To map the distribution of transcriptionally engaged Pol II genome-wide, we performed global run-on sequencing (GRO-seq) experiments using nuclei from three stages of wild-type animals (embryos, starved L1 larvae, and L3 larvae) and dosage-compensation-defective embryos. In GRO-seq reactions, engaged polymerases are allowed to transcribe (run-on) short distances (100 nucleotides) and incorporate affinity tags into their nascent RNAs under conditions that prohibit new initiation (Core et al., 2008). Tagged transcripts are affinity purified, amplified, sequenced, and aligned to the genome to map engaged Pol II (Figure 1A–E).

Figure 1 with 12 supplements see all

Download asset Open asset

Figure 1—source data 1

Aligned reads for GRO-seq, GRO-cap, and ChIP-seq experiments.

The number of reads from each replicate of GRO-seq, GRO-cap, and ChIP-seq that uniquely aligned to the Caenorhabditis elegans genome are listed.

https://doi.org/10.7554/eLife.00808.004

Download elife-00808-fig1-data1-v1.xls

Figure 1—source data 2

Annotation of transcription start sites for protein-coding genes.

https://doi.org/10.7554/eLife.00808.005

Download elife-00808-fig1-data2-v1.xls

Figure 1—source data 3

Annotation of transcription start sites for non-coding RNAs.

https://doi.org/10.7554/eLife.00808.006

Download elife-00808-fig1-data3-v1.xls

The two GRO-seq biological replicates for each stage had high statistical correlation throughout the genome (Spearman correlation, ρ > 0.94) and across gene bodies (Spearman correlation, ρ > 0.98) (Figure 1—figure supplements 1–3 and Figure 1—source data 1). Gene expression levels calculated from GRO-seq data correlated well with expression data from microarrays and RNA-seq experiments (Figure 1—figure supplement 4). For the majority of expressed genes, we found continuous GRO-seq signal upstream of the WormBase (WB)-annotated transcription starts (Figure 1A,B), suggesting that GRO-seq reactions contain nascent RNAs with true 5′ ends.

To map TSSs unambiguously, we performed a series of enzymatic selections on our GRO-seq run-on RNAs to capture only those RNAs with a 5′ cap, the 7-methyl guanosine residue added shortly after transcription initiation (Rasmussen and Lis, 1993). This modified GRO-seq procedure, called GRO-cap (Figure 2), enabled us to map TSSs with nucleotide resolution by tracking nascent RNAs prior to trans-splicing. Use of nascent RNAs without size selection also reduces the background from RNAs that are capped post-transcriptionally and increases the probability of identifying TSSs from promoters with low Pol II occupancy.

Figure 2

Download asset Open asset

For genes that lack trans-splicing, the site of maximum GRO-cap signal was coincident with both the 5′-most GRO-seq signal and the WB-annotated start (Figure 1C), confirming that GRO-cap and GRO-seq together permit high-confidence mapping of TSSs. For trans-spliced genes with no previously identified TSSs, strong GRO-cap and GRO-seq signals were found upstream of the WB-annotated starts, and TSS calls were supported by uninterrupted GRO-seq signal between the maximum GRO-cap signal and the WB start (Figure 1A).

In total, a TSS was identified for 31.7% (6353 genes) of all C. elegans protein-coding genes from at least one of the three developmental stages examined (Figure 1—source data 2). Of our TSS calls, 77% are for genes shown previously to be trans-spliced (Figure 1—source data 2) (Allen et al., 2011). Plotting the average GRO-seq and GRO-cap signals from each developmental stage relative to the TSSs from the same stage revealed the vast improvement in gene models (Figure 1F,G and Figure 1—figure supplements 5, 6 and 7A–C). Independent GRO-cap reactions from the same developmental stage gave very similar results (Figure 1G and Figure 1—figure supplement 7D).

TSSs were also annotated for the majority of genes encoding short non-coding RNAs such as snoRNAs, 21 U-RNAs, and microRNAs, including TSSs for the five polycistronic microRNA clusters (Figure 1C,E and Figure 1—source data 3) (see ‘Materials and methods’). The TSSs for the 21 U-RNAs were 2 bp upstream from the mature RNA (Figure 1—figure supplement 9).

In addition, the primary TSSs and gene composition were determined for many multigenic transcription units called operons in which the polycistronic pre-mRNAs are processed to monocistronic mRNAs through 3′ end formation and trans-splicing using SL2 RNA (Figure 1D,E) (Blumenthal, 2012). Identification of the TSS for one operon showed it to include a 5′ gene not previously ascribed to it (Figure 1D). TSSs for several operons showed the TSS to be far downstream (>2 kb) of the WB start, a result we also found for genes not included in operons (Figure 1—figure supplement 10A,B). Within operons some genes have independent promoters to transcribe their pre-mRNAs (Allen et al., 2011), and their TSSs were determined, including snoRNA genes within introns of internal genes (Figure 1E and Figure 1—figure supplement 10A,C). In general, we used conservative statistical criteria for calling TSSs, and additional TSSs can be identified by visual inspection of the data.

The new TSS calls revealed promoters to be further upstream from the 5′ ends of mature mRNAs than previously thought, as demonstrated by heat maps showing the GRO-seq signal or GRO-cap signal of re-annotated genes relative to WB starts or TSSs (Figure 1F,G and Figure 1—figure supplement 7A–D) and histograms showing distances between TSSs and WB starts or SL1 trans-splice acceptor sites (TSA) (Figure 1—figure supplement 11A,B). The TSS-to-TSA, called the outron, was previously thought to be 50–500 bp. We found instead that outrons can be as long as 14 kb and have a median of 260 bp and mean of 753 bp (Figure 1—figure supplement 11B). Fully 59% of outrons are longer than 200 bp, 21% are longer than 1 kb, and 2.3% are longer than 5 kb (e.g., Figure 1—figure supplement 8A–C).

Multiple lines of evidence indicate that the GRO-seq signal between newly called TSSs and previously identified TSAs reflects legitimate outrons rather than independent overlapping upstream transcripts. First, 3′ UTRs or polyA signals are rare in outrons of >1 kb in length, indicating the engaged Pol II is not from independent polyadenylated transcripts. From 565 such outrons, only 1.4% had an identified 3′ UTR in the Jan et al. (2011) study, and 0.7% had a 3′ UTR in the Mangone et al. (2010) study. Furthermore, only 3.5% had a polyA site (Mangone et al., 2010). Second, regions corresponding to long outrons have a continuous ChIP-chip signal from antibodies enriched for either the ser2 phosphorylated form of Pol II or the hypo-phosphorylated form of Pol II (Figure 1—figure supplement 8A–C). These results and the restricted run-on length of ∼100 nucleotides (Core et al., 2008) indicate that the GRO-seq signal corresponds to bound Pol II in vivo and is not an artifact of the nuclear run-on (NRO) reactions in vitro extending beyond the 3’ ends defined in vivo (Figure 1—figure supplement 8A–C and Figure 3A). Third, a heat map of individual genes showing the GRO-cap signal relative to TSSs reveals that a dominant TSS contributes the majority of the vast GRO-cap signal (Figure 1G). Together these observations strongly support the argument that the GRO-cap signal paired with the continuous GRO-seq signal from WB starts defines true TSSs. Recently published studies of TSSs in C. elegans that used cutoffs for TSS calls of either 1 kb upstream (Gu et al., 2012) or 200 bp upstream (Chen et al., 2013) of WB starts identified some outron TSSs but could not identify TSSs for a large class of genes with longer outrons (see ‘Discussion’).

