Convergent transcription confers a bistable switch in Enterococcus faecalis conjugation

Abstract

Convergent gene pairs with head-to-head configurations are widespread in both eukaryotic and prokaryotic genomes and are speculated to be involved in gene regulation. Here we present a unique mechanism of gene regulation due to convergent transcription from the antagonistic prgX/prgQ operon in Enterococcus faecalis controlling conjugative transfer of the antibiotic resistance plasmid pCF10 from donor cells to recipient cells. Using mathematical modeling and experimentation, we demonstrate that convergent transcription in the prgX/prgQ operon endows the system with the properties of a robust genetic switch through premature termination of elongating transcripts due to collisions between RNA polymerases (RNAPs) transcribing from opposite directions and antisense regulation between complementary counter-transcripts. Evidence is provided for the presence of truncated RNAs resulting from convergent transcription from both the promoters that are capable of sense–antisense interactions. A mathematical model predicts that both RNAP collision and antisense regulation are essential for a robust bistable switch behavior in the control of conjugation initiation by prgX/prgQ operons. Moreover, given that convergent transcription is conserved across species, the mechanism of coupling RNAP collision and antisense interaction is likely to have a significant regulatory role in gene expression.

Convergent transcription from two opposing promoters of partially overlapping genes on opposite strands of DNA gives rise to transcripts with potential sense–antisense interactions in the overlapping region. Such convergent transcription is widespread in eukaryotes including the mouse (1), human (2), Drosophila melanogaster (3), and Arabidopsis thaliana (4) genomes and a number of prokaryotes such as Listeria monocytogenes (5), Mycoplasma pneumoniae (6), and Escherichia coli (7). We postulate that convergent, overlapping gene organization provides two-layered regulation by a combination of transcriptional interference and antisense regulation through RNA: RNA interactions between complementary transcripts (8).

Transcription from promoters of convergent overlapping genes gives rise to a finite probability that opposing elongating RNA polymerases (RNAPs) collide head-on, thus exerting a suppressive effect on transcription. Such suppressive influence of one transcriptional activity on a second transcriptional activity occurring in cis is referred to as transcriptional interference (9), as has been reported in prokaryotic (10, 11) and eukaryotic systems (12, 13). Furthermore, the two transcripts, having complementary sequence in the overlapping region, may exert antisense regulation through RNA: RNA interactions (8, 14). Given the frequent occurrence of sense–antisense transcripts in both prokaryotic and eukaryotic genomes, convergent transcription may play a significant role in gene regulation by both mechanisms. However, there have been few attempts to document the relative effect of each mechanism in control of transcription of overlapping, convergent genes.

The present study investigates the role of transcriptional interference arising due to RNAP collision and antisense regulation due to expression of complementary transcripts from operons prgQ and prgX, in plasmid pCF10 that regulates the conjugative transfer of antibiotic resistance between Enterococcus faecalis donor and recipient cells (15). Promoter P_Q drives expression of the prgQ operon, which encodes pheromone inhibitor peptide, iCF10 (AITLIFI) as well as the pCF10 conjugation machinery (16). Approximately 223 bp downstream of the prgQ start site, the convergent promoter P_X drives expression of the prgX operon which encodes repressor PrgX regulating P_Q transcription by binding to the operators XBS 1 and 2 of promoter P_Q (Fig. 1 A–C and Fig. S1 A and B)_. In addition to the full-length prgX transcript (denoted X), transcription from P_X also produces a 104-nt noncoding RNA, Anti-Q, by processing or termination within the 5′ end of the prgX transcript (17). The interaction of Anti-Q with nascent prgQ transcripts affects folding of prgQ RNA and results in a short 380-nt terminated Q_S transcript (at IRS1, Fig. 1A) that encodes iCF10, but is incapable of inducing conjugation (18).

Fig. 1.

