The poly(A) signal, without the assistance of any downstream element, directs RNA polymerase II to pause in vivo and then to release stochastically from the template.

Genes encoding polyadenylated mRNAs depend on their poly(A) signals for termination of transcription. Typically, transcription downstream of the poly(A) signal gradually declines to zero, but often there is a transient increase in polymerase density immediately preceding the decline. Special elements called pause sites are traditionally invoked to account for this increase. Using run-on transcription from the nuclei of transfected cells, we show that both the pause and the gradual decline that follow a poly(A) site are generated entirely by the poly(A) signal itself in a series of model constructs. We found no other elements to be involved and argue that the elements called pause sites do not function through pausing. Both the poly(A)-dependent pause and the subsequent decline occurred earlier for a stronger poly(A) signal than for a weaker one. Because the gradual decline resembles the abortive elongation that occurs downstream of many promoters, one model has proposed that the poly(A) signal flips the polymerase from the elongation mode to the abortive mode like a binary switch. We compared abortive elongators with poly(A) terminators and found a 4-fold difference in processivity. We conclude that poly(A) terminating polymerases do not merely revert to their prior state of low processivity but rather convert to a new termination-prone condition.

Transcription termination by RNA polymerase II is predominantly of two sorts, both poorly understood. On the one hand are polymerases, newly dispatched from the promoter, which have failed to acquire the modifications necessary to become adequately processive (1). These polymerases tend to dissociate from the DNA within a few hundred base pairs of the promoter and are sensitive to the presence of simple termination elements in the DNA (2)(3)(4)(5). The presence of any such element near the promoter thus serves to terminate these polymerases of low processivity (2). On the other hand, some fraction of the polymerases leaving the promoter of any actively expressed gene do become appropriately modified and go on to transcribe the entire gene in a highly processive manner. For most genes, namely those whose RNAs are destined to become cleaved and polyadenylated, the poly(A) signal is a required element for the termination of this processive transcription (6,7). Such termi-nation is referred to as poly(A)-dependent termination. Both in vivo and in vitro, the poly(A) signal alone is sufficient for signaling the polymerase to stop processive transcription (8,9).
Both the establishment and disestablishment of processive transcription by RNA polymerase II appear to be intricate processes. Transcription complexes leaving the promoter are not initially of sufficient processivity to enter into productive transcription (1). In part, this reflects underphosphorylation of the C-terminal domain of the largest subunit of RNA polymerase II (5). In addition, shortly after leaving the promoter (10), elongation complexes acquire the negative transcription elongation factors DRB 1 sensitivity inducing factor (DSIF) and negative elongation factor (NELF), which induce the polymerase to pause. Relief of pausing and the establishment of processivity is brought about by the action of P-TEFb (positive transcription elongation factor b), which phosphorylates both DSIF and the C-terminal domain (11,12). However, the efficiency with which processivity is established varies with the promoter and the physiological context (3,13,14). The promoterproximal termination of those polymerases on which high processivity has not been conferred has been called abortive elongation (15).
Productive elongation is brought to an end by poly(A)dependent termination (6). At the heart of this mechanism is the problem of how the poly(A) signal communicates with the polymerase to direct it to terminate. For many years the favored model was that cleavage at the poly(A) site released a signal telling the polymerase to terminate (7, 16 -18). However, now it is clear that poly(A) signaling can occur in the absence of poly(A) site processing both in vivo (19) and in vitro (9). Therefore, the signal to stop transcription is delivered at some point during the assembly of the cleavage and polyadenylation apparatus before processing itself takes place.
Termination does not occur at the position of the poly(A) signal on the DNA but at a variable distance downstream. Often, polymerase density decreases gradually downstream of the poly(A) site, as assessed by run-on transcription (20 -27). In many instances this decrease in polymerase density is preceded by a region in which the polymerase density is higher than it is over the body of the gene (25)(26)(27)(28)(29)(30)(31)(32)(33). This has generally been interpreted to indicate the presence of polymerase pausing in the region, but the basis for this pausing is not understood.
Just as the mechanism of poly(A) signaling is not clear, neither is the mechanism of the release step that follows. It has been proposed that regardless of the mechanism of poly(A) signaling, its effect is to flip a binary switch that returns the polymerase to its prior state of low processivity (18). This idea is consistent with the observation that termination downstream of a poly(A) site, like abortive elongation, often occurs gradually (20 -27). However, this idea has never been tested.
Because poly(A)-dependent termination is, by its nature, a gradual process, a variety of auxiliary elements downstream of the poly(A) site often are also pressed into service to assist in termination (6). For example, some promoter elements are designed to repulse encroaching polymerases that have crossed an upstream poly(A) site but have not yet terminated and might otherwise lead to transcription interference (34 -36). Other elements assist in termination in conjunction with a transcript cleavage activity of unknown function (37). Still other elements assist in termination while exhibiting a polyadenylation enhancement function of unknown mechanism (6,27). The latter elements have been called pause sites because it was thought necessary first to pause the polymerase to give the poly(A) signal time to act (6,7,17).
Because of the complexities attending the wide variety of auxiliary elements involved, we have chosen to focus on the core poly(A) signal alone to gain a better understanding of the basal mechanism of poly(A)-dependent termination. Previously, we have shown that the core poly(A) signal by itself can direct efficient termination in vivo, a form of poly(A)-dependent termination that we have referred to as being poly(A)-driven (8). We have also shown in vitro that this signaling from the poly(A) signal to the polymerase does not depend on cleavage at the poly(A) site (9). Here we test the idea that poly(A) signaling operates a binary switching mechanism that converts polymerases to the same low processivity characteristic of polymerases near the promoter. We find that polymerases in the process of terminating downstream of a poly(A) site are several times more processive than abortively elongating polymerases proximal to a promoter. This shows that the poly(A) signal does not simply trigger a reversion of the polymerase to a prior state but rather that it orchestrates a transition to a new set of activities. We identify one of these activities as pausing. The core poly(A) signal directs all polymerases to pause downstream of the poly(A) signal in a way that does not depend on the underlying DNA sequence. Thus, pausing downstream of the poly(A) signal reflects an intrinsic property of poly(A) signaling that does not depend on the presence of any special pause sites.

MATERIALS AND METHODS
Transfection and Nuclear Harvest-COS cells were passed to 35-mm plates 1 day prior to transfection at about 50 -70% confluency with 1 g of DNA using FuGENE 6 (Roche Molecular Biochemicals) according to the manufacturer's instructions. Transfected cells were fed after 1 day and then harvested after approximately 2 days (except for the work depicted in Fig. 1) by scraping and lysing with a solution containing 0.5% IGEPAL (Sigma), 10 mM Tris (pH 7.4), 10 mM NaCl, and 3 mM MgCl 2 . Nuclei were pelleted at 6000 rpm, resuspended in 16 l of 50 mM Tris (pH 8.1), 5 mM MgCl 2 , 40% glycerol, and 0.1 mM EDTA and stored at Ϫ80°C. Wild type and mutant plasmid constructs were always transfected and assayed in parallel.