Figure 3 with 2 supplements see all

Download asset Open asset

Several other noteworthy features of promoters emerged from this comprehensive mapping of C. elegans transcription units and TSSs. (1) Many trans-spliced genes have multiple TSSs (Figure 3A), suggesting that trans-splicing has removed the selective pressure to form promoters with a single TSS. (2) Genes can use different TSSs across developmental stages, indicating developmental stage-specific regulation of transcription initiation (Figure 3B). (3) DNA sequences flanking the newly annotated TSSs have strong sequence conservation across nematode species (Figure 3—figure supplement 1B) and also have evolutionarily conserved core promoter elements, including the TATA-box (worm consensus TATAWAWR) (Figure 3C and Figure 3—figure supplements 1A and 2A,B) and the initiator element (Inr) (worm consensus YCAYTY) (Figure 3D and Figure 3—figure supplement 2C), both of which facilitate formation of the Pol II pre-initiation complex (Juven-Gershon and Kadonaga, 2010).

Three prominent features of transcription emerged. Accumulation of Pol II at 3′ ends of genes is abundant (Figure 4A–C and Figure 4—figure supplement 1A,B), as is divergent transcription from promoters lacking upstream divergent genes (Figure 4D–F). In contrast, promoter-proximal RNA Pol II pausing in C. elegans is rare under normal growth conditions, unlike in other metazoans, as shown later in ‘Results’.

Figure 4 with 3 supplements see all

Download asset Open asset

Figure 4—source data 1

Gene expression, and 5′ pausing and 3′ pausing data for protein-coding genes.

https://doi.org/10.7554/eLife.00808.024

Download elife-00808-fig4-data1-v1.xls

Pol II 3′ accumulation, likely caused by slow 3′ end formation and RNA processing (Gromak et al., 2006), is positively correlated in C. elegans with the expression level of the gene (Figure 4—figure supplement 1A and Figure 4—source data 1) and is more extensive in C. elegans than in Drosophila (Figure 4A and Figure 4—figure supplement 1B). Multiple peaks of 3′ accumulation within a gene help identify genes having several isoforms with distinct 3′ ends (Figure 1—figure supplement 10D). The prevalence in C. elegans of polycistronic operons requiring extensive RNA processing to produce monocistronic mRNAs caused us to ask whether 3′ Pol II accumulation was positively correlated with the gene’s position in an operon and hence the level of RNA processing. First and middle genes in an operon require two forms of co-transcriptional RNA processing at the 3′ end to generate monocistronic mRNAs, polyadenylation of the upstream gene, and trans-splicing of the downstream gene, while the terminal gene requires only polyadenylation at the 3′ end. First and middle genes showed more 3′ accumulation than terminal genes or genes not in operons (Figure 4B,C). In other words, 3′ pausing is longer at genes with another gene to process just downstream. The 3′ Pol II accumulation at terminal genes and genes not in operons was equivalent, yet greater than in Drosophila genes. Lastly, 3′ accumulation was similar between terminal genes of operons and monocistronic genes undergoing trans-splicing, and both gene sets had greater 3′ accumulation than monocistronic genes lacking trans-splicing, which nonetheless had more 3′ accumulation than Drosophila genes (Figure 4—figure supplement 2A). Cues triggering trans-splicing, particularly operon-specific trans-splicing, appear to facilitate 3′ accumulation and perhaps predispose C. elegans Pol II to greater 3′ pausing genome-wide, whether or not a gene resides in an operon. It should be noted that Pol II accumulation at 3′ ends does not overlap with U-rich regions at 3′ ends. Therefore, the high GRO-seq signal is not due to selective enrichment of U-rich RNAs (Figure 4—figure supplement 2B,C).

GRO-seq readily detects divergent transcription from promoters. In C. elegans, divergent transcripts are short and initiated 75–150 bp upstream from TSSs of promoters lacking upstream divergent genes (Figure 4E and Figure 4—figure supplement 3A–C). The frequency of upstream divergent transcription in C. elegans appears to be intermediate in degree between that in mammals, where it occurs at the majority of active promoters (Kapranov et al., 2007; Core et al., 2008; Seila et al., 2008), and that in Drosophila, where it occurs only rarely (Nechaev et al., 2010; Core et al., 2012) (Figure 4F). For both mammals and worms, promoters with TATA elements rarely support divergent transcription (Figure 4—figure supplement 3D,E) (Core et al., 2012). These results underscore fundamental similarities and differences in the architecture, evolution, and function of eukaryotic promoters.

Analysis of transcriptionally engaged Pol II in wild-type vs dosage-compensation-defective embryos provided a robust assessment of the genome-wide impact of disrupting dosage compensation. To assess dosage compensation, we disrupted sdc-2 (sex determination and dosage compensation), the central XX-specific factor that triggers assembly of all DCC components onto X and induces hermaphrodite sexual differentiation by repressing the autosomal male sex-determining gene her-1 (Figure 5A) (Dawes et al., 1999). Without sdc-2, DCC subunits fail to bind to X, and her-1 is expressed, causing XX embryos to become severely masculinized and die from overexpression of X-chromosome genes (Figure 5A). The pivotal role of sdc-2 in dosage compensation made its depletion the most effective way to disrupt dosage compensation, although depletion of any DCC condensin subunit causes similar XX lethality and elevation of X gene expression, as shown previously by our genome-wide measurements of gene expression (Jans et al., 2009).

Figure 5 with 3 supplements see all

Download asset Open asset

Figure 5—source data 1

Genome-wide changes in gene expression caused by the disruption of dosage compensation.