The prgQ–prgX genetic locus controls conjugation of pCF10 plasmid in Enterococcus faecalis. (A) Convergent promoters P_Q (red) and P_X (green) drive expression of the antagonistic prgQ and prgX operons, respectively, with a 223-bp overlap. Transcripts (indicated by thick arrows) from the inducible P_Q promoter include 380 nt Q_S RNA terminating at an inverted repeat sequence (IRS1) and 530 nt Q_L RNA that results in expression of downstream conjugation-related genes. Transcripts from constitutive antisense P_X promoter include noncoding Anti-Q and full-length X RNA coding for PrgX protein. (B and C) Promoter P_Q in the repressed state (B) and the derepressed state (C). (D) Model for the prgQ–prgX genetic switch. Successful transcription gives rise to full-length Q and X RNA (>223 nt). Failed transcription, upon collision of converging RNAP, gives rise to various sizes of truncated RNA (<223 nt) indicated by dashed arrows. Some truncated RNAs (Q_AR and X_AR) have the potential to exert antisense interaction with full-length counter-transcripts (not to scale).

P_Q is normally in a repressed state whereas P_X is constitutive (Fig. 1B). The switch of P_Q between the repressed state (conjugation off, Fig. 1B) and the derepressed state (conjugation on, Fig. 1C) is controlled by competing interactions of PrgX with inhibitor iCF10 and an inducer, cCF10, a chromosomally coded heptapeptide (LVTLVFV) that is produced predominantly by recipient cells (16) because in donor cells, the pCF10-encoded PrgY protein sequesters endogenous cCF10 (19, 20). In the absence of recipient cells and thus low ratios of cCF10 to iCF10, iCF10-bound PrgX tetramers cause a DNA loop between XBS 1 and 2 sites, stabilizing a repressing complex at P_Q (Fig. 1B) (21, 22). In the derepressed state, the cCF10 imported from the environment via membrane protein PrgZ (23) displaces iCF10, resulting in the disruption of PrgX-iCF10 tetramers and the DNA loop (Fig. 1C). This disruption leads to increased levels of prgQ transcription, which serves to titrate the Anti-Q RNA, and results in expression of a longer 530-nt transcript called Q_L (Fig. 1A), capable of inducing conjugation by a post-transcriptional mechanism (24).

In this report we used a mathematical model to demonstrate that convergent transcription of the prgX/prgQ system plays a regulatory role in endowing the system with a bistable switch-like behavior, a characteristic of many robust biological switches, including regulation of competence in Bacillus subtilis (25), lysogeny and lytic state in bacteriophage λ (26), and during cell fate determination in Xenopus laevis oocytes (27). We provide experimental evidence that RNAP collision gives rise to shorter truncated transcripts that exert an antisense effect on transcripts of prgX and prgQ. Both transcriptional interference as a result of RNAP collision and antisense regulation are necessary for the prgX/prgQ genetic switch to function in a robust manner.

Results

Modeling the prgQ–prgX Genetic Network: Coupled Effect of RNAP Collision and Antisense Regulation.

A model for conjugation control by the prgX/prgQ operons is shown in Fig. 1D. The proposed model is based in part on previously described mechanisms (15, 18). We included additional regulatory effects contributed by transcriptional interference caused by RNAP collision and antisense effects caused by truncated RNA produced as a result of transcription interference. The underlying hypothesis is that RNAP collision causes failure in transcription, thus suppressing the transcription rate from both promoters. A second element is the generation of truncated transcripts that are released upon RNAP collision. We consider the hybridization of these short RNAs to their complementary counter-transcripts.

A two-part mathematical model was formulated to describe the mechanism and to quantitatively evaluate the behavior of transcription from P_Q and P_X in response to inducer cCF10 (9). The first part is a discrete model that enumerates the frequency of RNAP collision and the generation of truncated RNA (Figs. S2 and S3). The second part is an ordinary differential equation (ODE)-based mathematical model on the mass balance of the components of the pCF10 system (Fig. 1D, Fig. S4, and Tables S1–S6).

