. The authors used RBPDB and ATtRACT to define the RBP library, which clearly does not include all the RBPs. There are other resources for RBPs (for example, Tuschl, 2014). RBP research community, including in "Ray, D. et al., Nature, 2013, 499, 172-177", "Pierre C. et. al., Cell, 2012, 150, 5, 1068-1081", "Daniel D. et al., Molecular Cell, 2018, 70, 854-867", "Basak E.et al., Mol Syst Biol. ,2019, 15, e8513" and "Girolamo G., et.al., Database, 2016, Volume 2016 As the reviewer mentioned, the paper published by Tuschl et al in 2014 reported a census of 1,542 RBPs, which contained RNA-related RBDs identified by hidden Markov models with RNA-related functions. This work predicted a large number of potential RBPs with RBD, but direct experimental evidence of their interaction with RNA was not provided. And other RBPs' databases were based on CLIP-seq datasets, for example starBase (http://starbase.sysu.edu.cn/index.php) and POSTAR2 (http://lulab.life.tsinghua.edu.cn/postar/index.php) only focus on a small fraction of RBP (36 and 171,respectively). Therefore, we chose RBPDB and ATtRACT databases to define the RBP library with consideration of coverage as well as reliability. Figure S1h, RNAs were not fully digested.

Based on
Our response: According to this figure, the specific bands of 18S and 28S RNA disappeared upon RNase treatment indicating the intact RNA molecules were fully degraded. The small fragments of RNA would not completely disappear thus resulting in smears in the lane. In this situation, the intact structure of RNA molecules no longer existed, resulting in the loss of interaction activity between RBP and RNA.

Referee 2:
A global screening identifies chromatin-enriched RNA-binding proteins and the transcriptional regulatory activity of QKI5 during monocytic differentiation by Ren et al.
Ren et al. perform a systematic proteomics screen to identify RNA binding proteins that are associated with the chromatin in three different cell lines. Among the candidate list they decide to focus on seven proteins that have been linked to haematopoiesis. For these proteins they perform genome-wide studies on RNA and DNA binding. From these experiments they find that QKI5 is binding to promoter regions. There it acts independent of its role as RNA binding protein and controls critical monocytic differentiation-associated genes. The overall topic of the interplay between Chromatin and RNA biology is very relevant and the finding are interesting. However, in the current state the quality of the genome-wide datasets and analysis are difficult to judge and will require some additional experiments and analyses before publication. For more details see comments below.
Our response: Thanks for the comments and suggestions. We added necessary controls in key experiments according to the reviewer's advices and had responded to all the questions as follows. Comments: 1) The authors perform CLIP experiments for several RBPs. However, it is difficult to evaluate the quality of these experiments. This is of particular interest for QKI5 CLIP since the authors report binding patterns that contradict previous studies. Are the antibodies used for Immununoprecipitation (IP) specific to QKI5? The authors should show that for the different IPs the protein of interest is pulled down and no contaminations. For QKI5 it would be required to perform the CLIP IP in a QKI5 knockdown/knockout background to show that the purified QKI5-RNA complexes are lost.
Response: We thank the reviewer for this suggestion. We have added IP controls for antibodies we used in ChIP/CLIP-seq and results showed that the antibodies could specifically enrich the indicated proteins (Supported file 2-1).
As for quality controls of QKI5 CLIP-seq, we conducted CLIP-seq in QKI5 knockdown THP-1 cells as the reviewer suggested. First, QKI5 protein was successfully knocked down as shown below (Supported file 2-2: a). Next, the CLIP-IP result showed that QKI5 in control group was immunoprecipitated by the antibody while rarely been immunoprecipitated in QKI5 knockdown (QKI5 KD) group (Supported file 2-2: b). Furthermore, we analyzed the CLIP-seq data of control/QKI5 KD groups respectively, which showed that QKI5 CLIP peaks acquired in control group were much more than that from QKI5 KD group (7337 vs. 669), demonstrating that RNAs bound by QKI5 could hardly be immunoprecipitated after the loss of QKI5 (Supported file 2-2: c). Besides, the overlapped peaks of the two datasets were very few, indicating the inconsistency between the 2 CLIP-seq datasets (Supported file 2-2: d), also the enrichment level of non-overlapped peaks from control group was higher than the overlapped peaks (Supported file 2-2: e). In addition, the read counts in control group were more than that in QKI5 KD group (Supported file 2-2: f). The IGV results also demonstrated that the CLIP peaks on target sites disappeared after QKI5 knock-down (Supported file 2-2: g). Since there's very little RNA could be enriched in QKI5 KD group, we concluded that the QKI5 CLIP peaks we have obtained were specific.
Also, it would be required to report some statistics in the methods section: How many reads remained per sample after mapping. Has PCR duplicate removel been performed? How many replicates have been performed (At least 2-3 would be required!)? How many binding sites are there for the different proteins? Are they reproducible between the replicates?
