Temporal regulation of origin activation is widely thought to explain the pattern of early- and late-replicating domains in the Saccharomyces cerevisiae genome. Recently, single-molecule analysis of replication suggested that stochastic processes acting on origins with different probabilities of activation could generate the observed kinetics of replication without requiring an underlying temporal order. To distinguish between these possibilities, we examined a clb5Δ strain, where origin firing is largely limited to the first half of S phase, to ask whether all origins nonspecifically show decreased firing (as expected for disordered firing) or if only some origins (“late” origins) are affected. Approximately half the origins in the mutant genome show delayed replication while the remainder replicate largely on time. The delayed regions can encompass hundreds of kilobases and generally correspond to regions that replicate late in wild-type cells. Kinetic analysis of replication in wild-type cells reveals broad windows of origin firing for both early and late origins. Our results are consistent with a temporal model in which origins can show some heterogeneity in both time and probability of origin firing, but clustering of temporally like origins nevertheless yields a genome that is organized into blocks showing different replication times.

DNA replication in eukaryotic cells is a complex enterprise that appears to be optimized for anything but speed. Although each chromosome has multiple origins that are capable of supporting replication initiation, the number of origins that actually initiate DNA synthesis (“fire”) varies. Some origins are efficient, firing in virtually every cell in a population, while others are inefficient, firing in only a subpopulation of cells. Thus, the potential density of active origins is greater than the actual density. Furthermore, those origins that fire appear not to do so simultaneously. A few, well-characterized origins have been shown to fire later in S phase (Ferguson et al. 1991; Friedman et al. 1997; Yamashita et al. 1997). Therefore, the density of origins, their efficiencies of activation, the times at which they fire, and the relative rates at which forks move should determine how long it takes to fully replicate a eukaryotic genome and the order in which it is replicated.

Recent analysis of Saccharomyces cerevisiae chromosome VI replication by in vivo labeling and single-molecule analysis by DNA combing (Czajkowsky et al. 2008) suggested that there is no obligate order of origin firing along any single chromosomal DNA molecule and that the observed temporal pattern of replication for a population of molecules could be explained largely by variable probabilities of origin firing without the need to invoke temporal staggering of initiations at different origins. A similar suggestion has been made regarding origin firing in fission yeast and metazoans (Rhind 2006). DNA fiber images are compelling, but their analysis is not entirely straightforward, requiring large sample sizes and different pulse/chase regimens to capture the complexity of the population. Nevertheless, they do raise questions about the conclusions of previous work, which all pointed to the existence of a temporal firing pattern. In fact, the results of some previous studies could be interpreted either way—as a temporal firing pattern or as population aggregates of origins with different efficiencies. For example, mapping of single-stranded DNA (ssDNA) after hydroxyurea treatment identified origins that are either unchecked or checked by the checkpoint protein Rad53p (Feng et al. 2006) largely along the lines of what previously had been classified as early- vs. late-replicating origins, respectively. However, the relative abundance of ssDNA at different origins matched more closely with origin efficiency than with replication time for the few origins for which efficiency had been measured.

We started by considering the two extreme possibilities: (1) that, as proposed for Schizosaccharomyces pombe (Patel et al. 2006), origin firing time is essentially random, with the more efficient origins in the genome firing in a larger proportion of cells and therefore on average replicating earlier in S phase, and (2) that origins are programmed to fire at different times in S phase, with efficient firing potentially being the property of any temporal class of origins. One way to distinguish between these extreme possibilities is to use mutations that affect the temporal availability of an initiation factor. For example, in a mutant where origin initiations can occur only early in S phase, the temporal activation model predicts that origins that normally fire in the first half of S phase should fire normally, while origins programmed to fire late in S phase would be unable to fire and therefore be replicated passively from forks from the earlier-activated origins (Figure 1A, top). The predicted mutant phenotype, in this model, is that some origins should show normal activation and normal kinetics of replication, while other origins should show reduced activation and delayed replication (Figure 1, B and C, top). In contrast, the stochastic, disordered firing model (Figure 1A, bottom) predicts that if initiations in the mutant could occur only in the first half of S phase, then with no set firing time, the observed efficiency of all origins would be reduced (due to passive replication across these origins in the second half of S phase), and replication kinetics would be slowed across the genome (Figure 1, B and C, bottom).

One such mutant is clb5Δ. The CLB5 gene product is one of nine yeast cyclins that activate and regulate the cyclin-dependent kinase CDK (Cdk1p, CDC28) (Mendenhall and Hodge 1998). During a normal S phase, CDK activated by Clb5p or a second, short-lived, cyclin Clb6p, stimulates origin firing (Mendenhall and Hodge 1998; Jackson et al. 2006). When CLB6 is absent, S phase proceeds with no apparent defect, indicating that Clb5p-CDK alone is sufficient to direct a normal round of replication (Schwob and Nasmyth 1993). In the absence of CLB5, however, S phase is substantially longer than in wild-type cells (Epstein and Cross 1992; Kühne and Linder 1993; Schwob and Nasmyth 1993; Donaldson et al. 1998b), indicating that Clb6p-CDK on its own cannot fully substitute for Clb5p-CDK.

We showed previously that cells lacking CLB5 suffer a significant decrease in the firing efficiency of some origins. Among the handful of origins tested, the defect appeared greatest for those that had been classified as late-S activated (Donaldson et al. 1998b), consistent with the expectation for the temporal program model. However, left open were the possibilities that some origins cannot be activated by Clb6p or that some of the activation seen in the clb5Δ mutant could be ascribed to the other B-type cyclins, such as Clb3p or Clb4p. Furthermore, the two-dimensional (2-D) gel analysis did not address whether initiation at “early” origins in the mutant proceeded with quantitatively normal kinetics, a key discriminator between the temporal program model and the disordered firing model.

Here, we have extended the 2-D gel analysis by examining origin firing in synchronous S-phase cultures of wild-type and clb5Δ strains and have examined the genomewide kinetics of replication in the two strains. We find that multiple zones, together composing over half of the genome, suffer significant delays in replication in a clb5Δ mutant, while the remainder of the genome is largely unaffected, consistent with the temporal program model and not consistent with the simplest interpretation of the disordered firing model. CLB5-dependent regions (CDRs) occur on all chromosomes and in blocks up to hundreds of kilobases in size. In general, CDRs correspond to regions of the genome that on average replicate late in S phase (Raghuraman et al. 2001; Alvino et al. 2007). Introduction of a stable version of Clb6p (Jackson et al. 2006) advanced the replication timing of CDRs and restored a temporal pattern of replication that is indistinguishable from wild type, confirming that Clb5p and Clb6p are equally capable of directing CDK to the correct targets and validating the premise underlying our test of the models. Our results best support a model in which each origin has a distribution of firing times centered about a mean that reflects the time of maximum probability of its activation. Mean firing times range from early to late S, but because origins have overlapping temporal distributions of initiation, this model would preclude an obligate order of firing. Nevertheless, clustering of origins with approximately similar times of activation leads to a genomic organization into temporal blocks of early and late replication.


