SAF-A promotes origin licensing and replication fork progression to ensure robust DNA replication Free

DNA replication in eukaryotic genomes initiates from discrete sites termed DNA replication origins. Potential replication origin sites are defined by stepwise assembly of a protein complex, the pre-replication complex (pre-RC), during G1 phase of the cell cycle (Fragkos et al., 2015). During pre-RC formation, the origin recognition complex (ORC) and CDT1 cooperate to load the heterohexameric minichromosome maintenance complex (MCM), leading to ‘origin licensing’ (McIntosh and Blow, 2012). MCM plays a critical role when DNA replication initiates at each origin, forming the central component of the replicative helicase (Fragkos et al., 2015). Cells monitor the level of replication licensing and prevent cell cycle progression if an insufficient number of sites are licensed (Feng et al., 2003; Lau et al., 2009; Nevis et al., 2009; Shreeram et al., 2002; Zimmerman et al., 2013). This ‘licensing checkpoint’ mechanism appears to be compromised in cancer cells (Feng et al., 2003; Lau et al., 2009; Nevis et al., 2009; Shreeram et al., 2002; Zimmerman et al., 2013).

A recent study has demonstrated that replication licensing is impacted by the state of chromatin packaging. The histone methyltransferase SET8 (also known as PR-SET7 or KMT5A) can stimulate origin licensing at specific sites (Tardat et al., 2010), but also prevents over-licensing by enhancing chromatin compaction as cells exit mitosis (Shoaib et al., 2018). SET8 is responsible for the methylation of histone H4 at lysine 20 (H4K20) and for maintaining chromatin compaction at the M-G1 boundary (Shoaib et al., 2018). Replication licensing is therefore impacted both by local chromatin changes and broader changes occurring at the chromosome domain level. How chromatin packaging status affects origin licensing and the subsequent steps in DNA replication is still not fully understood, but there is a well-established connection between chromatin packaging and the temporal programme of replication (Fu et al., 2018; Gilbert, 2010; Gilbert et al., 2010), in which euchromatin domains containing active genes generally replicate early in S phase, whereas heterochromatic, highly packaged domains containing mainly inactive genes replicate late. Replication timing of some domains is modulated during development, often reflecting changes in gene activity (Hiratani et al., 2008). The replication timing programme is established at early G1 phase, which coincides with chromatin decompaction and chromatin remodelling as cells exit M phase (Shoaib et al., 2018), suggesting that the dynamic controls over chromatin structure imposed as cells exit mitosis determine the replication potential and subsequent replication timing of local chromatin domains (Dimitrova and Gilbert, 1999; Dimitrova et al., 2002).

Chromatin packaging also impacts on replication progression, with processive replication of heterochromatin regions requiring local decompaction (Chagin et al., 2019). Recent studies also highlight numerous ‘difficult-to-replicate’ regions (Cortez, 2015; Gadaleta and Noguchi, 2017), including DNA–protein complexes, repetitive DNA such as centromeres and telomeres, and secondary DNA structures. Replicating such regions requires support by specific proteins, without which replication tends to fail, leading to genome instability (Cortez, 2015; Gadaleta and Noguchi, 2017) and the formation of fragile sites (Boteva et al., 2020). These observations highlight the importance of modulating chromatin structure during the replication process.

Scaffold-attachment factor A (SAF-A; also known as heterogeneous nuclear ribonucleoprotein U, HNRNPU) is an RNA- and DNA-binding protein that modulates chromatin structure by tethering chromatin-associated RNA (caRNA) to chromatin (Creamer et al., 2021; Fackelmayer et al., 1994; Kiledjian and Dreyfuss, 1992; Nozawa et al., 2017; Sharp et al., 2020). SAF-A oligomerisation contributes to decompacted chromatin (Creamer et al., 2021; Nozawa et al., 2017), and depletion of SAF-A causes global chromatin condensation (Fan et al., 2018). A super-resolution microscopy study also implicates SAF-A in the establishment of correct chromatin structure, and SAF-A has been shown to regulate both active chromatin and X-chromosome inactivation (Hasegawa et al., 2010; Lu et al., 2020; Smeets et al., 2014). SAF-A interacts and colocalises with proteins that define chromatin domain boundaries, namely CTCF and cohesin, and plays a role in defining boundaries of topologically associated domains that form smaller units of chromosome organisation (Fan et al., 2018; Zhang et al., 2019). Although SAF-A promotes open chromatin, depletion of SAF-A has fairly minor effects on transcription (Nozawa et al., 2017). SAF-A is, however, reported to be implicated in alternative splicing (Xiao et al., 2012; Ye et al., 2015), mRNA stability (Yugami et al., 2007) and nuclear retention of mRNA (Huang et al., 2021).

Association of SAF-A with chromatin is cell cycle-regulated: SAF-A is associated with chromatin throughout interphase but is removed from chromatin in M phase (Sharp et al., 2020). Regulated dissociation of SAF-A from the mitotic chromosome, triggered by phosphorylation of SAF-A by Aurora kinase B, is essential for proper progression of mitosis (Douglas et al., 2015; Sharp et al., 2020). SAF-A reassociates with chromatin as cells exit from mitosis, implicating SAF-A in chromatin decompaction at this cell cycle stage.