G-less Run-on Transcription Assays (8)-For routine assays the stored nuclei were mixed with an equal volume of transcription buffer and incubated for 30 min at 30°C. Final concentrations were 280 mM (NH 4 ) 2 SO 4 , 2 mM MgCl 2 , 5 mM Tris (pH 7.5), 470 M ATP, 470 M UTP, 118 M 3Ј-MeO-GTP, 2 mM dithiothreitol, 5 M CTP, 30 Ci of [␣-32 P]CTP, and 1 unit of RNase inhibitor. Then cold CTP was added to 1 mM for 12 min followed by 10 units of DNase I for 20 min and then 15 units of T1 RNase and EDTA to 1 mM for 30 min and finally 36 g of proteinase K and SDS to a final concentration of 0.5% for 20 min, all at 30°C. RNA was extracted with TRIzol (Invitrogen), precipitated with isopropanol and 2 g of tRNA, washed with 70% ethanol, and resuspended in 50 l of 50 mM Tris (pH 7.5) and 1 mM EDTA. Finally, T1 RNase (500 units) was added again, and after incubation for 30 min at 45°C the RNA was extracted with phenol/chloroform and precipitated with isopropanol and 3 g of glycogen. The RNA pellet was washed with 70% ethanol, resuspended in 7 M urea, heated for 5 min at 90°C, chilled to 0°C for 2 min, and then run on an 8% polyacrylamide gel. The gels were dried and analyzed using a PhosphorImager and ImageQuant software. For the data in Fig. 3, the 5 M cold CTP was omitted during labeling, and the RNA was purified using RNeasy columns (Qiagen). The RNA was eluted in 50 l of 10 mM Tris and 0.1 mM EDTA (pH 8)  Table I. B, wild type and mutant versions of the IA͗LC 9 ͘ construct were transfected into COS cells. Nuclei were harvested at various times after transfection and assayed for termination by G-less cassette analysis. We have used Photoshop to normalize the gel lanes with respect to the 131-nt cassette. This facilitates visual comparison by removing the effects of variations in transfection efficiency and sample recovery. The numerical values are based on phosphorimagery (Amersham Biosciences). and then incubated with 500 units of T1 at room temperature for 15 min. Finally, the buffer was evaporated, and the RNA pellet was resuspended in 7 M urea and analyzed as above. Both methods gave similar results.

The G-less Cassette Assay for Transcription Termination-
This assay utilizes nascent transcript pulse labeling (run-on transcription) for detecting transcriptionally engaged polymerases in isolated nuclei. The assay allows us to determine the polymerase density within individual G-less cassettes placed at intervals along a transfected template. Any transcription termination in the region between two cassettes is detected as a reduced polymerase density in the downstream cassette (8). For example, consider the construct shown in Fig.  1A. Poly(A)-driven termination directed by the SV40 late poly(A) signal, L, can be detected by comparing the polymerase density in a region upstream of L (within the 131-bp G-less cassette) to that in a region some distance downstream (within the 174-bp G-less cassette). Fig. 1B displays the results of an experiment in which plasmids containing either wild type or mutant L were transfected into COS cells and then assayed by run-on transcription of the isolated nuclei 2.5-4.5 days later. The results show that the polymerase density is much lower in the post-cassettes located downstream of the wild type poly(A) signal (odd numbered lanes) than in the post-cassettes downstream of the mutant (even-numbered lanes). As illustrated in Fig. 1B, we quantitate this poly(A)-dependent deficit in polymerase density by first normalizing each post-cassette signal to its own pre-cassette to control for transfection efficiency and sample recovery, and then we express the normalized wild type post-cassette signal as a percentage of its corresponding mutant run in parallel. Post-cassette/pre-cassette ratios for any given construct are reproducible for samples transfected and assayed in parallel but can vary from one transfection to another. However, wild type/mutant ratios for any given wild type-mutant pair remain consistent across experiments.
We have previously pointed out that a theoretical advantage of the cassette assay over traditional hybridization analysis is that polymerase arrest between the cassettes cannot masquerade as termination (8). For example, were a polymerase to become immobilized within L, the blockade would eventually give rise to a stack of accumulated polymerases that would extend back into the pre-cassette (see Fig. 1A). This increased polymerase density in the pre-cassette would ordinarily lead to an increased yield of pre-cassette transcripts during run-on transcription under the usual conditions. If this were followed by hybridization analysis, the conventional interpretation of the resulting data would be that termination had occurred because the polymerase density was less downstream than upstream of the poly(A) site. In contrast, the run-on transcription carried out for G-less cassette analysis uses 3Ј-MeO-GTP in place of GTP during the transcription so that any polymerases that become backstacked into the cassette in vivo cannot exit in vitro for lack of GTP. Because the polymerases cannot move, the pre-cassette signal should decrease rather than increase as a result of the stack, and the erroneous impression of termination is avoided.
The above scenario depends on efficient trapping of the polymerases in the cassettes because of GTP starvation during the run-on in vitro. We have confirmed that GTP starvation is stringent under our conditions. In a control experiment the overall level of [␣-32 P]CTP incorporation rose more than 100fold when GTP rather than 3Ј-MeO-GTP was used during run-on transcription of the nuclei (data not shown). The scenario outlined in the previous paragraph also assumes, for the hypothetical arrested polymerase, that sufficient time has elapsed following transfection for a polymerase stack to form that would reveal the presence of the stalled polymerase. The results of Fig. 1B are consistent with this assumption and show that even after a long period following transfection there is no evidence that a stack has begun to form. In this experiment the harvesting of nuclei was delayed considerably beyond our average time of 2 days post-transfection. If the low post-cassette signal for the wild type were due to immobilized rather than terminated polymerases, one would expect to see evidence of a growing stack behind the immobilized polymerases. As this backstacking invades the pre-cassette in vivo, the run-on signal obtained subsequently in vitro in the presence of 3Ј-MeO-GTP for this cassette should decrease, causing both the post-cassette/pre-cassette and the wild type/mutant ratios to increase with time. Fig. 1B shows that, compared with analysis at 2.5 days post-transfection, there is no tendency for either of these ratios to increase over the subsequent 2 days. These controls indicate that the decreased polymerase densities observed for the wild type post-cassettes in the experiments described below reflect termination of transcription rather than immobilization of the polymerases without release.