Strains used:

All strains are derivatives of A364a. BB14-3a (MATa bar1 ura3-52 trp1-289 leu2-3,112 his6) is a CDC7 derivative of RM14-3a (Donaldson et al. 1998a; McCarroll and Fangman 1988). HM14-3a (wild type) is a URA3 derivative of BB14-3a. The clb5Δ mutant strain is described elsewhere (Donaldson et al. 1998a). CW1 (CLB5 CLB6−HA), CW2 (clb5Δ CLB6−HA), and LD3 (clb5Δ clb6Δ100-HA) were derived from HM14-3a or the isogenic clb5Δ mutant strain by integration of the modified CLB6 alleles at the native CLB6 locus. The HA tags were derived from pFA6a-3HA-TRP1 (Longtine et al. 1998). All cultures were grown at 30°. GA14-3a (MATa bar1 cdc7-1 trp1-289 leu2-3,112 his6 ρ0) was produced by culturing KK14-3a (Raghuraman et al. 2001) in the presence of ethidium bromide (G. Alvino, B. Brewer and M. K. Raghuraman, unpublished results) and was grown at 23°.

Dense isotope transfer:

Dense isotope transfer experiments were performed as described (Donaldson et al. 1998a; McCarroll and Fangman 1988), increasing sample volumes 9- to 20-fold. For the time-course experiments with KK14-3a and the HA-tagged strains, a single set of collections was performed for each strain. For the pooling experiments, two separate experiments were performed, transfer A and transfer B.

Determination of genomic percentage of replication:

DNA was extracted from pooled or discrete timed samples of dense isotope transfer experiments as described elsewhere (Brewer and Fangman 1987; Huberman et al. 1987; Rose and Winston 1990). The DNA was digested with EcoRI and fractionated in CsCl density gradients. The fractions were slot blotted and probed with the BB14-3a (ρ+) or GA14-3a (ρ0) genomic DNA probe to identify fractions containing DNA and to determine the proportions of heavy-heavy (HH) and heavy-light (HL) DNA in each pooled sample.

Microarray production and analysis:

All microarrays display the collection of S. cerevisiae open reading frames (ORFs). The discrete timed samples were hybridized to Agilent yeast oligo microarrays (“Agilent arrays”) at the University of Washington Center for Expression Arrays according to the manufacturer's specifications. The pooled S-phase samples were hybridized to microarrays from the Fred Hutchinson Cancer Research Center DNA Array Lab (“FHCRC arrays”; GEO platform accession no. GPL1914, http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GPL1914). Production, scanning, and analysis procedures for those arrays are described elsewhere (Fazzio et al. 2001).

Microarray target DNA preparation, labeling, and microarray hybridization scheme:

Microarray target DNA was obtained by isolating HH and HL DNA from CsCl gradients. For the pooled experiments, “single-copy” DNA standards were isolated from α-factor-arrested G1 cells of the appropriate strain in a similar manner. The target DNA (HH, HL, or single copy; 1–2.5 μg) was concentrated by ethanol precipitation and labeled for hybridization to microarrays essentially as described elsewhere (http://cmgm.stanford.edu/pbrown/protocols/4_genomic.html), using 15 μg of random hexamers as primers and 50 units Klenow fragment (3′ → 5′ exo−; New England Biolabs, Beverly, MA). When labeling was complete (2–4 hr at 37°), the reactions were mixed with 60 μg tRNA and purified on 1-ml Sephadex G50 columns equilibrated with 10 mm Tris-HCl pH 8.0, 1 mm EDTA. Following purification, targets were combined for hybridization, mixed with 20 μg poly(dA)–poly(dT), and ethanol precipitated. For hybridizations to the FHCRC arrays the labeled DNA targets were resuspended in 26 μl 3× SSC (45 mm Na-citrate, 450 mm NaCl) and SDS was added to 0.3% prior to hybridization. For hybridizations to the Agilent arrays, samples were prepared and hybridized according to the manufacturer's specifications.

For both the wild-type and the clb5Δ mutant strain, HH and HL DNA from the pooled S-phase collections were labeled separately with Cy3-dUTP (Amersham Biosciences, Piscataway, NJ), combined with the single-copy DNA standard (labeled with Cy5-dUTP), and hybridized to separate microarrays. The two microarrays (HH plus standard and HL plus standard) compose one hybridization set. “Reciprocal” hybridization sets, in which HH and HL DNAs were labeled with Cy5-dUTP and the standard was labeled with Cy3-dUTP, were also produced. The two hybridization sets for each strain constitute one data series. Series A contains wild-type and clb5Δ data from dense isotope transfer A, and series B contains data from transfer B.

For the analyses of discrete timed samples from KK14-3a and the HA-tagged strains (CW1, CW2, and LD3), HH and HL DNAs were differentially labeled with Cy5-dUTP and Cy3-dUTP and hybridized to the same microarray. Reciprocal hybridizations were performed here as well.

Microarray data collection and normalization:

Microarray data were extracted from the FHCRC arrays with GenePix 4.0 software (Axon Instruments, Foster City, CA). ORFs were excluded from further analysis if the DNA standard signals were not three times above background or if they differed significantly within a hybridization set. In addition, a set of 140 repetitive ORFs (supplemental Table 1) was excluded from further analysis to avoid spurious results due to cross-hybridization. The microarray signals were also locally normalized by SNOMAD (Colantuoni et al. 2002). Hybridization data were extracted from the Agilent arrays according to standard procedures.

For all experiments, global normalization was applied to correct for the number of cells that actually underwent replication in the analyzed cultures (determined from either the maximum percentage of small- and medium-budded cells or the maximum percentage of replication achieved by the culture) and for the relative amounts of total HH and HL DNA within the pooled samples (Alvino et al. 2007).

Mathematical analysis of microarray hybridization data:

Data processing was performed essentially as described previously (Raghuraman et al. 2001; Alvino et al. 2007). The intensity ratios of HH/standard and HL/standard (pooled sample experiments) or total HH and HL intensities (discrete sample experiments) for each ORF were used to find the percentage of replication values:MathEach value was plotted to the center of the corresponding ORF and values for the intervening regions were determined at 1-kb intervals by linear interpolation. The interpolated values were averaged over a 12-kb sliding window (moving in 1-kb increments) and used as a target for Fourier convolution smoothing (Raghuraman et al. 2001).

Comparison of wild-type and clb5Δ replication profiles from pooled samples:

The variation in percentage of replication between pooled wild-type and clb5Δ replication profiles was determined from the absolute difference in average percentage of replication between the two strains at 1-kb intervals. The wild-type vs. clb5Δ difference value for each coordinate i was converted to units of standard deviation (Z scores), usingMathwhere xi is the difference value for coordinate i, μs is the average difference in percentage of replication overall for the same-strain comparisons (μs = 1.327, obtained by taking the same-strain absolute difference at each coordinate and then computing the mean of these same-strain difference values), and σs is the same-strain standard deviation (σs = 0.6604). Standard Q-Q plots were used to confirm normality for the distribution of same-strain comparison difference values. On the basis of an estimated false discovery rate of 0.005 (Benjamini and Hochberg 1995), the threshold for significance was set at Z = 4.886. Coordinates were deemed inconclusive if the same-strain comparisons yielded Z scores greater than the threshold. Of the remaining coordinates, those where the wild-type to clb5Δ comparisons yielded Z ≥ 4.886 were designated as CDRs and those with Z < 4.886 as “non-CDR.” CDR loci were ranked as categories 1, 2, or 3 (Z scores of 4.886–9.15, 9.15–16.335, and >16.335, respectively).