Genes were separated into different gene sets based on their length, origin, and chromosome (X vs autosome) to compare GRO-seq gene expression between the sdc-2 mutant and control RNAi embryos. Shown are the total number of genes, the median sdc-2 mutant/control RNAi expression ratio, the average sdc-2 mutant/control RNAi expression ratio together with the standard error of the mean, and the number of genes in each set that are more highly expressed in each condition. Across numerous protein-coding gene sets, the X chromosome is more highly expressed and the autosomes are slightly less expressed in the sdc-2 mutant. Furthermore, when X-linked genes are significantly changed in expression, they are almost exclusively increased in expression. Each gene set is separated into two sets, one containing all genes and the other containing genes that are significantly changed in expression as determined by analysis with DESeq (p<0.05) (Anders and Huber, 2010). For the first two lists (labeled with ‘≥250 bp’), average GRO-seq gene expression was calculated from the beginning to the end of either the WormBase (WB) model or the newly annotated transcription start site (TSS) gene model. WB genes had to be expressed at greater than 1 RPKM (reads per kilobase per million), and have at least 250 uniquely mappable bases in both sets. For the next set (labeled with ‘WormBase WS230 Genes ≥1.1 kb’), average GRO-seq expression was calculated for genes greater than 1.1 kb, with the first and last 300 bp of the gene excluded. The level of expression had to be ≥1 RPKM for WB genes, and have at least 250 uniquely mappable bases for a gene to be included. For the final two sets of genes (labeled with ‘≥1.1 kb’ and ‘≥1.5 kb’), expression was calculated for genes of the indicated length that have a newly annotated TSS, with the first and last 300 bp of the gene excluded. A gene had to have at least 250 uniquely mappable bases for it to be included. RNA polymerase II transcribed microRNAs are controlled by dosage compensation, while RNA polymerase III transcribed tRNAs are not. Average GRO-seq gene expression from sdc-2 mutant and control RNAi embryos was compared across ncRNAs. For microRNAs, expression values were calculated from the full length of the WB ‘primary transcript’ or re-annotated TSS gene models. For tRNAs, expression values were calculated from the beginning of the ‘mature transcript’ to 50 bp downstream of the stop. Because tRNAs are highly repetitive and transcription of highly transcribed tRNAs continues downstream of the stop, the extra 50 bp was included to increase the unique mappability of each tRNA. For a gene to be considered for analysis, it had to have at least 25 bp of uniquely mappable DNA, and to have an average expression of at least 1 RPKM in both control RNAi and sdc-2 mutant embryos. The median and mean sdc-2/control expression levels show that X-linked microRNAs are more susceptible to dosage compensation than autosomal microRNAs. X-linked tRNAs are decreased slightly in expression in the sdc-2 mutant, suggesting that its expression is not controlled by dosage compensation.

https://doi.org/10.7554/eLife.00808.029

Download elife-00808-fig5-data1-v1.xls

Severe disruption of sdc-2 function was achieved by treating an sdc-2 partial loss-of-function XX mutant with sdc-2 RNAi [sdc-2(y93, RNAi)] (Figure 5—figure supplement 1). GRO-seq data showed that expression of her-1 (Figure 5B) and protein-coding genes on X (Figure 5C) were elevated in sdc-2 mutants, while expression of genes on autosomes was mildly reduced (Figure 5—figure supplement 2A–D). her-1 was de-repressed 12.7-fold, and the increase in GRO-seq signal was uniform from the promoter to the 3′ end, suggesting that SDC-2 controls sex determination by reducing Pol II recruitment or initiation at the her-1 promoter in XX embryos (Figure 5B).

The median change of gene expression in sdc-2 mutant vs control embryos was an increase of 1.63- to 1.67-fold for X-linked genes of all lengths and a slight reduction of 0.79- to 0.81-fold for autosome-linked genes (Figure 5—source data 1 and Figure 4—source data 1). For the subset of X and autosomal genes whose expression was statistically different (p≤0.05) between mutant and control embryos, 99% of X-linked genes had elevated expression in the mutants, while 63–66% of autosomal genes had reduced expression (Figure 5—source data 1 and Figure 4—source data 1). The reduced autosomal gene expression was robust to normalization procedures that counteract potential complications from increased X expression (see ‘Materials and methods’).

Our GRO-seq experiments also provided the first indication that dosage compensation controls the expression of small non-coding RNAs. We found most embryonically expressed X-linked microRNAs to be dosage compensated (Figure 5D and Figure 5—source data 1, and ‘Materials and methods’), while X-linked tRNAs were not, implying that RNA polymerase II is broadly sensitive to dosage compensation, and RNA polymerase III is insensitive (Figure 5—source data 1).

Previous studies showed the DCC binds to sequence-specific DNA recruitment sites on X and disperses to promoter regions of actively transcribed genes (Ercan et al., 2009; Jans et al., 2009; Pferdehirt et al., 2011), but the lack of precise TSS calls had prevented an accurate alignment of DCC binding sites with promoters. Mapping of new TSSs relative to DCC binding sites called from our ChIP-seq experiments performed for this comparison showed the peak of DCC binding to be immediately upstream of the TSS (Figure 5E). Although the DCC binds to promoter regions, prior studies showed that DCC binding to a gene is neither necessary nor sufficient for the compensation of that gene, and not all genes are dosage compensated (Jans et al., 2009). We assessed the generality of that conclusion by comparing DCC binding upstream of a TSS to the increase in gene expression caused by disrupting sdc-2. Our results confirmed and extended the original conclusion (Figure 5—figure supplement 3A,B). Thus, the DCC can act at a distance to control gene expression, and DCC binding intensity is not a proxy for the dosage compensation status of the gene.

To determine the step of transcription controlled by the DCC, we compared the distribution of the GRO-seq signal along X and autosomal genes between control and dosage-compensation-defective embryos. The change in Pol II distribution expected from disrupting dosage compensation differs according to the step of transcription affected by the DCC. If the DCC restricts Pol II recruitment, a uniform increase in engaged Pol II would be expected from the promoter to the 3′ end of genes of mutants. If the DCC reduces transcription by preventing the release of Pol II promoter-proximal pausing, or by reducing transcription elongation at stages downstream of the pause site, an increase in the level of engaged Pol II would be expected in the gene body and 3′ end in mutants, with the increase beginning more promoter proximal for a mechanism that controls pausing.

Promoter-proximal pausing of transcriptionally engaged Pol II is a rate-limiting step of transcription in metazoans (Adelman and Lis, 2012) observed at ∼ 40% of active genes in mammalian cells (Core et al., 2008) and more than 60% of active genes in Drosophila (Core et al., 2012). To determine whether control of 5′ pausing was a plausible mechanism for dosage compensation, we calculated 5′ pausing ratios (GRO-seq signal for 5′ end/gene body) for genes in each developmental stage using TSSs calls for the respective stage (Figure 4—source data 1). In contrast to Drosophila and mammalian genomes, very few genes (0.38%, 15 of 3975) of wild-type C. elegans embryos showed evidence of 5′ pausing (Figure 4—source data 1, Figure 6A,B, and Figure 6—figure supplement 1A–D). The paused genes were not enriched on X (Figure 6C). Moreover, the number of 5′ paused genes was not decreased in sdc-2 mutants (Figure 4—source data 1 and Figure 6D,E).

Figure 6 with 1 supplement see all

Download asset Open asset

Nonetheless, the GRO-seq assay is capable of detecting 5′ pausing in C. elegans, since genes not paused in embryos become paused in L1s deprived of food (Figure 6A and Figure 6—figure supplement 1A–C). In starved L1s, we found 7.7% of genes (166 of 2133) to exhibit 5′ pausing, and most were on autosomes (Figure 4—source data 1). A prior genome-wide analysis of Pol II binding in L1 animals cultured with and without food discovered an enrichment of Pol II only at some promoters of food-deprived larvae, consistent with 5′ proximal pausing (Baugh et al., 2009). Our results confirm their proposal that Pol II pausing can be induced by food deprivation. These cumulative results make it highly implausible that 5′ pausing control is the mechanism of X-chromosome dosage compensation in embryos, and they are consistent with the lack of a C. elegans negative elongation factor complex (NELF), which contributes, in part, to pausing in other organisms (Yamaguchi et al., 1999).