There remains the formal possibility that the value of 20% for the wild type/mutant ratio reflects a 5-fold increase in the speed of transcription downstream of a poly(A) signal rather than termination. A 5-fold increase in transcription speed would give rise to a 5-fold decrease in polymerase density, which would be interpreted as termination. However, methods of analysis other than run-on transcription indicate clearly that the poly(A) signal leads to halting transcription rather than to an increase in the speed of the polymerase (34, 38 -41). Quantitation of Polymerase Processivity in Vivo-One goal of this study was to evaluate the long standing hypothesis (18) that the poly(A) signal provokes termination in the manner of a binary switch by returning processive polymerases to their initial promoter-proximal state of low processivity. A prediction of this model is that the terminating polymerases downstream of the poly(A) site would have the same partial processivity as the abortively elongating polymerases proximal to the promoter. We therefore set out to devise a method for quantitating the processivity of transcribing polymerases in vivo.
The G-less cassette assay is easily adapted to provide a quantitative measure of processivity. We have previously suggested that RNA polymerase II disengages stochastically from the template during poly(A)-driven termination (8). Subsequently, it was shown for Escherichia coli RNA polymerase that, after potentiation by factor, the RNA is released according to first order kinetics (i.e. stochastically) (42). Although this rate of release varies with template sequence, there is a characteristic probability of release at each position along the template (43). Assuming that poly(A)-driven termination follows these same principles, one predicts a relationship like that shown in Fig. 2A, where polymerases gradually dissociate from the template after crossing the poly(A) signal. Indeed, poly(A)dependent termination is often observed to be gradual (20 -26). Thus, using distance along the template as a proxy for time, one can plot the number of polymerases remaining at each position along the template versus the distance of that position past the poly(A) site. Assuming stochastic dissociation and making the approximation that the template is homogeneous, one predicts a monophasic exponential decrease in polymerase density with distance ( Fig. 2A). If this decline in polymerase density takes place across a G-less cassette window (as in Fig.  2A), the characteristic slope can be estimated by taking the difference between the logs of the polymerase densities determined for the two cassettes and then dividing by the distance between them.
The processivity of abortively elongating polymerases can be determined similarly (Fig. 2B). In this case, the polymerase density begins to decrease immediately downstream of the promoter. If poly(A)-driven termination operates like a binary switch causing processive polymerases to revert to the abortive elongation mode, then the slopes obtained as shown in Fig. 2, A and B should be similar.
It is convenient for us to adopt a uniform definition for processivity. If polymerases decay from the template by a simple monophasic exponential process as shown in Fig. 2, then we can think of processivity in terms of half-life as the distance along the template required for half the polymerases to terminate (D1 ⁄2 ). Note that, given such a relationship, the measurement of D1 ⁄2 can be made using any cassette window interval along the slope of declining polymerase density as long as sufficient transcription remains to be measured.
Polymerases Terminate Stochastically as a Single Homogeneous Class Downstream of the SV40 Early Poly(A) Signal-To measure polymerase density as a function of distance downstream of a poly(A) signal, we constructed a series of plasmid templates in which the cassette window was systematically varied in length (Fig. 3A). To minimize DNA sequence effects, we varied these lengths by repetition of a short piece of randomly chosen spacer DNA (see Table I). This randomly chosen DNA shows no evidence of containing any special DNA elements (8,44), and control experiments show that several repetitions of this spacer segment behave similarly to the bacterial chloramphenicol acetyl transferase DNA sequence when placed in a cassette window (data not shown). To prevent any unique initial effects of the poly(A) signal or of the spacer DNA from complicating our analysis, we placed the poly(A) signal in front of (rather than within) the cassette window followed by 1 unit of spacer DNA also in front of the cassette window. To restrict our attention to poly(A)-dependent effects only, all measurements were made using paired templates having wild type or mutant poly(A) signals (see Fig. 3A), and only the differences between them were analyzed. For this series of experiments, the SV40 early poly(A) signal, E, was used. In all cases the poly(A) signal was separated from the promoter by more than 2 kb of DNA to ensure that only fully processive polymerases would encounter the poly(A) signal.
The results in Fig. 3B, lanes 1-4 show that the relative polymerase density in the 174-nt post-cassette of the wild type templates decreases gradually as the post-cassette is placed progressively farther from the poly(A) site. In all cases the wild types were accompanied by their respective mutants, but the data for both are shown only for the longest cassette window (Fig. 3B, lanes 4 and 5). The wild type/mutant ratios are plotted in Fig. 3C. The data show that the exponentially decreasing polymerase density fits a simple first order relationship, having a processivity of 403 bp. Within experimental error the semi-logarithmic plot yields a straight line, indicating that all of the polymerases crossing the poly(A) site are converted into a single kinetic class of stochastically terminating molecules.
Interestingly, both wild type and mutant versions of the C 1 construct, which has the shortest cassette window, exhibit equal polymerase densities in their 174-bp post-cassettes. This is evident from Fig. 3C, which shows that for construct 1 the wild type has 100% of the polymerase density of the mutant in the 174-bp cassette. Thus, although the 174-bp cassette in construct 1 lies over 500 bp downstream of the poly(A) cleavage site, no detectable termination has yet occurred by this assay. We will return to this point later.
Promoter-proximal Abortively Elongating Polymerases Are Much Less Processive than Poly(A) Terminating Polymerases-Next, we determined the processivity of promoter-proximal polymerases ( Fig. 4) for comparison with the results described above for polymerases terminating downstream of a poly(A) site ( Fig. 3). For this purpose a series of constructs was prepared in which the 131-bp cassette was placed almost immediately downstream of the starting point of transcription instead of downstream of a poly(A) signal. As before, the 131-bp cassette was followed, at increasing distances, by the 174-bp cassette ( Fig. 4A). Fig. 4B shows that the polymerase density within the 174-bp cassette drops rapidly as the cassette window is lengthened, indicating that most polymerases undergo premature termination. Nevertheless, from Fig. 3B, lane 5 it is clear that in the absence of a functional poly(A) signal this same promoter launches polymerases that are sufficiently processive to reach the end of the 174-bp cassette in the A 3 EC 1 ͗C 9 ͘) construct nearly 4 kb downstream. Thus, in agreement with previous work (2,3), polymerases proximal to the promoter comprise a mixture having both high and low processivities. This inhomogeneity is reflected in the plot of these data shown in Fig. 4C. The semi-logarithmic termination profile for the ͗C n ͘ series is not a straight line but a curve. The initial slope of this curve is more than four times as steep as the relationship for poly(A) terminating polymerases in Fig. 3C. This suggests that polymerases aborting near the promoter are much less processive than polymerases undergoing termination downstream of a poly(A) site.