Characterization of origins from microarray profiles at intervals through S phase:

We compared each wild-type replication profile to a background profile produced by hybridizing the “light” shoulder of the peak of unsubstituted genomic DNA purified from a CsCl gradient, which identifies DNA fragments that contaminate the HL region of the density transfer gradients because of higher A + T content or small size, or both (Alvino et al. 2007). We determined the standard deviation for the background profile and used it to assess the significance of each local maximum in the S-phase samples. An origin is considered to be active if it produces a local maximum (in all replicates for any of the S-phase samples) that is six standard deviations above the background sample and three standard deviations above the adjacent minima. While such stringent criteria allow us to remove spurious peaks that arise from genomic fragments with unusually high A + T content, they also likely exclude some inefficient origins in the genome. In a few cases the distance between two adjacent origins prevented either origin from scoring as a significant maximum because the minimum between them was high, yet it was clear from the surrounding region that one or the other (or both) of these origins was active. These origins were included if the minima flanking the two origins were 3 standard deviations below the average maxima of the two origins. Two hundred origins were identified by these criteria.

A subset of 64 origins was analyzed for their initiation time and efficiency. To reduce the contribution of passive replication, only origins that were >50 kb from their flanking origins were included. At each sampling interval the percentage of replication at the two flanking local minima was averaged and subtracted from the percentage of replication of the maxima (origin). This value is then compared to the value for the previous sample time (or the background control in the case of the 10-min S-phase sample) to determine the change in origin activation across S phase.

2-D agarose gel electrophoresis:

For asynchronous samples, cells were harvested from early-log-phase cultures. For synchronous samples, cells were grown to early-log phase, arrested with α-factor, and released into S phase for collection of timed samples. DNA was prepared as described previously (Huberman et al. 1987; Brewer et al. 1992). First-dimension gels (0.4% agarose) were run for 20–22 hr and second-dimension gels (1–1.1% agarose) were run for 4.5 hr at 4° in the presence of ethidium bromide (Friedman et al. 1995). The fragments analyzed are ARS501 [chromosome (chr.) V, 547,714–550,909], ARS607 (chr. VI, 196,963–201,291), ARS1413 (chr. XIV, 248,983–252,999), ARS1414 (chr. XIV, 277,747–281,720), ADE2 ARS (chr. XV, 562,300–567,264), and ARS1502 (chr. XV, 599,062–603,953).

Flow cytometry:

Cell samples for 2-D gel analysis were assayed for cell-cycle progression by bud emergence and DNA content (by flow cytometry). Preparation of samples for flow cytometry was performed as described previously (Nash et al. 1988) with some modification. Each sample was pelleted, resuspended in sterile water for sonication, and fixed overnight at 4° in 70% ethanol. Fixed cells were washed once in 50 mm sodium citrate (pH 7.4), resuspended in sodium citrate, and treated with 0.25 mg/ml RNAase A for 1 hr at 50° or 55°. Samples were further treated with proteinase K (1 mg/ml) for an additional hour at 50° or 55°. The cells were then pelleted and resuspended to appropriate concentrations in SYTOX Green stain (Molecular Probes, Eugene, OR) in the sodium citrate solution. Flow cytometry was performed on a BD FACScan using CellQuest software (BD Biosciences).

Western blot analysis:

Total protein was extracted from asynchronous and timed, synchronous samples as described previously (Koning et al. 2002). SDS–PAGE and electrotransfer were performed according to standard methods. HA-tagged proteins were identified using a mouse anti-HA primary antibody (CRP), an HRP-conjugated goat anti-mouse secondary antibody (Amersham), and the ECL detection system (Amersham).

Statistical analysis:

Unpaired t-test and ANOVA analyses were performed using commercial software packages (KaleidaGraph v3.6, Synergy Software, Wolfram Research Mathematica v5.2, or Microsoft Excel).


Reduced origin firing in the clb5Δ mutant:

If origins indeed have characteristic early and late times of initiation within S phase, then those origins that had been classified as late firing specifically should show reduced firing efficiency in clb5Δ cells, in which Clb6p-kinase is active only in the early part of S phase, whereas truly early-firing origins should not be affected by the mutation (Figure 1, top). In contrast, the “disordered firing” model predicts that all origins should show some reduction in firing (Figure 1, bottom). Two-dimensional agarose gel analysis of a set of origins had previously shown that late origins are indeed specifically reduced in efficiency (Donaldson et al. 1998b) but those studies had been performed on asynchronous populations of cells, leaving open the possibility that at least some origin activation was stimulated by later-appearing Clbs such as Clb3p or Clb4p. Therefore, we collected cells from synchronous cultures at discrete times following release from G1-phase arrest and used 2-D gel electrophoresis to compare origin firing in wild-type and clb5Δ cells.

Figure 1.—

Models for pattern of origin activation in S phase. The temporal program model and the disordered firing model predict different outcomes for origin firing in clb5Δ cells. If a temporal program of origin activation is responsible for the pattern of replication across S phase, then in wild-type cells some origins would show the highest probability of firing early in S phase, while other origins would show the highest probability of firing late in S (A, top). In clb5Δ cells, the earliest origins would have access to active Clb-kinase but the latest origins would not. As a result, in 2-D gels the late origins would show reduced overall firing in the mutant, while early origins, firing at a time when Clb-kinase was still active, would fire at their normal levels (B, top). Consequently, chromosomal replication profiles would reveal normal replication kinetics for the earliest origins (asterisks in C, top). If instead the pattern of replication for a population of cells is produced by disordered firing of origins that have different overall probabilities of firing (A, bottom), then initiation should show no specific temporal tendency in wild-type cells. In clb5Δ cells, 2-D gels for both high- and low-affinity origins would show reduced firing because Clb-kinase is active only for the first portion of S phase (B, bottom), and replication profiles would show delayed replication at all origins (C, bottom).

In wild-type cells, ARS607, previously classified as early firing, produced bubble-shaped intermediates 10 min earlier than the late-origin ARS1502 (Figure 2). This same 10-min lag is seen in clb5Δ, but with a notable decrease in the proportion of bubble structures at ARS1502 but not at ARS607 (compare 30-min CLB5 and clb5Δ samples; Figure 2, A and B). Similar results were observed for origins ARS1413, ARS1414, and the ADE2 ARS, all previously classified as late (data not shown). These observations confirmed in a qualitative sense that early and late origins respond differently to the lack of Clb5p as predicted by the temporal program model. However, we could not rule out the possibility that there was a change in the efficiency of some early origins as well. To ask if these observations held true genomewide and to obtain a statistically robust measure of changes in replication kinetics, we employed microarray-based analysis of genome replication in wild-type and clb5Δ cells.

