Gene Loops Enhance Transcriptional Directionality

See allHide authors and affiliations

Science  02 Nov 2012:
Vol. 338, Issue 6107, pp. 671-675
DOI: 10.1126/science.1224350


Eukaryotic genomes are extensively transcribed, forming both messenger RNAs (mRNAs) and noncoding RNAs (ncRNAs). ncRNAs made by RNA polymerase II often initiate from bidirectional promoters (nucleosome-depleted chromatin) that synthesize mRNA and ncRNA in opposite directions. We demonstrate that, by adopting a gene-loop conformation, actively transcribed mRNA encoding genes restrict divergent transcription of ncRNAs. Because gene-loop formation depends on a protein factor (Ssu72) that coassociates with both the promoter and the terminator, the inactivation of Ssu72 leads to increased synthesis of promoter-associated divergent ncRNAs, referred to as Ssu72-restricted transcripts (SRTs). Similarly, inactivation of individual gene loops by gene mutation enhances SRT synthesis. We demonstrate that gene-loop conformation enforces transcriptional directionality on otherwise bidirectional promoters.

Eukaryotic genomes are ubiquitously transcribed, generating an extensive network of noncoding RNAs (ncRNAs) (1, 2). Most ncRNAs are made by RNA polymerase II (Pol II), which can initiate transcription nonspecifically and bidirectionally on nucleosome-depleted chromatin (35). Although this promiscuous transcription is partly restricted by rapid transcript degradation (6, 7), we demonstrate that actively transcribed genes adopt a gene-loop conformation that reduces aberrant transcription by focusing Pol II into productive mRNA synthesis (see the supplementary materials and methods). Gene-loop formation depends on both promoter-associated transcription factors and polyadenylation complex (pAC) factors (811) such as Ssu72, localized at the 5′ and 3′ ends of genes (12, 13). On the basis of quantitative 3C analysis, we initially confirmed that mutation of Ssu72 (ssu72-2) prevents gene-loop formation across FMP27 (Fig. 1A). We also detected an increase in promoter-associated antisense ncRNA and increased Pol II density over the FMP27 promoter region in ssu72-2 (Fig. 1, B and C). Furthermore, we observed unanticipated genetic interactions between either Ssu72- or pAC-associated Pta1 and the nuclear exosome component Rrp6, which is responsible for the degradation of many ncRNAs, especially cryptic unstable transcripts (CUTs) (fig. S1) (6, 7). Taken together, our initial results indicate that the loss of gene-loop formation by inactivation of Ssu72 results in the production of aberrant ncRNAs that are stabilized in Δrrp6 mutant cells.

Fig. 1

Ssu72 inactivation abrogates FMP27 gene loop and enhances antisense transcription. (A) Graphical representation of 3C interaction levels in ssu72-2 versus the wild type across FMP27. Red stars show significant 3C interaction. The positions of 3C primers are indicated, as are reverse transcription quantitative polymerase chain reaction (RT-qPCR) amplicons. For 3C analysis, primer 1 (anchor) was combined sequentially with downstream primers 2 to 7. Error bars represent SEM. (B) RT-qPCR analysis of FMP27 mRNA and ncRNA in ssu72-2 versus the wild type. Error bar represent SEM. (C) Pol II profile (ChIP-seq) across the FMP27 promoter region (12) in ssu72-2 versus the wild type.

We next compared the effect of mutating RRP6 and SSU72 alone or together on the genomic profile of coding and ncRNAs. Total RNA from wild-type (WT), ssu72-2, Δrrp6, and double ssu72-2Δrrp6 strains grown at 32°C (semipermissive conditions) was hybridized to strand-specific Saccharomyces cerevisiae tiling arrays. The profiles obtained confirmed that loss of Rrp6 causes accumulation of CUTs, especially from bidirectional promoters (7). However, ssu72-2 mutation alone or in combination with Δrrp6 gave rise to many additional ncRNAs (Fig. 2A).

Fig. 2

Ssu72 inactivation leads to ncRNA transcription (SRTs). (A) I to IV: Genomic transcription across 28 kilobases of chromosomes 2 and 4 (x axis) for the Watson (W, top) and the Crick (C, bottom) strands. For the whole genome, see Normalized signal intensities are shown for profiled samples (y axis). Triplicate data are shown for the wild type (1 to 3), ssu72-2 (1 to 3), Δrrp6 (1 to 3), and ssu72-2Δrrp6 (1 to 3) strains. Red vertical lines represent inferred transcript boundaries. Nucleosome positions [green tracks (darker for higher occupancy) (25) and genome annotations are shown in the center: annotated ORFs (blue boxes), ncRNAs (orange boxes), and TSSs (arrows)]. (I) ncRNA0151, (II) ncRNA0397, (III) ncRNA0524, and (IV) ncRNA4353 represent promoter-associated SRTs. (B) Distribution of relative distances of 605 SRTs (upper panel) versus 1982 CUTs (lower panel) to their nearest ORF TSS (red line). Dashed lines (350 bp) indicate the cut-off position used to define intergenic cryptic transcription–sharing ORF promoters. (C) Distributions of gene expression levels are shown for downstream (down) and upstream (up) ORFs of tandem genes [three categories: downstream promoter pSRT (blue), pCUT (purple), or no pncRNA (green)]. Downstream ORFs with pncRNAs are significantly higher expressed than those without (P = 0.001 [pSRT] and P < 2.2 × 10–16 [pCUTs]). No significant association was found for upstream ORFs. (D) Pol II profiles for ssu72-2 (red) and the wild type (blue) around the ORF promoter in tandem genes that are more than 400 bp apart. Solid lines indicate median Pol II occupancy; shaded areas denote 25 to 75 percentiles. Pol II occupancy increases upstream of TSSs in ssu72-2.

