In yeast, many tandemly arranged genes show peak expression in different phases of the metabolic cycle (YMC) or in different carbon sources, indicative of regulation by a bi-modal switch, but it is not clear how these switches are controlled. Using native elongating transcript analysis (NET-seq), we show that transcription itself is a component of bi-modal switches, facilitating reciprocal expression in gene clusters. HMS2, encoding a growth-regulated transcription factor, switches between sense- or antisense-dominant states that also coordinate up- and down-regulation of transcription at neighbouring genes. Engineering HMS2 reveals alternative mono-, di- or tri-cistronic and antisense transcription units (TUs), using different promoter and terminator combinations, that underlie state-switching. Promoters or terminators are excluded from functional TUs by read-through transcriptional interference, while antisense TUs insulate downstream genes from interference. We propose that the balance of transcriptional insulation and interference at gene clusters facilitates gene expression switches during intracellular and extracellular environmental change.https://doi.org/10.7554/eLife.03635.001
A DNA double helix is made up of two DNA strands, which in turn are made of molecules that are each known by a single letter—A, T, C, or G. The sequence of these ‘letters’ in each DNA strand contains biological information.
Genes are sections of DNA that can be ‘expressed’ to produce proteins and RNA molecules. To express a gene, the DNA strands in the double helix must first be partially separated so that one of them can be used as a template to build an RNA molecule in a process called transcription. Either of the DNA strands in a helix can be used as an RNA template, but contain different genes and are read in opposite directions. One of the two strands is called the ‘sense’ strand, the other the ‘antisense’ strand.
The RNA molecule does not transcribe a whole DNA strand; instead, it transcribes a section of DNA, known as a transcription unit, which contains at least one gene. The end of a transcription unit is marked by certain signals that stop transcription. However, some transcription units in a DNA strand overlap, so there must be some way that the transcription machinery can sometimes ignore these stop signals.
The activity of some genes is linked to the activity of their immediate neighbours. Furthermore, some genes are expressed in different amounts in response to changes in environmental conditions. Researchers have previously suggested that there must be some form of switch that controls when these genes are expressed.
Nguyen et al. now engineer start and stop signals at a neighbouring pair of genes, called HMS2 and BAT2, in yeast. When one gene is switched on, the other is switched off and which gene is active depends on the diet of the yeast cells.
On the antisense DNA strand opposite to HMS2 is another gene, SUT650. Nguyen et al. show that when this gene is transcribed, the transcription of HMS2 on the other DNA strand is blocked. This has the knock-on effect of turning on BAT2. Conversely, transcribing HMS2 switches off SUT650 and BAT2 because the end of HMS2 overlaps with the beginning of both SUT650 and BAT2. Switching between different genes relies on loops that physically link the start and stop signals of the gene to be transcribed while ignoring the start and stop signals for neighbouring genes.
Proteins called transcription factors can bind to DNA and affect whether a gene is transcribed. Nguyen et al. found that a transcription factor that binds near the start of the HMS2 gene helps to control which DNA strand is transcribed. When transcription factors do not bind to the start of HMS2, antisense transcription—and the expression of SUT650—occurs instead.
Overall, Nguyen et al. show that the transcription process itself makes up part of a switch that can control the expression of several genes on both the sense and antisense strands of a DNA double helix. This may also explain how many other, more complex, gene networks are activated in response to changes in the environment.https://doi.org/10.7554/eLife.03635.002
Genome-wide mapping of RNA transcripts in the budding yeast Saccharomyces cerevisiae has revealed an extensive array of coding and non-coding transcripts, giving rise to a genome that is heavily interleaved. Individual genes can possess multiple, overlapping transcripts, in the sense and the antisense orientations with respect to the pre-mRNA, as well as poly-cistronic transcripts, which span neighbouring genes (Kapranov et al., 2007; Berretta et al., 2008; Pelechano et al., 2013). Genome-wide mapping of nascent transcription, using techniques such as NET-seq, shows that transcription commonly extends into and over the intergenic regions of both convergent and tandemly arranged genes (Churchman and Weissman, 2011). For tandemly arranged genes, transcription into the promoter of the downstream gene would be expected to interfere with its expression by a variety of mechanisms, including modifying the local chromatin environment and interference by removal of transcription factors (Martens et al., 2004; Martianov et al., 2007; Hainer et al., 2011). Furthermore, extensive transcription antisense to (Venters and Pugh, 2009; Murray et al., 2012), and into the promoter of (Perocchi et al., 2007; Xu et al., 2009), the canonical coding transcript is also implicated in modulating gene expression by similar mechanisms (Hongay et al., 2006; Camblong et al., 2007; Uhler et al., 2007; Houseley et al., 2008; Pinskaya et al., 2009; Xu et al., 2011; van Werven et al., 2012; Castelnuovo et al., 2013). An interleaved genome with overlapping transcription units requires that polyadenylation and transcription termination signals in the sense and antisense orientations are by-passed but it is not clear how this is achieved. In addition, questions are commonly raised about whether transcription of these overlapping transcription units is contemporaneous.
It is now clear that much transcription is organised into biologically relevant temporal windows within phenomena such as the metabolic cycle (Tu and McKnight, 2006). Indeed, periodic or cycling expression of genes can be detected using fluorescent reporters or dual-labelled RNA FISH in cultures of asynchronous cells (Laxman et al., 2010; Silverman et al., 2010), or in the absence of cell division (Slavov et al., 2011). This periodic expression is a result of synchronization of respiratory and glycolytic activities into robust oscillations in oxygen consumption, characterized by phase-specific transcript signatures involving over 3000 genes, known as the Yeast Metabolic Cycle (YMC) (Klevecz et al., 2004; Tu et al., 2005, 2007; Soranzo et al., 2009; Slavov et al., 2011; Cai and Tu, 2012). In the long-period YMC, a single cell alternates between periods of high (oxidative (OX) phase) or low oxygen consumption (reductive building (RB) and charging (RC) phases), the residence time in each phase being nutrient-dependent. For exponentially growing cells in batch culture, the majority of cells in the population will be in the OX phase of the YMC (Slavov et al., 2011). Transcript levels for genes that cycle in the YMC will change as the cell moves through these phases; at any time, some cells in an asynchronous population will contain a transcript and some will not. These shifts through transcriptional states are robust but not invariable, as YMC-regulated genes switch on and off in response to cues prompted by both regulated and erratic changes in the intracellular and extracellular environment.
In addition to the partitioned gene expression patterns during the YMC, genomic spatial arrangements also contribute to regulatory relationships between genes, albeit on a smaller scale (Cohen et al., 2000; Lee and Sonnhammer, 2003; Batada et al., 2007). While genes displaying functional relatedness, such as the GAL genes, are regulated similarly by possessing the same cis-acting sequences (i.e. the UASGAL), co-expression of clustered genes in S. cerevisiae is largely independent of similarly controlled transcription, gene orientation, and/or shared regulatory sequences. However, members of adjacent gene pairs or clusters are more likely to belong to the same functional pathway than expected by chance (Cohen et al., 2000; Lee and Sonnhammer, 2003; Batada et al., 2007). The mechanism by which these clustered genes are regulated remains largely elusive.
Here we show that overlapping transcription, in both the sense and antisense orientations, constitutes an additional layer of regulation at clustered genes by managing state-switching in response to environmental change. We use a simple carbon source shift coupled with NET-seq to define genes whose transcription increases or decreases >threefold, after transfer from glucose (GLU)- to galactose (GAL)-containing media and show a remarkable enrichment for genes whose transcripts cycle during the YMC. The majority of these genes have no functional associations with transcription factors that might mediate repression or induction in GLU or GAL, but are organised in clusters and subject to overlapping sense and antisense transcription. We exemplify this mode of gene regulation, state-switching by transcriptional interference and insulation, at the HMS2:BAT2 tandem gene cluster. By engineering promoters and terminators at HMS2:BAT2, we demonstrate the formation of alternative transcription units underlies state-switching. We suggest that overlapping transcription and the formation of alternative transcription units, associated with temporally segregated gene expression, will be a general feature of gene regulation.