Analysis of GRO-seq data comparing transcriptionally engaged Pol II in control vs dosage-compensation-defective XX embryos revealed a uniform increase in the level of engaged Pol II across the entire length of X-linked genes, from TSSs to 3′ ends, in the sdc-2 mutant. This conclusion was reached both by metagene analysis (Figure 7A) and by the analysis of individual genes exhibiting a range of overexpression in the mutant (Figures 5C and 7C, and Figure 7—figure supplement 1). Analysis of individual genes averts complications in interpretations that might arise from averaging of data.

Figure 7 with 4 supplements see all

Download asset Open asset

For the metagene analysis, the ratio of engaged polymerase in mutant vs control was consistently elevated by at least 1.6-fold across scaled X-linked genes, from TSSs to 3′ ends (Figure 7A). This elevation in the GRO-seq signal was also readily apparent in metagene analyses of X-linked genes subdivided by quartiles of expression (Figure 7—figure supplement 2) or by gene length (Figure 7—figure supplement 3). In contrast, metagene analysis of autosomal genes from the same data sets showed only a slight decrease in engaged Pol II across their lengths (Figure 7B), consistent with the slight decrease in autosomal gene expression in mutants (Jans et al., 2009).

Heat maps displaying the log₂ ratio of mutant to control GRO-seq signal across individual X-linked (Figure 7C) and autosomal (Figure 7D) genes confirmed the uniform elevation of transcriptionally engaged Pol II across X-linked genes and the reduction across individual autosomal genes of sdc-2 mutants. Together, these results demonstrate that the dosage compensation mechanism reduces X-linked gene expression in C. elegans hermaphrodites by restricting either the recruitment or initiation of Pol II. This conclusion is strongly supported by the analysis of elongation density indices (average GRO-seq signal across the last 75% of genes/average signal across the first 25%) in mutant vs control embryos for genes on X and autosomes. The ratios of indices between control and mutant embryos were not significantly different for genes on X compared to autosomes, indicating that dosage compensation does not preferentially affect Pol II elongation (Figure 7E).

Two further lines of evidence show Pol II recruitment to be the predominant step of transcription targeted by the dosage compensation process. First, re-analysis of our previous genome-wide ChIP-chip analysis of Pol II binding (Pferdehirt et al., 2011) relative to the new TSSs showed an approximately twofold increase in the occupancy of hypo-phosphorylated Pol II at the promoters of X-linked genes in sdc-2 mutant vs control embryos (Figure 7F). DNA-bound hypo-phosphorylated Pol II at promoters is enriched for non-initiated Pol II. The increase in Pol II occupancy at promoters of mutants (measured by ChIP) was nearly equivalent to the increase in post-initiated Pol II (measured by GRO-seq), indicating that Pol II promoter recruitment is rate-limiting when dosage compensation is active.

Second, in sdc-2 mutants, the levels of sense and upstream divergent transcription were elevated coordinately at promoters on X lacking upstream divergent genes (Figure 7G and Figure 7—figure supplement 4A). At those promoters, the level of hypo-phosphorylated Pol II was also elevated about twofold in mutant vs control embryos, as assayed by ChIP (Figure 7—figure supplement 4B). Since we found increased Pol II occupancy in mutants and observed virtually no promoter-proximal pausing in control embryos, escape from pausing into productive elongation, or any other form of post-initiation regulation, cannot be the cause of elevated transcription at the divergent promoters in sdc-2 mutants. Furthermore, an increase in transcription initiation by bound polymerases cannot account for these results, but an increase in Pol II recruitment can. Our combined experiments demonstrate that C. elegans dosage compensation controls X-chromosome-wide gene expression predominantly by reducing Pol II recruitment to promoters of hermaphrodites.

In organisms that equalize X-chromosome gene expression between the sexes by a dosage compensation process, the question arises as to whether the compensated level of X-chromosome expression is equivalent to or half of the expression from the two sets of autosomes. The answer to this question has been controversial, although evidence has mounted in favor of a mechanism to balance total expression between X chromosomes and the two sets of autosomes based on measurements of accumulated transcripts (Xiong et al., 2010; Deng et al., 2011; Disteche, 2012; Lin et al., 2012; Deng et al., 2013; Jue et al., 2013). Our GRO-seq experiments have addressed this question in the most definitive way to date by measuring the levels of nascent transcripts prior to co-transcriptional processing. We show that in wild-type C. elegans embryos, X and autosomes have nearly equivalent levels of total gene expression, and that levels of transcribing Pol II are uniform across X and autosomal genes (Figure 8A). In dosage-compensation-defective mutants, the level of X expression and engaged Pol II exceeds that of autosomes by 1.7-fold, from the TSSs to the 3′ ends (Figure 8B). These results demonstrate the existence of a separate mechanism in C. elegans to elevate the intrinsic rate of transcription from the X chromosomes of both sexes, so that after dosage compensation, X chromosomes and the two sets of autosomes have equivalent expression. Our experiments provide evidence that it, like dosage compensation, works at the level of controlling Pol II recruitment.

Figure 8

Download asset Open asset

GRO-cap has advantages over TSS mapping strategies that use total accumulated RNA or short-capped (sc) RNA as the starting material. The use of nascent RNAs without base hydrolysis or size selection, as in GRO-cap, enriches the proportion of 5′ capped RNAs within the starting RNA population, reduces the level of false TSS calls from RNAs that are capped post-transcriptionally (Fejes-Toth et al., 2009), and increases the probability of identifying TSSs from promoters that are transcribed at low levels. Most importantly, the GRO-cap strategy permits reliable assignment of TSSs by pairing TSS calls with uninterrupted GRO-seq signal for transcriptionally engaged Pol II between the GRO-cap TSS and the previously annotated 5′ end. Multiple lines of evidence showed that GRO-cap signal combined with continuous GRO-seq signal from WB starts defines true TSSs.

Two reports of nematode TSSs have recently been published. The first used two approaches to identify the transcription starts (Gu et al., 2012). The first, called CapSeq, used total accumulated RNA as the starting material for the enzymatic enrichment of RNAs with 5′ caps and set the cutoff for TSS calls to be within 1 kb upstream and 100 bp downstream from previously annotated 5′ ends. As a consequence, CapSeq did not identify TSSs for a large class of protein-coding genes. TSSs of small processed non-coding RNAs were also difficult to identify by CapSeq. The second approach, called CIP-TAP, used scRNAs and was equivalently effective as GRO-cap for identifying TSSs of small non-coding RNAs.

The second report also used scRNAs to map TSSs and set the cutoff for TSS calls to be no greater then 200 bp upstream of the previously annotated 5′ ends, thereby also not defining TSSs for a large class of genes (Chen et al., 2013). The authors found numerous clusters of Pol II initiation upstream of their calls and classified many of these initiation events as enhancer-like chromatin signatures based on overlap with bound transcription factors. Based on comparison with our data (e.g., Figure 1—figure supplement 12A,B), we propose that a proportion of the upstream enhancer-like signatures are TSSs giving rise to full-length transcripts. Two clear examples of genes are shown in Figure 1—figure supplement 12B,C, where the single TSS (either 2534 bp or 2878 bp upstream of the WB start) called from GRO-cap and GRO-seq data was classified as an enhancer by Chen et al. (2013).