Ssu72 is involved in the transcription termination of small nucleolar RNAs, as is clearly revealed by the widespread appearances of extended transcripts for these genes in ssu72-2 (fig. S2A, I) (14). The profiles also unveil a role of Ssu72 in transcriptional termination of CUTs, as many show 3′ extensions in the ssu72-2Δrrp6 double mutant compared with Δrrp6 (fig. S2A, II). Ssu72 inactivation also leads to increased initiation of new cryptic transcripts. Like CUTs, Ssu72-restricted transcripts (SRTs) often run in a divergent orientation from bidirectional promoters. We detected some SRTs in the single ssu72-2 mutant strain and others only in combination with RRP6 deletion (Fig. 2A). The array data demonstrated the presence of 605 SRTs in addition to the expected 1982 CUTs (Fig. 2B), as validated in specific cases (fig. S2B).

CUT and SRT initiation is associated with mRNA transcription start sites (TSSs) (Fig. 2B) (6, 7). To focus on promoter-associated ncRNAs (pncRNAs), we selected CUTs and SRTs that are positioned between tandem open reading frames (ORFs) (hereafter referred to as pCUTs and pSRTs). 678 pCUTs and 135 pSRTs initiate antisense transcription between tandem ORFs. Promoters that generate a divergent pncRNA tend to express more mRNA (down ORF) (Fig. 2C). In contrast, we found no correlation in mRNA expression level (up ORF) with downstream-positioned ncRNAs. We further showed that SRT expression is not due to loss of nuclear pre-mRNA down-regulation–dependent CUT termination or differential RNA stability effects (fig. S3, A and B). Finally, a genome-wide Pol II occupancy profile of the ssu72-2 mutant (12) revealed a distinct peak upstream of the TSS, which is absent in the wild type (Fig. 2D), as well as a higher Pol II occupancy over SRT transcript regions in the mutant compared with the wild type (fig. S3C). Overall these results established that the loss of Ssu72 promotes de novo initiation of pSRTs.

Publicly available genome-wide data revealed that pSRT-associated promoters are especially depleted of histone H4 acetylation (15), implying a more repressed transcriptional state also shown in four selected pSRT-producing promoters (Fig. 3A). Loss of Ssu72 seems to relax this repressed chromatin structure by promoting histone acetylation and consequent pSRT expression. A genome-wide analysis of S. cerevisiae nascent transcripts reported a potentially similar connection between ncRNA levels and histone deacetylation (16). Loss of histone deacetylase Rco1 (in Rpd3S complex) known to contribute to H4 deacetylation in gene 3′ regions also increased antisense transcription, suggesting its potential role in promoter directionality. However, antisense transcripts may derive from antisense initiation at gene 3′ ends (17). We compared pSRTs to antisense ncRNA induced in Δrco1 [Rco1-restricted transcripts (RRTs)] by generating transcriptome profiles for Δrco1 and Δrco1Δrrp6 matching our ssu72-2 profiles. To distinguish between transcripts arising from gene 5′ or 3′ ends, we selected tandem genes separated by either more or less than 400 base pairs (bp) (fig. S4A). We showed that RRT expression (in regions where pSRTs are also detected) in Δrco1Δrrp6 versus Δrrp6 is clearly greater in close tandem gene configurations than in distant ones, which indicates that RRTs are produced from gene terminator regions. We therefore performed a metagene analysis on tandem gene pairs more than 400 bp apart that have a pncRNA arising between them. SRTs peak near the TSS, whereas RRTs align with the transcription termination site (TTS) (Fig. 3B) also validated for specific tandem and divergent gene pairs (fig. S4B). The terminator association of RRTs fits with the known gene 3′ end association of Rpd3S (18, 19). Collectively we show that, contrary to previous interpretation (16), antisense RRTs are terminator-derived, whereas SRTs are promoter-derived. Ssu72 thus enforces promoter directionality. We also detected a small but significant trend of decreased expression in ssu72-2 for tandem genes that generate pSRTs (fig. S5), which suggests that the loss of promoter directionality results in decreased genic transcription. Because Ssu72 is required for gene-loop formation, we tested whether other gene-loop–associated factors similarly act to restrict pncRNA synthesis. Inactivation of TFIIB (Sua7) or other pAC components (Pta1, Rna14, and Rna15) has been shown to restrict gene-loop formation (10, 11). Similarly we show that their inactivation caused an increase in pncRNA in a range of S. cerevisiae genes (fig. S6).

