A colocalization of the replisomes themselves was proposed in the factory model of DNA replication (6) and demonstrated cytologically (7). Instead of the left and right replisomes racing away from each other until they meet at the terminus (Fig. 1C), this model proposed that the replisomes have stationary positions in the center of the cell (7). Much as in the trombone model, colocalization is facilitated through looping of the DNA, which is fed through the factory (Fig. 1D). This model organizes replication to the cell's advantage (8, 9). Extruding the two daughter chromosomes in opposite directions should greatly facilitate chromosome partitioning (10–12). Furthermore, resources such as dNTPs, Ssb protein, and β-clamps can be concentrated in one place, improving the speed and efficiency of replication. Surveillance by repair proteins also becomes much easier if replication occurs in a single location.

As with the trombone model's colocalization of the two polymerases, colocalization of the two replisomes into a factory provides an opportunity for coordination, ensuring that the replisomes arrive at the terminus at the same time. Lemon and Grossman (7) proposed as part of the factory model that the left and right replicative hexameric helicases were joined in a dodecamer to achieve coupling of the replisomes. Together with the trombone model, this proposal leads to a strict coupling of all four polymerases: all synthesize DNA at the same rate, and blocking one leading-strand polymerase would cause all to stop. We term the obligatory coupling of left and right replisomes the coupled replisome factory model (Fig. 1E Left).

A second possibility is a facultative coupling of replisomes, a semiindependent replisome factory model. Here, the replisomes move in lock step, but uncoupling occurs if one encounters an obstacle (Fig. 1E Center). This process is similar to the uncoupling of the leading polymerase in response to blocked lagging-strand replication (4, 5). In a third possibility, the independent replisome factory model, there is no coupling between replisomes despite colocalization, and the replisomes may move at different rates (Fig. 1E Right).

DNA Combing. Cultures of 15T- (19) were grown, labeled with BrdUrd, harvested, and prepared for combing in agarose plugs (20, 21). Plugs were digested overnight by using β-agarase (NEB, Beverly, MA). Glass microscope slides were treated with 3-aminopropyltriethoxysilane (Sigma) as described (14) except that they were used within days of preparation and stored in ambient air. DNA was bound to silanated glass and processed as described (20).

Biotinylated FISH probes for oriC and digoxigenated probes for the whole genome were prepared by random priming (BioPrime kit, Invitrogen). Surfaces with combed DNA were treated with Blocking Reagent (Roche Molecular Biochemicals) for 2 h at 37°C, denatured at 90°C for 90 s, hybridized with FISH probes overnight, and washed, blocked, and treated with fluorescent avidin and antibodies (14, 20). Avidin and antibodies were from Vector Laboratories except mouse anti-BrdUrd (Becton Dickinson) and sheep anti-digoxigenin (Roche Molecular Biochemicals).

Slides were examined by using a Zeiss Axioplan 2 fluorescence microscope. We counted molecules in which labeling occurred after initiation as well as those in which BrdUrd labeling overlapped with the origin. We analyzed only those molecules with BrdUrd signals that ended within the molecule and not at a break and that had oriC FISH signals with clearly recognizable orientations. The measured length of BrdUrd tracks in microns was converted to kilobases by using the oriC FISH signal as an internal standard. To establish the 95% confidence limit for differences in BrdUrd track length, we randomly selected several molecules and measured the size of every gap within BrdUrd tracks. Ninety-five percent of 59 measured gaps were <13.6 kb; the expected error in position of the end of the BrdUrd signal is half this number, or 6.8 kb, because each spot of BrdUrd is effectively surrounded by a “half gap” on each side.

Synchronization of Bacterial Cultures. Cultures bearing the dnaC2 allele, which prevents initiation of replication but not elongation at the restrictive temperature, were grown in either M9 plus 0.2% glucose and 0.2% casamino acids. They were synchronized by incubation at the restrictive temperature (38°C or 42°C) for 90 or 120 min, followed by 2-fold dilution with medium to give a final temperature of 30°C, allowing replication to initiate. Samples (10–20 ml) were removed, treated with NaN₃ (0.3 M), and frozen in liquid nitrogen.

In the experiments with AB2937 and AB2932, the temperature was shifted back to the restrictive temperature 14 min after initiation to prevent reinitiation. In the experiment shown in Fig. 5, the glucose-containing medium was replaced by filtration with medium containing arabinose at the time of initiation. In other experiments (Table 1), the shift to arabinose occurred 30–90 min before initiation.

Microarray Procedures and Data Analysis. Preparation of microarrays and fluorescent labeling of genomic DNA was as described (22). WPC2, MG1655dnaC2, and AB2932 samples were always paired with reference DNA from the same experiment, which was taken from the culture before initiation. Because of the possibility of stalled forks at the ectopic ter, we used either MG1655dnaC2 or AB2932 reference DNA in experiments with AB2937.