Although 5′ promoter-proximal pausing in metazoans is a common rate-limiting step of transcription that is highly regulated to control gene expression (Adelman and Lis, 2012), we found it to be rare in C. elegans under normal growth conditions. However, in L1 larvae deprived of food, we found 5′ pausing at 7.7% of genes with TSS calls. A prior study using ChIP-seq discovered the accumulation of Pol II at promoters of starved larvae and proposed that 5′ promoter-proximal pausing was responsible for Pol II accumulation (Baugh et al., 2009). Our results confirm that interpretation. Two factors promote 5′ pausing in metazoans, negative elongation factor (NELF) and DRB sensitivity-inducing factor (DSIF), although the relative contribution of each is not fully understood (Adelman and Lis, 2012; Yamaguchi et al., 2013). C. elegans appears to lack NELF, suggesting either that DSIF is sufficient in the infrequent cases of pausing or that the core promoter complex (Kwak et al., 2013) or another not-yet identified negative elongation factor participates.

The 3′ accumulation of Pol II has been documented previously in flies and humans using GRO-seq (Core et al., 2008, 2012). We found 3′ Pol II accumulation to be more extensive in nematodes than flies. Pol II 3′ accumulation is likely caused by slow 3′ end formation and RNA processing (Gromak et al., 2006). Because C. elegans has numerous polycistronic operons requiring extensive RNA processing to produce monocistronic mRNAs (Blumenthal, 2012), we were able to discover a strong positive correlation between the amount of 3′ Pol II accumulation and the amount of RNA processing. Curiously, though, even genes that were not part of operons exhibited greater 3′ accumulation than genes in flies. This observation raises the question of whether RNA processing in C. elegans is generally slower than that in other organisms to accommodate extensive trans-splicing, or whether C. elegans accommodates operons by imposing additional regulation on Pol II to enhance its 3′ pausing at all genes to assess whether to halt transcription or continue elongation.

Gene expression in metazoans is controlled by diverse regulatory mechanisms that function over widely different distances. Some mechanisms act locally on individual genes, while others such as dosage compensation function across large chromosomal territories or along entire chromosomes to regulate a large set of genes coordinately. In general, the step of transcription controlled by long-range mechanisms is not understood. In our study, multiple lines of evidence supported the conclusion that C. elegans dosage compensation regulates gene expression along X primarily by reducing the recruitment of RNA Pol II to the promoters of hermaphrodite X-linked genes.

First, regulation of 5′ promoter-proximal pausing cannot be the mechanism underlying C. elegans dosage compensation. If the DCC reduced X-chromosome gene expression by increasing 5′ promoter-proximal pausing, numerous genes on X would be paused in wild-type embryos, and disruption of dosage compensation would reduce the level of pausing. GRO-seq experiments showed instead that 5′ promoter-proximal pausing is rare in wild-type XX embryos. The few genes that exhibited 5′ pausing were not enriched on X chromosomes relative to autosomes, and the level of pausing was not decreased in dosage-compensation defective mutants.

Second, Pol II-mediated transcription elongation is not preferentially affected by the DCC. The ratios of elongation density indices (average GRO-seq signal in last 75% of a gene/average GRO-seq signal in first 25% of a gene) calculated for genes on X chromosomes and autosomes in control vs dosage-compensation-defective embryos were not significantly different between X-linked and autosomal genes, indicating that Pol II-mediated transcription elongation is not selectively changed on X chromosomes of mutants.

Third, the level of transcriptionally engaged Pol II assayed genome-wide by GRO-seq in control vs dosage-compensation-defective XX embryos revealed a uniform increase in engaged Pol II across the entire length of X-chromosome genes, but not autosomal genes, in mutants. Hence, the DCC controls X-chromosome gene expression in XX animals by reducing either the recruitment or initiation of Pol II. This conclusion was validated both by metagene analysis and by analysis of hundreds of individual X-linked genes showing different levels of de-repression in dosage-compensation-defective mutants.

Fourth, genome-wide quantification of Pol II occupancy by ChIP plotted relative to the new TSSs showed an increase in the hypo-phosphorylated form of Pol II at promoters of dosage-compensation-defective embryos vs control embryos that was equivalent to the increase in post-initiated Pol II measured by GRO-seq. Since DNA-bound hypo-phosphorylated Pol II at promoters is enriched for non-initiated Pol II, these results indicate that Pol II promoter recruitment is rate limiting when dosage compensation is active.

Our combined experiments reveal that the primary mechanism by which the C. elegans dosage compensation process reduces X-chromosome gene expression by half in XX embryos is to limit Pol II recruitment to promoters of X-linked genes. Our study does not eliminate the possibility of a minor repressive influence acting through another step of transcription or through a post-transcriptional mechanism such as RNA stability.

How might condensin reduce Pol II recruitment to the promoters of X-linked genes by approximately twofold? Our current and prior (Jans et al., 2009) studies showed that DCC binding to the promoter of a gene is neither necessary nor sufficient to elicit repression of the gene. Hence, the DCC influences gene expression over long distance, likely by imposing changes in higher-order chromosome structure. Clues to such a DCC function were suggested originally by the simultaneous discovery of the DCC and condensin’s biochemical properties in vitro as an ATPase that alters DNA topology (Chuang et al., 1994; Kimura and Hirano, 1997; Hagstrom et al., 2002; Hirano, 2012; Piazza et al., 2013) and its canonical roles in vivo of compacting and resolving mitotic and meiotic chromosomes for proper chromosome segregation (Hirano and Mitchison, 1994; Hagstrom et al., 2002; Chan et al., 2004; Hirano, 2012; Piazza et al., 2013). However, mechanisms of DCC function are perhaps best deduced from its non-canonical roles in vivo of regulating interphase chromosome structure (Bauer et al., 2012) and meiotic crossover recombination (Mets and Meyer, 2009; Wood et al., 2010; Aragon et al., 2013). Condensin II in Drosophila induces axial compaction of interphase chromosomes, globally disrupts inter-chromosomal interactions, and promotes dispersal of peri-centric heterochromatin (Bauer et al., 2012). These activities serve to compartmentalize the interphase nucleus into discrete chromosomal territories. Furthermore, nematodes carrying mutations that disrupt condensin I, which shares four subunits with the DCC, display elongated chromosomal axes during meiotic prophase and exhibit chromosome-wide changes in the distribution of double strand breaks and crossovers (Mets and Meyer, 2009), providing strong evidence that DCC subunits control intra-chromosomal structure.

These cytological, biochemical, and genetic observations of condensin function suggest that chromosome structure might affect Pol II recruitment in several ways, which are not mutually exclusive. Chromosome compaction and associated topological changes in DNA could broadly affect promoter accessibility of Pol II and the regulatory factors that recruit or stabilize it, thereby reducing Pol II recruitment in a quantitatively similar manner at different sites. Compaction of interphase chromosomal territories could reduce the local concentration of bound transcription activators and hence bound Pol II. Changes in intra-chromosomal interactions could alter the relationships between distal regulatory regions and their target promoters, thereby limiting Pol II recruitment. These and other models are topics of future studies.