SAF-A also localises to DNA damage sites swiftly after γ-ray irradiation (perhaps to modulate chromatin structure), and then at a later stage appears to be excluded from damage sites (Hegde et al., 2016). Interestingly, the expression of the SAF-A gene tends to increase in a wide range of cancers, particularly breast invasive carcinoma (The Cancer Genome Atlas; https://www.cancer.gov/tcga). This increased expression suggests that SAF-A contributes to the formation or survival of cancer cells in a dose-dependent manner. Conversely, SAF-A loss-of-function alleles are linked to developmental disorders, including microcephaly (Durkin et al., 2020; Leduc et al., 2017; Yates et al., 2017). Overall, these observations suggest a positive role for SAF-A in promoting cell proliferation. While roles of SAF-A in mitosis have been investigated (Sharp et al., 2020), its contribution to cell proliferation during interphase, particularly to DNA replication, has not been studied.

Here, we investigate the effects of SAF-A on DNA replication. We show that SAF-A protein is required for full replication licensing in the G1 phase of the cell cycle, and depleting SAF-A leads to increased spacing between replication origins. We find, moreover, that replication fork progression is compromised in cells depleted for SAF-A, and that SAF-A protein plays a role in defining the boundaries of early and late replication domains in the genome-wide replication programme. Loss of these functions leads to spontaneous replication stress and increases cellular entry to quiescence, explaining the need for SAF-A for normal cell proliferation.

To assess the general impact of SAF-A on subnuclear organisation of chromatin, we examined the distribution of DNA within nuclei of hTERT-RPE1 cells treated with siRNA targeting SAF-A (siSAF-A) (Fig. 1A). hTERT-RPE1 is a non-cancer cell line derived from retinal pigment epithelial cells immortalised by expression of human telomerase (hTERT; Bodnar et al., 1998). Using super-resolution microscopy to examine very thin sections, we found that control nuclei showed relatively homogeneous DNA density distribution (i.e. mostly intermediate DAPI signal intensity; green in Fig. 1A), with smaller areas of higher (yellow and red in Fig. 1A) or lower (cyan and blue in Fig. 1A) DNA density. In siSAF-A cells, the DNA density distribution revealed larger areas with low DNA density (blue in Fig. 1A) interspersed with high DNA density areas (red in Fig. 1A), indicative of a more polarised distribution with sections of genomic DNA densely packed in abnormally compact domains. Unbiased classification of ‘DAPI high’ and ‘DAPI low’ areas in each nucleus (Fig. 1B) confirmed the formation of larger ‘DAPI low’ areas in siSAF-A nuclei, with chromatin confined into smaller areas. These data suggest that SAF-A promotes proper dispersal of chromatin distribution within nuclei and prevents the formation of over-compacted chromatin. This microscopic observation is consistent with SAF-A function in maintaining correct chromatin architecture as revealed by microscopy (Creamer et al., 2021; Nozawa et al., 2017) and Hi-C chromatin conformation capture methods (Fan et al., 2018).

Cells depleted of SAF-A have been reported to show proliferation defects (Nozawa et al., 2017), but the exact nature of the defect has not been studied in detail. We examined the cell proliferation and DNA replication profiles of cells depleted of SAF-A. siSAF-A cells showed a significant and reproducible reduction in cell proliferation rates compared with those of hTERT-RPE1 cells treated with control siRNA (siControl; Fig. 1C), which is consistent with a previous report (Nozawa et al., 2017). Flow cytometry analysis of DNA content in asynchronous cultures (Fig. S1A), however, showed no specific cell cycle arrest point but did reveal a slight reduction of the S phase population, suggesting that loss of SAF-A may cause problems with DNA replication.

Cells depleted of SAF-A have been reported to be defective in recovery from replication inhibition by the DNA polymerase inhibitor aphidicolin (Nozawa et al., 2017). We therefore tested whether SAF-A-depleted cells also fail to recover from the DNA replication inhibitor hydroxyurea (HU), an inhibitor of ribonucleotide reductases that causes stalled replication forks. After treating siControl and siSAF-A cells with 4 mM HU for 24 h to cause early S phase arrest (Fig. 1D, 0 h), we examined recovery by monitoring DNA content and incorporation of a thymidine analogue ethynyl deoxyuridine (EdU). Control cells recovered from arrest efficiently and reached mid-S phase by 4 h after release from HU (Fig. 1D, siControl). In contrast, very few siSAF-A cells recovered to reach a similar stage by 6 h (Fig. 1D, siSAF-A). Assessment of EdU-positive cells further indicated that a reduced number of siSAF-A cells were able to resume DNA synthesis compared with siControl cells (Fig. 1E), and that the rate of DNA synthesis in EdU-positive siSAF-A cells was lower than that in siControl cells until 6 h after release (Fig. S1B,C). By 8 h after the release, the majority of siControl cells had finished DNA replication, whereas a notable fraction of siSAF-A cells were still synthesising DNA (Fig. S1B,C). These observations indicate that cells depleted of SAF-A have difficulty in recovering from HU treatment. Taken together, our results show that deficiency of SAF-A causes cells to be severely impaired in recovery from replication stress.