The question we address in this work is whether transcriptional interference can explain the switching on and off of genes and thus altered gene expression in response to environmental change. We used a change in carbon source by shifting exponentially growing cells from glucose- (GLU) to galactose-containing media (GAL) for 3 h and analysed the native transcripts associated with elongating RNA polymerase II (NET-seq) (Churchman and Weissman, 2011) to identify genes whose transcription is altered during this environmental change (Supplementary file 1A). 10.45% (551) of ORF-Ts (open reading frame—transcripts) showed a >threefold increase and 9.99% (527) showed a >threefold decrease in transcription in GAL relative to GLU (Supplementary file 1B). By comparing the NET-seq output with a microarray of Poly(A)+ RNA, we show that for the majority (≈88%) of genes, the change in transcript levels on the GLU to GAL shift reflects altered transcription rather than altered transcript stability (Supplementary file 1A). Gene ontology (GO) analysis revealed highly significant associations to growth, quiescence, and transcripts that cycle in the YMC, particularly during the oxidative (OX) and the reductive charging (RC) phases of the YMC (Figure 1A; Supplementary file 1C,D). The genes whose transcription decreases in GAL are enriched (467; 88.6% p < 1x10−5) for OX phase genes that are mainly involved in ribosome biosynthesis and growth (Figure 1B). By contrast, for genes whose transcription increases in GAL, there is significant enrichment (392; 71.1%, p < 1 × 10−5) for genes whose expression peaks in the reductive charging phase (RC) of the YMC (Figure 1B). These genes are associated with stress resistance, metabolism, and quiescence. 33.1% (891) of the 2691 annotated OX- and RC-regulated genes also show >threefold change on the GLU to GAL shift, supporting a shared regulatory mechanism between the YMC and the GLU to GAL shift (Figure 1D; Supplementary file 1C). In addition to ORF-Ts, many of the non-coding transcription events are regulated by the GLU to GAL shift. These include antisense transcription reads (to ORF-Ts), which increase in GAL compared to GLU (Figure 1E) and 25.7% of the 1772 annotated non-coding transcripts, the stable unannotated transcripts (SUTs), or cryptic unstable transcripts (CUTs), which show >threefold change in transcription on the GLU to GAL shift (Supplementary file 1E).
We used a computer simulation to obtain the expected number of occurrences of genome-wide features, by randomly shuffling genes, their orientation, YMC status, and expression levels (data from Supplementary file 1A, codes in Source code 1). Selected data from this analysis are shown in Supplementary file 1F. In summary, pairs of ORF-Ts in the convergent orientation occur more often than expected, while tandemly arranged ORF-Ts occur less often than expected. YMC genes are clustered, with 503 (16.9%) YMC genes flanked by non-cycling genes, 1431 (48.4%) with a single adjacent YMC partner either upstream or downstream and 1023 (34.5%) with a neighbouring cycling gene both upstream and downstream. Compared to more isolated genes, ORF-T clusters are more likely to show increased transcription in GLU or GAL than expected, suggesting that clustering is associated with some aspect of their regulation. In addition, ORF-Ts flanked by annotated SUTs or CUTs (in any orientation) occur more often than expected and show a twofold to fourfold higher median expression upon change from GLU to GAL than the average median expression obtained through the simulations. This is consistent with CUTs and SUTs playing a role in modulating ORF-T transcription, particularly on environmental change (Xu et al., 2011). Convergent overlapping arrangements occur three times more frequently than expected, whether the feature is a non-coding transcript (NC) such as a SUT or a CUT, or an ORF-T, with a median overlap of 92 bp for convergent ORF-Ts and 462 bp for ORF-Ts with non-coding transcripts. Thus clustering with overlapping transcription and transcriptional regulation are common features of the yeast genome.
As the genes whose transcription changed during the GLU/GAL shift are enriched for OX and RC YMC genes, we examined whether the nature of the flanking gene influences transcription, focusing on the tandem, and divergent combinations of these genes (Supplementary file 1F). OX genes, regardless of the nature (OX or RC) or orientation (divergent or tandem) of the upstream feature showed higher than expected transcriptional rates in GLU, but not in GAL, suggesting they are activated in GLU but not repressed in GAL. Similarly, RC genes are more likely to show higher than expected transcriptional rates in GAL. The behaviour of RC genes in GLU, however, does appear to be dependent on the orientation of the upstream OX gene. With a divergent upstream OX gene, the transcription of the downstream RC pair is not enriched or depleted in GLU. However, if the OX gene is in tandem, the RC gene is repressed in GLU. This raises the possibility that at the OX.RC tandem pair, transcription of the OX gene is mediating transcriptional interference and repressing the transcription of the RC gene in GLU. We note that the downstream gene in a tandem RC.RC pair also shows significant repression in glucose. To investigate further, we examined three different data sets: (i) three groups of tandemly arranged genes; OX.RC, RC.OX, and non-cycling (Supplementary file 1G); (ii) three sets of 20 consecutive ORF-Ts from Supplementary file 1A showing a sevenfold increase, no change or a sevenfold decrease on the GLU/GAL shift (Supplementary file 1H); and (iii) GLU/GAL-regulated genes with an annotated antisense CUT or SUT that is also GLU/GAL-regulated (Supplementary file 1I). From this analysis, we picked tandem gene clusters where the transcription of the target gene is regulated by GAL/GAL shift or not, and examined how neighbouring genes behave (Figure 1—figure supplements 1–4; Supplementary file 1H; Supplementary file 1J). The tandem partner of a GLU/GAL-regulated gene is often also regulated by the GLU/GAL shift and the YMC, while non-regulated genes are less likely to be in tandem and if they are, are surrounded by genes that are also not regulated by the GLU/GAL shift or the YMC. Common features of the regulated tandem clusters include reciprocal transcription in GLU and GAL, reciprocally cycling transcripts in the YMC, reciprocal AS transcription to OX genes, di-cistronic transcripts, and/or antisense transcripts spanning promoters that could mediate temporal transcriptional interference and thus facilitate cycling transcription. We chose the OX.RC pair HMS2:BAT2 for further study, as all the above characteristics are present and because the relatively abundant HMS2 antisense transcript, SUT650, can be detected experimentally in both GLU and GAL (Figure 1F,G).
We examined the relationship between HMS2 and BAT2 (transcripts A and D respectively). BAT2 transcripts peak in the RC phase of the YMC and in GAL, reciprocal to HMS2 sense transcripts, which peak in the OX phase of the YMC and in GLU (Figure 1F,G and Figure 2—figure supplement 1A–C). Interestingly, there is also a reciprocal relationship between HMS2 and its antisense SUT650 (transcripts A and B respectively). Dual-labelled RNA FISH revealed a degree of temporal separation in the production of the HMS2 and SUT650 transcripts. In multiple analyses, no examples of FISH signals for both HMS2 and SUT650 at the site of transcription in the nucleus are observed in either GLU or GAL. Cells with nascent transcripts (arrows, Figure 2A) often have the opposite transcript type or both SUT650 and HMS2 transcripts in the cytoplasm, suggesting they are changing state. 35.5 ± 7.9% of the cells contained both transcripts in GLU, 22.8 ± 10.5% of all cells lacked both the SUT650 and HMS2 sense transcripts. The remaining cells contained either HMS2 or SUT650 transcripts but not both (Figure 2A). For cells containing a single type of transcript, more cells contained antisense transcripts in GAL than GLU while more cells contained sense transcripts in GLU than GAL (Figure 2B). Thus, the increase in levels of SUT650 transcripts in GAL reflects an increase in the number of cells expressing SUT650 and a concomitant decrease in the number of cells expressing HMS2, rather than a change in the levels of transcripts within a fixed number of cells in the population. We conclude that the expression of HMS2 and SUT650 is temporally separated, similar to the PHO84 sense and antisense transcripts (Castelnuovo et al., 2013). This suggests potential co-regulation between BAT2 and SUT650 and reciprocal regulation of HMS2 and BAT2.
A number of different methods revealed heterogeneity at both the 5′ and 3′ ends of the HMS2 and SUT650 transcripts and longer RNA species (C1 and C2) on the HMS2 sense strand extending through the coding region of BAT2 and beyond, consistent with low-level read-through transcripts initiating at the HMS2 promoter (Figure 2C–E and Figure 2—figure supplement 1D,E). These read-through transcripts increase upon loss of nuclear exosome (rrp6Δ) (Figure 2D). Thus HMS2 is at the head of a contiguous gene cluster including the downstream genes BAT2 and YJR149W (Figure 2E). The 3′ UTR of the HMS2 transcript is of variable length. Polyadenylation (pA) occurs either 85 nt (proximal) or 426 nt (distal) downstream from the HMS2 stop codon. The distal pA site is 45 nt upstream of the BAT2 initiation codon, such that this longer transcription event is likely to interfere with initiation of BAT2 transcription. Additional transcriptional interference might be mediated through the production of the longer HMS2-BAT2 di-cistronic RNA species.