The discrete model simulated RNAP moving along the overlapping DNA between P_Q and P_X (SI Text and Fig. S2). Upon RNAP binding to the promoter, a delay of 10 s is allowed for formation of the elongation complex (28) before RNAP moves at a velocity of elongation at 50 bp/s, which has been reported as the average velocity of RNAP in E. coli in the presence of pausing (29). The maximal RNAP binding rate at a derepressed P_Q was set at 0.1/s on the basis of RNAP initiation rates estimated for the P_bla system in E. coli (30) and the PR-PRE system in bacteriophage λ (10). Using experimentally determined relative strengths of promoters P_Q and P_X under repressed and derepressed states in Fig. 2, the RNAP binding rates at P_Q and P_X under these two states were calculated. RNAP collision is considered to occur when converging RNAPs from P_Q and P_X are separated by 60 bp (RNAP footprint). This result takes into consideration occlusion of P_Q due to passage of RNAP originating from P_X and vice versa. Our experimental data suggested that RNAP collision did not result in complete abolition of transcription on the basis of a 10% level of X transcript observed upon derepression; we thus assume that 90% of RNAP collisions result in transcriptional termination (31), releasing short transcripts terminated at the point of collision (SI Text and Fig. S2). Movement of RNAP is not hindered by binding of PrgX to DNA (Fig. S5A). Successful passage of RNAP across the overlapping region (223 nt) without collision generates the full-length Q or X RNA. RNAP collision gives rise to a population of truncated RNA with varying length of <223 nt from P_Q and P_X (Fig. 1D). Previous work showed that stem loops at 46–51 nt and 80–83 nt of P_X transcript respectively interact with stem loops at 173–178 nt and 156–161 nt of P_Q transcript (Fig. S3) (32). On the basis of these data and in silico RNA structural analysis using Sfold (33), we postulate that truncated P_X RNAs (including Anti-Q RNA) between 104 nt and 223 nt in length and truncated Q RNA between 180 nt and 223 nt are capable of RNA:RNA interaction with their counter-transcripts Q and X, respectively, due to the presence of stem loops (Fig. S3). These subpopulations of truncated RNAs capable of antisense effects are collectively denoted as Q_AR (180–223 nt) and X_AR (104–223 nt) RNA from P_Q and P_X promoters, respectively (Fig. 1D).

Fig. 2.

Interference exerted on P_Q expression due to convergent transcription. (i) pCJ1 (OG1Sp), single P_X promoter construct; (ii) pBK1-25 (OG1Sp), single P_Q promoter construct in derepressed (D) state; (iii) pBK1-25 (100-5), single P_Q promoter construct in repressed (R) state; (iv) pBK1 (OG1Sp), convergent promoter construct measuring expression from derepressed P_Q in the presence of P_X; and (v) pBK1 (100-5), convergent promoter construct measuring expression from repressed P_Q in the presence of P_X. The lacZ mRNA expression levels (normalized to gyraseB) shown are an average of three independent qPCR experiments.

From the discrete model simulation the rates of generation of Q, X, Q_AR, and X_AR species were calculated (Fig. 3 A and B) for simulation using the ODE model (SI Text and Fig. S2). A set of eight ODEs describes the balance of RNA species Q_S, Q_L, X, X_AR and Q_AR, intracellular iCF10 and cCF10, and extracellular iCF10 (Eqs. S25–S32), considering the rate of production, degradation, and dilution due to volume expansion caused by growth as well as interactions among these components (SI Text). For RNA species, interaction between sense transcripts (Q and Q_AR) and antisense transcripts (X and X_AR) following second-order kinetics (32) was also considered (SI Text). Hybrid RNA duplex between X RNA and Q_AR is considered to be unavailable for translation (Fig. 1D) (8). The formation of RNA duplex between X_AR and nascent Q RNA gives rise to formation of Q_S RNA (18, 32). Nascent Q transcripts that do not terminate proceed to elongate and form Q_L. Translation of Q_S and Q_L leads to production of the secreted inhibitor peptide iCF10 (15). Transport of iCF10 and cCF10 across the membrane protein PrgZ is considered to follow first-order kinetics (23). The model assumes that due to cCF10-sequestering activity of PrgY protein, the effect of endogenous cCF10 in donor cells is negligible (20).

Fig. 3.