We tested whether depletion of SAF-A impacts DNA replication in the absence of exogenous stress by measuring DNA synthesis rate based on pulse labelling nascent DNA with EdU followed by flow cytometry analysis. Cells depleted of SAF-A showed a significantly reduced percentage of EdU-positive cells (Fig. 1F; 37.2% in siControl and 24.7% in siSAF-A). Moreover, the EdU-positive population of siSAF-A cells showed a reduced DNA synthesis rate compared to that of siControl cells (Fig. 1G). This DNA synthesis defect of siSAF-A cells is not confined to a specific stage of S phase (Fig. S1D), suggesting that SAF-A function is required throughout DNA replication. Taken together, these results indicate that SAF-A is required for robust DNA replication without exogenous replication stress, and also supports the recovery of cells after replication stress.

Changes in chromatin due to SAF-A depletion could potentially affect multiple steps of DNA replication, including origin licensing, replication fork progression and fork restart. We decided to assess the requirement for SAF-A for each of these steps in a series of experiments.

Since SAF-A plays a positive role in open chromatin structure and prevents over-compaction (Fig. 1A,B) we hypothesised that SAF-A may play a positive role in stimulating origin licensing by promoting open chromatin. To test this hypothesis, we used a flow cytometry ‘3D licensing assay’ (Moreno et al., 2016) (Fig. 2A,B), which simultaneously measures MCM loading on chromatin and EdU incorporation (to assess cell cycle stage). In this assay (Fig. 2A), cells in G1 (red box), S (cyan box) and G2/M (orange box) phase can be clearly distinguished, and the amount of chromatin-associated MCM3 can be assessed in each cell cycle population (Fig. 2B). As clearly seen in Fig. 2B, siSAF-A-treated hTERT-RPE1 cells showed reduced levels of chromatin-associated MCM3 in individual cells both in G1 phase (red) and in cells entering S phase (left-hand part of cyan population). This observation indicates that siSAF-A cells show compromised levels of MCM loading in G1-phase cells that persist into S phase, suggestive of a defect in origin licensing.

We next tested whether SAF-A affects other licensing proteins. CDT1 interacts with MCM and assists its loading onto chromatin in G1 phase (Frigola et al., 2017; Zhai et al., 2017a): outside of G1 phase CDT1 is negatively regulated by geminin (Blow and Tanaka, 2005). In vitro studies show that CDT1 dissociates from MCM after the assembly of the double MCM hexamer on DNA (Zhai et al., 2017b). Consistently, flow cytometry analysis showed that CDT1 associates with chromatin predominantly in G1 phase (vertical spikes in Fig. 2C). We found that depletion of SAF-A in hTERT-RPE1 cells caused reduced chromatin association of CDT1 (Fig. 2C, siSAF-A), consistent with the reduced MCM licensing in G1 phase (Fig. 2B).

We next tested the chromatin association of ORC1 protein. ORC1 protein is a subunit of the ORC protein complex that initially defines MCM loading sites. ORC1 protein expression and stability is cell cycle-regulated so that it is present predominantly in G1 phase, helping to confine MCM loading to G1 phase of the cell cycle (Méndez et al., 2002; Ohta et al., 2003; Tatsumi et al., 2003). In the absence of an ORC1 antibody suitable for flow cytometry analysis, we made use of a HEK293-based cell line expressing FLAG-tagged ORC1 protein (Tatsumi et al., 2003) to analyse chromatin association of ORC1 (Fig. 2D). As expected, and consistent with previous reports (Hiraga et al., 2017), ORC1 chromatin association was detected predominantly in the G1 phase of the cell cycle (Fig. 2D, top panels). Depletion of SAF-A resulted in a reduction of chromatin-associated ORC1 (Fig. 2D, top-right panel). We also observed that CDT1 and MCM licensing were reduced when SAF-A was depleted in the HEK293 FLAG–ORC1 cell line (Fig. 2D, middle and bottom panels, respectively), similar to the effects in hTERT-RPE1 cells (Fig. 2B). In contrast, chromatin association of ORC2, which does not fluctuate during the cell cycle (Méndez and Stillman, 2000; Méndez et al., 2002), was not affected by SAF-A depletion (Fig. S2A). We confirmed the reproducibility of these effects in multiple independent experiments in both HEK293-derived cells and hTERT-RPE1 cells (Fig. S2B–D).

Reduced association of licensing factors in SAF-A-depleted hTERT-RPE1 cells was confirmed by western blot analysis of chromatin-associated proteins. Western blotting of chromatin-enriched fractions confirmed the reduced association of CDT1 protein (Fig. S2E, lanes 3 and 4) and ORC1 protein (Fig. S2E, lanes 7 and 8, top panel) after SAF-A depletion. CDC6, another protein required for replication licensing (Blow and Tanaka, 2005), however, did not show such a reduction in chromatin association (Fig. S2E, lanes 7 and 8, middle panel). Chromatin association of ORC2 was not affected by SAF-A depletion (Fig. S2E, lanes 11 and 12), consistent with the flow cytometry result (Fig. S2A).

The effects of SAF-A depletion on chromatin association of FLAG–ORC1 and CDT1 appeared very similar in HEK293 FLAG–ORC1 cells extracted with CSK buffer (which contains 100 mM NaCl, as compared to standard conditions for flow cytometry that involve extraction with only 10 mM NaCl) (Fig. S3). CSK buffer is widely used for biochemical preparation of chromatin-associated protein fractions in HEK293-derived cells. In hTERT-RPE1 cells, we noticed that CDT1 chromatin association appeared more salt sensitive (data not shown), suggesting that chromatin association may differ between cell types.