Microarrays were scanned with a Genepix 4000A or 4000B scanner (Axon Instruments, Union City, CA), and 16-bit TIFF images of each fluorescence channel were acquired and gridded with genepix 3.0. Spots were filtered to require that at least 40% of pixels in each spot be at least one standard deviation above background in both channels. The results were then normalized so that the terminus had a relative abundance of one, except for late time points where forks had reached the terminus; these were normalized by using the region behind each fork corresponding to replicated DNA. The data were sorted by their position on the chromosome and smoothed by using the lowess algorithm (23) supplied with the r language and environment, version 1.6.2 (24) by using the setting f = 0.075.

Our DNA combing experiments required a high degree of incorporation of the nucleoside analog BrdUrd to robustly visualize nascent DNA. We screened thymidine-requiring derivatives of the MG1655 and W3110 backgrounds, but none gave satisfactory results. However, strain 15T- (19), known for its ability to incorporate high levels of thymidine, was able to label sufficiently well for DNA combing.

We treated 15T- cultures growing at 37°C with a 10–15 min pulse of BrdUrd. A set of biotinylated probes was hybridized to the DNA to identify the location and orientation of oriC (Fig. 2A), and both the probes and BrdUrd were detected by immunofluorescence. In most experiments, we used a third fluorescent probe of digoxigenin-labeled genomic DNA to lightly counterstain the entire combed DNA molecule (Fig. 2 C and D Bottom). The counterstain verified that the ends of BrdUrd signals did not correspond to breaks in the DNA. We only considered molecules without genomic counterstains if the BrdUrd signal was entirely within the 52-kb oriC FISH pattern, as in Fig. 2B.

Three visualized DNA molecules are pictured in Fig. 2 B–D. None showed the symmetric pattern expected for coordination between replisomes. In Fig. 2B, only the very beginning of replication occurred during the labeling period. The BrdUrd signal extends 7.3 kb to the left of oriC, but there is no signal to the right. Thus, this molecule is unidirectionally initiated. In Fig. 2C, there is bidirectional replication, but the leftward replisome is 19 kb farther from the origin. Fig. 2D also shows bidirectional replication, but this time the rightward replisome was ahead by 16 kb.

We observed a total of 46 newly initiated DNA molecules (Fig. 3A). In 32 molecules, 70% of the total, the offset between replisomes was statistically significant (>6.8 kb; see Materials and Methods). This level of variability strongly supports the independent replisome factory model. Providing further support for this model, 24 of the 32 molecules with significant variation had more replicated DNA on the left side (Fig. 3B); this bias is highly significant (P = 0.007, binomial test). Overall, the mean offset is 9 kb to the left.

Whole-Genome Analysis of Replication with DNA Microarrays. To investigate whether the independent replisome model is generally true, we tracked replication by using genomic microarrays with more commonly used E. coli strains. We synchronized cultures with the dnaC2 allele, which confers temperature sensitivity for replication initiation (16, 27). Microarrays complement our combing results by tracking fork progression throughout elongation rather than only near initiation. In our first microarray analyses, we isolated genomic DNA from WPC2, a dnaC2 derivative of W3110 (16), just before and at 12 times after replication initiation. Each timed sample was hybridized competitively with preinitiation reference DNA to a microarray containing ≈95% of the ORFs in the E. coli genome. We smoothed the results with the lowess algorithm (23) (Fig. 4A) and then fit the data to a series of straight lines. On each side of the origin, the horizontal lines corresponded to prereplication and postreplication areas of the genome, whose abundance should be constant (Fig. 4B). The sloped lines represent the position of the replication forks. Because the synchrony was not perfect, the sloped lines are not vertical and had a width of 10–15 map minutes, corresponding to the window of time within which initiation occurred. We assigned the average fork position as the midpoint of the sloped line. Reinitiation started ≈30 min after the first initiation. It was well separated from the first wave of forks but poorly synchronized, and thus it was not analyzed.

We plotted the distance of each fork from the origin versus time (Fig. 4C). We draw three conclusions from these data. First, the average rate of replication is constant for both forks. The data for each replisome fit very well to a straight line (R² = 0.992 and 0.972 for the left and right forks, respectively). Second, the average speeds of the two forks are, within error, identical: 604 nt/s for the left replisome and 617 nt/s for the right at 30°C. Thus, the forks move at nearly the same speed throughout elongation. These values are consistent with the speed of 1 kb/s at 37°C previously measured for DNA polymerase III (1). Third, the left replisome was farther from the origin by ≈150 kb. Thus, the only consistent difference between the two replisomes occurs at or just after initiation, when the leftward replisome establishes a lead. This difference is compatible only with the independent model.

The raw microarray data used to generate Fig. 4C showed that WPC2 had acquired a second inversion around oriC in addition to the documented inversion in W3110 (28). To ensure that the initiation offset did not result from an undetected chromosomal rearrangement, we hybridized nonreplicating WPC2 genomic DNA to a microarray with nonreplicating genomic DNA from MG1655dnaC2 as a reference. We discovered that WPC2 contained a tandem chromosomal triplication on the right side of oriC (Fig. 7A, which is published as supporting information on the PNAS web site). This finding was confirmed by PCR (Fig. 7B). Our analysis, including calculation of offset size, takes these rearrangements into account; notably, the second inversion causes the orientation of oriC to be the same as in MG1655-derived strains, but not other W3110 derivatives.