Mathematical modeling predicts RNAP collision and antisense regulation are required for bistability in the prgQ–prgX switch. (A and B) Frequency of various sizes of truncated and full-length RNA as a result of RNAP collision in the repressed (RNAP firing rates f_Q,R = 0.011 s⁻¹ from P_Q, and f_X= 0.012 s⁻¹ from P_X) and derepressed (f_Q,D = 0.1 s⁻¹ and f_X= 0.012 s⁻¹) state from P_Q (A) and P_X (B). (C) Steady-state level of Q_L at different cCF10 concentrations shows characteristic bistable switch behavior. (D) Effect of decoupling RNAP collision (RC) and antisense regulation (AR) on the steady-state response of Q_L to cCF10.

The ODE model was solved to obtain a steady-state solution to a fixed concentration of extracellular cCF10 (Fig. 3 C and D). The parameter values were obtained from the literature or experimental data (Tables S1 and S2). To test the sensitivity of the steady-state behavior to the parameter values, the parameters in the model were lumped into six dimensionless parameters (Tables S3 and S4) to reduce the model to a set of six nondimensional ODEs (Eqs. S48–S53). The values of three of these dimensionless numbers were determined from experimental data (SI Text), whereas the remaining three were obtained from literature and subjected to sensitivity analysis (Table S6). In the dimensionless model, the term corresponding to RNA interaction between truncated RNAs X_AR and Q_AR was eliminated as it did not affect the steady-state behavior of the system. Simultaneous parameter space search for the remaining three dimensionless parameters demonstrated that steady-state behavior shown in Fig. 3 C and D existed over at least two orders of magnitude (Fig. S4A and Table S6). The details of model derivation are described in SI Text.

P_Q Activity in Presence and Absence of Convergent Transcription.

The strength of P_Q in the presence (repressed state) and absence (derepressed state) of repressing PrgX complexes was quantified by measuring steady-state expression levels of the lacZ transcript from plasmids containing either a single P_Q promoter or both convergent promoters with the lacZ reporter fused downstream of IRS1 (Fig. 2). Two kinds of host strains were used: (i) OG1Sp, a wild-type host strain; and (ii) 100-5, a derivative of OG1Sp that provides PrgX in trans from the chromosome at levels similar to wild-type pCF10 plasmid (Fig. S1B). By combining the plasmids and host strains a set of five conditions was created as shown in Fig. 2. The intrinsic promoter strength of P_Q (derepressed state) in the absence of convergent transcription is 8.3-fold higher than that of the constitutive P_X (Fig. 2, ii vs. i). The repression of P_Q by supplying PrgX in trans caused a 9-fold reduction in expression in the absence of convergent transcription (Fig. 2, iii vs. ii). In the repressed state, P_Q has almost identical strength to P_X (Fig. 2, i vs. iii). In the derepressed state P_Q becomes the stronger promoter (Fig. 2, i vs. ii). Whereas convergent transcription from P_X caused only a modest (∼15%) reduction of transcripts from P_Q in the derepressed state (stronger P_Q than P_X; Fig. 2, ii vs. iv), it reduced P_Q transcript levels by ∼90-fold in the repressed state (similar strength of P_Q and P_X; Fig. 2, ii vs. v). These results show that convergent transcription greatly enhances the effect of repression on P_Q.

Model Simulation and Bistable Behavior.

The discrete model was used to predict the relative abundance of transcripts under repressed and derepressed conditions as a result of RNAP collision (Fig. 3 A and B and Tables S1 and S2). Under the repressed condition, a higher number of transcription events are predicted to give rise to full-length X RNA, compared with the depressed condition, whereas the opposite is predicted for full-length Q RNA. On the other hand, a broader distribution of Q_AR and X_AR RNA is predicted in the repressed state compared with the derepressed state. In the derepressed state, the 8.3-fold stronger P_Q not only increases Q transcription but also knocks off most RNAP initiated from P_X at loci proximal to the P_X promoter. As a result, RNAP collision results in only a moderate decrease in Q transcription, but a rather large decrease in X transcription.