In summary, analysis of transcripts around HMS2 and BAT2 supports (i) an overlapping, reciprocally related and temporally segregated, sense–antisense pair (SAP) at HMS2, (ii) reciprocally expressed sense transcripts at HMS2 and BAT2, and (iii) low-level read-through di-cistronic transcripts (HMS2:BAT2 and SUT650:RPS4A). We ask if overlapping transcription in the sense and antisense orientations at HMS2 contributes to the reciprocal regulation of HMS2 and BAT2, by dissociating one transcription unit from the other.
Insertion of the ADH1 transcription terminator (ADH1t) into the middle to HMS2, to separate HMS2 transcription from BAT2, allows us to test directly whether the HMS2 sense transcription over the HMS2:BAT2 intergenic region represses BAT2 transcription (Figure 3A). We confirmed that ADH1t efficiently terminates the HMS2 transcript (Figure 3B,C) and that there are no detectable transcripts over the HMS2 3′ region (Figure 3D). The strain with ADH1t inserted showed a ≈twofold increase in BAT2 transcript levels compared to WT in both GLU and GAL (Figure 3B,C) supporting a role for HMS2 sense transcription in repressing the BAT2 promoter. We conclude that HMS2 sense transcription interferes with BAT2 transcription.
Insertion of the ADH1t into the HMS2 coding region resulted in a ≈fourfold increase in levels of a truncated HMS2 sense transcript (AA) (Figure 3A,B,D) and blocks SUT650 antisense transcription, resulting in the truncated antisense transcript (BA) (Figure 3A,D). We confirm there are no antisense transcripts over the 5′ region of HMS2 (Figure 3B,E). In the absence of SUT650, expression of the HMS2:ADH1t transcript remains sensitive to the change in carbon source, showing a similar decrease to full-length HMS2 1 h after transfer from GLU to GAL (Figure 3B,C). This suggests that the carbon source responsive signal is received at the HMS2 promoter and that reciprocal switching of HMS2 and SUT650 is a function of sense transcription. However, as the levels of HMS2 sense transcripts are higher without the antisense transcript, we asked if antisense transcription plays a role in limiting the number of cells containing sense transcript using RNA FISH (Figure 3F). When grown in glucose, optimal conditions for HMS2 sense transcription, we observed an increase in the number of HMS2:ADH1t cells containing sense transcript compared to WT (Figure 3G), consistent with a role for antisense transcription in reducing the number of sense transcription initiation events, leading to a decrease in the number of cells in the population that contain the sense transcript. We conclude that SUT650 antisense transcription does not influence the environmental responsiveness of HMS2 transcription but acts to limit the number of cells in the population that respond.
Next, we asked whether transcription of SUT650 antisense might prevent HMS2 sense transcription from interfering with BAT2, thus insulating BAT2. To ablate SUT650 whilst ensuring that the regulatory elements in the flanking sequences between HMS2 and BAT2 remained intact, we removed the transcription start sites (TSS) for SUT650 located between +953 and +1038 relative to the HMS2 ATG, near the end of the HMS2 coding region. We chose to do this by substituting the complete coding region of HMS2 with that of URA3, to ensure a natural ORF between the HMS2 promoter and terminator, rather than mutating the regions containing multiple TSSs for SUT650 within the ORF (Figure 4A). We note that insertion of the URA3 ORF could affect transcript stability, although the nature of the promoter may be the primary determinant of transcript stability (Bregman et al., 2011; Trcek et al., 2011), so we focused on transcript size and relative abundance. We observed two clear effects. First, loss of the antisense transcript (B), which could be explained either by introduction of a unidirectional terminator in the URA3 ORF coding region or loss of TSSs for SUT650 (Figure 4B–D and Figure 4—figure supplement 1). Second, an increase in the read-through transcripts (C1U) relative to the HMS2:URA3 sense transcript (AU) (Figure 4B–D). We confirmed these were read-through transcripts by inserting a terminator (T) at the 3′ region of the HMS2:URA3 hybrid gene (Figure 4A) and observing loss of the read-through transcripts and increased levels of a shorter transcript (AUT) (Figure 4E,F). We note that transcripts initiated at the HMS2 promoter remain sensitive to a change in carbon source in the absence of antisense transcription (Figure 4D), as we observed using HMS2:ADH1t (see Figure 3) and consistent with a regulatory input at the promoter. The increase in the read-through transcript (C1U) relative to the HMS2:URA3 sense transcript (AU) in the absence of antisense transcription suggests a role for antisense transcription in attenuating sense transcription and preventing interference at the BAT2 promoter. Consistent with this, a twofold decrease in BAT2 sense transcript is observed in the antisense-less HMS2:URA3 strain relative to WT in GLU, concomitant with a threefold increase in AU and C1U (Figure 4B,C). Taken together, the effect of blocking (Figure 3) or increasing (Figure 4) HMS2 transcription on levels of BAT2 transcript supports transcriptional interference by HMS2 at the BAT2 promoter and a reciprocal relationship between HMS2 and BAT2. We sought evidence for transcriptional interference resulting from increased levels of AU and C1,2U. Chromatin immunoprecipitation (ChIP) analysis at the HMS2:BAT2 intergenic region reveals reduced initiation-related modifications (H3K4me3 and H3K56ac) and increased elongation-related modifications (H3K79me3 and H3K36me3) in the HMS2:URA3 hybrid gene, consistent with transcriptional interference leading to repression of BAT2 transcription initiation (Figure 4G). We conclude that SUT650 antisense transcription (i) changes the proportion of cells expressing HMS2 and (ii) controls the amount of overlapping and read-through transcription from HMS2 and thus levels of BAT2 transcript. We envisage that at any one time in a single cell, formation of the HMS2 and SUT650 transcription units (TUs) is mutually exclusive, but the formation of the SUT650 TU and the BAT2 TU is not. Thus formation of the SUT650 transcription unit insulates BAT2 from interference by HMS2 transcription, as the HMS2 TU cannot form when SUT650 is expressed.
To address whether HMS2 transcription influences SUT650 antisense transcription, we replaced the HMS2 promoter with the inducible GAL1 promoter (pGAL1) (Figure 5A). When cultured in glucose, the GAL1 promoter is repressed but becomes induced about 2 h after transfer to galactose (Figure 5B). SUT650 is normally induced during the GLU-GAL shift (see Figure 1). However in the pGAL1:HMS2 strain, an antisense transcript (BG) is evident in GLU and remains until the sense transcript is induced by GAL (Figure 5C,D). This suggests that sense transcription represses the antisense transcript. We used the inducible pGAL1:HMS2 system in conjunction with the Anchor Away technique (Haruki et al., 2008) to deplete the nucleus of the Gal4p activator during growth in GAL. Removal of Gal4p by treatment with rapamycin (Rap) for 1 h, after activation of pGAL1:HMS2 for 3 h with GAL, abolishes the HMS2 transcript (AG) and leads to restoration of the antisense transcript (BG) (Figure 5E,F). This confirms that the reduction in the antisense transcript is a consequence of sense transcription, rather than the change in condition (GLU to GAL). To determine whether antisense transcription had any effect on pGAL1:HMS2 sense transcript induction rates, the antisense--less HMS2:URA3 construct was placed under the control of the GAL1 promoter (Figure 5A,B). The sense transcript profile in poly(A)+-enriched RNA was examined using a probe for the HMS2:BAT2 intergenic region (probe IG), a region transcribed in both constructs. No changes in induction rate were apparent, with the sense transcript only detectable after 2 h in both strains, signifying that antisense transcription does not impair sense activation kinetics (Figure 5B). Rather, this finding is compatible with our observation that without antisense, more cells would express the sense transcript (see Figure 3F,G). Activation of the pGAL1:HMS2 sense transcript results in up to 3.5-fold repression of BAT2 (Figure 5F), consistent with pGAL1:HMS2 sense polyadenylation at 426 nt, just upstream of the BAT2 ATG, interfering with BAT2 transcription initiation in galactose (see Figure 2—figure supplement 1). We conclude that formation of the HMS2 sense transcription unit represses the formation of the SUT650 and BAT2 transcription units, regardless of conditions.