Since the western blot analysis reported above suggests some reduction in total CDT1 and ORC1 levels (Fig. S2E, lanes 1 and 2 for CDT1, lanes 5 and 6 for ORC1), we investigated expression of these proteins per cell using flow cytometry. CDT1 expression was not affected by SAF-A depletion in hTERT-RPE1 cells. In FLAG–ORC1 cells depleted of SAF-A, we found that levels of FLAG–ORC1 and CDT1 were somewhat reduced (Fig. S4), although results were variable between experimental repeats (compare top two rows in Fig. S4). Overall, we conclude that reduced ORC1 and CDT1 expression may contribute to phenotypes of SAF-A depletion but cannot fully account for the reduced licensing levels, particularly in hTERT-RPE1 cells.

Overall, the data presented show that SAF-A promotes the G1 phase chromatin association of several origin licensing components, including loading of MCM itself.

Impaired replication licensing in cells depleted of SAF-A suggests there will be a reduced number of potential replication origins available for activation. Therefore, we next tested whether a reduced origin frequency is observed on chromosomes in SAF-A-depleted cells, by measuring inter-origin distances (IODs) using single-molecule DNA fibre analysis.

To detect origin activation on single DNA molecules, nascent DNA was sequentially labelled with thymidine analogues 5-chloro-2′-deoxyuridine (CldU) and 5-iodo-2′-deoxyuridine (IdU), as illustrated in Fig. 3A. Analogue incorporation was analysed by immunostaining of DNA fibres stretched by molecular DNA combing (Bianco et al., 2012). Replication origins can be identified as illustrated in the top panel of Fig. 3B, with the mid-point between divergent replication forks assigned as a replication origin. In these experiments, increased distance between replication origins is indicative of fewer active origins. We found that depletion of SAF-A caused an increase in IOD compared with that of the control (Fig. 3B), suggesting that the number of active origins is indeed reduced by SAF-A depletion. This reduction in origin activation frequency probably reflects inefficient origin licensing.

Stalling of DNA replication forks due to replication stress causes activation of nearby dormant origins (Ge et al., 2007), believed to protect cells from replication stress by guarding against the formation of unreplicated stretches between two stalled or collapsed replication forks (Blow and Ge, 2009; Kawabata et al., 2011). Since the licensing defect of SAF-A-depleted cells might affect the number of available dormant origins, we assessed whether cells depleted of SAF-A activate dormant origins normally (Fig. 3B,C). siRNA-treated cells were incubated for 4 h with 0.1 mM HU to slow replication forks. At the end of the HU treatment, nascent DNA was labelled with CldU and IdU as shown in Fig. 3A. Under this condition, replication forks continued to progress but with significantly reduced speed (Fig. 4A). As expected, HU treatment induced the activation of dormant origins near stalled forks, evidenced by a reduction in IOD and appearance of very short IODs below 20 μm (below blue dashed line in Fig. 3C), both in siControl and siSAF-A cells (Fig. 3C). In cells depleted of SAF-A, however, short IODs (in the range 0–30 kb) occurred at reduced frequency compared to siControl cells (Fig. 3C), suggesting impaired dormant origin activation. Overall, these data confirm that the number of active DNA replication origins is reduced in cells depleted of SAF-A, and that SAF-A is required for activation of dormant origins at normal frequency under replication stress.

SAF-A depletion leads to decreased cellular DNA synthesis rate in unperturbed S phase (Fig. 1F,G), as well as reduced origin licensing (Fig. 2) and activation (Fig. 3). However, it was unclear whether reduced origin activation fully accounts for the decreased cellular DNA synthesis. To explore whether altered DNA replication fork speed also affects DNA synthesis when SAF-A is deficient, we investigated replication fork speed using the same DNA combing technique as depicted in Fig. 3A. The lengths of IdU tracts in stretched DNA molecules were taken as a proxy for DNA synthesis rate. The average replication fork speed was not affected by SAF-A depletion (Fig. 4A, compare siControl and siSAF-A). However, we repeatedly detected wider variance of replication fork speed in siSAF-A cells treated with HU, compared with that in siControl cells treated with HU (Fig. 4A, compare siControl+HU and siSAF-A+HU), suggesting that in SAF-A-depleted cells replication fork speed is less tightly regulated during replication stress.

We next tested whether SAF-A depletion affects the processivity of DNA replication forks. If processivity is high, the rate of DNA synthesis will stay consistent through the CldU and IdU labelling periods, and the log₂ value of the ratio of IdU tract length to CldU tract length [log₂(IdU/CldU)] is expected to be close to 0 (for example, Fig. 4B siControl). Frequent pause or collapse of forks will lead to a wider spread in log₂(IdU/CldU) values. We found that cells depleted of SAF-A show wider spreading of the log₂(IdU/CldU) values under HU-treated conditions, indicating increased probability of fork slowing, pause or collapse in cells depleted of SAF-A (Fig. 4B; note that forks pausing or collapsing in the CldU labelling period were not counted – because they produce only CldU labelling, they are indistinguishable from termination sites). This result suggests that SAF-A is required to support processive DNA synthesis under replication stress. The nascent DNA labelling experiments demonstrate that SAF-A is required for robust replication fork progression, as well as to support origin licensing.

