Technical Comments

Response to Comments on “Drosophila Dosage Compensation Involves Enhanced Pol II Recruitment to Male X-Linked Promoters”

Science  19 Apr 2013:
Vol. 340, Issue 6130, pp. 273
DOI: 10.1126/science.1232874

Abstract

Ferrari et al. and Straub and Becker wrongly claim that an error in the computational analysis calls into question the conclusions of Conrad et al. All the available evidence, including the reanalyzed genomic data, show that the conclusions and the key message of the study remain unchanged: RNA polymerase II recruitment to male X-linked promoters is an important regulatory step during dosage compensation.

The Technical Comments by Ferrari et al. (1) and Straub and Becker (2) raise two main issues. The first relates to the error in the chromatin immunoprecipitation sequencing (ChIP-seq) analysis, and the second to the overall importance of our study. Below, we address the two issues in turn, along with other points. We demonstrate that the arguments of Ferrari et al. and Straub and Becker are misguided.

We acknowledge the error in the original analysis. Briefly, in each sample, we smoothed the ChIP/input signal by calculating averages in sliding windows across the entire genome (3). The value in each window was then multiplied by the square root of the number of bins containing usable ChIP-seq reads in order to reduce the contribution of windows containing fewer “good” bins. This is a standard technique for processing ChIP microarray (ChIP-chip) and ChIP-seq data in individual conditions (4, 5). The error arose specifically in this study because of the comparison in ChIP-seq data between samples; the log2(IP/input) signals were subtracted from each other when computing the differences in RNA polymerase II (Pol II) occupancies between X and autosomal genes in male and female samples. The multiplication factors would have cancelled out had they been applied to the IP/input values. However, because the factors were applied to the log2(IP/input) values, their effects were not cancelled out, thus affecting the magnitude of the difference between the samples. The correct difference between male and female Pol II occupancies is 1.2-fold rather than the 1.9-fold as originally reported. The factor itself has little effect on the analysis owing to the high data quality. Additionally, the y axes of the metaprofiles in Conrad et al. (3) should have been labeled “average smoothed Pol II log2(IP/input)” because they represent the processed log2(IP/input) values, not an enrichment of ~30,000-fold of ChIP signal over input. We sincerely apologize for these mistakes.

The key message in Conrad et al. is that regulation of Pol II function at gene promoters makes an important contribution to dosage compensation. This is an important concept because earlier work in the field has focused on transcription elongation as the main mechanism, excluding possible regulation at the promoter level. In contrast to this assumption, we demonstrate that the effect of dosage compensation is apparent at the beginning of X-linked genes.

Ferrari et al. and Straub and Becker question the conclusions of our study because of the data-processing error. Their arguments are incorrect and prone to exaggeration: Multiple lines of evidence —including Ferrari’s coauthors’ own global run-on sequencing (GRO-seq) data [i.e., Park and Kuroda (6)]—support our original conclusions.

(i) Having corrected the analysis (i.e., applying the factor to IP/input values or removing the factor), we find a 1.2-fold difference in X-linked Poll II promoter associations between males and females (Fig. 1 shows corrected results). Despite the change in magnitude, the effect at the promoter is still extremely significant. Ferrari et al. and Straub and Becker obtain near-identical results using different analysis pipelines.

Fig. 1 Enhanced Pol II recruitment to make X-linked promoters.