If the polymerases leaving the promoter are a simple mixture of two homogeneous populations, one of very high and another of very low processivity, then it should be possible to deconvolute their contributions to the termination profile. This would allow us to isolate and examine the properties of the abortively elongating population of polymerases. The closed symbols in Fig. 4C represent the normalized post-cassette polymerase densities of the ͗C n ͘ series expressed as a percentage of that for a control plasmid, A 3 ͗C 0 ͘, whose polymerases are assumed to be completely processive in the region of the cassette window. The sequence upstream of the pre-cassette in A 3 ͗C 0 ͘ is identical to that in the A 3 EC 1 ͗C n ͘ series of Fig. 3 except for lacking the 245-bp EC 1 portion. Thus, the polymerases traversing the cassette window in A 3 ͗C 0 ͘ are highly processive because they have already traversed 2.2 kb of A 3 DNA upstream of the precassette. Moreover, they should travel efficiently from one cassette to the next because the window itself is nothing more than the pre-and post-cassettes placed next to each other. The ͗C n ͘ series curve in Fig. 4C appears to level off at ϳ10%, suggesting that after all of the abortively elongating polymerases have been cleared from the template, there remain about 10% of the polymerases that are highly processive. Subtracting the contribution of these processives from each point on the curve should reveal the termination profile for the abortive population.
The open symbols in Fig. 4C show that, after removing the estimated contribution of the processive polymerases from the promoter-proximal data (see legend to Fig. 4), the points that remain fit a straight line. This suggests that the abortives constitute a single homogeneous polymerase class. The slope of the termination profile yields a processivity of 92 bp for the abortives. This matches the initial slope of the curve for the ͗C n ͘ series as a whole (Fig. 4C) and is more than four times as steep as the slope given by the A 3 EC 1 ͗C n ͘ series for poly(A)driven termination in Fig. 3C. Thus, both abortively elongat-ing and poly(A) terminating polymerases behave as homogeneous populations, but they differ from each other dramatically in their degree of processivity. Therefore, we conclude that the poly(A) signal does not simply return RNA polymerase II to a default state of low processivity characteristic of polymerases close to the promoter.
Polymerases Pause before Releasing-We pointed out earlier in presenting the data of Fig. 3C that in the first construct of that series there was no difference in post-cassette polymerase density between the wild type and mutant versions of the template. In this construct the 174-bp post-cassette is some 500 bp downstream of the poly(A) site. We were surprised that the polymerase density this far downstream of a functional poly(A) signal had not yet begun to decrease relative to that in the mutant. To investigate this further, we wished to assay polymerase densities closer to the poly(A) site. In the constructs of Fig. 3A, polymerase densities close to the poly(A) site cannot be assayed because the poly(A) signal lies upstream of the cassette window. We therefore prepared the constructs shown in Fig. 5A in which the same poly(A) signal lies within the cassette window. Fig. 5B shows, in agreement with the results of Fig. 3, that polymerase densities in the 174-bp post-cassette do not begin to decrease significantly until the post-cassette is moved several hundred base pairs downstream of the poly(A) site. Also in agreement with Fig. 3, the plot in Fig. 5C shows that the polymerases on the A 3 ͗EC n ͘ constructs (Fig. 5) decay from the template with a processivity of 386 bp, very similar to that of the polymerases on the A 3 EC 1 ͗C n ͘ constructs (Fig. 3). Thus,

FIG. 3. Poly(A)-driven termination occurs stochastically downstream of the SV40 early poly(A) signal.
A, maps, drawn to scale, of the constructs used for panels B and C. Plasmid construction is described in Table I. Note that the SV40 early poly(A) signal contains two hexamers, both of which must be mutated to completely inactivate the signal (16). The downward arrows indicate the poly(A) cleavage site. B, some gel lanes from a G-less cassette analysis. Lanes 1-4 show the decreases in post-cassette polymerase density that accompany increases in cassette window length downstream of an active poly(A) signal. Each wild type sample was accompanied by its corresponding mutant, but only the mutant for the longest cassette window is shown (lane 5). Images have been normalized as for Fig. 1. C, termination profile for the A 3 EC 1 ͗C n ͘ series. The data were obtained by transfection of the entire series on three separate occasions. Note that polymerase density in the post-cassette is expressed for wild type as a percentage of that for the mutant. This is conceptually equivalent to the percentage of the original as shown in Fig. 2A but normalizes out any non-poly(A)-dependent effects on elongation that may be present in these constructs. The line is an exponential fit to the data using KaleidaGraph. termination downstream of the SV40 early poly(A) signal is similar in two different plasmid contexts.
The most notable feature in Fig. 5C is that the polymerase density in the post-cassette of the wild type template is actually higher than that of the mutant (i.e. Ͼ100% in Fig. 5C) when the post-cassette is close to the poly(A) signal. This indicates that the polymerases pause transiently a short distance downstream of the poly(A) site, thereby increasing their density on the template. We emphasize here that this pausing can only be in response to the poly(A) signal and cannot be the result of any special pause element in the DNA. First, the only difference between the wild type and mutant templates is two point mutations in each of the two poly(A) signal hexamers (as shown in Fig. 3A). Second, the pausing cannot be attributed to an element within the spacer sequence, because the increased polymerase density is detected within the G-less cassette. Third, the excess polymerase density within the post-cassette placed 200 -300 bp downstream of the wild type poly(A) site (Fig. 5C) cannot be attributed simply to an element within the cassette, because the same cassette with the same flanking sequences placed a few hundred base pairs farther downstream shows a deficit rather than an excess of polymerases. Moreover, this same pausing phenomenon on the wild type template was observed (data not shown) for a construct in which the precassette was longer than the post-cassette and (except for a cloning junction) contained the entire sequence of the postcassette nested within it. Thus, polymerases crossing the postcassette were encountering the same sequence for the second time but behaving differently (i.e. pausing) on account of the upstream poly(A) signal. Therefore, we conclude that the poly(A) signal induces a change in the state of the polymerase that causes it to become susceptible to pausing as it moves downstream. This pausing occurs regardless of the underlying DNA sequence, and when a G-less cassette is placed within the interval where pausing occurs we can detect it with our assay.
We desired independent confirmation that the rise above 100% in the plot of Fig. 5C corresponds to pausing. It may seem paradoxical to propose that run-on transcription, which requires polymerase movement, might provide the means to detect paused polymerases, which are unable to move. However, paused polymerases, at least those in promoter-proximal positions, are relieved of their pause by the high salt concentrations IA͗LC n ͘ series and respective mutants (n ϭ 0, 1, 2, 3, 6, 9) The intron-containing PstI-RsaI fragment, I, of pRL-SV40 (Promega) was inserted into StuI-cut pAP͗cat͘ (Fig. 6A.1 of Ref. 9) and then re-excised as a SphI-HpaI fragment, IA, which was used to replace segment A 3 in SphI (sticky) and KpnI (blunted)-cut A 3 ͗LC 6 ͘. This change was then propagated to the rest of the series via the exchange of SphI-SmaI segments. For the respective mutants, the hexamer of L was mutagenized from AATAAA to AgTAcA.