There are two species of antisense transcript during induction of pGAL1:HMS2; a long species (BG1), dominant until 30 min after the shift to GAL, and a shorter species (BG2), also detectable after 30 min of growth in GAL and which becomes the dominant antisense transcript form after 1 h (Figure 5C). 3′end RACE revealed alternative polyadenylation sites as one cause for the observed size difference, with the long form terminating upstream of the Gal4p binding sites at approx. 500 nt upstream of the ATG in the GAL1 promoter and the shorter form terminating midway through the GAL1 promoter, 195 nt upstream of the GAL1 ATG, and downstream of the Gal4p binding sites (Figure 5—figure supplement 1). Upon glucose repression of induced pGAL1:HMS2, levels of the long form of antisense transcript were recovered (Figure 5G). Antisense transcript shortening after transfer to GAL is an inherent property of the natural SUT650 antisense transcript (Figure 5H). To ask if this was a function of transcription factor binding to the promoter, we used the pGAL1:HMS2 construct and showed that shortening of the antisense transcript is independent of Gal4 binding at pGAL1 and remains for 3 h in GAL in the absence of sense transcription (Figure 5I). This rules out a potential roadblock by Gal4 as an explanation for the use of different antisense polyadenylation/termination sites in pGAL1. We conclude that SUT650 antisense transcription is normally repressed by HMS2 transcription but remains stably expressed in GLU or GAL, varying in the use of polyadenylation sites, when sense transcription is inhibited.
To see if loss of transcription factors associated with the native HMS2 gene, Cbf1 (MacIsaac et al., 2006), and Fkh1 (Venters et al., 2011), influence the balance of sense and antisense transcription at HMS2, we used cbf1Δ, fkh1Δ, and fkh2Δ strains, regulators of nutrient availability, oxidative growth, and the cell cycle (Mitchell and Magasanik, 1984; Mellor et al., 1990; Kent et al., 1994; Zhu et al., 2000; Morillon et al., 2003; Tsankov et al., 2011). While under standard growth conditions in GLU these mutant strains showed normal expression of HMS2 and SUT650, growth in nutrient-limited GLU-based medium depleted for tryptophan resulted in loss of the HMS2 transcript and accumulation of the short isoform of the SUT650 antisense transcript (B2) in the cbf1Δ and fkh1Δ strains (Figure 6A). In contrast, the fkh2Δ strain showed reduced levels of SUT650 and increased HMS2 under these growth conditions. This supports transcription factors controlling the balance of sense, and thus antisense transcription, at this locus during the cell cycle (Fkh factors) (Granovskaia et al., 2010) and during oxidative growth in metabolic cycles (Cbf1) (Tsankov et al., 2011). We ablated a number of non-essential transcription factors, including Gln3, with putative binding sites at the 3′ region of HMS2 (MacIsaac et al., 2006) but observed no significant change in levels of SUT650, HMS2, or BAT2 transcripts with the exception of a small increase in the HMS2:BAT2 read-through transcript (C) in a gln3Δ strain in minimal medium (Figure 6—figure supplement 1). We propose that transcription factors signal to the HMS2 promoter (Input) to mediate state-changing (antisense-dominant to sense-dominant) in response to environmental conditions (Figure 6B). As we could find no direct evidence for regulation of SUT650 either by TFs (Figure 6 and Figure 6—figure supplement 1) or by the GLU to GAL shift (see Figure 5) but only by transcriptional interference resulting from activation of HMS2 transcription, we examined OX and RC genes genome-wide to look for common themes in their regulation. We asked what factors are significantly enriched (p < 0.01) at promoters of RC and OX genes, using a large data set of 202 factors (Venters et al., 2011). We observed marked differences (Supplementary file 1K). Of particular interest are the predominance of chromatin modulators and components at RC gene promoters, suggesting that RC promoters are particularly sensitive to chromatin-mediated repression (Tirosh and Barkai, 2008). These features, coupled with the distinct behaviour of OX and RC genes in GLU and GAL, particularly OX.RC genes in tandem (Supplementary file 1F), lead us to propose chromatin-mediated repression of RC genes by transcriptional interference. In addition, antisense transcription into OX promoters, for example by SUT650 at HMS2, may also limit OX gene transcription.
We assessed a role for a number of essential transcription factors using the Anchor Away technique (Haruki et al., 2008) but only upon nuclear depletion of TFIIB (Sua7) did we observe an increase in the HMS2:BAT2 read-through transcript (C) (Figure 6C). However, this occurred together with reduced levels of HMS2 (A), BAT2 (D), and SUT650 (B) transcripts, as expected upon depletion of an essential transcription factor. To explore this further, we used a well-characterised allele of SUA7 (sua7-1) (E62K) (Pinto et al., 1994), associated specifically with the 3′ region of genes, polyadenylation/transcription termination and higher order structures (gene loops) in chromatin (Medler et al., 2011). Remarkably, we observed an increase in HMS2 transcript (A) and sense read-through transcripts (C), coupled with a decrease in the SUT650 antisense transcript (B) (Figure 6D). Thus, TFIIB may have a specific role in determining the antisense transcription unit by establishing directionality of transcription over the region (Tan-Wong et al., 2012). For instance, by determining an antisense transcription unit, TFIIB aids in sense transcription termination, preventing ‘read-through’ over the intergenic region between HMS2 and BAT2. The capacity to read through transcription regulatory regions such as promoters and terminators is an essential component of a model for gene regulation by transcriptional interference such as that developed for the HMS2:BAT2 locus. To test this experimentally, we inserted a new transcription unit into HMS2 to disrupt the HMS2 and SUT650 transcription units and asked what mutations would restore transcription between the beginning and end of these TUs.
The HMS2 locus was engineered by insertion of the pTEF.KanMX.TEFt expression cassette (Longtine et al., 1998) at position +650 bp (Figure 7A). In this construct (HMS2:TEF:Kan), there are two promoters in tandem (HMS2 and TEF) and two terminators in tandem (TEF and HMS2). This is sufficient to disrupt the HMS2 and SUT650 transcription units. The SUT650 antisense promoter is functional and produces a truncated antisense transcript (BK) that ends in the vicinity of TEFt (Figure 7A,B). Thus TEFt functions bi-directionally to terminate SUT650 (BK), the sense transcript from pTEF, expressing the kanamycin resistance (KanMX) transcript (K), and the sense transcript (AK) initiated at the HMS2 promoter and terminating at TEFt (Figure 7A,B). pTEF is a well-characterised (Steiner and Philippsen, 1994) strong bi-directional promoter and produces an antisense transcript (E) extending into pHMS2. To confirm this, we inserted ADH1t directly upstream of pTEF (Figure 7A). This resulted in increased levels of a truncated sense transcript (AA) and considerably reduced levels of antisense transcript (E) (Figure 7B).
We reasoned that by introducing mutations into the TEF promoter, we would be able to ask how disabling pTEF affects the HMS2 and SUT650 transcription units. Using the HMS2:TEF:Kan construct as a starting point, mutations were introduced into the TEF promoter, including deletions of putative transcription factor binding sites (Figure 7C and Figure 7—figure supplement 1). A long deletion (HMS2.TEFdel1:Kan), removing all the TEF promoter sequences upstream of the TATA box, results in a significant reduction in expression of the KanMX cassette itself (transcript K) and in levels of the antisense transcript (E) (Figure 7D,E and Figure 7—figure supplement 1). Remarkably, in this strain, we observed a new antisense transcript from the 3′ end of HMS2 that reads through TEFt (BpTEFdel) (Figure 7C–E). We confirmed that the BpTEFdel transcript spans the 3′ region of HMS2 using the H3AS probe (Figure 7D). Thus, the disabled weak pTEFdel1 resulted in reduced functionality at TEFt allowing BpTEFdel, initiating at the SUT650 antisense promoter, to read through TEFt, the KanMX cassette and into the HMS2 promoter (Figure 7C–E). This experiment illustrates (i) the conditional nature of the bi-directional TEF terminator and (ii) the remarkable plasticity of interleaved transcription units. We suggest that the pTEF.KanMX.TEFt expression cassette is strongly expressed (compare levels of transcripts AK and K in Figure 7E), effectively preventing the formation of the HMS2:SUT650 transcription units. When the TEF promoter is disabled, reducing expression of the pTEF.KanMX.TEFt expression cassette, the SUT650 antisense transcription unit can form, resulting in the appearance of the long antisense transcript. Alternatively, reduced KanMX sense transcription decreases the ability of a terminator to stop antisense transcription from the SUT650 antisense promoter. This illustrates the complex relationships that control levels of expression in the interleaved yeast genome (Figure 7F).