Figure 2.—

Kinetics of origin activation assessed by 2-D gel electrophoresis. DNA was collected from (A) wild-type cells or (B) clb5Δ cells at intervals after release from α-factor for analysis by 2-D gel electrophoresis. Total DNA content and cell-cycle progression are indicated by flow cytometry at each sampling time. Shown are the gels for the “early” origin ARS607 (Friedman et al. 1997; Yamashita et al. 1997) and for the “late” origin ARS1502. Origin activation is directly proportional to the bubble arc signal (solid arrowhead) and inversely proportional to the ascending Y arc signal (open arrowhead). The relative amount of DNA loaded on each gel is assessed by the intensity of the 1N spot (asterisk). The 30-min sample (dashed box) reveals a major difference between wild type and clb5Δ in initiation at ARS1502.

Deducing replication kinetics from pooled S-phase samples:

Replication kinetics of individual restriction fragments can be assessed using a modification of the Meselson/Stahl experiment (Figure 3A; McCarroll and Fangman 1988). In this technique, cells are grown for at least seven generations in “heavy” medium containing 13C and 15N to label both DNA strands with the heavy isotopes (HH). The cells are arrested with α-factor in G1 and then transferred to light medium (containing standard isotopes). After release from α-factor, replication generates DNA that is hybrid in density with one heavy strand and one light strand (HL). A single S phase can be examined by creating a “pool” of cells from samples collected throughout S phase (Figure 3A). Hybridization to microarrays of HH and HL DNA from such pooled samples produces a single replication profile that, for each strain, displays the relative replication timing for all loci (Raghuraman et al. 2001), greatly simplifying comparisons between strains and allowing rigorous statistical tests for differences.

Figure 3.—

Wild-type and clb5Δ replication profiles for chromosome XV. (A) Cultures synchronized in “heavy” medium with α-factor are released into “light” medium. Individual samples collected at regular intervals during S phase are pooled and HH and HL DNAs are separated in a gradient of CsCl. In the pool, early replicated fragments have proportionally more HL DNA than late replicated fragments. (B) The HH and HL DNAs are labeled with the same fluorophore and each is hybridized to an array with a genomic DNA sample isolated from G1-arrested cells, labeled with a second fluorophore, as a normalization control. The HH and HL signals are used to calculate percentage of replication values for all ORFs. (C) Wild-type and clb5Δ mutant replication profiles for chromosome XV from two separate experiments (transfer A and transfer B). Percentage of replication values are plotted against the corresponding chromosomal coordinates and smoothed mathematically to produce replication profiles of each chromosome. The relative replication time of a coordinate shows an inverse relationship with its percentage of replication value, so points higher on the y-axis replicate earlier than do points lower on the plot. Local maxima define potential origins of replication, while local minima indicate fork termination zones. Each data series contains two profiles (from separate microarray hybridizations) for each strain. The locations of origins ARS1501, ADE2 ARS, and ARS1502 are indicated. (Supplemental data 1 contains replication data for all chromosomes.) (D) The average absolute difference in percentage of replication is shown for same-strain comparisons and for wild-type to clb5Δ comparisons. All comparisons are between profiles of the same data series. The difference values are in units of standard deviation (Z score) from the mean same-strain comparison value. The dashed line indicates the threshold for significance (Z = 4.886). The bar below the difference plots displays the Z scores for comparisons of wild-type and clb5Δ profiles. Coordinates where these Z scores are greater than the threshold are CLB5-dependent regions (CDRs) and are shown in yellow, orange, or red. Yellow indicates the smallest significant Z scores and red indicates the greatest Z scores. Non-CDR coordinates are shown in blue.

We created pools of S-phase cells from two density transfer experiments (called “transfer A” and “transfer B”), including in these pools cell samples that were collected during the interval starting when wild-type genomic DNA replication was 10% complete and ending when it was 80% complete. Within the pool, the earliest-replicating fragments are expected to be mostly hybrid in density (HL) and thus have a very high cumulative ratio of HL:HH DNA. In contrast, the latest-replicating fragments are expected to be mostly HH in density and have a very low ratio of HL:HH DNA (Figure 3A). Fragments that on average replicate at intermediate times in S phase should have density ratios between these two extremes and in proportion to their actual average times of replication. Thus, for each probed locus across the genome, the relative amounts of HH and HL DNA within the pool are expected to reveal the relative time at which the locus replicates during S phase. The higher the percentage of replication value of a locus, the earlier it replicates on average during S phase (Raghuraman et al. 2001). It is important to note that these experiments measure replication time, not firing time—an origin that fires early in S phase but fires only in a few cells in the population could show a population-average later time of replication that primarily reflects its passive replication.

We hybridized the pooled samples to microarray slides printed with >6000 yeast ORF probes (FHCRC arrays; see materials and methods). To control for slide-to-slide variation we labeled both the HH and the HL DNAs from the pooled sample with the same fluorophore (Figure 3B) and hybridized them to separate arrays along with equal amounts of a single-copy DNA standard isolated from G1-arrested cells (labeled with a second fluorophore). We used the relative amounts of HH and HL hybridization to find percentage of replication values for all ORFs. From these profiles we interpolated percentage of replication values for nodes spaced at 1-kb intervals and used a Fourier convolution smoothing (FCS)-based algorithm (Raghuraman et al. 2001) to minimize experimental “noise” and return a finalized percentage of replication value for each node across the genome (Figure 3C; see supplemental data 1). Data from the endmost 12 kb of each chromosome were excluded to avoid artifacts introduced by the smoothing process (Feng et al. 2006).

Repeating the hybridizations for both experiments gave us a total of four profiles to compare (Figure 3C and supplemental Figure 2; gray and black profiles). Qualitatively, the four profiles show that the microarray data are very reproducible between hybridizations and experiments for wild-type cells. To validate our pooling technique we compared the average profile from our four replication profiles of wild-type cells to those of Raghuraman et al. (2001) and Yabuki et al. (2002) (supplemental Figure 1). The three studies produce replication profiles that are very similar in their general features: the positions of major peaks and valleys and the relative heights of peaks. In addition, previously characterized origins including ARS1501, the ADE2 ARS, and ARS1502 correspond to local maxima in each experiment, indicating that they were detected as active origins in all three studies (supplemental Figure 1).

Large regions of delayed replication in clb5Δ cells:

Although it begins at the correct time within the cell cycle, S phase in the clb5Δ mutant requires nearly 50% more time for completion, presumably because origins are able to fire only in the first half of S phase, when a sufficient amount of Clb6p is present. We sought to determine whether the extension in S-phase length could be ascribed to delayed replication of particular regions of the genome (as would be expected if just a late temporal class of origins were affected; see Figure 1) or whether there was a slowdown in replication across the entire genome (as would be expected if origin firing were unordered and therefore a random half of the origins remained unfired in different cells as they ran out of Clb6p). Since our pooling method reports only relative replication timing, pooling all clb5Δ mutant samples collected over the whole of S phase—as we did for wild type—would fail to reveal a genomewide slowdown in S phase. Therefore, we pooled only those clb5Δ S-phase samples collected during a time period equivalent to the wild-type S-phase interval, thereby excluding the final 20–30% of the clb5Δ S phase from further analysis. As a result, if there are particular regions of the clb5Δ genome that fail to replicate within the span of a normal S phase, those regions would report severely decreased percentage of replication values in the microarray analysis. In contrast, if the clb5Δ cells suffer a general slowdown in S-phase progression, we would see depressed values of percentage of replication throughout the genome when compared to wild type.