Using the ODE model we next examined the effect of RNAP collision and antisense regulation on the steady-state level of Q_L in response to the signaling molecule cCF10. The level of Q_L RNA is an indicator of the state of conjugation, because it encodes genes of conjugation machinery. Steady-state solution of the model shown in Fig. 3C demonstrates a characteristic bistable response of Q_L to cCF10 concentration. Multiple steady states reside in the S-shaped section of the curve. Two are stable steady states corresponding to on (upper) and off (lower) states, respectively. The unstable steady state in the middle is not observed experimentally. For the system at an off state, as cCF10 concentration increases it moves along the lower stable steady state (Q_L < 1.4 nM) until cCF10 reaches 3.7 ng/mL, and then the system undergoes a sharp transition to an on state (conjugationally competent) corresponding to a high level of Q_L (>3.8 nM). Conversely if the system is initially at an on state, it remains at an on state till cCF10 concentration decreases below 3.2 ng/mL where it rapidly changes to an off state. The system is thus marked by well-separated on (cCF10 > 3.7 ng/mL) and off states (cCF10 < 3.2 ng/mL).

We evaluated the relative contribution of RNAP collision and antisense regulation to bistable behavior by eliminating RNAP collision (RC) or antisense regulation (AR) or both from the model (Fig. 3D). AR was eliminated from the model by setting the rate constant for RNA interaction to zero, whereas RC was eliminated by setting the transcription rates of Q and X RNA to RNAP firing rates from P_Q and P_X, respectively, and X_AR and Q_AR concentration to zero (Table S5). Bistability is observed only when both RC and AR are present (RC/AR). When AR is absent and RC is present (RC/AR⁻) or both RC and AR (RC⁻/AR⁻) are eliminated, no bistability is seen (Table S6). We also examined the case where RC is only partially removed by reducing the RNAP falling-off rate upon RNAP collision. As the falling-off rate decreases to 50%, the steady-state behavior transitions from a bistable to a ramp-like response (Fig. S4B). The results thus hinted that the 104-nt Anti-Q RNA, produced as a result of X RNA processing, alone is insufficient for the bistable behavior. This result is indeed the case for RC⁻/Anti-Q, where the bistable response to cCF10 is lost and a ramp-like transition is observed (Fig. 3D).

Inverse Expression and Response to Pheromone.

We next measured the levels of transcripts from P_X and P_Q promoters in strain OG1Sp carrying wild-type (WT) plasmid pCF10. Under repressed conditions the Q_L transcript level was much lower than the X transcript level even though the strengths of P_Q and P_X are similar. Upon induction with high levels of cCF10 (50 ng/mL), the Q_L transcript level increased >200-fold, whereas the X transcript decreased ∼9-fold (Fig. 4A). The opposite trend in their response to derepression is consistent with model predictions (Fig. 3C and Fig. S4C). The level of Q_L RNA observed was lower than that seen in the lacZ reporter system (Fig. 2). This result could be attributed to a high degradation rate for Q_L. This notion is consistent with model simulation, where a 10-fold decrease in Q_L degradation rate abolishes bistable behavior (Fig. S4D).

Fig. 4.

Inverse relationship of Q_L and X expression and presence of truncated RNAs in the prgQ–prgX locus. (A) Q_L and X RNA levels measured by qPCR using X-specific and Q_L-specific primers from OG1Sp (pCF10) under repressed and derepressed conditions. (B) Levels of Q_L RNA in response to different concentrations of cCF10. Data shown in A and B are averages of three independent experiments (error bars are SDs from mean values). (C) (Upper) Map for the single-stranded RNA probes Q1, Q2, X1 and X2 used for Northern blotting experiments. (Lower) Northern blots to detect transcripts from P_Q and P_X in cells carrying WT pCF10, in either the derepressed (D) or the repressed (R) state. Probes and lanes: Q1, lanes 1 and 2; Q2, lanes 3 and 4; X1, lanes 5 and 6; and X2, lanes 7 and 8. (D) Quantitative estimation of intensity of bands for truncated P_Q and P_X transcripts in lanes 1 and 2 and 5 and 6 normalized to intensity of Q_S + Q_L and X, respectively, using ImageJ software (National Institutes of Health).

A dose/response curve of Q_L transcript to cCF10 showed that Q_L expression remains low below 0.5 ng/mL of cCF10, followed by an 8.3-fold increase at 5 ng/mL of cCF10 (Fig. 4B and Fig. S4E), characteristic of a switch from the off to the on state as predicted by the model (Fig. 3C). Flow cytometric data using GFP protein expression as an indicator of on state confirm that the system transitions from an off to an on state between 0.5 and 5 ng/mL (Fig. S4F).

