Main

We previously used DNA combing to analyse the replication dynamics along the AMPD2 (adenosine monophosphate deaminase 2) region3 in Chinese hamster fibroblasts highly amplified for this locus (Supplementary Fig. 2). Mutant lines were selected for their resistance to coformycin, an inhibitor of AMPD2, which depletes cells of guanine derivatives4. In mutant line 471, replication forks proceed at about 1.3 kilobases (kb) min-1 (referred to below as ‘fast’). The density of initiation events is low, and about 80% of the events observed within the locus take place at oriGNAI3, a previously characterized replication origin5,6,7, the remaining events being scattered between five other discrete origins. In contrast, forks travel at about 0.6 kb min-1 in line 474 (referred to below as ‘slow’). Initiation events occur at high density and seem evenly distributed between the five origins and oriGNAI3 (ref. 3). Because the selection of mutant lines relied on unbalanced nucleotide pools, we reasoned that abnormal pools of DNA precursors might persist in cells of line 474 (‘474 cells’). Indeed, the simple addition of adenine and uridine to the culture medium (referred to below as A + U) was sufficient to convert the replication pattern of those slow cells to that observed in fast cells3 (Fig. 1a). Conversely, hydroxyurea-induced nucleotide starvation slows fork speed in cells of line 471 (‘471 cells’) and results in the redistribution of initiation between all six origins of the locus. Altogether, these results reveal a highly controlled switch in origin usage when cells face variations in growth conditions2,3, offering a model with which to address the question of how and when origin choice is established.

Figure 1: The density of active origins depends on replication speed.
figure 1

a, Combed DNA molecules (DNA stained in blue) from 474 cells grown in normal or in A + U medium. A schematic representation of the replication patterns is shown below each molecule. b, Localization of initiation events (one dot per event) along the AMPD2 region in 471 cells grown in normal medium or challenged for 2 h or 6 h with 75 ng ml-1 aphidicolin (APC). ME, number of molecules with multiple initiation events; n, number of molecules analysed. c, DNA molecules (as in a) from 471 cells labelled with IdU, then with CldU with or without APC (75 ng ml-1), as described in ref. 27. Left: combed molecules. Initiation events occurring during: the first pulse (i), the second pulse on DNA regions that were not replicating during the first pulse (ii), and the second pulse on DNA regions that did start to replicate during the first pulse (iii). Right: percentage of events observed in each category; n = 450.

PowerPoint slide

Full size image

We first determined whether the initiation pattern is controlled by nucleotide pools or by fork movement. Fast 471 cells were treated with aphidicolin (75 ng ml-1), a DNA polymerase inhibitor that does not interfere with the pools. Analysis of combed DNA molecules from cells labelled with iododeoxyuridine (IdU) then with chlorodeoxyuridine (CldU) showed that replication speed decreased to about 0.4 kb min-1 but cell cycle progression remained roughly normal under these conditions (Supplementary Fig. 3). The distribution of initiation events along the AMPD2 locus was studied by combining fluorescence in situ hybridization (FISH) with detection of IdU and CldU on newly synthesized DNA3. We observed a strong decrease in oriGNAI3 prominence 2 h after the addition of aphidicolin, and initiation events seemed evenly distributed between the different origins after 6 h. Loci showing more than one initiation event increased correspondingly (Fig. 1b). Note that oriGNAI3 still fired, at least on half the molecules that we observed (Supplementary Fig. 4a); loss of its prominence therefore results partly from an increased efficiency of the other origins.

We also performed a global analysis of the replication dynamics in 471 cells that were labelled with IdU in normal medium (fast) and then with CldU with or without aphidicolin. Comparison of the lengths of the IdU and CldU tracks showed that fork speed decreased almost instantaneously on the addition of aphidicolin (Supplementary Fig. 4b). The distribution of initiation events occurring during each labelling period was studied by focusing on 200-kb DNA regions (Fig. 1c). As expected, the same frequency of initiation was observed during the first pulse whether or not the cells were treated with aphidicolin during the second pulse (Fig. 1c, i). In cells not challenged by aphidicolin, initiation during the CldU-labelling period was observed with one-tenth the frequency on regions already labelled by IdU than on regions free of IdU signal (Fig. 1c, compare ii with iii). This confirms that the density of initiation events is low when replication forks travel fast. By contrast, in cells treated with aphidicolin, the probability of firing during the second pulse was similar on both types of region (Fig. 1c, compare ii with iii), confirming that the density of initiation events is high when forks travel slowly.

Taken together, our results show that fork speed itself, rather than the nucleotide pools, controls the pattern of initiation. They further show that, within 30 min, the cells start to compensate for the decrease in fork speed by mobilizing latent origins, which are thus able to change their fate within S-phase. Because licensing is prevented after the onset of S phase8, we conclude that the pre-initiation complex is present on these latent origins.