Given its effect on replication origin activation and fork progression, we examined whether SAF-A is important for DNA replication timing. To enable detection of changes that might not be evident in a population analysis (such as increased variability that does not affect the average replication time of a locus), we examined the replication timing programme in single cells. Briefly, we used a recently described method (Miura et al., 2020; Takahashi et al., 2019) in which single mid-S-phase cells are collected by cell sorting based on their DNA content, then next-generation sequencing (NGS) library preparation and copy number sequencing are carried out for each individual single cell. As a control, we carried out similar analysis using a pool of 100 mid-S-phase cells. The relative copy number of 200-kb segments was calculated based on the number of sequencing reads, normalised against reads obtained from G1-phase cells.

We compared the replication timing profiles of 33 single mid-S-phase hTERT-RPE1 cells for siControl, and 25 single mid-S-phase cells for siSAF-A (Fig. 5A). We found that, as previously proposed, replication timing of single siControl cells generally reflects A/B compartment distribution as determined by hTERT-RPE1 cell Hi-C analysis (Darrow et al., 2016; Miura et al., 2018). While in siSAF-A cells the overall replication timing profiles are largely similar to those of siControl cells, we found that the boundaries of the replication timing domains are less uniform. For example, in the regions shown magnified at the bottom of Fig. 5A, siControl cells show clear, fairly uniform boundaries between unreplicated (blue) and replicated (red) domains across the 33 analysed cells. In siSAF-A cells, in contrast, the boundary position shows more variation between single cells, resulting in a lack of clear boundaries when viewed across the population. The ‘RT changes’ plot in Fig. 5A shows the difference between single siControl and siSAF-A cells in average replication timing. Statistical comparison of single-cell replication timing between siControl and siSAF-A cells confirms the impression that boundaries are blurred (see ‘−log₁₀P’ plot in Fig. 5A), with peaks indicating regions showing significant difference between the siControl and siSAF-A profiles, generally coinciding with timing domain boundaries. We identified 420 ‘−log₁₀P’ peaks across the genome, of which 173 overlap with replication timing domain boundaries (RT boundaries). A one-tailed Fisher's exact test found that the co-occurrence of ‘−log₁₀P’ peaks and RT boundaries is statistically significant (P=2.22×10⁻¹⁸), whereas no statistical significance was found if genomic locations of ‘−log₁₀P’ peaks were randomly shuffled (P=0.99; 76 co-occurrences out of 420).

Depletion of SAF-A does not lead to any clear trend in replication timing changes genome-wide (with similar numbers of regions becoming earlier or later; Fig. 5B, i), nor does it cause a consistent shift in timing at all RT domain boundaries (Fig. 5B, ii). If, however, we consider all regions showing significant difference in replication time between siControl and siSAF-A cells (defined by −log₁₀P>3), then SAF-A depletion results in some tendency towards earlier replication timing (Fig. 5B, iv), although the changes vary substantially in direction and magnitude with some sequences replicating later than normal. Furthermore, at RT boundaries where siControl and siSAF-A cells show substantial differences, the effect of SAF-A depletion is noticeably bimodal, with the majority of such boundary regions replicating earlier but some replicating later (Fig. 5B, iii). Our finding that many RT boundaries are sensitive to SAF-A depletion is consistent with the proposed function of SAF-A in defining chromatin domain boundaries (Fan et al., 2018).

We also noticed that chromosomal locations that show replication timing shift in siSAF-A cells tend to coincide with genomic locations where A/B compartment and replication timing patterns are discordant or ‘disagree’ (see Fig. S5A for specimen loci). For statistical comparison, we picked the top 10% of genomic segments showing later replication timing in siSAF-A cells compared to siControl cells (‘EtoL’ sites), and the top 10% of segments showing earlier replication in siSAF-A cells (‘LtoE’ sites). These EtoL and LtoE sites show a significantly higher proportion of discordance between compartment and replication timing than the genomic average (Table 1). Moreover, at these loci, the observed replication timing shift in siSAF-A cells tends to bring replication timing into alignment with the A/B compartment pattern (Fig. S5B; Table 1).

To quantitatively confirm the variability in replication timing between individual siSAF-A cells, we compared the number of NGS reads per 200 kb sliding window (i.e. tag density) at 40-kb intervals (Miura et al., 2020). In the pool of 100 mid-S-phase siControl cells, distribution of the tag density forms two overlapping peaks (Fig. 5C, left), representing unreplicated (left peak) and replicated (right peak) portions of the genome. The separation of these peaks in the 100 pooled cells means that (1) unreplicated and replicated domains are distinct in each cell and (2) this distinct pattern is essentially conserved in the 100 cells. In other words, the replication timing programme is well-conserved in these 100 siControl cells. In contrast, tag density from the 100 mid-S-phase siSAF-A cells does not show clear peak separation (Fig. 5C, right). Note, however, that we do see clear separation of two peaks when analysing single siSAF-A cells at mid-S phase (Fig. S5B), similar to single siControl cells, indicating that our single-cell analysis does effectively distinguish unreplicated and replicated domains. Therefore, the poor peak separation of tag density in the pool of 100 mid-S-phase siSAF-A cells is due to compromised conservation of the replication timing programme between single cells.

Although the overall replication timing profiles appear to be fairly similar (Fig. 5A) and only a limited number of loci show altered replication timing (Fig. 5B, Table 1; Fig. S5A), t-SNE clustering analysis (van der Maaten, 2014; van der Maaten and Hinton, 2008) of the distribution of early and late domains in single cells showed a clear separation of the siControl and siSAF-A populations (Fig. 5D), indicating that the genome-wide replication timing programme is indeed altered in siSAF-A cells.