Dosage compensation strategies differ across species. Mammals inactivate one of the two female X chromosomes, flies double expression of the single male X chromosome, and nematodes halve expression of both hermaphrodite X chromosomes. A central question is whether the molecular mechanisms underlying these diverse forms of chromosome-wide transcriptional regulation are the same or different.

In Drosophila melanogaster, dosage compensation is achieved by the male-specific lethal (MSL) complex that binds along the single X chromosome of males to double transcription of X-linked genes (Gelbart and Kuroda, 2009; Conrad and Akhtar, 2012). The complex contains two long non-coding RNAs and five proteins, including the H4K16 histone acetyltransferase MOF (Hilfiker et al., 1997; Conrad et al., 2012a) and the H2BK34 ubiquitin ligase MSL2 (Wu et al., 2011). The MSL complex was proposed to regulate gene expression by controlling transcription elongation (Smith et al., 2001). This model was subsequently supported by genome-wide mapping of MSL proteins and MSL-dependent H4K16ac to the bodies of male X-linked genes, with a bias toward 3′ ends (Alekseyenko et al., 2006; Gilfillan et al., 2006; Legube et al., 2006), and GRO-seq experiments in S2 cells with or without RNAi of the msl2 gene (Larschan et al., 2011). While a recent study analyzing genome-wide Pol II occupancy in males and females by ChIP-seq experiments suggested an alternative view that fly dosage compensation operates at the level of Pol II recruitment or initiation (Conrad et al., 2012b), a mathematical computation error discovered in this study (Ferrari et al., 2013; Straub and Becker, 2013) rendered the results insufficient to distinguish between the competing models. The elongation model received recent further support from the discovery that dosage compensation is disrupted by impairing the function of SPT5, a transcription elongation factor that co-localizes with the MSL complex on male X chromosomes and interacts physically with the MSL complex (Prabhakaran and Kelley, 2012). Thus, the weight of evidence strongly favors enhancement of transcription elongation as the primary mechanism of fly dosage compensation.

Not only does the overall dosage compensation strategy differ between worms and flies, the underlying molecular mechanism appears to differ as well. In worms, reduction in X-chromosome gene expression is primarily achieved by reducing recruitment of Pol II to promoters, while in flies, elevation in X-chromosome gene expression is primarily achieved by facilitating Pol II transcription elongation. Multiple solutions have evolved to coordinately control gene expression across an entire chromosome.

In many species, the evolution of sex chromosomes to be the primary determinants of sexual fate resulted in males having one X chromosome and females having two. Such chromosome sex-determining mechanisms had the potential to cause two problems in gene expression, an imbalance in X-chromosome gene expression between the sexes and an imbalance in gene expression between the male X chromosome and his two sets of autosomes (Disteche, 2012). X-chromosome dosage compensation strategies such as the Drosophila strategy solved both problems by doubling transcription of the single male X chromosome. An alternative solution would be to co-evolve two mechanisms, one to increase expression of X chromosomes in both sexes, thereby preventing hypo X expression in males, and a second to decrease total expression from female X chromosomes to prevent hyper X expression relative to female autosomes and male X chromosomes. Nematodes and mammals evolved the second strategy for equalizing X-chromosome expression between the sexes. It has been controversial whether these organisms also evolved a strategy to elevate X expression in both sexes and thereby balance expression between X chromosomes and autosomes (Xiong et al., 2010; Deng et al., 2011; Disteche, 2012; Lin et al., 2012; Deng et al., 2013; Jue et al., 2013).

The bulk of evidence now favors the presence of a mechanism to up-regulate X-chromosome expression in males and females of both organisms (Deng et al., 2011, 2013; Jue et al., 2013), and our results provide the most compelling evidence to date for C. elegans. Our approach of examining nascent transcripts in mixed-stage embryos by GRO-seq conferred three advantages. First, it enabled us to quantify transcription specifically from somatic embryonic cells and thereby avert the complication of quantifying maternally contributed germline transcripts that contaminate mature embryo mRNA. Second, we could examine nascent transcription at a stage for which most germline-specific mechanisms of gene regulation would have been erased. Third, we could quantify transcriptionally engaged Pol II across the entire length of a gene, starting from the bona fide TSS, and assess the step of transcription controlled by the process. We found that a major part of the mechanism to increase X expression in C. elegans is to increase the level of Pol II recruitment to promoters, as was shown recently for mammals (Deng et al., 2013). Together, these results reinforce the evolutionary importance of balancing gene expression across all chromosomes of a genome.

The following nematode strains were used:

• Wild-type strain N2

• sdc-2 mutant, sdc-2(y93,RNAi)

• Control RNAi, N2(L4440 RNAi).

Bacteria for use in feeding RNAi were prepared by growing HT115 bacteria bearing an RNAi plasmid (sdc-2 or the L4440 negative control) overnight in TB and ampicillin, inducing for 2 hr with 1 mM IPTG, pelleting, and resuspending in 1 vol (wt/vol) of LB with 20% glycerol. For RNAi treatment, embryos were harvested from gravid hermaphrodites and allowed to hatch off for 24 hr. NG agar plates supplemented with 1 mM IPTG and 100 μg/ml carbenicillin were spotted with super-induced RNAi bacteria and allowed to induce further overnight at 25°C. Hatched off L1 larvae were spotted on the RNAi plates and grown at 20°C until gravid.

For wild-type samples, embryos were harvested from gravid hermaphrodites grown at 20°C on concentrated HB101. Embryos were allowed to hatch off and starve for 24 hr at 20°C. Worms were fed and grown to L3 stage under liquid culture (1 worm/μl; 10 mg/ml HB101) for 34 hr at 20°C.

Animals were collected from whichever stage was desired. After washing twice with M9 buffer, animals were washed with cold nuclear isolation buffer (250 mM sucrose, 10 mM Tris-HCl (pH 7.9), 10 mM MgCl₂, 1 mM EGTA, 0.25% NP-40, 1 mM DTT, protease inhibitors, 4 U/ml SUPERaseIn [AM2696; Ambion, Grand Island, NY]). Animals were resuspended in nuclear isolation buffer (embryos and starved L1 in 3 vol, L3 in 1 vol), and dripped into liquid nitrogen to freeze. Starved L1 and L3 samples were ground under liquid nitrogen by mortar and pestle. Larval samples, post-grinding, and embryo samples were dounced with a Kontes 2 ml glass dounce to release nuclei. Douncing and collection of nuclei was performed for up to six rounds as follows: dounce with 10X pestle A, 10X pestle B, 5 min centrifugation at 100×g, removal of nuclei-containing supernatant, and addition of an equal volume of nuclear isolation buffer to the pellet. Nuclear isolation was monitored each round to determine effectiveness and when it was complete. The pooled supernatant was centrifuged for 5 min at 1000×g to pellet nuclei. The nuclear pellet was washed with nuclear freezing buffer (40% glycerol, 50 mM Tris-HCl (pH 8.3), 0.1 mM EDTA, 5 mM MgCl₂, 1 mM DTT, protease inhibitors, 4 U/ml SUPERaseIn). Approximately 1 × 10⁸ nuclei were resuspended in 100 µl nuclear freezing buffer and stored at −80°C until GRO-seq reactions were performed.

