Transcription Analyses with Heteroduplex trp Attenuator Templates Indicate That the Transcript Stem and Loop Structure Serves as the Termination Signal*

DNA sequences that control transcriptional termination by prokaryotic RNA polymerases normally con- tain an inverted repeat, or self-complementary se- quence, about 10 base pairs upstream from the site of RNA chain termination. Point mutations that interrupt this self-complementarity can reduce or eliminate RNA chain termination. We have constructed heteroduplex DNA templates using wild type and mutant attenuators for the Escherichia coli trp operon to probe the relative contributions of the two DNA strands in the termination process. Transcription analyses show that only the sequences in the transcribed DNA strand determine whether or not a heteroduplex terminator can function. This result strongly supports a model in which the formation of a stem and loop structure in the nascent RNA transcript is the signal for transcriptional termi- nation. prokaryotic eukaryotic genomes Terminator sites most often are used to restrict transcription to defined genetic units. However, many such sites can also act conditionally to allow either termination or transcriptional readthrough and are termed attenuator sites (3, 4). The bio-chemical mechanisms by which DNA sequences at termination sites act to stop transcription or to permit readthrough are not well understood. Bacterial termination sites show certain sequence similarities 30-40 base pairs upstream from the base that determines the 3’OH terminus of the completed transcript (defined here as base -1). In particular, the RNA sequences from -1 to -9 are often rich in uridine residues, for strong terminators that are effective in vitro even in the absence of accessory termination factors. Addi-tionally,

DNA sequences that control transcriptional termination by prokaryotic RNA polymerases normally contain an inverted repeat, or self-complementary sequence, about 10 base pairs upstream from the site of RNA chain termination. Point mutations that interrupt this self-complementarity can reduce or eliminate RNA chain termination. We have constructed heteroduplex DNA templates using wild type and mutant attenuators for the Escherichia coli trp operon to probe the relative contributions of the two DNA strands in the termination process. Transcription analyses show that only the sequences in the transcribed DNA strand determine whether or not a heteroduplex terminator can function. This result strongly supports a model in which the formation of a stem and loop structure in the nascent RNA transcript is the signal for transcriptional termination.
Transcriptional termination sites are important regulatory loci in both prokaryotic and eukaryotic genomes (1-4). Terminator sites most often are used to restrict transcription to defined genetic units. However, many such sites can also act conditionally to allow either termination or transcriptional readthrough and are termed attenuator sites (3, 4). The biochemical mechanisms by which DNA sequences at termination sites act to stop transcription or to permit readthrough are not well understood. Bacterial termination sites show certain sequence similarities 30-40 base pairs upstream from the base that determines the 3'OH terminus of the completed transcript (defined here as base -1). In particular, the RNA sequences from -1 to -9 are often rich in uridine residues, especially for strong terminators that are effective in vitro even in the absence of accessory termination factors. Additionally, the DNA sequences from about -10 to -40 normally contain an inverted repeat sequence from 3 to 12 bases in length (1-4). Since these sequences are self-complementary in the RNA transcript, they can interact to form a duplex stemloop structure in either the RNA or either DNA strand. The inverted repeat sequences at a terminator play an essential * This work was supported by National Institutes of General Medical Sciences Research Grant GM12010. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
$ To whom reprint requests should be sent.
role in termination since point mutations in either arm of the repeat which interrupt the self-complementarity of the sequences can reduce or eliminate termination at that site (2, 3). Similarly, mutations that alter the AT-rich sequence proximal to the termination site can also reduce or eliminate termination (5, 6). Many workers have adopted the ad hoc hypothesis that the critical step in termination involves formation of a stem-loop structure in the nascent RNA transcript which acts somehow to prevent further RNA chain elongation by the RNA polymerase in the DNA-RNA complex, and indirect evidence from several kinds of experiments is consistent with this view (3-5). However, there is no direct evidence that an RNA duplex structure is involved. Elongation by RNA polymerase is unaffected by addition of single-stranded or duplex RNA (7,8) or polynucleotide analogues (9), and termination is not eliminated in high salt concentrations which would be expected to eliminate binding of a duplex structure to most protein binding sites (10). Furthermore, models can be designed in which elongation is blocked due to formation of a stem-loop structure in the nontranscribed DNA strand, or through formation of a DNA-RNA complementary duplex between the nascent RNA and the complementary region of the nontranscribed DNA strand (11). These latter two models can be distinguished from the RNA duplex model since they require a role for the DNA sequences in the nontranscribed DNA strand, while the RNA duplex model does not.