We propose a two-state model for transcription at HMS2 and BAT2 that combines the effects of temporal separation, overlapping interfering transcription able to bypass typical termination and promoter signals, and insulation from interference by the formation of an antisense transcription unit (Figure 7G). There are two states: a sense-dominant state (blue transcripts) and an antisense-dominant state (orange transcripts) with respect to HMS2 (see Figures 6B and 7G). Transcription factors (Input) at the HMS2 promoter regulate sense transcription and loss of input results in the antisense-dominant state. SUT650 transcription insulates BAT2 from interference by HMS2, reinforcing the antisense dominant state-switch. We propose that transcription in the HMS2 sense orientation transmits the state-defining regulatory signals to neighbouring genes through mechanisms of cis-acting transcriptional interference, thus repressing BAT2, YJR149W, and SUT650. By these means, reciprocal regulation in cycles or on environmental change could be achieved.
More generally, how far a transcription event extends into the regulatory sequences of its SAP, or the regulatory sequences of an upstream or downstream gene may determine whether that region is competent to make a state-switch or not and respond to environmental or metabolic cues. Our genome-wide analysis supports many more examples of the type of organisation we observe at HMS2:BAT2. Moreover, the HMS2:BAT2 cluster is conserved in a range of yeast strains expected of a functional linkage. There are many loci at which di-cistronic read-through transcripts are evident (Pelechano et al., 2013), raising the possibility that transcriptional interference leading to reciprocal expression of genes in tandem arrays is a much more widespread phenomenon in yeast.
A di-cistronic transcript means that RNA polymerase has to read through the region determining polyadenylation and termination (the terminator). Moreover, antisense transcription arising from divergent promoters may have to bypass the terminator of the upstream gene. This work indicates the capacity of a terminator to function is conditional on the strength of the proximal promoter. We suggest that this reflects the role for promoters in defining the beginnings and ends of TUs (El Kaderi et al., 2009; O'Sullivan et al., 2004); stronger promoters preferentially selecting the proximal terminator.
The role of Sua7 in preventing read-through transcription is intriguing given its association with the 3′ region of genes, polyadenylation/transcription termination, and higher order structures in chromatin (Medler et al., 2011). This raises the possibility of an antisense transcription unit-specific function for TFIIB at the 3′ ends of genes, characterised here using the sua7-1 allele. A simple scenario in which 3′end TFIIB directs transcription by promoting both antisense transcription and physically terminating sense transcription is compatible with the interference/insulation model proposed here. It is also plausible that alternative higher order structures linking the sense or antisense promoters to one of a possible number of terminators could also explain dynamic state-switching between sense and antisense or multi-cistronic transcripts (Murray et al., 2012) (Figure 7G). Only one structure would exist in each cell at any one time leading to temporal separation of the different TUs. By this model only the juxtaposed regions would function in transcription initiation and termination, allowing other promoters or terminators between these regions to be ignored (Figure 7F,G). In either case, our data support the idea that the functions of these linked regulatory elements are coupled through the act of transcription. Thus, the interleaved architecture of the yeast genome not only lends itself to high degrees of transcriptional overlap, but more importantly, to an equally high degree of transcription unit interdependency by co-opting transcription itself as a regulatory mechanism.
While this study has been predominantly conducted using a single locus approach in yeast, the implications of this work could be far reaching. For instance, establishing wider, biologically relevant transcriptional landscapes through sense/antisense state-switching could be pertinent to aspects of chronobiology such as circadian rhythms (Kramer et al., 2003; Feng and Lazar, 2012), the mitotic/meiotic cell cycles (Morris and Vogt, 2010), and other developmental processes in higher eukaryotes. This work places the sense/antisense relationships in a broader biological context and sheds light on how transcription might work as the nexus in more extensive networks that are organized in both spatial and temporal dimensions.
All experiments were performed at least in duplicate to ensure that the trends observed were reproducible. Briefly, unless otherwise stated, cells were grown to exponential phase in rich medium (YP) supplemented with 2% glucose (YPD) or 2% galactose (YPGal). The strains of S. cerevisiae used for this study are shown in Supplementary file 1L. The Anchor Away technique exploits the high affinity interaction between the FK506 Binding Protein (FKBP12) and FKBP12 Rapamycin binding (FRB) domain of human mTOR protein in order to deplete target proteins from the nucleus in a time-dependent manner (Haruki et al., 2008). Control and treated cells were harvested after 1 h of DMSO or rapamycin exposure. Total RNA extracts were acquired by hot phenol extraction and enriched for polyadenylated transcripts using Oligotex‐dT as directed by the manufacturer (QIAGEN NV.). The concentration of RNA extract was measured and strandardised to 1 µg/μl using a nanodrop spectrophotometer. The protocol for rapid amplification of cDNA 3′ ends (3′ RACE) used in this study was modified from the methods provided by the 3′ RACE System for Rapid Amplification of cDNA Ends kit (Invitrogen Carlsbad, CA). Primers used are shown in Supplementary file 1M. Northern blotting was done as described in Murray et al. (2012). In vitro transcription with T7 RNA polymerase and radiolabelling was used to create strand-specific probes for Northern blotting. Primers are shown in Supplementary file 1N. Hybridization of RNA to strand-specific, high resolution tiling arrays was performed as described in Perocchi et al. (2007). Data analysis was performed by members of the Lars Steinmetz group at the European Molecular Biology Labs in Heidelberg, Germany (http://steinmetzlab.embl.de/cgi-bin/viewMellorLabArray.pl?showSamples=502_Glu1Vs503_Gal1&type=heatmap&gene=hms2). Native elongating transcript sequencing, (NET-seq) was done and analysed as described in Churchman and Weissman (2011). To assess the genome-wide significance of our observations, all genes in Supplementary file 1A were characterised according to their orientation with respect to their upstream (5P divergent, 5P tandem) or their downstream gene (3P, convergent), their own, and their neighbouring genes′ gene type (ORF, CUT, SUT, other) and YMC status (OX, oxidative; RB, reductive building; RC, reductive charging; NA/NC, non-cycling) and the number of genes in each category and median expression level in glucose and galactose determined. Genes that were located more than 1,000 bp from their neighbouring genes were excluded from the analysis. Computer simulations were performed by shuffling the strand location, gene type, YMC status, and expression levels of individual genes on every chromosome while keeping distances between genes and the total number of genes on each strand and of each type constant (Source code 1–MATLAB codes). Expression levels were simulated by randomly selecting recorded expression levels. After each simulation run, the number of genes and the median expression levels were calculated for each category. Simulations were run 10,000 times. Overlapping genes were defined as being positioned on the opposite strand and having 3′ ends covering the same region of the genome. Enrichment analysis for factors was performed using the published data set by Venters et al. (2011). The dual-labelled single molecule RNA Fluorescence In Situ hybridization (FISH) protocol was adapted from Zenklusen and Singer (2010). Fluorescently labelled DNA probes to detect sense or antisense transcripts are shown in Supplementary file 1O. Image acquisition was performed with a DeltaVision CORE: Wide-field fluorescence deconvolution imaging microscope. Using a 100x objective lens, 31 ‘z-stacks’ were collected. For each stack, an exposure time of 0.5 s for DAPI and 1 s for Cy3 (TRITC) and Cy5 filters was applied. Images were captured using a CoolSNAP HQ camera (Photometrics, Tucson, AZ). To facilitate image analysis, 3D data sets were compressed into 2D images by using a maximum projection, using hms2Δ cells to determine threshold signal for expression. Images displayed as 2D compressions of z-stacks 12–22 (11 stacks), which included most of the nucleus and cytoplasm. Error bars represent standard error of the mean. Chromatin immunoprecipitation (ChIP) was performed as described by Morillon et al. (2003, 2005). Primers used for RT qPCR are displayed in Supplementary file 1P.
Driving the cell cycle through metabolismAnnual Review of Cell and Developmental Biology 28:59–87.https://doi.org/10.1146/annurev-cellbio-092910-154010
Bimodal expression of PHO84 is modulated by early termination of antisense transcriptionNature Structural & Molecular Biology 20:851–858.https://doi.org/10.1038/nsmb.2598
Gene looping is conferred by activator-dependent interaction of transcription initiation and termination machineriesThe Journal of Biological Chemistry 284:25015–25025.https://doi.org/10.1074/jbc.M109.007948
Intergenic transcription causes repression by directing nucleosome assemblyGenes & Development 25:29–40.https://doi.org/10.1101/gad.1975011
Genome-wide transcription and the implications for genomic organizationNature Reviews. Genetics 8:413–423.https://doi.org/10.1038/nrg2083
Chromatin structure modulation in Saccharomyces cerevisiae by centromere and promoter factor 1Molecular and Cellular Biology 14:5229–5241.