A 3 ͗EC n ͘ series and respective mutants (n ϭ 0, 1, 2, 3, 6,9) The L segment in the A 3 ͗LC n ͘ series (n 0) was excised by SmaI-BamHI digestion and replaced with a HpaI-BamHI fragment containing the SV40 early poly(A) signal (Fig. 2 of Ref. 59). The SmaI-HpaI cloning junction extended the upstream G-less cassette by 5 bp. For n ϭ 0 all DNA between E and the cassette was removed by BamHI-BglII digestion. For the mutants, the two AATAAA hexamers in E were mutagenized to AATAGC and TTTAAA by site-directed mutagenesis.
A 3 EC 1 ͗C n ͘ series and respective mutants (n ϭ 1, 3, 6, 9) The EC 1 segment was cut from A 3 ͗EC 1 ͘ by AluI-BglII digestion and inserted into the XhoI site of the A 3 ͗LC n ͘ series from which L had been deleted by SmaI-BglII digestion. The mutants were as for the A 3 ͗EC n ͘ series.
typically used in run-on transcription reactions (15,(45)(46)(47)(48). Thus, our routine assays would detect all polymerases located within cassettes, both paused and elongating. We decided to carry out run-on transcription under conditions of low salt concentration, reasoning that the paused polymerases would not be detected under these conditions (48), thereby allowing us to restrict our attention to elongating polymerases only. The predicted outcome of excluding paused polymerases from analysis is that the wild type/mutant ratio should no longer exceed 100% because the paused polymerases on the wild type template will not be visible. Moreover, this reasoning predicts that run-on transcription at low salt concentrations (for which the paused polymerases in the post-cassette on the wild type template would not give a signal) should always yield wild type/ mutant ratios that are equal to or less than those obtained from high salt run-on transcriptions (for which the paused polymerases on the wild type template would give a signal). The wild type version of A 3 ͗EC 1 ͘, which is construct 2 in Fig.  5A, displays the greatest excess in post-cassette polymerase density of the A 3 ͗EC n ͘ series, as can be seen in Fig. 5C. To determine whether this polymerase excess in the wild type is due to paused polymerases in the post-cassette, we first transfected wild type and mutant A 3 ͗EC 1 ͘ into several batches of cells. We then isolated the nuclei and carried out run-on transcriptions for G-less cassette analyses at a series of decreasing salt concentrations. Fig. 6A shows that the 174-nt post-cassette signal intensity decreases substantially for the wild type relative to that of the mutant as the salt concentration is decreased. Thus, at 280 mM (NH 4 ) 2 SO 4 , a high salt concentration typical of run-on transcriptions, the normalized 174-nt post-cassette intensity for the wild type is slightly greater than that for the mutant. In contrast, at 28 mM (NH 4 ) 2 SO 4 the wild type postcassette intensity is considerably less than that for the mutant. Thus, reducing the salt concentration affects the outcome of wild type and mutant run-on transcriptions differently. The results are consistent with the interpretation that many of the polymerases in the 174-nt cassette of wild type but not mutant A 3 ͗EC 1 ͘ are paused in vivo. Such polymerases would be detected under conditions of high salt run-on transcription but would fail to elongate during run-on transcriptions at a low salt concentration. Fig. 6B summarizes the results of a number of such low-salt experiments and compares them with the high-salt data presented previously in Fig. 5C. Fig. 6B shows, as predicted by the pausing model, that when attention is restricted to elongating polymerases only (low salt), the wild type polymerase densities in post-cassettes close to the poly(A) signal no longer exceed those of the mutant (i.e. do not exceed 100%). Moreover, throughout the termination profile, wild type to mutant ratios are lower when the wild type and mutant run-on transcriptions are carried out at low rather than high salt concentrations. This indicates that even at distances greater than 1 kb downstream there are still polymerases on the wild type template that have paused in vivo in response to the upstream poly(A) signal and are unable to resume transcription in vitro at low salt concentrations. Indeed, a dashed line (Fig. 6B) Table I. Except for a cloning junction in the cassette window, members of this series differ from members of the series in Fig. 3A only by lacking all but 59 bp of the sequence between the promoter and the 131-bp G-less cassette. B, gel lanes from a G-less cassette analysis of the ͗C n ͘ series. Images have been normalized as for Fig. 1. C, termination profile for the ͗C n ͘ series. For each member of the series, the post-cassette signal was first normalized to its own pre-cassette signal and then expressed as a percentage of the comparable value obtained for A 3 ͗C 0 ͘ run in parallel. Readthrough of the cassette window in A 3 ͗C 0 ͘ is assumed to be 100%, because the 2.2-kb A 3 sequence preceding the pre-cassette should screen out the abortively transcribing polymerases. The data points for the ͗C n ͘ curve are the average and range of values obtained following transfection on two separate occasions. The data points for the ͗C n ͘ abortives were obtained by subtracting 9.85 from each normalized ͗C n ͘ post-cassette signal and then expressing this as a percent of the normalized A 3 ͗C 0 ͘ post-cassette signal from which 9.85 had also been subtracted. The exact amount to subtract (i.e. 9.85) was chosen so as to give the best exponential fit to the resulting points. ence of essential elements (AAUAAA hexamer and the GU-rich element), these two poly(A) signals are unrelated at the sequence level. Their modular arrangement is also unrelated. SV40 late consists of a poly(A) signal core flanked on both sides by strong enhancing elements (49), whereas SV40 early consists of two interdigitated cores but no additional elements (16). Fig. 7 shows that the SV40 late poly(A) signal, L, resembles the SV40 early poly(A) signal, E, in all of the significant functional properties discussed so far. First, Fig. 7A shows that polymerases downstream of L display a monophasic exponential decrease in polymerase density along the template, very much like E in Figs. 3C and 5C. Thus, L, like E, converts the polymerases into a single homogeneous class of stochastically terminating polymerases. Second, the processivity downstream of L is 447 bp, similar to the 386-and 403-bp processivities downstream of E (Figs. 3C and 5C). Therefore, the poly(A)terminating polymerases downstream of L and E resemble each other closely but differ from the much less processive abortively elongating polymerases proximal to the promoter (D1 ⁄2 ϭ 92 bp; Fig. 4C). Third, run-on transcription of wild type and mutant L at high and low salt concentrations reveals a significant fraction of paused polymerases downstream of wild type L that contribute to the run-on signal at high but not at low salt (Fig. 7B). This also is as described earlier for E (Fig.  6B). Moreover, this pausing leads to a polymerase density proximal to L on the wild type template that exceeds that of the mutant when examined by high salt run-on transcription (Fig.  7), also like E (Fig. 5C). Finally, a dashed line fitted to the low salt data (Fig. 7B) suggests, as for E (Fig. 6B), that a similar constant proportion of the polymerases are paused in vivo at all points downstream of the poly(A) signal on the wild type template. Thus, both L and E drive termination by what appear in the end to be identical pause-release mechanisms.