We again performed two separate dense isotope transfer experiments (transfer A and transfer B) for clb5Δ cells and performed two sets of hybridizations per experiment (Figure 3C and supplemental Figure 2; orange profiles). The four profiles for clb5Δ cells are again strikingly similar to each other but show large regions that differ from the wild-type profiles (for example, see region 550–850 kb, Figure 3C) in that they are all of lower percentage of replication than wild type, indicating that they were delayed in the clb5Δ mutant. We call these regions CDRs. It is particularly noteworthy that some chromosomal segments show kinetics of replication that are indistinguishable between wild type and clb5Δ, indicating that origin activation in large blocks of the genome proceeds with the same kinetics in clb5Δ as in wild type as predicted by the temporal program model (Figure 1).

To establish a systematic and statistically stringent method to identify CDRs we calculated the average absolute difference in percentage of replication between wild-type and clb5Δ profiles at each genomic coordinate. We also performed “same-strain” comparisons to assess variation within each experiment by comparing the two wild-type profiles (obtained from the replicate array hybridizations) and the two clb5Δ profiles. In addition to controlling for spurious hybridization, these comparisons also provide a baseline for our analysis: any position where the clb5Δ vs. wild-type comparisons gave significantly different values from the corresponding same-strain comparisons was scored as having significantly different replication timing in the clb5Δ mutant. To establish a significance cutoff, we converted all clb5Δ vs. wild-type difference values to Z scores (units of standard deviation) by normalizing them to the same-strain mean and standard deviation (Figure 3D, supplemental data 2). We defined all coordinates where the wild-type vs. clb5Δ comparison gave Z ≥ 4.886 as CDRs since they depend on Clb5p for normal replication time (Figure 3D, horizontal dashed line). The greater the Z score, the greater the difference in percentage of replication value between wild type and clb5Δ (indicated in the color-coded horizontal bar in Figures 3D and 4 and in supplemental Figure 2). All CDRs have lower percentage of replication values in clb5Δ than in wild type; i.e., there are no regions in the genome that replicate significantly earlier in the mutant than in wild type. For simplicity of presentation and analysis, we used singular-value decomposition to combine the four replication profiles for each strain into two “gold standard” profiles (supplemental Figure 2) (D. Collingwood, unpublished data).

Figure 4.—

Genomic organization of Clb5-dependent regions. Heat maps of Z scores resulting from comparisons of wild-type and clb5Δ profiles are shown for all coordinates. Inconclusive coordinates are those where the same-strain comparisons give Z scores greater than the significance threshold (found in small regions on chromosomes I, VI, VIII, XII, and XVI). Genomic locations of 200 origins that were identified by ssDNA mapping (Feng et al. 2006) as being Rad53 checked or unchecked are indicated by upward- and downward-pointing triangles, respectively. Only origins that met our criteria for significant initiation in a normal S phase (Figure 6 and supplemental information) are shown.

CDRs and non-CDRs generally occur in large blocks (Figure 4) that are much larger than the average ARS spacing (35 kb), consistent with a prediction that a significant extension to S phase could occur only if origins that depend on Clb5p are spatially separated from those that do not. Over half of the genome (56.8%) falls into CDRs, including the region on chromosome XV described above and many subtelomeric coordinates (Figure 3 and supplemental Figure 2). Each of the late origins characterized by Donaldson et al. (1998b) resides in CDR regions, consistent with the prediction that their replication was expected to be delayed by the absence of Clb5p. Conversely, none of the 16 centromeres, all of which on average replicate early in S phase (McCarroll and Fangman 1988; Raghuraman et al. 2001), or the early origins studied by Donaldson et al. (1998b) fall within CDRs. The CDR blocks show a striking congruence to the large blocks of delayed and slowed replication seen during replication stress in a mutant lacking the Isw2 chromatin remodeling complex (Vincent et al. 2008), underscoring the potential role of chromatin in the regulation of replication kinetics (Vogelauer et al. 2002; Donaldson 2005).

Early degradation of Clb6p is responsible for origin inefficiency in the clb5Δ mutant:

Work by Jackson et al. (2006) revealed that Clb5p and Clb6p differ in their stability during the cell cycle. While they are both synthesized in G1, Clb6p is rapidly degraded as cells enter S phase. Clb5p persists through the remainder of the cell cycle until it is degraded by the anaphase promoting complex in mitosis (Jackson et al. 2006). We have attributed the appearance of CDRs in the clb5Δ mutant to the temporal limitations of the active CDK and not to a potential difference in substrate specificity of the two CDKs. Consistent with this idea, Gibson et al. (2004) found that increasing the gene dosage of CLB6 in a clb5Δ strain shortened the length of the S phase in the clb5Δ strain and improved the firing efficiency of two late-replicating origins, ARS603 and ARS1011. However, these results do not rule out the possibility that some origins have a higher affinity for Clb5-CDK than for Clb6-CDK but nevertheless can be satisfied by an increased concentration of Clb6-CDK in the absence of Clb5p. If the two cyclins were truly interchangeable, with no substrate specificities for different origins, then we reasoned that expressing a stable version of Clb6p in the clb5Δ mutant would be expected to restore the normal pattern of origin activation. On the other hand, if the specificities of the two cyclins for their targets at origins varied among different origins, we might expect to see alterations in the temporal pattern of replication even if the stable version of Clb6p was able to rescue the phenotype of lengthened S phase.

To distinguish between these two possibilities we created a stable version of Clb6p by removing the first 100 amino acids from the N terminus (clb6Δ100-HA, Figure 5A) (S. Haase, personal communication; Jackson et al. 2006). This deletion removes key phosphorylation sites that target the protein for rapid ubiquitination and proteolysis, but does not destroy its ability to interact with CDK (Jackson et al. 2006). To monitor Clb6p abundance we tagged the C terminus of both wild-type Clb6p and Clb6pΔ100 with a triple hemagglutinin (3× HA) tag. Both of these modified alleles were placed under the CLB6 promoter in its endogenous location in both the CLB5 and the clb5Δ mutant strains. As expected from previous work (Jackson et al. 2006), full-length Clb6p−HA protein appears ∼20 min after releasing cells from the α-factor arrest and disappears rapidly after the onset of S phase; the truncated version (Clb6pΔ100) appears at the comparable time after release from the α-factor arrest and persists at high levels over S phase (Figure 5A).