All NRO reactions and bead enrichment steps for GRO-cap were carried out as described above, with the exception that the RNA for GRO-cap was not base hydrolyzed. After the first bead binding, the TruSeq RNA 3′ Adapter (RA3) (Illumina part # 15013207) was ligated to the 3′ end of the NRO RNA. First, 50 pmol of the 3′ adapter were mixed with NRO RNA and 2 μl 50% PEG 8000, brought to 14 μl with DEPC water, incubated at 70°C for 3 min and put on ice for 2 min. Then, 2 μl of 10× T4 RNA Ligase I buffer, 2 μl 10 mM ATP, 1 μl SUPERaseIN, and 1.5 μl T4 RNA Ligase I (M0204; NEB) were added and reaction incubated at 22°C for 4–6 hr. The reaction was then brought to 100 μl with binding buffer and subjected to a second round of bead enrichment.

After the second bead enrichment, 5′ mono-phosphate RNAs were selected against in two successive steps. First, to selectively degrade RNAs with a 5′ mono-phosphate, NRO RNA was resuspended in 16.5 μl DEPC water, mixed with 0.5 μl SUPERaseIN, 2 μl 10× Terminator reaction buffer A, and 1 μl of Terminator 5′ phosphate-dependent exonuclease (TER51020; Epicentre), and incubated at 30°C for 1 hr. The reaction was extracted and precipitated using the standard method (above), and resuspended in 10 μl DEPC water. Second, 5′ mono-phosphate RNAs were dephosphorylated to prevent their participation in subsequent ligation reactions. For this, the RNA was then mixed with 1 μl SUPERaseIN, 14.5 μl DEPC water, 10× Antarctic Phosphatase buffer, 1.5 μl of Antarctic Phosphatase (M0289S; NEB), and incubated at 37°C for 30 min. The reaction was brought to 200 μl with 10 mM Tris-HCl (pH 7.5), 5 mM EDTA, and heat inactivated at 65°C for 5 min. The reaction was then extracted and precipitated using the standard extraction method and resuspended in 10 μl DEPC water.

The 5′ capped RNAs then were prepared for ligation and the final library preparation steps. The NRO RNAs were then split in half and the experimental sample was treated with tobacco acid pyrophosphatase (TAP): 1 μl SUPERaseIN, 15 μl DEPC water, 3 μl 10× TAP buffer, and 1 μl TAP (T19500; Epicentre), with incubation at 37°C for 1 hr. The control reaction was treated identically except for the addition of TAP. The reaction was brought to 200 μl and then extracted and precipitated using the standard method. The TruSeq RNA 5′ Adapter (RA5) (Illumina part # 15013205) was ligated to the 5′ end of the NRO RNA. First, 50 pmol of the 5′ adapter were mixed with NRO RNA, and 2 μl 50% PEG 8000, brought to 14 μl with DEPC water, incubated at 70°C for 3 min and put on ice for 2 min. Then, 2 μl of 10× T4 RNA Ligase I buffer, 2 μl 10 mM ATP, 1 μl SUPERaseIN, and 1.5 μl T4 RNA Ligase I (M0204; NEB) were added and reaction incubated at 22°C for 4–6 hr. The reaction was then brought to 100 μl with binding buffer and subjected to a third round of bead enrichment. After the third enrichment, samples were reverse transcribed, amplified and PAGE purified as described (Core et al., 2008), and quantified before submission for sequencing.

Libraries were sequenced with Illumina’s HiSeq 2000 platform. Reads were required to have passed the CASAVA 1.8 quality filtering to be considered further. To remove reads containing the RT-PCR adapter, we used cutadapt version 0.9.5 (http://code.google.com/p/cutadapt/) with the following command (cutadapt -a TGGAATTCTCGGGTGCCAAGG -a AAAAAAAAAAAAAAAAAAAA -z -O 15 -e .1 --minimum-length=30). The remaining reads were trimmed to 30 bp in length and aligned uniquely to the C. elegans WS230 genome using bowtie’s default settings (version 0.12.7), which permit two mismatches in the first 28 bp.

Because the 5’ base in each read most closely identifies the location of transcriptionally engaged RNA polymerase prior to the run-on, we created pile-ups using only the first base of each read. For GRO-seq, the pile-up of reads mapping to both strands was normalized by the number of millions of reads that mapped uniquely to the genome and was multiplied by 1000 to obtain RPKM (reads per kilobase per million). For GRO-cap, reads were normalized by the same method but were not multiplied by 1000, hence RPM (reads per million). As expected, the GRO-cap signal is enriched at TSSs compared to the GRO-seq signal.

Further analysis of our GRO-seq data using an alternative normalization strategy confirmed that X-linked genes are increased in expression, and autosomal genes are decreased in expression in the sdc-2 mutant. This new analysis evaluated the original normalization strategy. Because expression of genes on the X chromosome is elevated in sdc-2 mutants, it is difficult to determine whether the autosomes are changed in expression in the sdc-2 mutant. The proportion of autosomal reads relative to total reads per experiment is lower in the sdc-2 mutant (75.9%) than in the control RNAi (85.7%). To normalize the autosomal expression between the two conditions, we divided the average expression of all genes in the sdc-2 mutant by the scaling factor of this difference (75.9%/84.715% = 0.885). We then investigated changes in the sdc-2 mutant compared to the control for our most inclusive gene set: WB WS230 genes that are >250 bp in length (see Figure 5—source data 1). With this alternative normalization, the median sdc-2/control gene expression ratios increase for both X (1.67 becomes 1.88) and autosomes (0.81 becomes 0.91). This procedure changed the proportion of X-linked genes that are more highly expressed in the mutant (genes increased in expression by 1.5-fold or greater) from 64.9% to 78.4%. The original normalization showed that fewer autosomal genes increase by 1.5-fold than decrease by 1.5-fold (5.2% compared to 26.7%, respectively), and the alternative normalization shows the same trend (8.4% compared to 16.5%). Therefore, the same conclusions about X and autosomal gene expression can be made with either normalization strategy.

We computationally identified all 30mers in the WS230 genome. After passing these sequences through the cutadapt parameters outlined above, we aligned the remaining sequences uniquely using bowtie. We mapped the 5′ base to determine whether a read beginning at that base pair can be aligned uniquely.

Genome annotation files were downloaded from WB. Protein coding genes that were labeled as ‘Coding_transcript’ and ‘gene’ were extracted from the WS230 annotation file. Genes encoding microRNAs and snoRNAs that were labeled as ‘miRNA_primary_transcript’ and ‘snoRNA_mature_transcript’, respectively, were extracted from the WS225 annotation file. tRNAs predicted by ‘tRNAscan-SE-1.23’ were extracted from the WS230 annotation file. The WB remap and unmap tools were used to convert DNA coordinates between releases, thereby ensuring that all analyses matched the correct genome versions.