We then studied the reciprocal situation, namely the specific silencing of some origins when 474 cells are shifted from slow to fast conditions. A global analysis of the replication dynamics showed that fork speed reached the maximum value (about 1.8 kb min-1) 15 min after A + U addition (Supplementary Fig. 5a). To assess the distribution of initiation events, we focused our analysis on long DNA molecules (Fig. 1a). After only 2 h in A + U, the density of initiation events decreased by 30% and remained stable at later time points (Supplementary Fig. 5b). At each time point, initiation events seemed confined to a restricted region along any individual molecule that we observed (Fig. 1a). This is consistent with previous work in yeast and mammalian cells showing that origins are functionally organized as clusters9,10,11. Strikingly, the total length of the labelled regions relative to that of the DNA molecules was constant (Supplementary Fig. 5b). Hence, modulation of the density of initiation according to replication speed occurs mostly at the level of individual clusters, a finding consistent with the results obtained in aphidicolin-treated 471 cells (Fig. 1c).

We also determined how the replication pattern evolved along the AMPD2 locus after shifting 474 cells to A + U medium (Supplementary Fig. 6). We found that the proportion of loci displaying close initiation events (separated by less than 90 kb) decreased from 80% to about 50% within 6 h. However, it took about 16 h before oriGNAI3 prominence was established, roughly coinciding with the doubling time of these cells. Because we observed cells in only a narrow window of S phase (the locus replication time), this correlation suggested that the cells have to go through a complete cell cycle in fast medium to reprogram the relative efficiency of the origins. We tested this hypothesis by selecting mitotic 474 cells (slow) and then replating them in fast conditions. We specifically analysed cells at the first and second S phases after replating (Fig. 2). In both cases, closely spaced initiation events were observed on about 50% of the molecules displaying several events, as in unsynchronized fast 474 cells. The initiation events remained evenly distributed between all the origins of the locus during the first S phase, and oriGNAI3 became prominent at the second S phase.

Figure 2: oriGNAI3 prominence appears only during the second S phase after shifting 474 cells to fast conditions.
figure 2

Left: FACS analysis of cells growing exponentially, treated with nocodazole, and recovered after mitotic shake-off. The lower box contains cells in G1 + S + G2; the upper box contains mitotic cells. The percentage of cells in each group is indicated. Right: distribution of initiation events along the AMPD2 locus during the first and the second S phases after replating in A + U medium (symbols are as in Fig. 1b). Below, multiple initiation events (ME) are separated into two categories depending on the inter-origin distance.

PowerPoint slide

Full size image

We conclude that remodelling of the initiation pattern by an increased speed occurs in two steps. First, some origins are randomly silenced, probably because they are passively replicated by forks emanating from origins activated earlier. Next, if conditions of high replication speed are maintained, an additional mechanism commits some origins, such as oriGNAI3, to fire preferentially during the following S phase. This origin hierarchy remains stable in the absence of further speed variations3.

Replicon size, which is dictated by the spacing of active origins, has long been correlated with the length of chromatin loops12. We therefore studied the size of the loops surrounding the nuclear matrix (Supplementary Fig. 7) and found that they were periodically remodelled during the cell cycle. Furthermore, in G1 nuclei, their mean size increased twofold in fast 474 cells in comparison with slow ones. To some extent, the correlation held in S and G2 phases, even though the difference was less striking. The study of fast and aphidicolin-treated 471 cells confirmed these observations (Supplementary Fig. 8), establishing a correlation between fork speed and loop size. The existence of the nuclear matrix remains controversial13. However, a functional relationship between replication and attachment to an operationally defined matrix is now well documented14. For example, in Chinese hamster ovary (CHO) cells, a matrix attachment region (MAR) is required for the maintenance of plasmids not integrated in the chromosomes15. Moreover, all the origins of the AMPD2 locus co-localize with MARs16, similarly to many origins characterized in vertebrates17. Here we further show that, depending on the growth conditions, halos with different and specific sizes could be obtained from the same cells, salt-extracted and further treated in parallel.

We then determined whether the spatial distribution in the halos of oriGNAI3, oriA, oriB and a non-origin sequence depended on the replication dynamics (Fig. 3a). In small halos seen with G1 nuclei of slow 474 cells, all four sequences were similarly distributed between the matrix and the loops. In large halos obtained with G1 nuclei of fast 474 cells, whereas oriB, oriA and the non-origin sequence still seemed distributed, oriGNAI3 localized preferentially at the matrix. In G2 nuclei of these cells, oriGNAI3, like oriB, was randomly distributed in the halos (Fig. 3a). The very same distribution of oriGNAI3 was observed in fast 471 cells (Supplementary Fig. 8a). This origin therefore strikingly relocalizes from the matrix to the loops between G1 and G2 and from the loops to the matrix during mitosis or early G1 in cells grown in fast conditions. By studying cells sorted by fluorescence-activated cell sorting (FACS), we found that oriGNAI3 relocalized from the matrix to the loops during S phase (Fig. 3a and Supplementary Fig. 7e). This agrees with previous studies showing that newly synthesized DNA moves away from the base of the loops in mammalian cells18 and that origins are pulled out of the replisome on fork progression in yeast cells19. We then determined when oriGNAI3 relocalized from the loops to the matrix by performing a time-course analysis of oriGNAI3 and oriB in fast 474 cells released from a mitotic block (Fig. 3b). OriGNAI3 was reattached to the matrix as early as 1 h after release, suggesting that the switch occurs in mitosis. This interpretation is supported by previous data showing that origins can be reset in mitotic Xenopus egg extracts in a topo-II-dependent manner20,21 and that resetting correlates with significant remodelling of chromatin loops21. In addition, specific movements of DNA sequences during mitosis were observed in vivo in CHO cells22. Our data do not account directly for the observation that initiation sites are determined in G1, at a step called the origin decision point23,24. However, the attachment of origins to the matrix may be a prerequisite for the further selection of initiation sites.