Stochastic Termination, the Basal Poly(A)-dependent Mechanism-
We have found, in agreement with the original proposal of Logan et al. (18), that poly(A) signals trigger the conversion of processive RNA polymerase II into a new state of greatly reduced processivity. When the poly(A) signal is followed by a simple repetitive sequence in the DNA, polymerase density downstream exhibits a simple monophasic exponential decrease. This resembles a first order reaction in which the polymerases in their new state of reduced processivity exhibit a characteristic probability of disengaging from the template at each position along the way. This probability can be expressed in terms of a half-life, which we have adopted as a convenient measure of processivity. On our templates the processivity of terminating polymerases downstream of two very different poly(A) signals is about 400 bp; that is, the polymerase density is reduced by about half every 0.4 kb. The exponential decrease does not depend on any downstream elements in the DNA, but  Table I. B, gel lanes from a G-less cassette analysis of the A 3 ͗EC n ͘ series. Images have been normalized as for Fig. 1. Only wild type constructs were used in this experiment so that the entire series could be transfected and assayed in a single experiment. The quantitation in panel C is based on additional experiments carried out according to our usual procedure in which wild type templates and their respective mutants were run in parallel. C, termination profile for the A 3 ͗EC n ͘ series. Each data point is the mean Ϯ S.D. of several (minimum three, average six) transfections carried out on separate occasions. The line is an exponential fit to the data. the value of the half-life can be modulated by DNA sequence (8). In both of these respects (i.e. first order dissociation and modulation by sequence) poly(A)-terminating RNA polymerase II resembles E. coli RNA polymerase that has been potentiated to terminate by factor (42,43). This appears to be the basal termination mechanism directed by the core poly(A) signal.
In many in vivo contexts the basal mechanism is presumably adequate. However, genes that are closely spaced may require additional elements to hasten disengagement (6). Thus, just as the core promoter of a gene can be embellished with elements that enhance or repress its activity (50), so also the core poly(A) signal can be accompanied by elements that modify its ability to direct termination. Most elements that have been described serve to enhance the termination function (6), but work on the SV40 early poly(A) signal, described below, illustrates that inhibitory activities also play a role.
In elegant work a number of years ago (16,36,51), Connelly and Manley showed that both an intact SV40 early poly(A) signal core and a downstream protein binding site (CCAAT) were required for efficient termination of transcription in their constructs. Termination normally occurred within several hundred base pairs of the CCAAT element, but point mutations in this site resulted in readthrough transcription that continued unabated for more than 3 kb (36). Thus, the poly(A) signal was unable to effect termination unassisted by the additional element. Yet we have found in this study, as well as previously (8,9), that the core SV40 early poly(A) signal efficiently stops transcription both in vivo and in vitro. The simplest explanation for this discrepancy is that there is still another element, yet to be identified, that prevents termination in the absence of the CCAAT site. This would resemble the situation described in the introduction whereby DSIF and NELF bind to the polymerase, rendering further elongation contingent on the action of P-TEFb. Thus, our data together with those of Connelly and Manley indicate that the basal termination activity of the SV40 early poly(A) signal was inhibited by some element contained within the sequences of their construct and that the role of the CCAAT site was largely to reverse this inhibition.
A similar situation was recently reported by Dye and Proudfoot (37). Working with the human ␤and ⑀-globin genes, they FIG. 6. Paused polymerases resume transcription at high but not low salt concentrations in vitro. A, gel lanes from a G-less cassette analysis carried out on construct 2 of Fig. 5A as a function of salt concentration. Run-on transcription was as described under "Materials and Methods" except that the final concentration of (NH 4 ) 2 SO 4 was varied as shown. Images have been normalized as for Fig. 1 Table I. The map for IA͗LC 9 ͘ is shown in Fig. 1A. Each data point is the mean Ϯ S.D. of several (minimum three, average four) transfections carried out on separate occasions. The line is an exponential fit to the IA͗LC n ͘ data. Also shown are the data points from Fig. 5C. B, low salt [28 mM (NH 4 ) 2 SO 4 ] termination profile for the IA͗LC n ͘ series superimposed on the high salt data for the same series from panel A. Except for the data point with no error bar, the low salt values are the average and range obtained from transfections on two separate occasions. The dashed line is an exponential fit to the low salt data. found that termination for these genes also did not occur without the assistance of auxiliary downstream elements. As in the case of the SV40 early poly(A) signal discussed above, it seems likely that the role of the downstream globin elements is to neutralize the effects of another inhibitory element located elsewhere in the construct. This predicts that the poly(A) signal cores of these globin genes will drive termination if removed from their native context and placed in an environment devoid of auxiliary elements.
Our experience in observing efficient termination driven by a variety of different poly(A) signals in a variety of different contexts (this study as well as Refs. 8 and 9 and other data not shown) indicates that the default capability of the core poly(A) signal, unassisted, is to drive efficient transcription termination. The fundamental mechanism depends on functions that are all orchestrated by the poly(A) signal itself. As illustrated by the example of the SV40 early poly(A) signal, the involvement of additional elements reflects functional modifications superimposed on the basal mechanism. The situation may resemble that of abortive elongation in which polymerases are rendered unusually susceptible to termination by a variety of unrelated elements (2,5,13). The great variety of elements is not indicative of great complexity but rather reflects the simple fact that abortive polymerases are of very low processivity and respond similarly to many different kinds of impediments along the way. Thus, such elements alter the pattern of abortion, but their actions are mechanistically distinct from the basic cause of the abortive state.
Stochastic Termination Resembles but Is Different from Abortive Elongation-One goal of this study was to evaluate the hypothesis that poly(A)-dependent termination reflects the flipping of a binary switch. This model was based on the obvious similarity between abortive elongation and the gradual loss of polymerases that is often observed downstream of poly(A) sites (18). We compared the kinetics of termination by abortively elongating polymerases with the kinetics of termination as driven by poly(A) signals. We found that polymerases terminating downstream of a poly(A) site (D1 ⁄2 ϭ 386 -447) were actually more than four times as processive as abortive polymerases proximal to the promoter (D1 ⁄2 ϭ 92). Moreover, each constituted a single homogeneous class, indicating that abortively elongating and poly(A) terminating polymerases represent distinctly different states of the enzyme. Thus, the poly(A) signal does not trigger the polymerase to revert to its former state of low processivity, but rather it drives it into a new state unique to poly(A)-dependent termination.
