Cohesin not only links sister chromatids but also inhibits the transcriptional machinery’s interaction with and movement along chromatin1–6. In contrast, replication forks must traverse such cohesin-associated roadblocks to duplicate the entire genome in S phase. How this transpires is unknown. Through single-molecule analysis, we demonstrate that the replication factor C (RFC)-Ctf18 clamp loader1, 7 controls the velocity, spacing, and restart activity of replication forks and is required for robust acetylation of cohesin’s Smc3 subunit and sister chromatid cohesion. Unexpectedly, we discovered that cohesin acetylation itself is a central determinant of fork processivity, as cells lacking the Eco1-related acetyltransferases Esco1 or Esco28–10 (including those derived from Roberts Syndrome patients, in whom ESCO2 is biallelically mutated11) or expressing a non-acetylatable form of Smc3 also harbor slow-moving replication forks. This defect was a consequence of cohesin’s hyperstable interaction with two regulatory cofactors, Wapl and Pds5a12, 13; removal of either cofactor allowed forks to progress rapidly without Esco1, Esco2, or RFCCtf18. Our results reveal a novel mechanism for clamp loader-dependent fork progression, mediated by the post-translational modification and structural remodeling of the cohesin ring. Loss of this regulatory mechanism leads to the spontaneous accrual of DNA damage and may contribute to the abnormalities of the Roberts Syndrome cohesinopathy.
The ring-shaped cohesin complex not only catenates sister DNAs but also acts as a structural barrier to other chromatin-based transactions. For example, in budding and fission yeast cohesin slides to the 3’ ends of convergently arrayed genes and prevents readthrough transcription by RNA polymerase II2, 14. In mammals, cohesin accumulates at insulators and boundary elements and blocks activation of adjacent promoters4, 5. These sites occur about once per 25 kb in the human and mouse genomes, an interval severalfold smaller than the average somatic replicon15, and are occupied by cohesin throughout the cell cycle4, 5. Cohesin can also bind regions between insulators in a distributive manner and pair sister chromatids through these sites5. Hence, nearly all forks must replicate through at least one cohesin-associated region during S phase. In addition to mediating genome duplication, fork passage is thought to lead to entrapment of nascent DNA strands within the cohesin ring, thereby generating functional cohesion between sister chromatids1. Cohesion establishment also requires acetylation of cohesin’s Smc3 subunit by Eco1-related acetyltransferases8–10, but how this modification affects cohesin’s interaction with the replication fork remains unclear.
In addition to Eco1, cohesion establishment is aided by the replication factor C-Ctf18-Ctf8-Dcc1 (hereafter RFCCtf18) complex, which loads the polymerase processivity clamp PCNA onto replication forks in vitro and in vivo1, 7. Nevertheless, physical assays of replication fork dynamics have not uncovered this enzyme’s contribution during a normal S phase, nor is it known whether RFCCtf18 is important for Smc3 acetylation. To address these questions, human retinal pigment epithelial cells were subjected to two rounds of adeno-associated virus (AAV)-mediated homologous recombination, such that exon 2 of the DCC1 locus was either flanked by loxP sites or deleted outright (Supplementary Fig. 1). To inactivate RFCCtf18, DCC1flox/Δ cells were infected with an adenovirus expressing Cre recombinase (AdCre). The resulting DCC1Δ/Δ cells lacked detectable Dcc1 protein and had markedly reduced levels of Ctf18, but normal amounts of all other RFC subunits (Fig. 1a). Whereas RFCCtf18-deficient yeast grow robustly, DCC1Δ/Δ clones could not be isolated via limiting dilution (Fig. 1b), indicating that this complex is essential in mammals. To understand this requirement, DCC1flox/Δ cells were infected with AdCre en masse and monitored during serial passage. Within 7 divisions, DCC1Δ/Δ cells ceased to proliferate and exhibited hallmarks of premature senescence (increased size, flattened morphology, and expression of senescence-associated β-galactosidase; Fig. 1c,d). Senescence is commonly triggered by genotoxic stresses that activate the DNA damage response16; consistently, DCC1Δ/Δ cells accumulated high levels of Nbs1 foci during cell cycle exit (Fig. 1e). Unlike RFCRad17-deficient cells17, RFCCtf18-deficient cells activated Chk1 and imposed the G2/M checkpoint in response to genotoxic stress (Supplementary Fig. 2).