A genomewide oscillation in transcription gates DNA replication and cell cycleProceedings of the National Academy of Sciences of USA 101:1200–1205.https://doi.org/10.1073/pnas.0306490101
Genomic gene clustering analysis of pathways in eukaryotesGenome Research 13:875–882.https://doi.org/10.1101/gr.737703
Evidence for a complex of transcription factor IIB with poly(A) polymerase and cleavage factor 1 subunits required for gene loopingThe Journal of Biological Chemistry 286:33709–33718.https://doi.org/10.1074/jbc.M110.193870
CPF1, a yeast protein which functions in centromeres and promotersThe EMBO Journal 9:4017–4026.
Regulation of glutamine-repressible gene products by the GLN3 function in Saccharomyces cerevisiaeMolecular and Cellular Biology 4:2758–2766.
Gene loops juxtapose promoters and terminators in yeastNature Genetics 36:1014–1018.https://doi.org/10.1038/ng1411
Antisense artifacts in transcriptome microarray experiments are resolved by actinomycin DNucleic Acids Research 35:e128.
Characterization of sua7 mutations defines a domain of TFIIB involved in transcription start site selection in yeastThe Journal of Biological Chemistry 269:30569–30573.
Metabolic cycling in single yeast cells from unsynchronized steady-state populations limited on glucose or phosphateProceedings of the National Academy of Sciences of USA 107:6946–6951.https://doi.org/10.1073/pnas.1002422107
Metabolic cycling without cell division cycling in respiring yeastProceedings of the National Academy of Sciences of USA 108:19090–19095.https://doi.org/10.1073/pnas.1116998108
Sequence and promoter analysis of the highly expressed TEF gene of the filamentous fungus Ashbya gossypiiMolecular & General Genetics 242:263–271.https://doi.org/10.1007/BF00280415
Integrative Genomics Viewer (IGV): high-performance genomics data visualization and explorationBriefings in Bioinformatics 14:178–192.https://doi.org/10.1093/bib/bbs017
Two strategies for gene regulation by promoter nucleosomesGenome Research 18:1084–1091.https://doi.org/10.1101/gr.076059.108
Metabolic cycles as an underlying basis of biological oscillationsNature Reviews. Molecular Cell Biology 7:696–701.https://doi.org/10.1038/nrm1980
Cyclic changes in metabolic state during the life of a yeast cellProceedings of the National Academy of Sciences of USA 104:16886–16891.https://doi.org/10.1073/pnas.0708365104
A role for noncoding transcription in activation of the yeast PHO5 geneProceedings of the National Academy of Sciences of USA 104:8011–8016.https://doi.org/10.1073/pnas.0702431104
Antisense expression increases gene expression variability and locus interdependencyMolecular Systems Biology 7:468.https://doi.org/10.1038/msb.2011.1
Joaquin M EspinosaReviewing Editor; Howard Hughes Medical Institute, University of Colorado, United States
eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.
Thank you for sending your work entitled “Transcription mediated insulation and interference direct gene cluster expression switches during biological rhythms” for consideration at eLife. Your article has been favourably evaluated by James Manley (senior editor), a Reviewing editor, and 3 reviewers.
After a thorough discussion, the consensus among the reviewers and the editors was that this is an excellent paper reporting some remarkable findings, which would be worthy of publication in eLife after addressing the concerns listed below.
The metabolic cycle has emerged as a fundamental biological phenomenon that is critical for proper cell function. It has been appreciated that oxidative conditions are in conflict with reducing reactions, and vice versa. Consequently, cells have evolved mechanisms to temporarily separate these conflicting conditions, by oscillating both the oxygen consumption as well as the expression of ∼3000 genes. The remarkable precision of the synchrony in maintaining oscillating levels of mRNAs must involve intracellular events and communication, the basis of which remains obscure. This paper provides some solutions to this enigma by demonstrating how transcription of tandemly arrayed genes can regulate a bi-modal switch so that activation of one gene would repress the other and vice versa. This paper will serve the scientific community in various aspects. First, it will drive the attention of the readers to the complexity of the YMC, an issue that is still underrepresented in the literature. Second, it illustrates the critical roles played by transcription terminators in both terminating transcription and initiating transcription in the anti-sense direction. Third, transcription of anti-sense (AS) as a key modulator of gene expression, in general, and its involvement in the bi-modal switch, in particular. Fourth, and most importantly, the paper demonstrates, in many ways, the intricate relationship between transcription of sense, anti-sense and neighbouring genes in a manner that supports the bimodal switch.
The major concerns raised by the reviewers are:
1) Generality of findings. The impact of this work would certainly be higher upon defining the prevalence of the key regulatory mechanisms described. The paper focuses on two tandemly arrayed genes (HMS2/BAT2) whose expression switches during the YMC. They provide a detailed mechanism of the switching whereby transcription read-through into the downstream promoter downregulates transcription of the downstream gene and how AS transcription is involved. Their conclusion that the sense and AS transcription are in conflict, and that cells tend to separate the “sense state” from the “AS state” temporarily is by itself remarkable. However, how general is this mechanism? What fraction of YMC-regulated genes is adjacent or anti-sense? More than expected by chance? And then what fraction of adjacent or anti-sense transcripts is co-regulated in the YMC? Moreover, if the mechanism is more prevalent than just HMS2/BAT2, do the anti-sense transcripts always fall into opposite YMC categories, i.e., oxidative vs reductive? Is this locus even representative of other glu-gal clusters? If 10 loci are chosen at random (based on having one strong glu-gal-regulated transcript), do the neighbouring genes show potent regulation like the one well-explored region? Reviewers agreed that a more global assessment of the prevalence of the mechanism described would raise the impact significantly. However, it was noted that the bioinformatics involved for a global analysis would not be trivial. Thus, reviewers request that the authors manually mine the NET-seq data regarding transcript isoforms (like the data shown in Figure 2–figure supplement 1D) and examine whether transcription of one gene invades the promoter of its neighbouring gene, and whether the AS transcription is anti-correlated. The control should be the same number of tandemly positioned gene pairs that are chosen at random. If the authors have the bioinformatic means to take a whole genome view, all the better.
2) Weak links to YMC and circadian rhythms. Although the reviewers appreciated the novelty and elegance of the molecular mechanisms discovered, there was agreement that in its current form the manuscript does not make a convincing argument about general relevance to YMC or circadian rhythms and it was agreed that the writing, including the title, should be tempered to focus more on the mechanistic aspects and less on the YMC and circadian biology. It is proposed that the title be changed to “Transcription mediated insulation and interference direct gene cluster expression switches”.
3) The choice of sua7-1 is not well communicated. The reader of the “Results” section is puzzled why TFIIB was chosen to begin, and why this specific allele was used, in particular. Only in the Discussion the reader learns that sua7-1 is defective in loop formation.
4) Figure 3B shows that the insertion of ADH1t abolished the signal detected by probe H2. From the point of view of the AS transcription, this probe is located downstream of ADH1t. The authors concluded that ADH1t blocks SUT650 AS transcription. However, it is possible that ADHt acts in both orientation and simply terminated transcription and therefore H2 could not detect any transcript. If the authors know that ADH1t does not act as terminator in the reverse orientation, they should indicate it. Moreover, even if the ADH1t is not acting as terminator of the AS, it can introduce a destabilization element that leads to its rapid degradation. In order to support this important conclusion, the authors could use a decay mutant strain (e.g., rrp6) and probe the membrane to detect the region between the ADHt and SUT650 promoter.
5) The replacement of HMS2 ORF with URA3 requires some clarifications. Why didn't the authors simply remove the region around the TSS, leaving all other sequences unchanged? Why did they prefer to introduce the entire URA3? There are two issues that need to be addressed. (1) The possibility that URA3 sequences, in both orientations, introduce stability biases onto the RNAs. (2) The possibility that the URA3 sequence contains a terminator in the AS direction, explaining why the AS RNA was not detected by H1 probe, located downstream of this hypothetical terminator. These issues can be discussed or addressed experimentally.
6) In Figure 6–figure supplement 1, the authors used TFs with putative binding sites at intergenic region between HMS2 and BAT2 (no references were included). The membranes were probed only with the H1 probe. This case might be an opportunity to examine if any of these factors play a role in the transcription of BAT2, by probing the membrane with an appropriate probe. URA3 ORF can be replaced back to HMS2 ORF by homologous recombination and selection on 5-FOA, thus resulting in a precise deletion of just the region around the TSS. Maybe one of these TFs is required for BAT2 transcriptional repression by C1 transcription (e.g., transcription through the binding site might displace the TF).https://doi.org/10.7554/eLife.03635.026
1) Generality of findings. The impact of this work would certainly be higher upon defining the prevalence of the key regulatory mechanisms described.