We have devised an experiment to test the relative role of sequences in the two DNA strands in transcriptional chain termination, using point mutations in the Escherichia coli trp attenuator sequence that disrupt the complementarity of the inverted repeat sequence, and permit transcriptional readthrough in vitro (3). We construct heteroduplex templates in which the mutant terminator sequence is in one strand and the wild type terminator sequence is in the other (see diagram in Fig. 1). Transcription of both of the two possible heteroduplex templates should reveal whether one of the two DNA strands plays a dominant role in the process and, if so, which one.
We have employed three E. coli trp attenuator mutans, isolated and generously provided to us by Charles Yanofsky and his co-workers (12, 13). trp L153 is a GC --., AT transition at position 132 in the trp leader which alters the arm of the repeat sequence nearest to the termination site (Table I). It reduces termination in vitro from 95 to 26%. trp L29 and L75 are two mutations earlier in the leader sequence which have no effect on in vitro termination. They were chosen as controls for the possible effects of heteroduplex sites on transcription.  John Clark, University of California, Berkeley, CA) and then transduced to str" using a PI C1 g lysate prepared on E. coli strain W3110str". Transductions and matings were carried out by standard microbiological procedures (14).
An EcoRI fragment from pVH153 bearing a segment of the trp operon from the promoter to the trpC gene (15) was cloned into M13mp2 using procedures developed by J. Messing (16). After ligation of an EcoRI digest of pVH153 into EcoRI cut M13mp2 replicative form, the mixture was used to transform E. coli 71-18, giving from 5-10% white plaques on Xgal plates. A plate stock of M13 was prepared from the transformed cells and was heated at 70 "C for 10 min. This stock was used to infect E. coli F'amp ALD102 str" (multiplicity of infection = 0.1). and trp' str" ampR colonies were selected on minimal medium. Phage-producing colonies were characterized. M13mp2 trp ' (+), bearing the trp' RI fragment in the orientation with the transcribed strand in the viral sequence, was identified by its ability to hybridize to 3H-labeled trp mRNA (14) and by restriction mapping of the replicative form duplex. Strain M13mp2 trp' (-), bearing trp' in the opposite orientation, was characterized by its ability to hybridize to M13trp+ (+) (17) and by restriction mapping of the replicative form. Large amounts of both phage were prepared by growth of infected Famp ALD102 strR in minimal glucose medium (M9) overnight. Phage were purified by two precipitations with 4% polyethylene glycol in 0.5 M NaCl followed by CsCl equilibrium sedimentation (16).
Heteroduplex molecules containing a trp mutant in one strand and trp' in the other were prepared by two different methods. In Method A, plasmid DNA (50 pg) bearing a mutant trp RJ fragment was cleaved with EcoRI and then mixed with 100 pg of MI3 mp2trp+ (+) or M13 mpZtrp+ (-) in a final volume of 300 pl. The mixture was diluted with Hz0 (500 p1) and EDTA (100 pl, 0.25 M), and NaOH (100 pl, 1 M) was added to denature the duplex DNA. After 10 min at 20 "C, the solution was neutralized with 1 M Tris, pH 7.2 (200 pl), and 1.2 ml of 99% formamide was added. Hybridization was allowed to proceed for 4 h at 25 "C, then nucleic acids were precipitated with 0.5 M NaCl and 2.5 volumes of ethanol, and the products were dissolved in 0.01 M Tris, pH 8, 0.001 M EDTA (TE buffer). A sample was analyzed by electrophoresis on -0.7% agarose gel to determine the amount of renatured homoduplex, usually 10-20% as compared with the heteroduplex product. The original mixture was then digested with HpaII and the trp HpaII 570 fragment was isolated by agarose gel electrophoresis followed by extraction and purification using adsorption to glass beads (18).
In Method B, trp heteroduplex molecules were formed as in Method A and, after ethanol precipitation, were dissolved in 2 rnl of buffer containing 10 m M NaC1, 1 mM Na2HP04, pH 7.7, 0.1 mM EDTA. T4 gene 32 protein (final concentration, 125 pg/ml) was added and incubated at 25 "C for 10 min to complex single-stranded DNA.
The sample was then fiitered through a nitrocellulose filter (Schleicher and Schuell) presoaked in the same buffer. This leads to preferential retention of heteroduplex DNA molecules while renatured parental homoduplex molecules pass through the filter. (Al-1 mM EDTA. though single-stranded DNA containing molecules alone are expected to be retained on these filters, we consistently observed that a substantial fraction of Ml3-trp heteroduplex complexes passed through the filter in the absence of T4 gene 32 protein.) The filter was then washed with 2 ml of 10 mg/rnl of bovine serum albumin and the filter was placed upper surface down in a solution containing endonuclease HpaII. After 3 h at 37 "C, the excised restriction fragments were precipitated with 0.5 M NaCl and 2 volumes of ethanol. The trp HpaII 570 fragment was separated by agarose gel electrophoresis and was isolated by adsorption to glass beads as for Method A. Transcription of trp HpaII 570 fragments was carried out in a reaction (50 pl (20).