Our response: We have included the details for statistical analysis of eCLIP-seq data processing in Additional file 10: Table S9, each eCLIP-seq dataset had more than 100 million unique mapping clean reads for downstream analysis. For the eChIP-Seq experiment, we performed two replicates for each RBP. Due to the relatively strong correlation between the replicates (Additional file 1: Figure S2b, right), the two replicates were combined to perform peak finding and downstream data analysis. And we have included detailed data processing of eCLIP-seq in the Method section as described below: The hChe-RBPs CLIP-seq datasets were processed in accordance with previous studies (Van Nostrand EL, et al., Nat Methods, 2016, 13:508-514), and the eCLIP-seq data processing pipeline was available at "https://github.com/YeoLab/eclip". The Raw reads with distinct inline barcodes were demultiplexed using in-house scripts, and the 10-mer random sequence was appended to the reads name in bam file for later usage. Low quality reads and adapter sequence were trimmed by cutadapt (Martin M., et al., J EMBnet.journal, 2011, 17:3). Repetitive reads were removed by aligning reads with human repetitive element sequence on RepBase database (https://www.girinst.org/) by STAR. Cleaned reads were mapped to Homo sapiens genome (Ensembl GRCh38.p5) by STAR (Dobin A., et al., Bioinformatics, 2013, 29:15-21). PCR duplicate reads were removed by in-house script based on sharing identical random sequence. Two biological replicates were merged by SAMtools "merge" for following analysis. The CLIP-seq data processing statistics were list in Additional file 10: Table S9. Peaks calling and downstream data analysis were performed using clipper (Lovci MT., et al., Nat Struct Mol Biol, 2013, 20:1434-1442 ChIP peak annotation, comparison and visualization, 2015). The repeatability between two biological replicates was evaluated by Pearson correlation coefficient with read coverages for genomic regions per 1000bp, which were generated from Deeptools "multiBamSummary" (Ramírez F., et al., Nucleic Acids Res, 2016, 44:W160-165).
2) Considerations from comment 1 also hold true for the Chip-seq experiments. For example, in Figure 5C it looks like the read coverage for the QKI5 Chip is very low, with 5 or 10 reads forming a peak. Have these experiments been performed in replicates? This would be required and should be shown for some examples, at least as supp. Figure. Our response: We are confident that the ChIP-seq signals in our study were highly specific and reliable confirmed by both antibody-specificity verification and normative dataprocessing process with 2 replicates.
For the ChIP-Seq experiment, we performed two replicates for each RBP. The repeatability between two biological replicates was evaluated by Pearson correlation coefficient of reads coverage for genomic regions per 1000bp, which generated from Homer "getPeakTags". With the relatively strong correlation between the replicates (Supported file 2-3: a, left), the two replicates were pooled to perform peak calling and downstream data analysis.
For the peak visualization, we inputted a "bigwig" format file, which had been normalized the total number of reads to 1x10e7 for comparing different samples. The y-axis of IGV-plot represented the ChIP-fragment density, which was defined as the total number of overlapping fragments at each position in the genome. The ChIP fragment density was low because of the normalization methods, and the values represent relative reads coverage for comparing different samples rather than absolute reads coverage. We had provided the IGV-plot contained actual reads of binding sites on promoter regions of MOV10 and UACA genes which were QKI5 ChIP target genes verified in our work (Supported file 2-3: b).
3) The question if RBP binding leads to changes in transcript abundance upon knockdown of the RBP is very interesting. However, I am not convinced by the current analysis. To make it more convincing it would be required to make use of the dataset as a whole. Since the authors have RBP binding data for seven proteins as well as knockdown data for the same proteins they should compare all by all. Meaning calculating the odds ratio of all RBP binding sets versus all knockdown effects. This would be a nice internal control to show that the observed effects are specific.
Our response: Thank this reviewer for the advice. In the revised manuscript, we had calculated the odds ratio of all hChe-RBPs' binding genes vs. all differentially expressed genes (DEGs) generated by hChe-RBPs' knockdown effects. But the OR of all-by-all sets was 4.38 (p-value <2.2e-16) which indicated a strong relevance between the two datasets (data not shown). We realized that the 7 tested hChe-RBPs were screened out with hematopoiesisrelated functions thus their target genes might be partially overlapped causing functional relevance. And in this case, the OR of all-by-all sets was unsuitable to be the internal control. To test this hypothesis, we analyzed the overlapping ratio of the 7 RBPs' bound genes or DEGs upon each knockdown effect. The results showed that the pairwise overlapping ratios between each 2 of these 7 RBPs' ChIP-target genes could reach up to 0.77, meanwhile the overlapping ratios could be as high as 0.37 at DEG level (Supported file 2-4: a), indicating high overlapping between ChIP-binding sets and DEG sets which implied the functional relevance. Especially, SETD1A, also known as a histone methyltransferase, distributed broadly on genome covering up to 41% (17144/41480) of expressed genes in THP-1 cells (Supported file 2-4: a), thus overlapping largely with other RBPs' bound genes. Therefore, due to the high overlapping rates between SETD1A with other hChe-RBPs as an example, it is inappropriate to use this all-by-all strategy as the internal control. To strengthen the reliability of our analyses, we applied an IgG ChIP-seq dataset generated from wildtype THP-1 cells as a randomized control representing the background noise of non-specific binding signals which was irrelevant to the DEGs. By using this randomized control, the OR of QKI5, KHSRP and SETD1A were higher than the controls (grey columns represent the OR of IgG ChIP-seq vs. RBPs RNA-seq) confirming the correlation between their binding and the expression of target genes .