We agree with the reviewers about this point and had, in fact, already embarked on a computational genome-wide analysis. Part of this very large dataset has been analysed for this work and is presented in the revised version of the paper.
The paper focuses on two tandemly arrayed genes (HMS2/BAT2) whose expression switches during the YMC. They provide a detailed mechanism of the switching whereby transcription read-through into the downstream promoter downregulates transcription of the downstream gene and how AS transcription is involved. Their conclusion that the sense and AS transcription are in conflict, and that cells tend to separate the “sense state” from the "AS state" temporarily is by itself remarkable. However, how general is this mechanism?
From the point of switching sense and antisense states there is now one experiment in a recent paper from the Zenklusen lab supporting the idea of separation of the sense and antisense states in individual cells, in this case at PHO84 (Castelnuovo M, Rahman S, Guffanti E, Infantino V, Stutz F and Zenklusen D: Bimodal expression of PHO84 is modulated by early termination of antisense transcription. Nat Struct Mol Biol 20: 851-8). We also have as yet unpublished evidence that at GAL10, the sense and antisense transcripts are present in different cells during early induction. Given the FISH evidence for YMC regulated genes from the Singer and Botstein labs, we think that switching is likely to be common.
What fraction of YMC-regulated genes is adjacent or anti-sense? More than expected by chance? And then what fraction of adjacent or anti-sense transcripts is co-regulated in the YMC?
We have data in the revised manuscript addressing these issues (Tables S5 to S9). The antisense question is hard to address adequately due to the relatively poor annotation of the antisense transcriptome – being limited to CUTs and SUTs. There are 1772 annotated CUTs and SUTs but much more transcription antisense to genes is evident from the NET-seq analysis. A separate analysis suggests that 75% of all yeast genes are subject to some antisense transcription in the vicinity of the sense promoter that influences the chromatin organization around that promoter. Nevertheless, our analysis for this work suggests that YMC genes are much more likely to have antisense transcription, although the number of genes is relatively small because of the high thresholds (>3-fold change in GLU/GAL both the YMC gene and the SUT or CUT). 71 of the 206 genes (34.4%) with regulated sense and antisense have SUTs/CUTs whose transcription also changes 3-fold in the GLU/GAL shift. 26% of all SUTs and CUTs show a >3-fold change on the GLU/GAL shift. Thus regulated SUTs and CUTs are enriched opposite to ORFs that also show a > 3-fold shift.
Moreover, if the mechanism is more prevalent than just HMS2/BAT2, do the anti-sense transcripts always fall into opposite YMC categories, i.e., oxidative vs reductive?
Our selected gene study shows that for the 10 OX.RC tandem genes (like HMS2.BAT2), 9 (90%) of the OX genes have an antisense (annotated or not) that shows reciprocal transcription to the OX gene.
Is this locus even representative of other glu-gal clusters?
Yes, see Table S9 and Figure 1–figure supplements 1–4.
If 10 loci are chosen at random (based on having one strong glu-gal-regulated transcript), do the neighbouring genes show potent regulation like the one well-explored region?
Yes, see Table S9 and Figure 1–figure supplements 1–4.
Reviewers agreed that a more global assessment of the prevalence of the mechanism described would raise the impact significantly.
I hope that we have persuaded the reviewers that this is likely to be a more general phenomenon.
However, it was noted that the bioinformatics involved for a global analysis would not be trivial. Thus, reviewers request that the authors manually mine the NET-seq data regarding transcript isoforms (like the data shown in Figure 2–figure supplement 1D) and examine whether transcription of one gene invades the promoter of its neighbouring gene, and whether the AS transcription is anti-correlated. The control should be the same number of tandemly positioned gene pairs that are chosen at random. If the authors have the bioinformatic means to take a whole genome view, all the better.
We have used both approaches to provide the data in order to conclude this is a wider phenomenon.
2) Weak links to YMC and circadian rhythms. Although the reviewers appreciated the novelty and elegance of the molecular mechanisms discovered, there was agreement that in its current form the manuscript does not make a convincing argument about general relevance to YMC or circadian rhythms and it was agreed that the writing, including the title, should be tempered to focus more on the mechanistic aspects and less on the YMC and circadian biology. It is proposed that the title be changed to “Transcription mediated insulation and interference direct gene cluster expression switches”.
We are happy to do this and have removed this from the title and tempered our references to these mechanisms in the text.
3) The choice of sua7-1 is not well communicated. The reader of the “Results”" section is puzzled why TFIIB was chosen to begin, and why this specific allele was used, in particular. Only in the Discussion the reader learns that sua7-1 is defective in loop formation.
We agree with this point and have included more details in the Results section.
4) Figure 3B shows that the insertion of ADH1t abolished the signal detected by probe H2.
This is correct for the H2AS probe.
From the point of view of the AS transcription, this probe is located downstream of ADH1t.
This is correct.
The authors concluded that ADH1t blocks SUT650 AS transcription. However, it is possible that ADHt acts in both orientation and simply terminated transcription and therefore H2 could not detect any transcript.
This is correct.
If the authors know that ADH1t does not act as terminator in the reverse orientation, they should indicate it.
We do know that the ADH1t acts as a terminator in both directions. It does so in its native context (terminating the YOL086AW TU) and when used out of context in GAL10 (see Murray et al. 2012 NAR).
Moreover, even if the ADH1t is not acting as terminator of the AS, it can introduce a destabilization element that leads to its rapid degradation. In order to support this important conclusion, the authors could use a decay mutant strain (e.g., rrp6) and probe the membrane to detect the region between the ADHt and SUT650 promoter.
We have included two additional pieces of data in Figure 3 to address this. First we have data showing the effect of the decay mutant strain (rrp6) using probe H2AS. There are no unstable transcripts detectable in the rrp6 strain.
Second we use probe H3AS to show that SUT650 is prematurely terminated. Thus both HMS2 sense transcription and SUT650 antisense transcription are prematurely terminated in this strain.
5) The replacement of HMS2 ORF with URA3 requires some clarifications. Why didn't the authors simply remove the region around the TSS, leaving all other sequences unchanged? Why did they prefer to introduce the entire URA3?
We chose to completely replace the HMS2 ORF with the URA3 ORF so as to avoid problems with changing codons, deleting bases, or altering transcript stability while leaving the HMS2 flanking sequences unaltered.
There are two issues that need to be addressed. (1) The possibility that URA3 sequences, in both orientations, introduce stability biases onto the RNAs. (2) The possibility that the URA3 sequence contains a terminator in the AS direction, explaining why the AS RNA was not detected by H1 probe, located downstream of this hypothetical terminator. These issues can be discussed or addressed experimentally.
We cannot address this using experimental data as BY4741 is ura3Δ and thus the URA3 gene is not in our datasets to test for the presence of an early terminator. Given that the promoter of a gene is reported to determine sense RNA stability (Bregman A, Avraham-Kelbert M, Barkai O, Duek L, Guterman A and Choder M: Promoter elements regulate cytoplasmic mRNA decay. Cell 147: 1473-83; Trcek T, Larson DR, Moldon A, Query CC and Singer RH: Single-molecule mRNA decay measurements reveal promoter- regulated mRNA stability in yeast. Cell 147: 1484-97), we felt the URA3 ORF replacement was an appropriate strategy. We have discussed these potential problems in the text and how we have interpreted our data given these provisos.
6) In Figure 6–figure supplement 1, the authors used TFs with putative binding sites at intergenic region between HMS2 and BAT2 (no references were included). The membranes were probed only with the H1 probe. This case might be an opportunity to examine if any of these factors play a role in the transcription of BAT2, by probing the membrane with an appropriate probe. URA3 ORF can be replaced back to HMS2 ORF by homologous recombination and selection on 5-FOA, thus resulting in a precise deletion of just the region around the TSS. Maybe one of these TFs is required for BAT2 transcriptional repression by C1 transcription (e.g., transcription through the binding site might displace the TF).