Truncated RNA from P_Q and P_X.

To verify the existence of the truncated transcripts (X_AR and Q_AR) in pCF10 carrying donor (OG1Sp) cells, we used Northern blots with sets of probes complementary to the 5′ (Q1 and X1) and 3′ segments (Q2 and X2) within the overlapping region of the respective RNAs from each promoter (Fig. 4C, Upper). Shorter (truncated) P_Q (Fig. 4C, lanes 1–4) and P_X (Fig. 4C, lanes 5–8) transcripts within the overlapping region were indeed observed under both repressed and derepressed conditions. The Q1 probe detected truncated transcripts of ∼100-, 150-, and 200-nt sizes, with higher relative abundance in repressed cells (Fig. 4C, lane 2 vs. lane 1, and Fig. S5C). Similarly, the X1 probe detected truncated RNAs in the size range between 80 and 200 nt (Fig. 4C, lanes 5 and 6). The prominent band of ∼100 nt corresponds to the 104 nt Anti-Q RNA reported previously (17). The levels of truncated transcripts detected by probes Q1 and X1 (Fig. 4C, lanes 1 and 2 and 5 and 6) proximal to the 5′ end are higher than those detected by distal probes Q2 and X2 (Fig. 4C, lanes 3 and 4 and 7 and 8), consistent with the higher collision frequency near the 5′ end. The presence of a short 100-nt band detected by the X2 probe (Fig. 4C, lane 8) indicates an additional mechanism involved in processing of longer P_X transcripts. However, this truncated transcript is unlikely to have a regulatory role in the switch behavior, due to the absence of stem loops I and II required for antisense interaction with P_Q transcripts.

The high abundance of truncated RNA in the repressed state (Fig. 4C, lane 2 vs. 1 and lane 6 vs. 5, and Fig. S5C) is consistent with the prediction of the RNAP collision model (Fig. 3 A and B). The size of shorter RNAs observed falls within the range of Q_AR and X_AR considered in the model. However, it is possible that these short RNAs arise from degradation of full-length transcripts. In this case, the ratio of truncated RNA to full-length RNA would be similar under derepressed and repressed conditions. Conversely if these RNAs are the products of RNAP collision, we expect this ratio to increase for P_X transcripts in the derepressed state as X RNA decreases, whereas the ratio corresponding to P_Q transcripts should decrease because of a dramatic increase in Q (sum of Q_S + Q_L) transcripts upon induction. Quantification of Northerns in Fig. 4C (lanes 1 and 2 and 5 and 6) indeed shows that the ratio of truncated to full-length RNA increases for P_X transcripts and decreases for P_Q transcripts in the derepressed state compared with the repressed state (Fig. 4D).

Decoupling RNAP Collision and Antisense Regulation.

The relative contribution of RNAP collision and antisense regulation to switch response was evaluated by decoupling these effects using single or convergent promoter constructs in host strains OG1Sp and 100-5 for derepressed and repressed conditions, respectively (Fig. 5A). Three cases were analyzed: convergent transcription (RC/AR), no RNAP collision in absence of antisense RNA (RC⁻/AR⁻), and presence of antisense RNA supplied in trans (RC⁻/AR) using plasmid constructs shown in Fig. 5A.

Fig. 5.

Decoupling the effects of RNAP collision (RC) and antisense regulation (AR) in the prgQ–prgX locus. (A) Plasmid constructs containing either P_Q (red) or P_X (green) or both promoters used to decouple RC and AR effects (lanes indicated). Derepressed vs. repressed states are compared using strains OG1Sp (derepressed) vs. 100-5 (repressed). Antisense interactions in the absence of RC were assessed by supplying complementary RNAs in trans using the plasmids diagrammed on the Right. (B) (Upper) Transcripts from P_X probed with X1. (Lower) Transcripts from P_Q probed with Q1. Lanes and corresponding constructs/host strains used are as indicated in A.