Of course, not all polymerases leaving the promoter of an expressed gene are abortives. Our assay revealed, in fact, that the total population of polymerases proximal to the promoter constituted a mixture that gave a biphasic termination profile as shown by the curve in Fig. 4C. Although the biphasic nature of the profile was expected, we initially were surprised that the aborting polymerases accounted for 90% of the total. However, our plasmids carry the SV40 origin of replication (8), which is active in COS cells (52). Perhaps the high proportion of abortives reflects a template copy number effect as previously reported for Xenopus oocyte injection experiments (13,14).
The Poly(A) Signal Induces Pausing before Termination-The most remarkable observation in this study is the finding that the poly(A) signal, without the assistance of any downstream element in the template, causes the polymerase to pause before termination. Thus, our data support a pauserelease model for termination that is strikingly different from the conventional view. Traditionally, a role for distinct auxiliary pause elements in the poly(A)-dependent termination mechanism has been invoked (6,7,53). This reflected a belief that the poly(A) signal needed the extra time afforded by pausing imposed from the outside in order to communicate with the polymerase. However, we have shown here that the poly(A) signal itself enforces first the pause and then termination. No other element is required.
We wish to emphasize that all of the poly(A)-dependent effects reported in this study derive from comparisons of polymerase density between wild type and mutant templates. These templates differ only in that the mutant has two (SV40 late, Fig. 1A) or four (SV40 early, Fig. 3A) point mutations confined to the AATAAA hexamers of the poly(A) signals. Therefore, the effects that we interpret as pausing are unquestionably poly(A)-dependent effects.
Poly(A)-dependent pausing is indicated by our data in two ways. First, we find an excess in total polymerase density downstream of wild type poly(A) sites compared with their mutants (Figs. 5C and 7A). For two different poly(A) sites the polymerase density about 200 bp downstream is some 20% higher on the wild type than on the mutant template (Fig. 7A). A remarkably similar result was obtained recently in a comparison of the regions directly downstream of wild type and mutant mouse -s poly(A) sites, where there was also an excess of about 20% of polymerases for the wild type ( Fig. 6 of Ref. 27). These results are consistent with a pause-release mechanism for termination in which poly(A)-induced pausing is evident at locations close to the poly(A) site but is offset by the cumulative amount of polymerase release farther downstream.
Our second approach for detecting paused polymerases took advantage of their inability to elongate in vitro at low salt concentrations (15,(45)(46)(47). Using low salt run-on transcriptions, we found no excess polymerase density on wild type templates relative to mutant templates in the region 200 bp downstream of the poly(A) cleavage site (Figs. 6B and 7B), confirming our interpretation that the excess polymerase density on wild type templates is due to pausing. Moreover, the wild type/mutant ratio was consistently lower for run-on transcriptions carried out at low salt concentrations than for those at high salt for all positions along the template (Figs. 6B and 7B). These low salt termination profiles reveal an exponential decrease in the ability of polymerases to elongate that parallels and precedes the high salt profiles that describe a decrease in total polymerase numbers. This suggests that the stochastic event underlying the poly(A)-dependent termination profiles is pausing, not termination itself. The fact that the low salt and high salt relationships are parallel (within experimental error) shows that a relatively constant proportion of the polymerases along the wild type template is paused. This indicates that the paused complexes decay from the template (terminate) with similar kinetics whether pausing occurs close to or far downstream from the poly(A) site. The average horizontal distance separating the low salt and high salt profiles suggests an average pause duration that corresponds to the time required for the unpaused polymerases to transcribe ϳ220 bp of DNA.
Stochastic pausing resolves nicely some unexplained aspects of an elegant series of experiments reported by Osheim et al. (19) a few years ago. Using electron microscopic visualization of active genes, they probed the relationship between transcription and poly(A) signal activity on plasmids injected into Xenopus oocytes. Their results revealed plasmids on which transcription appeared to have stopped at various positions downstream of the poly(A) site. Behind the stoppage point on each plasmid was a dense array of transcribing polymerases extending back across the poly(A) site and up to the promoter. The authors interpreted these stop points as termination sites and assumed that all polymerases on each plasmid transcribed up to and then terminated at that same site. There were two puzzling features to these patterns. First, the positions of the stop points varied inexplicably from plasmid to identical plasmid within the same oocyte and did not correspond to any identifiable sequence features in the templates. Second, the molecular ruler that would direct all polymerases to transcribe exactly to the same arbitrarily chosen end point on each plasmid was a mystery. The idea of stochastic pausing directed by the poly(A) signal seems to provide a simple explanation for these observations. Polymerases pause randomly as directed by the poly(A) signal and then await release (or resumption of transcription). Because release is likely to follow pseudo first order kinetics, some paused polymerases will experience a greater lag before release than others. At any given time it will be the laggards of the moment that establish the distribution of pausing patterns observed on the plasmids. Note that the stop points observed on these plasmids do not necessarily correspond to sites of termination, because it has not been established that pausing events lead inevitably to release (see below).
The control experiments that we presented at the beginning of the present paper have shown that in our system there are no stably paused or arrested polymerases giving rise to any detectable polymerase backstacking. We might ask, then, whether the dense arrays of polymerases in the images of Osheim et al. (19) reflect backstacking. The answer is probably not, because their data show that the average polymerase density on plasmids that do not exhibit pausing is similar to that for plasmids that do. Nevertheless, for all images shown in their paper the polymerase density in the 3Ј-half of the array is equal to or greater than that in the 5Ј-half. This is consistent with a slight tendency of any polymerase at the head of the line, when paused longer than average, to impede the progress of those that follow. The duration of this pause cannot exceed the average reinitiation interval, however, otherwise the stacking would extend back to the promoter at steady state. On the other hand, the pause must be similar in duration to the reinitiation interval (the average polymerase spacing), or its effects would not be evident. This interval appears to be about 200 bp on the plasmids in Osheim et al. (19), a figure that is in remarkably close agreement with the estimated pause interval for our experiments (see above).
Poly(A)-dependent Pausing as a Checkpoint for Coupling Processing and Termination-Thus far we have established that separate pause sites are not required to slow down the polymerase to give the poly(A) signal time to act. Instead, we have found that the poly(A) signal on its own intercepts the polymerase within a couple hundred base pairs and causes it to pause. What then is the purpose of this kind of pause that is induced by the poly(A) signal itself? Clearly it is not designed to provide outside assistance to the poly(A) signal. We suggest that it serves as a final checkpoint designed to integrate the information on processing and the preparations for transport extant in the transcription factory at that time. In this way the appropriateness of both polyadenylation and termination at this juncture in transcription of the gene is assessed. Alternative outputs of this checkpoint would include proceeding with processing and termination, degrading the transcript, or resuming transcription. The last of these alternatives, resuming transcription, could conceivably lead to further rounds of pausing and checking until the polymerase either leaves the template or moves out of range of the poly(A) signal. A similar checkpoint model has proven attractive in accounting for events at the beginning of transcription (1,12,54). This model proposes that the pause imposed by DSIF and NELF serves as a checkpoint to ensure that proper capping (and perhaps other events) take place before P-TEFb is allowed to release the polymerase into processive elongation. Interestingly, DSIF travels with the polymerase the length of the gene as an elongation factor (55,56) and could conceivably be called upon by the poly(A) signal to do duty again as a mediator of the poly(A)-dependent pause.