Fig. 3

pSRTs initiate from bidirectional promoters. (A) I: Histone H4 acetylation (as a ratio with H3) compared over the intergenic region between ORF TSSs and divergent pSRT (blue) or pCUT (red) TSSs in WT strains. Intergenic regions of pCUTs show higher H4 acetylation levels than those of pSRTs. II: Chromatin immunoprecipitation analysis across the promoter regions of the indicated loci with WT- and ssu72-2-derived chromatin using anti-H4ac. Ssu72 inactivation caused H4 acetylation increase at all four loci. Telomeric region (TELV1) was used as a negative control. Error bars represent SEM. (B) Metagene analysis of ∆rco1∆rrp6 versus ∆rrp6 (green) and ssu72-2∆rrp6 versus ∆rrp6 (blue) differential expression levels for all antisense ncRNAs that initiate between tandem genes in relative position to the upstream gene TTSs and downstream gene TSSs.

Because gene loops require both an active promoter and functional polyA signals (PASs) (20), we tested the effect of terminating transcription on pSRT formation by directly replacing the PAS with an Rnt1 cleavage signal (RCS) that promotes efficient termination but not mRNA polyadenylation (21). Plasmid constructs containing CYC1 with transcription initiated on a GAL1 promoter and terminated by either a PAS or a RCS were transformed into Δrrp6 strain. After galactose induction, chromatin was subjected to 3C analysis (Fig. 4A, I). A clear peak of interaction between the promoter and a PAS (but not a RCS) was evident, confirming that RCS-mediated Pol II termination prevents gene looping. Next, we measured transcript levels of CYC1 mRNA and pncRNA in transformed Δrrp6 strains (Fig. 4A, II and III). The GAL1 promoter–associated pncRNA was enhanced in level when the CYC1 PAS was converted into a RCS due to loss of the PAS-dependent gene loop. In a genomic context, conversion of the MSN5 (which generates a pSRT) (Fig. 2A and fig. S2B) PAS into a RCS showed loss of gene looping and a threefold increase in pSRT production, mimicking the effect of Ssu72 inactivation (Fig. 4B). Finally, an integrated β-globin gene construct with a SV40 late PAS or mutated version (22) in human embryonic kidney 293 cells displays a gene-loop conformation with the wild type, but not the mutant PAS construct. Similarly we observed a threefold increase in the levels of divergent pncRNA with the mutated PAS (Fig. 4C), indicating a similar effect in a mammalian system.

Fig. 4

Gene-loop disruption by PAS mutation enhances divergent transcription. (A to C) I: Graphical representation of 3C interaction analyses for the CYC1 plasmid (A), MSN5 (B), and a mammalian β-globin gene construct (C), as in Fig. 1A. II: RT-qPCR analyses of mRNA versus pncRNAs. (A) III: Northern blot for CYC1 transcripts. Note that the CYC1 mRNA is smaller with RCS replacement, due to lack of a pA tail. Also, UTR1 downstream of CYC1 is inactive due to promoter deletion, denoted by the crossed arrow in (A, I). Error bars represent SEM. (D) The gene loop defines the transcription unit and promotes transcription directionality. Loss of the gene loop in ssu72-2 or with PAS mutation increases antisense pncRNA transcription. Gene loops involving Ssu72 (green ovals) may act to maintain nucleosome (yellow bubbles; ac denotes H4Ac) positions, preventing association of a second preinitiation complex (PIC) (red), leading to divergent transcription initiation.

Our results indicate that gene loops act to maintain the directionality of transcription. The loss of a mammalian gene’s PAS can directly influence the recruitment of transcription factors, with a consequent reduction in gene expression (22). PAS mutation has also been shown to increase levels of divergent transcripts (23). On the basis of our results, such effects are directly explicable by the loss of gene-loop formation and the potential to recycle factors from the terminator back to the promoter (see model, Fig. 4D). The role of Rpd3S in restricting antisense terminator transcripts (Fig. 3) clearly illustrates the importance of histone deacetylation in preventing inappropriate ncRNA synthesis. We predict that gene loops may similarly act to influence the recruitment of 5′ localized histone deacetylases such as Set3 (24). This would maintain promoters in a deacetylated, inactive state until gene activation selectively promotes transcription of genes rather than divergent pSRTs. We postulate that gene looping contributes to determining which transcription units are fully productive.

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Tables S1 to S8

References (2635)

References and Notes

  1. Acknowledgments: We thank B. Dichtl and J. Kufel for strains and H. Wijayatilake for FMP27 and β-globin 3C reagents. This work was supported by the Wellcome Trust (N.J.P.), the NIH and Deutsche Forschungsgemeinschaft (L.M.S.), European Molecular Biology Laboratory (J.B.Z., N.M.L., L.M.S.), and the Swiss National Fonds and European Molecular Biology Organization (J.C.). Genomic data are deposited at (E-TABM-936).
View Abstract

Stay Connected to Science

Navigate This Article