The remainder of the sample (3000 to 5000 cpm) was heated to 90 "C for 1 min and then loaded onto an 8% denaturing polyacrylamide gel in tris/borate/EDTA buffer at pH 8.3 containing 7 M urea. Gels were pre-electrophoresed for 2 h before use. Electrophoresis at 260 volts was continued until the xylene cyanol dye had reached the bottom of the gel. Gels were analyzed by autoradiography and also by counting slices from each track to measure the amount of radioactivity in each RNA species quantitatively.

RESULTS
We had originally planned to clone the trp' attenuator sequence as well as selected trp attenuator mutants into the vector M13mp2. Isolation of trp inserts in both of the two possible orientations for each d e l e would provide M13mp2 phage bearing single-stranded DNAs with trp inserts complementary to each of the two DNA strands. The trp' attenuator as well as the attenuator mutants of choice were available cloned on an EcoRI fragment of 7100 base pairs (15) bearing the trp promoter, attenuator, trp genes E, D, and part of trpC (trp POLED'C'). Combination of appropriate complementary strands would allow direct construction of both possible heteroduplex templates for each trp attenuator mutation.
The trp' fragment was cloned and M13mp2 trpA+ (+) and M13mp2 trpA+ (-) phage were identified bearing the transcribed and nontranscribed strands of the trp' POLEDC' RI fragment, respectively. However, no stable clones of trp attenuator mutant L153 were obtained, possibly because insertion of a strong unregulated promoter into a multicopy vector is toxic for the host cell (21). Consequently, we sought other methods for construction of the needed heteroduplexes.
If DNA of a plasmid bearing a trp attenuator mutation is cut with EcoRI, denatured with alkali, and annealed with excess M13trpA+ (+) or M13trpA' (-) DNA, the trpA+/trpAM and trpA'/trpA+ heteroduplexes are formed in the two respective reactions, together with small amounts of the original trpA'/trpM homoduplex (Fig.  1). The amount of the latter can be estimated by gel electrophoresis of the hybridization products. Cleavage of the products allows isolation of a trp HpaII 570 DNA fragment bearing the trp promoter/attenuator region. Transcription of this fragment gives RNA products of characteristic size arising from the trp promoter and terminating at the attenuator (140 bases) or reading through to the end of the fragment (260 bases; see Ref. 22). Inspection of when transcription takes place in the presence of purified trp repressor; repression depends on added tryptophan. Transcription of trp HpaII 570 fragments from trpAM plasmids R I bearing the L153 mutation gives predominantly the 260-base transcript (Fig. 2, track A ) and quantitative analysis shows about 7540% readthrough of the trp attenuator (Table 11, f rP part 11). Finally, transcription of heteroduplexes prepared with trpA mutations L29 (Fig. 2, tracks E and F ) or L75, which affect readthrough in vivo but not in vitro, gave low readthrough and gave normal amounts of the trp 140-base transcript, confirming that the presence of a heteroduplex base mispair earlier in the trp leader does not by itself induce readthrough in vitro.
While the results of Fig. 2 are qualitatively convincing, the presence of contaminating homoduplex template introduces the need for a substantial correction, especially in the case of the trpAM/trpA+ template. This correction depends on our reforms. This is difficult since the M13mp2trpA+/trpAM heteroduplexes do not migrate as sharp bands on agarose gels, but as diffuse smears. Consequently, the error in our estimation of the per cent readthrough for the trpL153/trpA+ hetwe have used a second method to prepare heteroduplex tem- allowed isolation of such templates free of homoduplex con-

140"
FIG. 2. Transcription of trp templates bearing homoduplex or heteroduplex attenuator regions. The figure shows an autoradiogram of transcription products from homoduplex and heteroduplex templates; electrophoresis was from left to right and the positions of the attenuated (140 nucleotides) and readthrough (260 nucleotides) trp specific transcripts are noted.The template used in each synthesis is indicated at right; the transcribed DNA strand is given second for each heteroduplex template, and is fully underlined, for example trpA+/trpLZ53 contains the L153 mutation in the transcribed strand. T r a c m larger than 260n and around 20011 are not trp promoter specific, since they are not suppressed by trp repressor. They may be transcripts initiated at the ends of DNA fragments. the transcription products read from the two heteroduplex templates prepared in this manner (Fig. 2) makes it clear that the heteroduplex bearing the mutation in the transcribed strand (trpA+/trpAM) gives predominantly the 260-base readthrough product (track D ) while the other heteroduplex (trpAM/trpA+) gives predominantly the 140-base normal attenuation product (track C). Quantitative analysis of the experiment, correcting for contamination of the heteroduplex templates with homoduplex (trpAM/trpAM) starting material, confirms this result (Table 11).