Unstable replication forks are a prime source of DNA damage in both prokaryotes and eukaryotes18. As an initial test of RFCCtf18’s contribution to fork stability, replication forks in newly generated DCC1+/Δ and DCC1Δ/Δ cells were transiently stalled with hydroxyurea (which depletes nucleotide pools) or aphidicolin (which inhibits replicative polymerases). RFCCtf18-deficient cells were hypersensitive to both replication inhibitors, as evidenced by their accelerated transit into senescence (Fig. 2a). To obtain a more precise view of replication dynamics, single-molecule analyses were performed on extended DNA fibers. Cells were sequentially pulse-labeled with the halogenated nucleosides iododeoxyuridine (IdU) and chlorodeoxyuridine (CldU), after which genomic DNA was gently extracted and stretched on glass slides, stained with IdU- and CldU-specific antibodies, and visualized microscopically (Fig. 2b). Unlike other assays for DNA synthesis, this technique affords specific and quantitative information on origin firing, replication fork movement, and fork recovery throughout the genome, based on examination of large numbers of individually labeled forks. Three effects of RFCCtf18 deficiency were evident in these assays. First, the velocity of individual forks decreased by more than 50% (0.7 ± 0.2 kb/min versus 1.7 ± 0.4 kb/min in DCC1+/Δ cells (mean ± s.d.); Fig. 2c), indicating a highly penetrant defect in replisome progression. Second, the interval between forks decreased significantly (from 38 ± 7.6 kb to 18 ± 4.5 kb; Fig. 2d), consistent with firing of dormant origins that usually are replicated passively19. Third, replication forks appeared to arrest or collapse at an elevated rate, as reflected by an excess of tracks labeled exclusively with IdU (Fig. 2e). As a further test, we measured the efficiency of fork recovery from an aphidicolin block, imposed between the IdU and CldU pulses. Over 30% of forks failed to restart under these conditions (Fig. 2e), thus explaining the sensitivity of DCC1Δ/Δ cells to replication inhibitors.
The role of the RFCCtf18 complex in cohesion establishment was also examined. As senescing cells are excluded from metaphase spreads, we abrogated this p53- and Rb-mediated response via expression of human papillomavirus E6 and E720. Bypassing senescence did not alter the requirement for RFCCtf18 in sustaining high fork velocity, origin dormancy, and replication restart (Supplementary Fig. 3), but did reveal precocious chromatid separation in over 40% of DCC1Δ/Δ cells, as well as structural aberrations (Fig. 3a,b). A similar mixture of cohesion loss and chromosome fragility occurs in cells expressing non-acetylatable Smc3 (ref. 8), suggesting that this modification may depend on RFCCtf18. To test this idea, cohesin was immunopurified and analyzed by quantitative immunoblotting with antibodies specific for acetylated lysine. Smc3 acetylation was reduced by 70% in DCC1Δ/Δ cells (Fig. 3c), despite proper targeting of cohesin and the Eco1-related acetyltransferases Esco1 and Esco2 to chromatin (Fig. 3d). Thus, RFCCtf18 promotes cohesion at a later step than the Cdc7 kinase and other pre-replicative complex components, which are required for cohesin’s deposition onto interphase chromosomes in Xenopus21. These data are compatible with a model wherein Esco1 and Esco2 are stimulated to acetylate cohesin through physical interactions with RFCCtf18-loaded PCNA1, 22.
Our results show that in the absence of RFCCtf18, replication fork processivity and Smc3 acetylation are both significantly diminished. We next asked whether these processes occur independently downstream of RFCCtf18 or are themselves linked. To this end, we analyzed replication dynamics in cells derived from patients with Roberts Syndrome/SC phocomelia (RBS), a severe hereditary disorder in which ESCO2 is biallelically mutated, resulting in weakened cohesion between sister centromeres11. Fork velocities were markedly reduced in cells from two unrelated RBS patients (Fig. 3e) and accompanied by spontaneous accumulation of DNA damage (Supplementary Fig. 4). Fork slowing was also observed in HeLa cells depleted of Esco2 or Esco1 via RNA interference (RNAi; see Fig. 4e). To test the contribution of cohesin acetylation more specifically, we analyzed replication forks in cells expressing wildtype (Smc3WT) or non-acetylatable Smc3 (Smc3AA) from a tetracycline-regulated promoter. Expression of Smc3AA alone reduced fork speeds from 1.5 ± 0.2 kb/min to 0.8 ± 0.1 kb/min (Fig. 3e), indicating that cohesin acetylation per se is required for processive DNA synthesis. In contrast, the expression of cohesin-regulated genes was insensitive to these manipulations (Supplementary Fig. 5 and Supplementary Table 1). We conclude that the acetylated form of cohesin specifically regulates replication fork progression, but is not required for transcriptional insulation.