To test the generality of the offset and its size, we performed repeated synchrony experiments with different strains. With a dnaC2 derivative of MG1655 (29), the strain used for genome sequencing (30), there was a 40-kb offset to the left. The mean for all MG1655 experiments was 51 kb (Table 1). We also used strains derived from W3110 with a single chromosomal inversion about oriC (28). As predicted, in these inverted strains, the offset switched sides; the rightward fork now had the lead, and its mean size in W3110-derived strains was 130 kb (Table 1). This result is consistent with the asymmetry of oriC or its nearby environment being the root of the offset between replisomes. One possible consequence of an offset in initiation is that one replication fork would arrive first at the terminus because of its head start. However, the offset size we observed should not be large enough to cause a large difference; the size of the region containing the central four ter sites is ≈200 kb (31).

Blocking One Fork Does Not Stop the Other. According to the independent replisome model, one fork can proceed even if the other is blocked. We tested this prediction directly by using strain AB2937, a dnaC2 derivative of KHG1005 (17). This strain has an ectopic ter site at lacZ to block the rightward replisome. ter sites bind Tus protein and the Tus–ter complex DnaB helicase and stall replication (32).

Our results might also be consistent with the precursor to the factory model, the train-on-tracks model (33) (Fig. 1C), where each replisome moves independently along its template as it copies DNA. However, there is an abundance of cytological evidence that favors localization of polymerases and other replication proteins as well as nascent DNA into foci for several bacterial species, including E. coli (34), Bacillus subtilis (7), and Caulobacter crescentus (35).

In fact, we see functionally independent yet colocalized replisomes as the best of both worlds. The cell receives the benefits of replisome colocalization, concentration of resources and repair factors in one place, and facilitation of segregation, while avoiding potential problems caused by coordination. Replication forks break down frequently (36), and it is inefficient for both forks to stop every time one encounters a problem. Additionally, replication near the terminus presents a unique problem; it becomes increasingly difficult to maintain a (–) supercoiled template as the size of that template decreases, so it is possible that forks stall more frequently there. If one fork continues while the other is stopped, this problem would be alleviated because torsional strain could more easily diffuse away from the active fork through the inactive one. Finally, in extremis, one fork may be able to get through the terminus region and complete the other fork's job if it has irretrievably halted. Consistent with this possibility, in a B. subtilis strain with a 900-kb insertion on one side of the chromosome, a significant number of forks from the other side penetrated beyond the first ter site after reaching the terminus region, whereas their counterparts were delayed by the insertion (37). Similarly, we observed a population of forks that bypassed the ectopic ter (Fig. 5).

In cells with multiple origins per chromosome such as eukaryotes, strict coordination of replication forks is also impossible. Origins are variably spaced and fire at different times (38), so one fork will terminate before the other (Fig. 6), particularly in the cases of DNA damage (36) and the rDNA repeats in Saccharomyces cerevisiae (39). Therefore, the commonly held view of replication as generally bidirectional is only true early in replication of a replicon; it must generally be unidirectional at the end.

Furthermore, the textbook view of replication initiation as occurring equally and symmetrically in each direction also must be modified. Leftward and rightward initiation are most likely two separate events. In many bacteria, phage λ, and budding yeast, origins are consistently asymmetric, with a replicator region that binds the initiator protein beside a zone where melting, replisome assembly, and initial priming occur (1, 40–43). In vitro evidence indicates that the two DnaB hexamers are loaded differentially (44). After E. coli DnaA has melted the AT-rich 13-mers on the left side of oriC, the leftward DnaB helicase is recruited. DnaB must then denature additional DNA before priming and further replisome assembly can occur, including the loading of the rightward helicase (44). In vivo, the first rightward primer was found to have a variable location that was sometimes outside and even to the left of oriC (45), and initiation at the λ replication origin showed a bias toward its melting side (40), hinting at the bias and stochasticity described here. The replication initiator complex must be assembled at origins carefully and precisely (46). We suggest that the leftward helicase is loaded first because it moves away from the initiator, whereas the rightward helicase will plow through the initiator–replicator complex, likely rendering it incompetent to direct further initiation events. In this model, the leftward helicase always starts first, and the fraction of DNA molecules with either rightward or no bias that we observed (Figs. (Figs.2D2D and and3)3) arose because the rightward replisome moved faster than its sister. An alternative possibility is that DnaA and other proteins bound at oriC slow the rightward replisome, but we consider this notion unlikely because replisomes did not pause specifically at the the datA locus, which binds up to 60% of the DnaA protomers normally in the cell (47), even when DnaA was overexpressed (22).