Transcription from P_X in the presence of convergent transcription (RC/AR) showed abundant truncated RNA from a repressed P_Q compared with a derepressed P_Q (Fig. 5B, lane 2 vs. lane 1). The longer (>700 nt), likely X transcript, was more abundant in the repressed condition, consistent with our model prediction (Fig. 3B and Fig. S4C). The distribution of transcripts was somewhat different from that in Fig. 4C because of different combinations of host cells and plasmid constructs.

The absence of convergent transcription (RC⁻/AR⁻) resulted in increased expression of both Anti-Q RNA (∼100-nt band) and longer X transcripts (Fig. 5B, lane 3), compared with the case with convergent transcription (lane 1), consistent with the negative effects of RNAP collision and antisense regulation on P_X expression. A 200-nt band observed in the RC⁻/AR⁻ case was not observed in the RC/AR case. This result could be caused by the presence of intrinsic pause sites, which were identified during in vitro transcription assays using conditions of RC⁻/AR⁻ (Fig. S5B). It is possible that in the RC/AR case, converging RNAP from P_Q knocks off elongating RNAP from P_X before reaching this locus.

The antisense effect (RC⁻/AR) was restored (Fig. 5B, lane 4) by providing excess amounts of shorter 253-nt RNA from P_Q in trans from a high copy number plasmid to the construct used in lane 3. The 253-nt RNA species contained the sequence of the overlapping region and was used to represent Q_AR species in the mathematical model. Compared with the RC⁻/AR⁻ case (Fig. 5B, lane 3), providing excess antisense Q_AR RNA reduced the expression of X and Anti-Q transcripts significantly. However, the extent of suppression of X (>700 nt) was lower than that observed in the RC/AR case, as the X transcript was still present even when large amounts of Q_AR RNA were supplied (Fig. S5D). Comparison of the intensity of X transcripts normalized to 5s RNA shown in lanes 3 (∼0.9) and 4 (∼0.3) (Fig. S5D) indicates that RNAP collision is responsible for two-thirds of the decrease of X transcription in the derepressed state, exerting a stronger effect on X transcription than antisense regulation. These results suggest that both RNAP collision and antisense regulation are required to achieve high degree of suppression of the X transcript upon derepression, although RNAP collision plays a more dominating role as predicted by the model (Fig. 3B). The presence of shorter RNA (100–223 nt) in lane 4 compared with lane 3 in Fig. 5B suggests that these are products of endonucleolytic cleavage within paired regions formed by the “kissing” interactions between the loop structures of Q_AR and P_X transcripts (Fig. S3) (32).

Similarly, during convergent transcription Q_L was expressed at higher levels in the depressed state than in the repressed state (Fig. 5B, lane 5 vs. lane 6), whereas higher abundance of truncated RNA was seen under the repressed state (Fig. 5B, lane 6). In the absence of RNAP collision and antisense regulation, longer Q_L and Q_S RNAs were expressed from P_Q, even under repressed state (Fig. 5B, lane 7, RC⁻/AR⁻). The lack of truncated RNA in this case in the absence of RNAP collision (Fig. 5B, lane 7) is consistent with our model that these RNAs are mainly a result of convergent transcription. When we provided excess amounts of antisense 104 nt Anti-Q RNA, representing X_AR species in the model, in trans from a high copy number plasmid (Fig. 5B, lane 8), Q_L expression decreased (Fig. 5B, lane 7 vs. lane 8); however, it was visible at longer exposures. Importantly, Q_L RNA was expressed in RC⁻/AR⁻ and RC⁻/AR cases even under repressed conditions. Intensity of bands corresponding to both X (Fig. 5B, lanes 1–4) and Q_L RNA in (Fig. 5B, lanes 5–8) normalized to 5s RNA shows that the high degree of suppression of X in the derepressed state (Fig. 5B, lane 1) and Q_L in the repressed state (Fig. 5B, lane 6) occurs only during convergent transcription (Fig. S5D). These results are consistent with model prediction.