Figure 5.—

A stabilized version of Clb6p rescues CDR origin efficiency. (A) HA-tagged versions of CLB6 and clb6Δ100 were introduced into the native CLB6 locus under the control of the endogenous promoter. Whole-cell protein extracts from asynchronous (A) and synchronous samples (collected at 10-min intervals after release from α-factor) were subjected to Western blot analysis. The top panel contains a 45-kDa HA-tagged protein as a marker (M). A cross-reacting yeast protein (*; Jackson et al. 2006) serves as a loading control. (B) Activation of two CDR origins (ARS501 and ARS603) in the HA-tagged strains was assessed by 2-D gel analysis of DNA preparations from asynchronous cultures. Black and white arrows point out robust and decreased bubble arcs, respectively. An arrowhead points to the ascending arc of Y's. (C) Replication profiles of chromosome XV in wild type (CLB5 CLB6−HA, green), clb5Δ mutant (clb5Δ CLB6−HA, blue) and a clb5Δ mutant with the stabilized version of Clb6p (clb5Δ clb6Δ100-HA, red) were generated by density transfer microarray analysis on samples collected at 10-min intervals during transit of cells through S phase. Only the 60-min samples are shown. The y-axes were adjusted vertically to maximize overlap of the three strains for the purpose of comparing the shapes of the replication profiles.

The stabilized version of Clb6p appeared to restore wild-type S-phase kinetics as judged by flow cytometric analysis (data not shown; Jackson et al. 2006). To determine what effect the Clb6Δ100-HA protein has on individual origins, we analyzed two CDR origins (ARS501 and ARS603) by 2-D gel electrophoresis of genomic DNA from asynchronous cultures of the three isogenic strains, CLB5 CLB6−HA, clb5Δ CLB6−HA, and clb5Δ clb6Δ100-HA. The intensity of the bubble arc (arrows in Figure 5B) relative to the intensity of the ascending arm of the Y arc (arrowheads in Figure 5B) serves as a measure of origin efficiency since the Y arc results from those cells in which the origin was passively replicated by a fork from an adjacent origin. Both origins suffer a decrease in initiation efficiency in the absence of Clb5p; however, efficiency is significantly restored by the introduction of the clb6Δ100-HA allele (Figure 5B). These results illustrate that Clb6p is capable of efficient activation of these two origins and that the decreased firing of these origins in clb5Δ cells is likely due to their late firing time and a lack of CDK activity in late S phase. Since the 2-D gel analysis was performed with DNA from asynchronous cultures, the results do not reveal when in S phase the restored efficient activation occurs, leaving open the possibility that the Clb6Δ100-HA protein does not activate origins at the normal time, but instead alters the normal progression of S phase in some unexpected manner. We therefore conducted Meselson/Stahl density transfer coupled with microarray experiments to investigate the effect of the stabilized Clb6p on the pattern of genome replication.

We harvested a late S-phase (60 min) sample from cultures of three strains: CLB5 CLB6−HA, clb5Δ CLB6−HA, and clb5Δ CLB6Δ100-HA. We then labeled the HL and HH DNAs with different fluorophores and cohybridized them to microarrays. A comparison of chromosome XV profiles shows striking differences in the shapes of the clb5Δ and wild-type replication profiles, reflecting precisely the differences seen in the pooled S-phase samples (Figure 5C; see specifically the region between 550 and 850 kb.) The shape of the clb6Δ100-HA profile, however, resembles that of the CLB5 CLB6−HA strain (Figure 5C). These results indicate that the stable allele of CLB6 not only rescues the initiation defect at previously identified late origins (Figure 5B) but also restores normal replication timing to origins in the CDRs in the clb5Δ mutant. It is possible that increased abundance of stabilized Clb6p could be compensating for low affinity of Clb6p-directed CDK for certain substrates. Nevertheless, because the pattern of replication across all regions of chromosome XV (and other chromosomes; supplemental Figure 3 and supplemental data 3) is indistinguishable from that seen for CLB5 CLB6−HA, we conclude that Clb5p and Clb6p are functionally redundant and do not provide qualitatively different specificities for CDK1 with regard to origin activation potential. Therefore, the delayed replication in CDRs seen in the clb5Δ mutant reveals the existence of an underlying temporal pattern of origin activation (Figure 1).

Characteristics of origins that reside in CDRs:

What can we learn about the origins that are replicated in the latter half of S phase? Comparison of wild-type and clb5Δ replication profiles (Figures 3 and 4 and supplemental Figure 2) gives us an opportunity to distinguish between origins that fire late in S phase and those that replicate late on average because they happen to be inefficient and therefore replicated passively in a large fraction of the cell population. Although origin locations are represented generally in replication profiles as local maxima, we found it difficult to reliably identify origins within the profiles from our pooling experiments (Figure 3C; supplemental Figure 2) because they contained “noise,” potentially caused by regions of low probe density on the microarrays or by the very high level of replicated DNA in the pooled samples (81 and 74% for wild type and clb5Δ, respectively). Several studies have defined potential sites of replication initiation on a genomewide scale at high resolution, using criteria such as binding of replication initiation factors (Wyrick et al. 2001), sequence conservation of intergenic regions among the sensu stricto species (Nieduszynski et al. 2006), and the persistence of ssDNA formation during hydroxyurea treatment of a rad53 mutant (Feng et al. 2006). However, these studies report potential sites of initiation, only a subset of which is actually used in an average S phase. We wanted to identify specifically those origins that contribute significantly to chromosome replication. To identify such origins we performed a synchronous density transfer experiment on isogenic cells with wild-type cyclins (carrying the cdc7ts allele to improve S-phase synchrony), taking samples every 5 min for microarray analysis (Figure 6, supplemental Figure 4, supplemental data 4).

Figure 6.—

Chromosome XV replication kinetics in a wild-type S phase. A synchronous density transfer experiment was performed on a CLB5 wild-type strain that contained a cdc7ts mutation. After synchronization with α-factor, cells were raised to the restrictive temperature (37°) to permit cells to accumulate at the G1/S border. S-phase progression in light medium was monitored by collecting cell samples every 5 min between 10 and 45 min (pink, red, yellow, green, cyan, blue, violet, and gray, respectively) after the return to 23°. A control profile that reflects A + T contamination of the HL DNA (replotted from Alvino et al. 2007) is shown as a solid gray landscape. Vertical gray lines mark the locations of the most active origins (see text). See supplemental Figure 4 for other chromosomes.

Using stringent statistical criteria (see materials and methods), we identified the 200 origins that contribute most prominently to chromosomal replication profiles. All but one of the 200 have been identified as sites producing ssDNA in a rad53 mutant in the presence of HU (see below) and all 200 are confirmed or likely ARSs or proARSs (Wyrick et al. 2001; Nieduszynski et al. 2006, 2007). The time in S phase at which an origin first reached significance varied among these 200 origins, with no new origins appearing after 25 min. [The locations and Rad53 status (see below) of the 200 origins are indicated in Figure 4. Replication profiles and times of first activation of origins are given in supplemental Figure 4 and supplemental data 5, respectively.] We compared the degree of delay—by categorizing the 200 origins into CDR classes 0, 1, 2, and 3, where CDR 0 shows no delay and CDR classes 1–3 show increasing amounts of delay and dependence on Clb5p—with the time at which they first appear as significant origins in the wild type (CLB5 CLB6 cdc7ts) S-phase experiment (first active at 10, 15, 20, or 25 min) (Figure 7A). The majority of origins that do not depend on Clb5p to maintain their on-time replication (CDR 0 origins) become active within the first 15 min of S phase (Figure 7A), whereas the majority of origins that depend on Clb5p for their on-time replication (CDR 1, 2, and 3) first become active in a normal S phase at ∼20 min (Figure 7A). In general, the more delayed the replication in the clb5Δ strain, the later the origins become active in a wild-type S phase.