We conclude that in cells depleted of SAF-A, the genome-wide DNA replication timing programme is less well-defined, becoming more ‘blurred’ and unstable, particularly at replication timing domain boundaries and in regions where A/B compartment and replication timing are normally discordant.

Our data suggest that DNA replication is aberrant at various stages in cells depleted of SAF-A, even without exogenous replication stress (Fig. 1F,G; Fig. S1D). Recent studies suggest that cells with incomplete DNA replication and/or DNA damage can progress through mitosis but may activate the p53-mediated G1 checkpoint in the subsequent cell cycle, leading to a transient quiescence of daughter cells (Arora et al., 2017; Barr et al., 2017). Such delayed progression can be monitored by examining expression of the CDK inhibitor p21 (also known as p21^WAF1 and CDKN1A). We tested the possibility that replication problems in siSAF-A cells lead to spontaneous quiescence by looking at the expression of p21. Cells depleted of SAF-A showed clear expression of p21 without any exogenous damage (Fig. 6A), whereas the expression of p21 was barely detectable in siControl cells. Flow cytometry (Fig. 6B; Fig. S6A) revealed that a significant proportion of siSAF-A cells with a ‘G1 phase’ DNA content showed p21 expression, suggesting that these cells are in quiescence (G0 phase). Interestingly, a fraction of G2-phase siSAF-A cells already expressed p21. Expression of p21 in G2-phase cells has been reported for cells that have undergone DNA damage and are destined to enter quiescence (Arora et al., 2017; Barr et al., 2017). The tendency of SAF-A-depleted cells to enter quiescence was also confirmed by measuring the phosphorylation of retinoblastoma (Rb, also known as RB1) protein. Rb is a negative regulator of the cell cycle and is phosphorylated in proliferating cells but remains unphosphorylated in quiescent cells (Giacinti and Giordano, 2006). Measurements of the cellular levels of Rb phosphorylation at Ser-807 and/or Ser-811 demonstrated that a significantly higher proportion of siSAF-A cells were in quiescence, as evidenced by dominance of cells with unphosphorylated Rb (Fig. 6C,D; Fig. S6B,C).

A previous study has demonstrated that depletion of SAF-A increases the proportion of cells showing diffuse localisation of the histone variant H2AX phosphorylated at its C terminus (referred to as γ-H2AX) (Nozawa et al., 2017). γ-H2AX has commonly been used as a DNA damage marker, but recent studies suggest that diffuse localisation of γ-H2AX within the nucleus is indicative of replication stress rather than DNA damage, whereas a more focal γ-H2AX localisation pattern represents DNA damage (Dhuppar et al., 2020; Moeglin et al., 2019). We assessed the impact of depletion of SAF-A with or without replication stress based on γ-H2AX localisation pattern. Fig. 7A shows a specimen image with ‘diffuse’ and ‘focal’ γ-H2AX localisation patterns. Without replication stress, few siControl cells exhibited either γ-H2AX pattern, with most cells showing no apparent γ-H2AX signal (Fig. 7B, siCont). Replication stress (induced by 3 h HU treatment) significantly increased the proportion of cells with ‘diffuse’ γ-H2AX (29%; Fig. S7A), consistent with the suggestion that diffuse γ-H2AX signal is indicative of DNA replication stress. In contrast, 29% of cells depleted of SAF-A had diffuse γ-H2AX even without HU treatment (Fig. 7B, siSAF-A), suggesting that depletion of SAF-A imposes replication stress on cells. In a separate experiment, we confirmed that virtually all cells with diffuse γ-H2AX signal were in S phase (Fig. S7B), and that a large fraction of S-phase cells had diffuse γ-H2AX signal when SAF-A was depleted (Fig. S7C).

Multiple protein kinases, including ATM, ATR and DNA-PK, have been implicated in γ-H2AX formation upon DNA damage (Rogakou et al., 1998; Wang et al., 2005). However, during replication stress, ATR is generally believed to be the kinase responsible for formation of γ-H2AX (Ward and Chen, 2001), although some reports implicate other protein kinases (Buisson et al., 2015; Chanoux et al., 2009; Serrano et al., 2013). To investigate which protein kinase is responsible for increased γ-H2AX in siSAF-A cells, we tested the impact of inhibiting ATM (using KU-60019), ATR (using VE-821) and DNA-PK (using NU-7441) on nuclear γ-H2AX signal in S-phase cells (Fig. 7C; Fig. S7D). Inhibition of ATR almost completely suppressed γ-H2AX induction upon SAF-A depletion (Fig. 7C, compare right halves of untreated and ATRi plots), consistent with the increased γ-H2AX signal in siSAF-A cells being caused by replication stress. Unexpectedly, ATM inhibition caused an increase in basal γ-H2AX signal even in control cells (Fig. 7C, compare left halves of untreated and ATMi plots), but did not notably impact the γ-H2AX levels in siSAF-A cells. Inhibition of DNA-PK had a slight effect on γ-H2AX levels. In conclusion, the increased γ-H2AX in SAF-depleted cells appears to be mediated mainly by ATR.

These data demonstrate that cells depleted of SAF-A suffer from constant replication stress, leading to more frequent (or more extended) quiescence than in control cells, which can at least partly explain the slower cell proliferation (Fig. 1C) and the reduced fraction of S-phase cells (Fig. 1F).