Figure 3: Large halos and preferential binding of oriGNAI3 to the nuclear matrix in G1 of ‘fast’ 474 cells.
figure 3

a, Exponentially growing cells at steady state. Left: halos from cells in G1 and G2. FISH was performed with probes for oriGNAI3 or oriB (green signals). DNA is shown in grey (reverse DAPI). MFHR, maximum fluorescence halo radius. The position of the stacks used for quantification of the FISH signal on all halos is shown: M is positioned at the halo-matrix border (0 to 1.5 µm from the matrix) and H1 to H5 are positioned every 1.5 μm. Right: histograms showing the MFHR and the percentage of FISH signals per stack. The non-origin sequence maps between oriD and oriF. b, Synchronized fast 474 cells in early G1. Histograms as in a. c, Synchronized 474 cells shifted from slow to fast conditions. Histograms as in a. Because the cells were partly desynchronized in the second G1 phase, they were sorted according to their DNA content. All error bars represent the s.e.m. of 35–50 halos.

PowerPoint slide

Full size image

To determine whether the pattern of initiation during the previous S phase sets loop size and origin localization in G1, we studied the fate of mitotic 474 cells (slow) that had been replated in either slow or fast conditions (Fig. 3c). At the first G1 phase after the shift to fast conditions, loop size and both oriGNAI3 and oriB localization remained the same as in slow cells. In contrast, at the second G1 phase, loop size increased and oriGNAI3 localized preferentially to the matrix. The study of the reciprocal situation, namely the fate of mitotic 471 cells (fast) replated in the presence of aphidicolin, confirmed that halos seemed remodelled only at the second G1 phase after the shift (Supplementary Fig. 8b). We conclude that the structural organization of the loops in G1 depends on the replication pattern in the previous S phase, and that the association of each origin with the matrix probably determines its probability of firing.

To account for the observation that the size of the loops in G1 depends on the replication dynamics during the previous S phase, we propose that replication marks, whose spacing depends on the density of initiation events, are present on the DNA until mitosis. We infer that such marks could be termination regions. Under fast conditions, large replicons would lead to distant marks that, in turn, organize large loops favouring the anchorage of origins with the highest affinity for the matrix. In slow conditions, the close DNA marks would give rise to small loops that offer all potential origins an opportunity to bind to the matrix. Finally, we postulate specifically that attached origins fire earlier than unattached ones, thereby establishing the flexible origin use observed in our experiments (Supplementary Fig. 1).

Methods Summary

Preparation of halos, and FISH

Halos were prepared as described12,21, with the following changes: nuclei were isolated by incubating cells in ice-cold NP40 buffer (0.5% Nonidet P40, 10 mM MgCl2, 0.5 mM CaCl2, 25 mM Tris-HCl pH 8.0) for 5 min. Nuclei were washed in ice-cold PBS, stained with 4,6-diamidino-2-phenylindole (DAPI; 2 μg ml-1) and sorted with a FACSVantage (Becton Dickinson) according to their DNA content. Sorted nuclei were spread on SuperFrost slides by Cytospin centrifugation before salt extractions as described21. Halos were then fixed for 10 min with 2% formaldehyde. FISH experiments were performed as described25. Maximum fluorescence halo radius was measured with Fluovision software (Imstar). Quantification of FISH signals was performed with ImageQuant v. 5.0 software (FujiFilm).

Cell synchronization procedure

Mitotic selection was performed as described26, after treatment of the cells for 3 h with nocodazole (200 nM). The quality of the mitotic cell preparations was verified by FACS analysis of MPM2-positive (mitotic protein monoclonal 2) cells. The cells were replated in fresh medium and then recovered after 5 h for the first G1 phase, after 12 h for the first S phase and after 29 h for the second S phase. To obtain cells in the second G1 phase, because of a partial desynchronization, the cells were recovered after 25 h and sorted by FACS with a FACSVantage according to their DNA content.

DNA combing

Combing was performed as described in ref. 3.