The original figure numbers from Conrad et al. (3) are shown in parentheses. (A) (figure 2A) (Top) Average Pol II occupancy in the promoter, middle, and 3′ end of X-linked (red) and autosomal (blue) genes in wild-type (WT) males. (Bottom) Relative difference between the two profiles (fold change marks are shown on the right-hand side of the plot). Thin gray lines represent the transcription start sites (TSS) and polyadenylation (PolyA) sites, respectively. Expressed genes with significant Pol II signals in their transcribed regions are included (254 X-linked and 1414 autosomal genes for the promoter and gene body; 96 X-linked and 406 autosomal genes for the 3′ end, because 3′ ends with neighboring genes were excluded). (B) (figure 2B) Same analysis for WT females. (C) (figure 4C) Same analysis for MSL2-RNAi males. (D) (figures 2E and 4F) Bar plots comparing mean Pol II occupancies between X and autosomal genes in male, female, and MSL2-RNAi samples at the promoter, middle, and 3′ end of genes. Relative occupancies are represented as percentage of deviation from an X versus autosomal ratio of 1. The mean X versus autosome Pol II occupancy ratios in WT males and females are 1.18, 1.21, and 1.22 for promoter, gene body, and 3′ end, respectively. Similarly, the mean X versus autosome Pol II occupancy ratios in WT males and MSL2-RNAi males are 1.15, 1.18, and 1.12. The P values indicate the statistical significance of the difference in occupancy ratios between samples (***P < 0.001; **P < 0.01; *P < 0.05; NS, not significant; permutation test). (E) (figure 3A) Average Pol II occupancy around the TSS (TSS –300 bp, TSS +500 bp) using expressed X-linked genes (red, 639 genes) and autosomal genes (blue, 3178 genes) in males and females. Bar plots are the same as in (D). (F) (figure 3B) Same analysis for stalled genes, defined as nonexpressed genes with significant Pol II occupancy at the promoter (395 X-linked and 1914 autosomal genes). The mean X versus autosome Pol II occupancy ratios in WT males and females are 1.19 and 1.18 in (E) and (F), respectively. The P values indicate the statistical significance of the difference in occupancy ratios between samples (***P < 0.001; permutation test). (G and H) Tables of P values for the promoter maximum ranking (G) and Pol II density index comparisons (H) between X and autosomes before and after correction of the erroneous data-processing step. The comparisons correspond to those originally presented in (3) in figures 2C, 4D, and 3, A and B for the promoter-rank analysis, and figures 2D and 4E for the Pol II density index analysis.

(ii) We performed all the original analysis using statistically robust, nonparametric methods that are independent of the actual fold-change difference. These include comparisons of ranking changes [figure 2C; figure 3, A and B; and figure 4D in (3)] and comparisons of elongation indices [figures 2D and 4E in (3)]. The results are analogous after correcting the ChIP-seq analysis (Fig. 1, G and H); Ferrari et al. also acknowledge this point.

(iii) A new permutation analysis of the ratios for the X to autosomal differences between males, females, and MSL2 RNA interference (RNAi) samples confirms the significance and robustness of our initial observations (Fig. 1, D to F). The observed differences at X-linked promoters arise because of measurable changes in Pol II occupancies and not by chance. Therefore, the ChIP-seq data provide clear evidence for enhanced Pol II recruitment to male X-linked promoters in Drosophila dosage compensation that is entirely independent of the original error in the analysis.

(iv) Our study contains a plethora of independent data supporting the observations from the ChIP-seq experiments. We performed ChIP-qPCR (quantitative polymerase chain reaction) validations of 23 genes (6 autosomal; 17 X-linked) that confirm enhanced Pol II recruitment at male X-linked promoters compared with females. We also measured the presence of initiated transcripts using the nascent RNA experiments for 30 genes (10 autosomal; 20 X-linked) showing increased RNA production at male X-linked promoters versus females. These results apply even to stalled X-linked genes, thus ruling out the possibility of contributions from sources other than increased Pol II association at the promoter.

Both Technical Comments suggest that a 1.2-fold increase in Pol II signals is too small to represent a biologically meaningful effect. This is a simplistic interpretation of the results. In hindsight, we acknowledge that we also put too much emphasis on the fold change; however, all the available evidence clearly indicates that the observed effects remain biologically and statistically significant. First, techniques such as GRO-seq and ChIP-seq involve extensive experimental handling of samples. Factors such as crosslinking efficiencies, antibody affinities, or the efficiency of enzymatic reactions all impact on the magnitude of the eventual read-out. Binding values are therefore relative measures of biological interaction frequencies rather than absolute quantifications. Second, both ChIP-seq and GRO-seq methods provide snapshots of Pol II occupancies on chromatin but cannot determine changes in Pol II turnover rates, again influencing the magnitude of the observed Pol II signal at a given locus. We acknowledge the limitations of the methodologies currently available, which is why we backed our findings using alternative methods.