Discussion

Our results demonstrate that a head-to-head convergent promoter system can accomplish two levels of regulation: (i) enhancing the degree of repression through RNAP collision and (ii) suppressing transcript levels through interactions with cis-encoded antisense RNA produced from the overlapping region. With RNAP collision alone, an 8.3-fold increase in P_Q strength is predicted to give rise to a 12-fold increase in successful transcription from P_Q (Fig. 3A) relative to the repressed state. Experimentally a much greater difference in Q_L transcript level was observed using both a reporter system (Fig. 2) and direct measurement of Q_L transcripts (Fig. 4A). This enhanced difference is accomplished by a greater degree of suppression of Q_L transcript level under the repressed conditions through the products of RNAP collision, namely truncated RNAs that have an antisense regulatory effect on transcripts from both P_Q and P_X. Using a combination of host strains and constructs to mimic conditions of RC and AR, we show that Q_L transcript level that was suppressed under repressed conditions with RC/AR became elevated when RC or both RC and AR were eliminated (Fig. 5B). Providing constant levels of Anti-Q RNA did not restore switch behavior, as demonstrated both experimentally (Fig. 5B, lane 8) and theoretically (Fig. 3D). The sharp difference in the level of Anti-Q/X_AR and Q species between the repressed and the derepressed state due to RC and AR contributes to the switch behavior.

Both mathematical modeling (Fig. 3 A and B) and experimental observations (Fig. 4C) indicate the presence of certain prominent sizes of truncated RNA. However, the stochasticity of RNAP firing from promoters is expected to blur the discrete nature of truncated RNAs, pointing toward the presence of potential RNAP pause sites within overlapping DNA, thus increasing the probability of collision at those sites. Such an effect will further strengthen the bistability due to enhanced effects of RNAP collision and antisense regulation (10).

Many biological switches have been shown to exhibit bistability in steady-state analysis of their system models (34, 35). The pivotal role of RC and AR in affording a bistable switch behavior is predicted by our model (Fig. 3 C and D). Bistability allows for a binary behavior. Because conjugation is an energy-intensive process, the donor cells must turn their conjugation system on only in response to true signal. A switch type of regulatory mechanism allows for a tighter control of conjugation.

Transcription interference has been reported to exert a suppressive effect on transcription from the weaker promoter in pR-pL and PR-PRE promoter pairs of bacteriophages 186 and λ, respectively (10, 11). In these studies, RNAP collision was the major contributor to the observed interference whereas antisense effects were concluded to be insignificant. Convergent transcription in the IME4 locus of Saccharomyces cerevisiae was shown to result in antisense and sense transcripts that were inversely related between diploid and haploid cell types (36). Again, in this case, the inverse relationship was ascribed to transcriptional interference but not to antisense regulation. Other convergent transcription systems have also been reported to have inverse transcript levels under different conditions, including the ecsit gene involved in mouse stem cell development (37) and the mogR locus of L. monocytogenes controlling cell motility (6). In a study of transcription of the Pot2 locus in rice blast fungus, truncated transcripts were observed that led to the identification of an antisense promoter (38). In silico analysis on the mouse genome and a follow-up study using an oligo-microarray revealed nearly 2,000 sense–antisense transcripts (39). The presence of multiple-sized transcripts without poly(A) appeared to hint at the presence of truncated transcripts.

The fact that convergent transcription is ubiquitous and has persisted in evolution (40) is perhaps an indication that such gene organizations confer fundamental mechanisms of gene regulation. We show that using a single repressor, convergent transcription allows for simultaneous regulation of two genes, giving rise to a complex, tightly regulated switch behavior. Critical to conferring the sophisticated control is the combined use of RNAP collision and antisense regulation. The collision frequency and the level of transcription suppression are highly dependent on many factors, including relative promoter strength, the length of overlap, and the exact sequence that causes RNAP pausing and interaction. With such a wide range of possible outcomes, using subtle structural tuning, convergent transcription may be highly adaptable to become a robust controller for many complex cellular events.

Materials and Methods

Details for strains, reporter constructs, medium, and growth conditions are provided in SI Text. Details for Northern blotting and qPCR are provided in SI Text. The numerical solution to stiff differential equations for steady-state analysis was obtained using the MATLAB function solve. For scanning parameter space, the MATLAB function polyxpoly was used for checking multiple steady states.