Taking all these observations together, our results demonstrate that SAF-A supports DNA replication by promoting origin licensing, fork progression speed and fork processivity, probably by modulating chromatin compaction to ensure the optimal structure for robust DNA replication.

Our investigation of the effects of SAF-A on DNA replication establishes that SAF-A promotes replication licensing (Fig. 2). Consistent with this effect, cells depleted of SAF-A showed increased origin spacing when compared to control cells, as well as reduced ability to activate dormant origins under replication stress (Fig. 3B,C). It has recently been demonstrated that the histone methyltransferase SET8, in contrast, limits replication licensing (Shoaib et al., 2018), presumably through its activity in histone H4K20 methylation. Given that SAF-A mediates the establishment of open chromatin structure (Creamer et al., 2021; Nozawa et al., 2017; Sharp et al., 2020), it therefore appears that the correct level of origin licensing requires an appropriate balance between oppositely acting cellular mechanisms that specify chromatin compaction.

In addition to its implication in chromatin structure, SAF-A is known to affect gene expression by modulating mRNA splicing, mRNA stability and mRNA localisation (Huang et al., 2021; Xiao et al., 2012; Ye et al., 2015; Yugami et al., 2007). We found a slight reduction in protein levels of FLAG–ORC1 and CDT1 (Fig. S4) that may contribute to the reduced licensing in cells depleted of SAF-A. SAF-A depletion appears to compromise replication licensing more severely than it does licensing factor expression, so we suspect that altered chromatin structure is the major determinant of reduced licensing in siSAF-A cells, with reduced protein expression an additional contributing factor.

Once origins have initiated, replication fork progression is also affected by SAF-A depletion, with fork speed less tightly regulated (Fig. 4A) and higher variance in fork processivity under replication stress (Fig. 4B). These changes may reflect an increased incidence of ‘chromatin obstacles’ in the absence of SAF-A, corresponding to hard-to-replicate sites that challenge the replication machinery (Gadaleta and Noguchi, 2017). It has been demonstrated that processive replication through heterochromatin regions is coupled with local chromatin decompaction (Chagin et al., 2019), so that chromatin over-compaction in the absence of SAF-A may increase the number of replication fork impediments, causing the observed inconsistent fork processivity (Fig. 4B). Interestingly, it is proposed that unactivated (‘non-CMG-assembled’) MCM proteins reduce replication fork speed to prevent DNA damage during DNA replication (Sedlackova et al., 2020). Reduced chromatin association of MCM in siSAF-A cells could be envisaged to increase the variance in fork processivity in this manner as well.

Despite having a moderate effect on origin licensing, depletion of SAF-A had a fairly mild effect on EdU incorporation levels without additional replication stress (Fig. 1F,G; Fig. S1D). This probably reflects the fact that under normal circumstances, MCM is loaded at a larger number of sites than will be utilised, so that a modest reduction in origin licensing has only a slight impact on cellular DNA replication dynamics under unperturbed conditions (Ge et al., 2007; Woodward et al., 2006). We find, however, that there is a stronger requirement for SAF-A in enabling cells to recover from replication stress (Fig. 1D,E; Fig. S1B,C). One possible explanation, based on the origin licensing defect of SAF-A-depleted cells, is that insufficient ‘dormant’ origins are available for activation to enable proper recovery from stress (Fig. 3B,C). Inadequate licensing that fails to provide enough dormant origins may lead to chromosome segments remaining unreplicated if incoming replication forks from both directions collapse under replication stress conditions (McIntosh and Blow, 2012).

Depletion of SAF-A led to increased γ-H2AX signal without exogenous damage (Fig. 7A,B). Aiming to identify the protein kinase responsible for the increased γ-H2AX, we treated cells with inhibitors against three candidate kinases (ATM, ATR and DNA-PK; Fig. 7C). The result confirmed that the increased phosphorylation of H2AX is mainly mediated by ATR, consistent with our inference based on the diffuse γ-H2AX localisation pattern that depletion of SAF-A causes replication stress.

Although depletion of SAF-A led to reduced overall licensing and increased origin spacing, we found that the genome-wide replication timing programme was not severely affected (Fig. 5A). This is perhaps consistent with the fact that each replication timing domain contains multiple replication origins whose activation is concomitantly regulated. Therefore, even if some licensed origins are lost from a replication timing domain, the domain can still retain its programmed replication timing, enabled by correctly regulated initiation at the remaining origins. We did, however, observe that the sharp boundaries that normally delineate replication timing domains tend to be blurred, showing considerably increased cell-to-cell heterogeneity (Fig. 5A,B). Such increased heterogeneity in timing domain boundaries will contribute to the poor separation of ‘early’ and ‘late’ replication peaks in the analysis of genome-wide tag density distribution (Fig. 5C), and is likely to be an important parameter in the separation of siControl cells and siSAF-A cells in t-SNE analysis (Fig. 5D). Interestingly, SAF-A protein has been reported to interact with chromatin domain boundary proteins, including CTCF and cohesion subunit RAD21 (Fan et al., 2018; Zhang et al., 2019), and to be involved in defining chromatin domain boundaries (Fan et al., 2018). We observed that depletion of SAF-A also tends to cause replication timing shift at loci where A/B compartment and early/late replication timing disagree (Table 1; Fig. S5A). Although in general the A/B compartment pattern corresponds well to the early/late replication pattern, there are some exceptional loci are where replication timing and A/B compartment patterns disagree. At some such loci, SAF-A is required for the establishment or maintenance of the replication timing pattern discordance with chromatin compartmentalisation.