Figure 7.—

Correlation of origin activation time with Clb5 dependence and Rad53 checked status. (A) Time of first activation of different categories of origins with respect to their Clb5 dependence. Time of first activation was determined for the 200 origins that were clearly identified as being active (see materials and methods, supplemental Figure 4, and supplemental data 5). (B) Correlation between Clb5 dependence of origins and their Rad53 status (checked vs. unchecked). (C) Change in origin activation across S phase. “Origin firing” is a conservative estimate of the amount of initiation occurring in each S-phase interval, obtained by subtracting the average increase in percentage of replication at two adjacent fork termination sites from the percentage of replication at the intervening origin (see materials and methods). Since these values are calculated by taking the difference between successive time samples, negative values are obtained when more cells have forks arriving at the termini from distal origins than are created by the origin between the termini. Shown are the averaged results for the unchecked, non-CDR origins (blue, n = 24) and the checked CDR origins (red, n = 30). Vertical bars indicate standard deviations.

We have previously classified origins as Rad53 “unchecked” (those that fire in HU regardless of the cellular status of Rad53) and Rad53 “checked” [those that are delayed when the Rad53 checkpoint responds to the presence of HU—that is, produce ssDNA in the presence of HU only if the cells contain the rad53 mutation (Feng et al. 2006)]. If Rad53 were an important player in executing the temporal program of origin firing, we would expect to see a correlation between those Rad53 checked origins and the origins that are in CDRs. Inspecting the locations of CDRs and Rad53 checked and unchecked origins (Figure 4) indeed reveals a general correlation: many origins that are “on time” despite the absence of Clb5p (CDR 0 origins) are Rad53 unchecked and many origins that are delayed in the absence of Clb5p are in the Rad53 checked category of origins. The results are summarized for each CDR category in Figure 7B and substantiate this general conclusion.

To analyze initiation across S phase we selected origins (total of 64) that are separated from their nearest neighbors by ≥50 kb. At this distance, the accumulation of HL DNA over the first 25 min of S phase at each of these origins is more likely to reflect initiation at that origin and less likely to be the consequence of passive replication from adjacent origins. By calculating the change in percentage of replication from one time in S phase to the next we can determine the interval in S phase in which the majority of origin activation takes place.

The earliest origins (unchecked, non-CDR) show a sharp peak of activation that is limited to the first few sampling intervals, while checked, CDR origins show a broad distribution of activity over several intervals (Figure 7C). By summing the change in percentage of replication over all of the intervals, we can also estimate efficiency. The early (non-CDR, Rad53 unchecked) origins are in general more efficient that the later origins (CDR, Rad53 checked), a difference that is likely due at least in part to passive replication obscuring origin activation at later times. Only 2 of the 64 origins can be described as being neither early nor late, but pan-S, akin to the pan-S activation class proposed for human replication origins (Jeon et al. 2005). This minority class of origins fits the characterization of “disordered, stochastic firing”—i.e., having no peak time of firing or having such a broad distribution about a peak that they appear disordered.

There are exceptions to the correlation between Rad53 checking and CDR status. Among the 200 most active origins some unchecked origins are delayed in the clb5Δ strain while some checked origins replicate on time. Many of the latter category can be explained by their proximity to early efficient origins. In these cases, forks from the early origins maintain on-time replication of the adjacent genomic region by passive replication. While that particular late origin might not fire in the clb5Δ mutant, its general time of replication is not greatly delayed. (The mean distance between the checked non-CDR origins and their closest unchecked non-CDR origin is ∼40 kb.) The former category—unchecked origins that are delayed in the clb5Δ mutant—could result from two scenarios:

  1. The origin is indeed early, but inefficient, relying on passive replication from nearby origins in a large fraction of cells. Consequently, any diminution in forks from outlying regions would cause that origin to appear to be Clb5 dependent. Many of the delayed but Rad53 unchecked origins fit this explanation. They are often at great distance from an unchecked, on-time origin. (The mean distance between the unchecked CDR origins and their closest unchecked non-CDR origin is 123 kb.)

  2. The origin is not restricted to a narrow window of S phase for its activation but is a pan-S origin, activated over a broad distribution of activation times in different cells in the population. In these cases, the origin would be detected in the ssDNA assay as an unchecked origin, but full activation potential cannot be realized in the clb5Δ mutant because of those cells in which the origin would be activated later.


Over the last few years, discussion in the literature (e.g., Rhind 2006) has questioned the general existence of temporal programming of origin activation in eukaryotes and has suggested that the budding yeast S. cerevisiae may be an exception to the rule in having such a program. A more recent report based on visual examination of stretched S. cerevisiae chromosome VI DNA fibers has called even that exception into question and has raised the possibility that origin firing in S. cerevisiae is disordered as well (Czajkowsky et al. 2008). Is the long-standing picture of programmed, temporally staggered origin activations really a misleading aggregate emerging from a population of cells that individually have heterogeneous patterns of origin activation—with what had previously been called early origins actually reflecting efficient activation (with no particular temporal tendency) and late origins reflecting inefficient activation?

Here, by examining the dynamics of replication in a clb5Δ mutant (where origin activation can occur only in the first part of S phase, driven by the short-lived Clb6p) we show that there are indeed temporal classes of origins in S. cerevisiae—some regions of the genome replicate with kinetics indistinguishable from those of a normal S phase, while other large blocks of the genome, containing what had been referred to as late origins, are severely delayed in replication. This distinction between origins is not a consequence of differences in substrate specificity between Clb5p and Clb6p, as the delayed replication phenotype can be rescued completely by the expression of a long-lived version of Clb6p (Figure 5; Jackson et al. 2006). These results are exactly as predicted for temporally staggered origin activation—those origins that fire early in S phase are able to fire normally in the clb5Δ strain (during a time when Clb6p is still available) while origins that would fire later in S phase are largely unable to fire as the cells run out of the rapidly degraded Clb6p. These conclusions are also fully consistent with studies showing distinct replication times for different single-origin plasmids, where variation in firing efficiency is not a factor (Ferguson et al. 1991; Ferguson and Fangman 1992; Friedman et al. 1996; Donaldson et al. 1998b).