In control experiments, transcription of trp HpaII 570 fragments cut from trp+ DNA or reconstituted by annealing M13trpA+ (+) and M13trpA+ (-) DNAs gives predominantly the 140-base transcript (Fig. 2, track B ) . Slicing of the gel and quantitative analysis indicated that 95% of the transcripts are terminated a t trpA+, in good agreement with other reports (12,22). Both the 140-base and 260-base transcripts disappear TABLE I1

Efficiencies of transcriptional termination in vitro with templates bearing homoduplex and heteroduplex trp attenuator sequences
The figures listed, indicating the percentage readthrough for each template, were calculated from the molar ratios of the runoff transcript to the attenuated product. Gels, such as that shown in Fig. 2, were analyzed by counting slices from each track to measure the amount of radioactivity in the 140n and 260n RNA species. Data from several independent experiments were averaged to obtain these figures. The figures in parentheses represent the percentage of readthrough for the L153 heteroduplexes after adjusting for contaminating renatured homoduplex in the template preparations (see "Experimental Procedures"). The methods used for preparing the heteroduplex templates are described under "Experimental Procedures." For heteroduplex templates, the second strand listed is the transcribed strand. tamination. Analysis of transcription from these templates (Table 11) c o n f i i s the fiiding that transcription termination at the trp attenuator is determined by the transcribed DNA strand. In addition, it appears that there is slightly more readthrough of the trp attenuator for both L153 heteroduplexes as compared to the homoduplex templates.

DISCUSSION
Our results show that the ability of the trpL153 attenuator mutation to suppress termination depends on its presence in the transcribed DNA strand. This result is the opposite of that predicted if termination were the result of the formation of a stem-loop structure in the nontranscribed DNA strand, or by models in which the nascent RNA chain must complex with the nontranscribed DNA strand to block transcription elongation. Hence, it would seem to rule out these mechanisms at least for strong bacterial terminator sites.The results is that expected if transcriptional termination at this site were determined by formation of a stem-loop structure in the nascent RNA transcript (2-4). Evidence against a major role in termination for the nontranscribed DNA strand has also been presented by Farnham and Platt (23).
One could imagine more complex models for termination in which there are essential base pairing interactions or hairpin structures formed within the transcribed DNA strand, as well as in the RNA product. Such models are made less likely by the demonstration that only 16 base pairs of DNA are open at the elongation site during normal elongation (11). While unwinding measurements have not yet been done for RNA polymerase actually at the terminator site, our results, taken with those of others (23, 24), strongly support a model in which the nascent transcript is displaced from the transcribed strand at a point less than 10 bases from the growing point and hence is free to form an RNA stem-loop structure, which can directly stop chain elongation by RNA polymerase. It has been shown that short self-complementary sequences can cause RNA polymerase to pause during elongation (25) even when termination does not occur, although this may not be true of all such sequences (26).
How does the RNA hairpin structure stop RNA chain elongation? One possibility is that there is a binding site on the enzyme specific for a short RNA duplex. Since RNA chain elongation is insensitive to polyanionic inhibitors (7-9) and since termination is not blocked at high salt concentrations (lo), this site cannot be ionic in nature but would have to be directed toward the ribose residues in the RNA, for example. An alternative possibility is that there is no such site on the enzyme, but that formation of the RNA hairpin itself forces a change in the conformation of the ternary DNA-RNA polymerase complex and blocks elongation (4). Resolution of these questions will require detailed mapping of the termination complex, potentially a difficult task since the enzyme appears to be rapidly released (27).
The slight increase in readthrough at the attenuator for both of the trpL153+trpA' heteroduplexes is interesting and not predicted by any of the current models for termination. Formation of the RNA stem-loop terminator structure depends on reformation of the DNA duplex and displacement of the nascent RNA chain. This occurs at a site about 10 base pairs from the growing end of the RNA chain, and the L153 mutation is located exactly at that point. Hence, one might speculate that the presence of a mispair in the DNA at this point could increase the probability of readthrough occurring before an effective RNA terminator structure can form, for example by allowing a slightly longer DNA-RNA hybrid to form. It would be quite interesting to study the effects of additional heteroduplex mispairs in the trp attenuator region to explore this phenomenon more closely.
The use of heteroduplex templates allows one to distinguish between several possible models for the action of regulatory sequences that are transcribed into RNA. In addition to prokaryotic terminators, such sites include DNA sequences that induce transcriptional pausing (26, 28) and potential terminator sequences for eukaryotic RNA polymerases I, 11, and I11 (29-32). It will be interesting to use this method to study the mechanism of action of some of these latter sites, especially since in some cases there are no evident RNA stemloop structures involved (26, 32).