How Smc3 acetylation affects cohesin biochemically is not well understood, but genetic evidence indicates that this modification acts in opposition to Rad61 (also called Wapl) and Pds5, which together bind cohesin and regulate its association with chromosomes10, 12, 13, 23–26. Strikingly, non-acetylatable human Smc3 copurified with increased amounts of both Wapl and Pds5a (Fig. 4a and Supplementary Fig. 6). Furthermore, this interaction was negatively correlated with Smc3 acetylation in synchronized cells (Fig. 4b and Supplementary Fig. 7). To determine if the excess stability of this interaction inhibits fork movement, RNAi was used to deplete Wapl and Pds5a in cells expressing Smc3AA (Fig. 4c,d) or concomitantly depleted of Esco1 or Esco2 (Fig. 4e and Supplementary Fig. 8). Strikingly, downregulating Wapl or Pds5a rescued fork progression in all instances but had no effect on its own, arguing that ring opening by the Wapl/Pds5a complex is not generally required for replication forks to move through cohesin-binding sites. Pds5a depletion also allowed forks in DCC1-null cells to move at normal rates (Fig. 4f–g). We conclude that the RFCCtf18 complex speeds chromosomal replication mainly by stimulating the acetylation-dependent remodeling of cohesin rings, rather than by activating DNA polymerases directly, at least in the presence of the replicative RFC.
Our analysis provides new and unexpected insights into the relationship between DNA replication and cohesion establishment. Previous studies did not detect an S phase delay in yeast ctf18 or eco1 mutants, leading to the conclusion that chromosomes are replicated in a similar fashion whether or not cohesion is established between nascent sister DNAs. In contrast, by applying sensitive single-molecule assays, we find that replication dynamics in human cells are highly attuned to the acetylation of Smc3 and possibly other substrates by Esco1 and Esco2. Notably, the fork’s response to deficiencies in cohesin acetylation is indistinguishable from the consequences of physically blocking the fork by other routes (e.g., nucleotide depletion, polymerase inhibition, or DNA alkylation), suggesting that a primary role of this modification is to switch cohesin from a configuration that obstructs the fork to one that permits its advancement. Furthermore, our genetic and biochemical data imply that this switch involves cohesin’s regulatory cofactors Pds5a and Wapl, rather than an intrinsic biophysical property of acetylated cohesin itself. To account for these observations, we propose that Wapl and Pds5a prevent the replication fork from sliding through the ring and/or interfere with cohesin’s stable embrace of the nascent sister DNAs (Supplementary Fig. 9). The possibility of an occlusive conformation is already suggested by Pds5’s bivalent interactions with opposite sides of the ring27, as well as flexion of the Smc1 and Smc3 coiled coils28. To neutralize this, the fork-associated RFCCtf18 complexes stimulate Esco1 and Esco2 to acetylate Smc3, causing the Wapl/Pds5a complex to dissociate. Cohesin could then adopt an open conformation that allows the fork to advance while stably trapping the nascent sister DNAs within the ring. This model would also explain the “anti-establishment” activities ascribed to Wapl and Pds5a10, 23, 25, 26, since hindrance of fork passage through the cohesin ring would not only slow chain elongation but also interfere with entrapment of sister chromatids. In addition to generating cohesion, entrapment may contribute positively to fork processivity by facilitating sister chromatid-dependent pathways of replication restart29.
Weakened centromere cohesion is a diagnostic hallmark of RBS11. However, we show here that this disorder is also associated with widespread slowing of the replication fork, presumably due to unopposed action of the Wapl/Pds5a complex on cohesin. Whether this chronic alteration in replication dynamics contributes to the abnormalities in RBS patients is unclear. However, DNA replication in general is known to be epigenetically coordinated with development. Whereas active origins are closely spaced during the initial cell cycles after fertilization, their density falls more than 10-fold during embryogenesis, resulting in a dramatic expansion of replicon size15. By impeding fork progression and reactivating dormant origins, defects in cohesion establishment may disrupt this epigenetic switch, causing replication stress and premature senescence or apoptosis of progenitor cells during development. Further work with in vivo models of RBS will be needed to test this hypothesis. Reducing fork velocity also results in the packaging of chromosomal DNAs into smaller than normal chromatin loops19. This effect has been suggested to involve changes in the replication-dependent modification of factors associated with matrix attachment regions and boundary elements, a proposal compatible with the known properties of cohesin4, 5. Its acetylation may therefore be required not only to produce cohesion locally, but also to shape mammalian genomes more globally.