How do our observations fit with the interpretations of single-molecule studies of replication in S. cerevisiae and S. pombe (Patel et al. 2006; Czajkowsky et al. 2008)? While our results are incompatible with the idea of truly disordered firing, variants of the stochastic firing model (Rhind 2006; Lygeros et al. 2008) are in fact compatible with temporally staggered initiation. We suggest that the apparent diversity of views regarding the presence or absence of a temporal program may be a false dichotomy that stems in part from a lack of consensus on terminology and the concomitant confusion in the interpretation of models. The term origin efficiency, for example, has been used in ways that could be taken to mean the probability of origin firing in a particular interval within S phase, the overall probability that an origin can fire at all, and the ability of an origin to recruit/use limiting initiation factors (e.g., Patel et al. 2006; Rhind 2006). In our view, origin efficiency is none of the above and is best reserved to describe an empirical value, the percentage of cells in a population that is observed to have fired a particular origin. A perhaps more biologically interesting property is origin competence—the percentage of cells in which an origin is biochemically competent to fire (Figure 8A). For example, poor loading of the origin recognition complex (ORC) or other components of the prereplicative complex (pre-RC) could result in some cells in the population not being able to fire an origin. An origin that is competent to fire nevertheless may be kinetically limited in its ability to complete the initiation reaction in a timely fashion—and therefore fail to fire in some cells in the population if it is replicated passively (i.e., gets run over by incoming forks) before it has a chance to fire (Figure 8A). The combination of origin competence and passive replication together must account for origin efficiency in all species, perhaps with variation in competence being more of a factor in some species compared to others.

Figure 8.—

Origin competence, efficiency, and the probability of firing. (A) Cartoon illustrating the difference between origin competence and origin efficiency. Some possible outcomes at three ARSs in a four-cell population entering S phase are depicted. An ARS is a potential origin; it may show 100% competence (red ring) if it successfully becomes biochemically competent to fire in every cell cycle, or it may show <100% competence if, for example, it is biochemically poor at assembling a prereplicative complex. In cells in which the origin failed to form a pre-RC, it would be replicated passively. Origins that are competent may nevertheless be passively replicated (dashed red ring) if an incoming fork arrives at the origin before it has a chance to fire. Origin efficiency is the percentage of cells in which the origin actually fires. (B) Distribution of firing times for an early and a late origin, deduced from kinetic replication curves of plasmids p13ARS and pRightDup, respectively (inset, replotted from Friedman et al. 1996). Because each plasmid contains just one origin, the incremental change in percentage of replication over the course of S phase (Δ % replication) is a direct indicator of the percentage of cells in which the plasmid origin had fired during that increment of S phase. Incremental changes in percentage of replication of an early (ARS305, green), a mid/early (ARS1, blue), and a late-replicating chromosomal sequence (R11, gray) are indicated in dashed lines as timing markers.

Thus, origins may be <100% efficient because they are not competent to fire in some cells, because they are replicated passively in some cells despite being competent to fire, or both: the observed % efficiency = % competence − % passive replication of the competent origins (Figure 8A). Single-origin plasmids (where passive replication is not a factor) illustrate differences in origin competence and/or kinetic limitation. For example, rDNA ARS plasmids are lost at a high rate (Kouprina and Larionov 1983) presumably because the rDNA ARS shows poor competence. In contrast, some single-origin plasmids are late replicating but nevertheless do not show elevated loss rates (Donaldson et al. 1998b)—indicating that their origins are highly competent but are perhaps slower to recruit initiation factors, as noted above for ARS1502 on the chromosome (Figure 2).

We therefore favor the view that each origin has a characteristic time when it shows the maximum probability of firing, with different cells in the population either not firing that origin at all or firing it at some distribution of times around that characteristic time. Different origins could show different breadths in their distributions of firing time. An origin that has rapid kinetics of binding to an initiation factor (for instance) might show a relatively tight distribution of firing early in S phase. An origin with less rapid binding would show a broader distribution with a lower mode. Variation in mean and distribution of firing times is clearly discernible in the replication kinetics of single-origin plasmids, where replication is a direct indicator of origin firing (Figure 8B; Friedman et al. 1996). Overlap in the firing-time distributions (Figures 7C and 8B) for different origins could result in apparent disordered firing in some chromosomes in the population. The 2-D gels of ARS607 and ARS1502 through a synchronous S phase (Figure 2A) also illustrate both the differences in mean firing time and the overlap in their distributions. Note that ARS607 shows normal levels of firing at 30 min in the clb5Δ S phase (Figure 2B), but ARS1502 at the same time shows reduced firing—consistent with the idea that the two origins kinetically are different in their ability to interact with Clb-CDK and suggesting that Clb-CDK could well be the limiting factor that leads to staggered firing. Finally, dynamic processes such as chromatin remodeling and cell-to-cell heterogeneity in chromatin structure could further shape the overall topography of replication in the population.

Variation in affinity for a limiting initiation factor (Rhind 2006; Lygeros et al. 2008) is one possible mechanism underlying the temporal differences between origins. If so, one might expect late-origin firing-time distributions to be wider and shallower than those for early origins (Figure 8B), which indeed is what we observe in the comparison of the earliest and the latest origins (Figure 7C). Such differences between origins could arise from intrinsic variation in their affinities for initiation factors (perhaps influenced by the local DNA sequence), variation in occupancy by ORC/pre-RCs, variation in chromatin compaction across the genome (Gilbert et al. 2004; Donaldson 2005), or some combination thereof. Indeed, the observation that relocating origins to ectopic sites in the genome can change their firing time (e.g., Ferguson and Fangman 1992) indicates that chromosomal context must matter. Regardless of mechanism, the resulting temporal variation in probability of firing is what we would call a temporal program of origin activation.

Despite the possibility of overlap in the distributions of origin firing times, the genome nevertheless is organized into non-CDR and CDR blocks (Figure 4) largely paralleling the blocks of origins described as firing or not firing in the presence of HU (Yabuki et al. 2002; Feng et al. 2006). Permutation tests in which origins are randomly labeled as CDR or non-CDR show that the actual genomic organization of origins into CDR and non-CDR clusters is very significantly different from a random distribution (data not shown). Clusters of origins with similar activation timing have been qualitatively observed in yeast and mammals (Hand 1975; Rivin and Fangman 1980; Reynolds et al. 1989; Friedman et al. 1996), perhaps reflecting an underlying organization of chromatin or higher-order chromosomal domains. Such clustering of temporally similar origins could result in a scenario where stochastic properties allow some disorder of firing within blocks while still maintaining a genome organization of fairly distinct temporal blocks—and in fact, large-scale temporal organization of the genome indeed has been discussed as a possibility (Czajkowsky et al. 2008). Some combination of reduced cluster size and dynamic reorganization of chromatin could result in blurring of the temporal program in other species.


We thank Christina Wilson, Anne Donaldson, Wenyi Feng, Kim Lindstrom, and Conrad Nieduszynski for technical assistance, strain construction, helpful discussions, and critical comments on the manuscript. We thank Zasha Weinberg for providing us with an algorithm for simulating replication profiles and Andy Marty for help with microarray hybridizations. We are grateful to Thomas Lumley for statistical assistance. This work was supported by National Institute of General Medical Sciences grant 18926 to B.J.B., M.K.R., and W.L.F. H.J.M. was supported in part by a National Science Foundation predoctoral fellowship and by a National Institutes of Health genetics training grant at the University of Washington.


  • 1 Present address: Cord Blood Registry, San Bruno, CA 94066.

  • 2 Present address: Sackler Institute of Graduate Biomedical Sciences, New York University, New York, NY 10016.

  • Communicating editor: N. M. Hollingsworth

  • Received July 23, 2008.
  • Accepted October 1, 2008.


View Abstract