Pausing and Termination Begin Earlier for Strong than for Weak Poly(A) Signals-The SV40 late poly(A) signal is several times stronger than the SV40 early signal (57). It was therefore striking that polymerases downstream of both L and E behaved so similarly (Figs. 6B and 7B). In particular we found no evidence for lower processivity downstream of L (D1 ⁄2 ϭ 447 bp), as might be expected for a stronger poly(A) signal, than downstream of E (average D1 ⁄2 ϭ 395 bp). However, one difference between these two poly(A) signals was apparent; events begin earlier for L than for E (Fig. 7A).
The high salt termination profiles for L and E are compared in Fig. 7A. There it can be seen that maximum pausing occurs at least 100 bp closer to the poly(A) site for L than for E. This suggests that stronger poly(A) signals intercept the polymerase quicker than weaker poly(A) signals. Consistent with this generalization, the position of maximum poly(A)-dependent pausing for the -s poly(A) signal mentioned above, like that for E, is more than 200 bp downstream of the cleavage site ( Fig. 6 of Ref. 27). Also like E, the -s poly(A) signal is considerably weaker than L (58).
Not only the maximum of pausing but also the very earliest events leading to termination occur closer to the poly(A) site for L than for E. The low salt termination profiles, from which the contributions of the paused polymerases have been removed, can be used to determine the point on the template at which the first detectable loss of elongating polymerases occurs. In These data can be compared with our previous results in which we studied the rate of commitment of L and E to cleavage and polyadenylation (59). There too we found that L was faster than E. Thus, L is faster both to commit to cleavage and polyadenylation and to pause the polymerase. This is consistent with the same cleavage and polyadenylation apparatus assembly process being responsible for both events.
What, Then, Are Pause Sites?-Because there is apparently not an obligatory requirement for distinct pause sites to assist the polymerase in recognizing the poly(A) signal, it is appropriate to ask how strong the evidence is that such sites exist. Alternatively, we can ask whether any of the elements currently known to enhance either poly(A)-dependent termination or cleavage and polyadenylation are known to act through pausing. It is important to keep in mind that the idea of pause sites, in the context of poly(A)-dependent termination, originated from the predictions of a widely quoted model (7,16) and do not refer to the known properties of any element.
Of the vertebrate poly(A)-dependent terminator elements said to function as pause sites, the best known are contained within 92 and 156 bp segments located downstream of the human ␣2 globin and C2 complement genes, respectively (30,60). The first to be characterized was the ␣2 element (30), which very likely acts as an auxiliary element in poly(A)-dependent termination (61). In one report, run-on transcription of the cloned ␣2 gene showed pausing in the neighborhood of this element but no termination that was poly(A)-dependent (30). In another report poly(A)-dependent termination was seen, but pausing was not (62). When placed downstream of a heterologous poly(A) site pausing was detected, but it was not determined whether this pausing was due to the ␣2 element or the poly(A) signal (30).
The other element, from the C2 gene, is also likely to be an auxiliary element in poly(A)-dependent termination (60,63). This element strongly activated a weak upstream poly(A) signal in an in vivo processing assay (60,63). For theoretical reasons this in vivo effect was ascribed to pausing (27,60,63), but an actual assay for pausing in vivo such as by run-on transcription has never been reported.
Both the ␣2 and C2 elements protect against transcriptional interference when placed upstream of a promoter (38). This effect was also ascribed to pausing (38). However, one would not expect pausing to affect the steady state flux of polymerases downstream as would be necessary to protect a promoter from interference. More likely, as previously described for the CCAAT sequence (51) these elements operate as weak poly(A)independent terminators when acting alone. Thus, whereas the ␣2 and C2 elements undoubtedly activate poly(A) signals (30,60,63) and probably also termination (30,60,61) and possibly even splicing (64), there is no consistent evidence that they actually pause RNA polymerase II in vivo or that their function is related to pausing.
Recent studies in vitro support the notion that the ␣2 and C2 elements activate cleavage and polyadenylation directly rather than through pausing, despite the conclusions of the authors (17,65) to the contrary. In a coupled transcription-polyadenylation system, the efficiency of processing was increased by polymerase arrest for 1-2 h next to one of these elements (17) but not by arrest adjacent to an irrelevant DNA-binding protein (65). Therefore, simply stopping a polymerase downstream of a poly(A) signal is of no intrinsic functional significance.
Rather, it appears to be the direct interaction of these elements with the transcription-processing apparatus that activates cleavage and polyadenylation. For example, the C2 element binds the multifunctional MAZ protein implicated in the activation, repression, and premature termination of transcription by RNA polymerase II (e.g. 35,66,67). Because RNA polymerase II is an intimate participant in the cleavage and polyadenylation reaction (68), the MAZ protein might be designed to modulate the contribution of the polymerase to processing as well. This may be related to the proposed role of MAZ in implementing fail-safe termination of transcription at promoters (35).
Very recently Peterson et al. (27) characterized a new element downstream of the mouse -s poly(A) signal. Like the ␣2 and C2 elements this new element substantially enhanced the activity of an upstream poly(A) signal in the in vivo processing assay. Although they called the new element a pause site because of its similarity to the so-called pause elements ␣2 and C2, their data reveal something quite different. In the absence of an upstream poly(A) signal, this element gave rise to a ϳ1.7-fold increase in polymerase density that persisted for 3.5 kb downstream. Thus, rather than inducing a transient pause, this element appears to facilitate the conversion of the elongation complex into a stable new state that elongates more slowly. Whether this property is related to its polyadenylation-enhancing property is unknown. Interestingly, purified E. coli RNA polymerase elongating on DNA in vitro can switch between stable states of differing elongation rates, and single molecule measurements have revealed a ratio of 1.7 between the speeds of the fast and slow forms (69).
Whether the ␣2 and C2 elements share the ability of the -s element to stably alter the elongation rate of the polymerase is not known. However, there is little or no evidence that any of these polyadenylation-enhancing elements act through pausing, although it is not ruled out that these or others may do so. The same appears to be true for similar elements, also called pause sites, in yeast (see discussion in Ref. 9). Thus, at present the poly(A) signal itself appears to be unique in its known ability to pause the polymerase at the end of the gene. We have suggested that this pausing serves as a check point. Perhaps auxiliary termination elements are designed to override this checkpoint function as needs demand.