To determine whether dosage compensation specifically affects transcription elongation across the X chromosome, we determined an elongation density index for each gene with a newly annotated TSS. We calculated the average GRO-seq signal across the last 75% of the gene and divided it by the average GRO-seq signal across the first 25% of the gene, excluding the first and last 500 bp of the gene from this calculation. To ensure that we averaged the GRO-seq signal over a sufficiently large number of bases, we required that genes be ≥2 kb in length. To avoid outlier ratios that can result from a low number of reads, genes with an average RPKM <1 in the first 500 bp, or the first 25% or last 75% of the remaining gene were excluded from the analysis. To reduce the possibility that the elongation density index was influenced by the 3′ accumulation of Pol II of an upstream gene, we required that genes lack another gene on the same strand within 1 kb of the TSS. We analyzed 481 X-linked genes and 1861 autosomal genes.

To compare GRO-seq signal across genes, we scaled genes to be the same length, allowing us to average the GRO-seq signal across them. To avoid small genes that could affect the sensitivity of our analyses, we required that genes be ≥1.5 kb in length. These genes were scaled to the same length as follows: the 5′ end (1000 bp upstream to 500 bp downstream of the TSS) and the 3′ end (500 bp upstream to 1000 bp downstream of the WB stop site) were not scaled, and the remainder of the gene was scaled to a length of 2 kb. We predicted that leaving the ends of the gene unscaled might allow us to better identify any effects that occurred at the ends of genes.

To investigate GRO-seq trends surrounding the TSS or across scaled metagenes, we plotted the average GRO-seq signal across these regions of interest. To do so, we totaled the strand-specific GRO-seq signal for every gene in the gene list at each base pair in the region. We then divided the total GRO-seq signal at each base pair by the number of genes that are uniquely mappable at that base pair. We then took a 25 bp moving window average of this average GRO-seq signal.

We used the Python package matplotlib to produce heat maps showing the GRO-seq expression of the TSSs of all genes, and the difference in expression between DCC mutant and control RNAi embryos. To show the GRO-seq signal at either the empirically determined TSS or WB start, the signal from each gene with a TSS call was plotted, one gene per row, and the GRO-seq signal was averaged across 15 bp windows. Genes were ordered from top to bottom with increasing distance between the WB start and the TSS called from GRO-cap. To show expression changes in the DCC mutant, we scaled each gene ≥1.5 kb to the same length as described above. We then totaled the GRO-seq signal from the DCC mutant embryos and separately from control RNAi embryos across 100 bp windows and calculated an average. The log₂ of the ratio was plotted for every 100 bp bin across the length of the metagene.

To calculate the relative level of upstream divergent transcription for each gene, we determined the maximum sense and antisense transcription in a 150 bp window within 500 bp of the TSS. Upstream divergent transcript data from Human and Drosophila samples were obtained (Core et al., 2012). We used R to plot kernel density estimations of the log₂ (sense/antisense) ratio for C. elegans, Human, and Drosophila. To determine how far upstream the closest upstream divergent gene was to each TSS, we searched for the closest transcript (protein coding, non-coding RNA, tRNA, or rRNA) upstream and antisense to the TSS. Prior to the search, the 6353 protein coding genes with new TSS calls were re-annotated with the most upstream TSS identified. To determine the upstream distance between the TSS and the start of upstream divergent transcription, we searched 500 bp upstream of the TSS to identify the position with the highest level of GRO-cap TAP+ minus the TAP− signal.

Wild-type N2 animals were grown on NG agar plates with HB101 bacteria. Mixed-stage embryos were harvested from gravid hermaphrodites, and cross-linked with 2% formaldehyde for 30 min. Cross-linked embryos were resuspended in 3 ml of FA buffer (150 mM NaCl, 50 mM HEPES-KOH (pH 7.6), 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 1 mM DTT, protease inhibitor cocktail) for every 1 g of embryos. This mixture was frozen on liquid nitrogen, then ground under liquid nitrogen by mortar and pestle. Chromatin was sheared by the Covaris S2 sonicator (20% duty factor, power level 8, 200 cycles per burst) for a total of 30 min processing time (60 s ON, 45 s OFF, 30 cycles).

To perform the ChIP reactions, extract containing approximately 2 mg of protein was incubated in a microfuge tube with 6.6 μg of anti-DPY-27 or random IgG antibodies overnight at 4°C. A 25 μl bed volume of protein A Sepharose beads was added to the ChIP for 2 hr. ChIPs were washed for 5 min at room temperature twice with FA buffer (150 mM NaCl), once with FA buffer (1 M NaCl), once with FA buffer (500 mM NaCl), once with TEL buffer (10 mM Tris-HCl (pH 8.0), 250 mM LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA), and twice with TE buffer. Protein and DNA were eluted twice with 1% SDS, 250 mM NaCl, 1 mM EDTA at 65°C for 15 min. After elution, sequencing libraries were prepared as published (Zhong et al., 2010) with minor changes: sequencing adapters were as described (Lefrancois et al., 2009) and adapters were ligated using the NEB Quick Ligation Kit (M2200).

Libraries were sequenced on the Illumina GA2 platform. After barcode removal, 32 bp reads were aligned uniquely to the C. elegans WS190 genome using bowtie. MACS (version 1.4) was used to call peaks and create pileups with DPY-27 ChIP as the treatment and random IgG ChIP as the control. To account for read depth, the ChIP signal was normalized to the total number of millions of reads that uniquely aligned to the genome. To correct for non-specific binding, the IgG signal was subtracted from the DPY-27 signal. The resulting ChIP-seq signal from two biological replicates was averaged at each base pair genome-wide.

Raw RNA Pol II ChIP-chip data from experiments using 8WG16 antibody (raised against the hypo-phosphorylated Pol II C-terminal domain) (Pferdehirt et al., 2011) was obtained from GEO (accession numbers GSM634580 and GSM634582). The ChIP signal was normalized to the GC content of individual probes using MA2C (Song et al., 2007). The average ChIP-chip signal surrounding the TSS was calculated using the sitepro script within the CEAS package version 1.0.2 (http://liulab.dfci.harvard.edu/CEAS).

To determine whether C. elegans genes contain known core promoter motifs such as the TATA-box and Initiator element (Inr), we performed motif searches using MEME (http://meme.ebi.edu.au/meme/intro.html). To identify a worm TATA-box, we searched for strand-specific motifs within a region 15–45 bp upstream of the TSS. To identify a worm Inr, we searched for strand-specific motifs within 10 bp of the TSS in either direction. These searches identified a TATA consensus of TATAWAWR, compared to TATAWAWR for yeast (Rhee and Pugh, 2012), and an Inr consensus of YCAYTY, compared to YYANWYY in humans and TCAYTY in Drosophila (Juven-Gershon and Kadonaga, 2010). To determine where these motifs lie, we calculated their distance from the TSS. For TATA, we calculated how far upstream the most 5′ base lies. For the Inr, we calculated how far the adenine lies from the TSS. In other organisms the adenine has been shown to be the +1 nucleotide in transcription; the location of the worm Inr relative to the TSS suggests that this is true in C. elegans.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank the Vincent J Coates Genomics Sequencing Laboratory for Illumina sequencing, D Stalford for assistance with figures, and T Cline for comments on the manuscript.

© 2013, Kruesi et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.