We apologise for not including references to the putative TF binding sites – this is now done. Putative factor binding sites, and even ChIP data with tagged factors, does not provide evidence that a particular factor is functional at a particular site, for a variety of reasons (Lenstra TL, Benschop JJ, Kim T, Schulze JM, Brabers NA, Margaritis T, van de Pasch LA, van Heesch SA, Brok MO, Groot Koerkamp MJ, Ko CW, van Leenen D, Sameith K, van Hooff SR, Lijnzaad P, Kemmeren P, Hentrich T, Kobor MS, Buratowski S and Holstege FC: The specificity and topology of chromatin interaction pathways in yeast. Mol Cell 42: 536-49, 2011). Indeed, many of the experiments linking transcription factors to changes in particular transcriptional responses at genes are likely to be artefacts resulting simply from altered growth rate as a result of the genetic intervention (O'Duibhir E, Lijnzaad P, Benschop JJ, Lenstra TL, Leenen D van, Groot Koerkamp MJA, Margaritis T, Brok MO, Kemmeren P and FCP H: Cell cycle population effects in perturbation studies. Mol Syst Biol. 10: 2014; Slavov N, Airoldi EM, van Oudenaarden A and Botstein D: A conserved cell growth cycle can account for the environmental stress responses of divergent eukaryotes. Mol Biol Cell 23: 1986-97, 2013), (and the different times spent in the OX and RC phases of the YMC). We think this criticism applies to all the experiments we have done with TF KO strains in Figure 6 and Figure 6–figure supplement 1 and have been careful not to over-interpret the data to infer that these TFs actually bind to the DNA. Rather, the data in Figure 6A showing the switching of the balance of HMS2 sense and SUT650 antisense suggest that ablation of some factors coupled to mild nutrient limitation leads to cells where transcription in the HMS2:SUT650 region is predominantly in sense or antisense orientations. This provides evidence for reciprocal switching between HMS2 sense and SUT650 antisense and suggests a signalling input regulates this only at the HMS2 promoter. We have a number of experiments in the paper to support this, particularly those in Figure 5. In summary these experiments show that expression of the antisense state (SUT650 and BAT2) is constitutive in the absence of sense transcription, regardless of growth conditions (GLU or GAL). Normally SUT650 transcription increases on the GLU to GAL shift but this reflects the reduction in HMS2 sense transcription in GAL. Without sense, SUT650 levels remain high in GLU. Similar arguments apply to BAT2. Elimination of sense transcript in the pGAL1:HMS2 anchor away experiment in Figure 5, by addition of Rapamycin to deplete nuclear Gal4 from cells in GAL, restores BAT2 transcript levels. Thus the primary BAT2 regulation is transcriptional interference by the sense transcription in GAL (in this experiment only as the inserted GAL1 promoter is GAL-regulated).
Our view is that the (semi-) quiescent RC phase in the YMC is likely to be the default and RC genes would be expressed unless a specific growth signal is received. HMS2 switches on and interferes with the transcription of SUT650 and BAT2 as long as the growth signal remains. If there is no growth signal, transcription of BAT2 and SUT650 is constitutive, as if cells are in the quiescent state that occurs naturally when nutrients are limited. In fact all the changes that occur when yeast naturally enter stationary phase due to nutrient depletion also occur in RC cells (Shi L, Sutter BM, Ye X and Tu BP: Trehalose is a key determinant of the quiescent metabolic state that fuels cell cycle progression upon return to growth. Mol Biol Cell 21: 1982-90). This does not rule out TFs in the regulation of RC genes, but it does suggest they are not condition-specific. This is supported by our two new genome-wide analyses now included in the manuscript (Table S10). We have examined the nature of the transcription factors’ association with the promoters of RC or OX genes (in GLU) from the Venters and Pugh work (2011) (now included in reference list). They looked at the association of 202 factors at yeast promoters. We used their dataset to ask whether OX and RC genes are enriched for particular factors, as both will be expressed in GLU-grown cells. We show significantly different (p<0.01) ChIP-Chip binding patterns for factors at the two groups of genes. The RC genes are enriched for cohesion, the CCR4/NOT complex, elongator, chromatin and chromatin remodelling factors and four TFs, (Fhl1, Ino4, Ume6 and Skn7). By contrast, OX genes are associated with GTFs, TEFs, TFIID, and the TFs Ifh1 and Rap1. Although “stress” related transcription factors may play a role at the RC genes, these data suggest chromatin-mediated regulation as a central feature for RC genes. Our data support TI-mediated chromatin changes at the BAT2 promoter.
So to the question of whether the putative TF binding sites between HMS2 and BAT2 regulate BAT2 expression or more simply, contribute to limiting HMS2 read-through (Figure 6–figure supplement 1). As it stands, our data, albeit quite crude, provides no evidence for a role for these factors in activation of either HMS2 or SUT650. We rule out TFs as roadblocks (at least Gal4 for SUT650 transcribed into the GAL promoter) but do see a role for Gln3 (only in minimal medium) in preventing transcript C (HMS2:BAT2 read through) and a reduction in SUT650 levels (supporting increased TI at the HMS2:BAT2 intergenic region). Given the literature supports a role for TFs and promoters in determining transcript stability, another explanation for the effect of loss of Gln3 is that transcript C is stabilised, but there is no change in transcription per se and thus no change in the transcript ratios. How RC genes are regulated is clearly complicated and beyond the scope of what we can do for a 2 month revision. We think the suggestion of the referee is excellent but have a number of reservations about this type of approach, especially as some of the putative TF binding sites are within the HMS2 ORF (similar arguments to the URA3 question above). Instead we have provided an Anchor-Away experiment to deplete nuclear Gln3 in Galactose (conditions expected to increase BAT2 and SUT650 expression). In the same experiment nuclear depletion of Sua7 results in loss of both BAT2 and SUT650 transcripts. We find no evidence that Gln3 is required for BAT2 or SUT650 expression in these conditions and the Sua7 control suggests this is not simple a result of the BAT2 transcript being particularly stable in GAL. These data are now included in Figure 6–figure supplement 1.https://doi.org/10.7554/eLife.03635.027
- Jane Mellor
- Lars M Steinmetz
- Jane Mellor
- Jane Mellor
- Lars M Steinmetz
- Tania Nguyen
- Ana Serra Barros
- Harry Fischl
- Françoise S Howe
- David Brown
- Ronja Woloszczuk
- Struan C Murray
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
The authors would like to thank Ilan Davis, Micron Oxford, and Dan Larsen for help with RNA FISH, Athar Ansari, and Mike Hampsey for strains containing the sua7-1 allele and helpful advice, Stirling Churchman for help and advice with the NET-seq technique, Anitha Nair for excellent laboratory support, Benjamin Schuster-Böckler for advice regarding the computer simulations, and the following for funding; Keble College and the Clarendon Fund (to T.N.) and a Wellcome Trust Strategic Award (091911) supporting advanced microscopy at Micron Oxford (http://micronoxford.com)
- Joaquin M Espinosa, Howard Hughes Medical Institute, University of Colorado, United States
© 2014, Nguyen 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.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
To characterize how protein-protein interaction (PPI) networks change, we quantified the relative PPI abundance of 1.6 million protein pairs in the yeast Saccharomyces cerevisiae across nine growth conditions, with replication, for a total of 44 million measurements. Our multi-condition screen identified 13,764 pairwise PPIs, a threefold increase over PPIs identified in one condition. A few ‘immutable’ PPIs are present across all conditions, while most ‘mutable’ PPIs are rarely observed. Immutable PPIs aggregate into highly connected ‘core’ network modules, with most network remodeling occurring within a loosely connected ‘accessory’ module. Mutable PPIs are less likely to co-express, co-localize, and be explained by simple mass action kinetics, and more likely to contain proteins with intrinsically disordered regions, implying that environment-dependent association and binding is critical to cellular adaptation. Our results show that protein interactomes are larger than previously thought and contain highly dynamic regions that reorganize to drive or respond to cellular changes.
The way in which multidomain proteins fold has been a puzzling question for decades. Until now, the mechanisms and functions of domain interactions involved in multidomain protein folding have been obscure. Here, we develop structure-based models to investigate the folding and DNA-binding processes of the multidomain Y-family DNA polymerase IV (DPO4). We uncover shifts in folding mechanism among ordered domain-wise folding, backtracking folding, and cooperative folding, modulated by interdomain interactions. These lead to "U-shaped' folding kinetics. We characterize the effects of interdomain flexibility on the promotion of DPO4-DNA (un)binding, which probably contributes to the ability of DPO4 to bypass DNA lesions, a known biological role of Y-family polymerases. We suggest that the native topology of DPO4 leads to a trade-off between fast, stable folding and tight functional DNA binding. Our approach provides an effective way to quantitatively correlate the roles of protein interactions in conformational dynamics at the multidomain level.