Ferrari et al. also note that enhanced RNA production is not apparent at X-linked promoters compared with autosomal ones in the male S2 RNA-seq data published by Adelman and colleagues (7). This observation is not surprising: Using Park and Kuroda’s GRO-seq data from S2 cells (6), we previously demonstrated that—in the absence of a biological reference, such as a male-specific lethal (MSL) RNAi or female data set—comparisons within individual samples cannot reveal the effects of dosage compensation. Indeed, variations in the behaviors of individual X-linked and autosomal genes mean that the effects of dosage compensation are masked by gene-specific effects, and different results are obtained depending on the gene set included in the analysis [figure S9 in (3)]. Such gene-specific differences are eliminated by comparing the behaviors of identical gene sets across conditions or sexes. Since Adelman’s short RNA data set comprises only untreated S2 cells, it cannot be used to examine dosage compensation.

To examine whether differences in short RNA production apply genome-wide, we reanalyzed Park and Kuroda’s GRO-seq data (6). We compared the untreated S2 samples with the MSL2 or green fluorescent protein (GFP)–RNAi samples using a robust, unbiased, and nonparametric analysis of rank orders (Fig. 2). These cross-sample comparisons show significant differences at promoters: Untreated versus MSL2-RNAi and GFP-RNAi versus MSL2 RNAi comparisons show increased X-linked promoter activity, but not untreated versus GFP-RNAi, demonstrating a genome-wide increase of promoter proximal RNA production at X-linked genes under dosage compensation. Therefore, the GRO-seq data also provide strong support for enhanced promoter activity on the compensated X chromosome.

Fig. 2 Comparison of ranked maximal promoter values for the GRO-seq data (4).

Box plots show Untreated versus MSL2-RNAi male (A), GFP-RNAi versus MSL2-RNAi (B), and untreated versus GFP-RNAi differences (C). The same gene set is used as in figure 2 of Conrad et al. (3). P values (Wilcoxon rank-sum test) are indicated in the box plots. The comparisons clearly display increased Pol II activity at the promoters of X chromosomal genes in S2 cells that are independent of gene-specific effects.

Finally, Ferrari et al. claim that our original reanalysis of their GRO-seq data was wrong because we did not include the last 500 base pairs (bp) of genes and genes between 500 bp and 1100 bp in length. This was originally done to maintain consistency with the other analyses in our paper. We recomputed the elongation index taking into account the last 500 bp of genes and genes longer than 500 bp. As expected, the results are analogous: (i) Most autosomal chromosome arms do not display any difference in elongation compared with the X chromosome; a recent study reported similar results using the processed GRO-seq data values provided on Park and Kuroda’s supplementary Web site (http://compbio.med.harvard.edu/Supplements/Nature11b.html) (8). (ii) Any differences in elongation are not MSL complex–dependent and therefore are unlikely to result from the dosage compensation mechanism.

Straub and Becker report a 10% difference in Pol II occupancies between the wild-type and MSL2 knockdown samples. It appears that the authors made the wrong comparison: similarly, as in our original analysis and as shown in Fig. 1D, MSL2-RNAi leads to a highly significant 87% reduction of the enhanced X-linked Pol II promoter activity observed during dosage compensation.

The error in the ChIP-seq analysis is clearly unfortunate, and we apologize for the misunderstanding that this may have caused. Apart from pointing out this mistake, however, neither Technical Comment provides any convincing evidence against the promoter-recruitment model. The corrected analyses—including those by Ferrari et al. and Straub and Becker themselves—and all the accompanying data clearly support the conclusion that regulation of Pol II activity at the promoter plays an important role in dosage compensation. This is indeed a key finding and of great importance for future investigations because promoter events have been ignored in the past. Nevertheless, further studies will be important to unravel whether or not early events at the promoter are coupled with other cotranscriptional events.

References and Notes

  1. Acknowledgments: This work was supported by a European Union–funded Network of Excellence EpiGeneSys grant awarded to A.A., N.M.L., and J.M.V.
View Abstract

Subjects

Navigate This Article