SAF-A has been implicated in inactivation of X chromosomes (Hasegawa et al., 2010; Lu et al., 2020; Smeets et al., 2014). The hTERT-RPE1 cells used for our timing analysis are female, but we did not observe any obvious impact of depleting SAF-A on X-chromosome replication timing (data not shown). However, subtle changes in the inactive X-chromosome replication timing might not be detected, given that our methodology did not separately analyse the two X chromosomes.

RNA appears to be a functional component of chromatin (Brockdorff, 2019; Michieletto and Gilbert, 2019; Rodriguez-Campos and Azorin, 2007). One recent study proposes that chromatin-associated RNA promotes open chromatin structures by neutralising the positive charges on histone tails (Dueva et al., 2019). The ‘polarised’ chromatin distribution we observed (Fig. 1A,B) may reflect a role for SAF-A in tethering RNA to decompact chromatin (Creamer et al., 2021).

In summary, we have demonstrated that SAF-A is required for robust DNA replication, both in unperturbed conditions and in the recovery from replication stress. Moreover, we show that depletion of SAF-A leads to spontaneous replication stress and increased quiescence. Elevated expression of SAF-A in a wide range of cancers (The Cancer Genome Atlas; https://www.cancer.gov/tcga) suggests that SAF-A is important for cancer cell survival, possibly through the management of replication stress in cancer cells (Gaillard et al., 2015; Macheret and Halazonetis, 2015). Overall, our findings reported here show that the promotion of robust DNA replication by SAF-A is crucial for its role in supporting cellular capacity for proliferation.

Cell lines hTERT-RPE1 (Bodnar et al., 1998) and HEK293 FLAG–ORC1 were as previously described (Tatsumi et al., 2003). Cells were checked for mycoplasma contamination at regular intervals.

All human cell lines were cultivated in synthetic defined media (described below) supplemented with 10% foetal bovine serum (tetracycline-free; Biosera), 100 U/ml penicillin and 100 μg/ml streptomycin in 5% CO₂, ambient O₂ and at 37°C. hTERT-RPE1 cells were generally cultivated in DMEM/F12 (Gibco), except for experiments involving dNTP analogue (EdU, CldU and IdU) labelling where DMEM (Gibco) was used. Other cell lines were cultivated in DMEM.

siRNAs used were: SAF-A siRNA, human HNRNPU (3192) ON-TARGETplus SMARTpool (Dharmacon Cat# L-013501-00-0005; Horizon Discovery); and control siRNA, luciferase (GL2) (Dharmacon Cat# D-001100–01; Horizon Discovery). Cells were transfected with 10 nM siRNA using Lipofectamine RNAiMAX reagent (Invitrogen, Thermo Fisher Scientific).

Primary antibodies used were: anti-SAF-A (mouse monoclonal, 3G6; Abcam, ab10297; 0.5 µg/10⁶ cells for flow cytometry and 1/10,000 for western blotting); anti-FLAG (mouse monoclonal, M2; Sigma-Aldrich, F-1804; 1/200 for flow cytometry); anti-MCM3 (goat polyclonal, N-19 IgG; Santa Cruz Biotechnology, sc-9850; 1/200 for flow cytometry); anti-CDT1 (rabbit monoclonal, EPR17891; Abcam, ab202067; 1/200 for flow cytometry); anti-ORC1 (mouse monoclonal, F-10 IgG1; Santa Cruz Biotechnology, sc-398734; 1/200 for western blotting); anti-ORC2 (rabbit polyclonal; Bethyl, A302-734A; 1/100 for flow cytometry); anti-p21 (rabbit polyclonal, C-19; Santa Cruz Biotechnology, sc-397; 1/200 for flow cytometry and western blotting); anti-histone H3 (rabbit polyclonal; Abcam, ab1791; 1/5000 for western blotting); anti-CldU [rat monoclonal anti-BrdU, BU1/75 (ICR1); Abcam, ab6326; 1/100 for DNA combing]; anti-IdU (mouse monoclonal anti-BrdU; BD Biosciences, cat# 347580; 1/100 for DNA combing); anti-ssDNA (mouse monoclonal IgG3, 16-19; Millipore, MAB3868; 1/100 for DNA combing); anti-γ-H2AX (rabbit monoclonal, 20E3; Cell Signaling Technology, #9718; used at 1/400 for immunofluorescence); Alexa Fluor 647-conjugated anti-γ-H2AX (mouse monoclonal, N1-431; BD Pharmingen, 560447; used at 5 µl per tube); phospho-Rb (Ser807/811) (rabbit monoclonal, D20B12 IgG; Cell Signaling Technology, #8516; 1/800 for flow cytometry).

Secondary antibodies used were: Alexa Fluor 647-conjugated donkey anti-rabbit IgG (H+L) (Abcam, ab150063); Alexa Fluor 488-conjugated donkey anti-rabbit IgG (H+L) (Abcam, ab150065); Alexa Fluor 488-conjugated donkey anti-mouse IgG (H+L) (Abcam, ab150109); Alexa Fluor 647-conjugated donkey anti-goat IgG (H+L) (Abcam, ab150135); Alexa Fluor 488-conjugated goat anti-mouse IgG (H+L) (Abcam, ab150117); Alexa Fluor 594-conjugated goat anti-rat IgG (H+L) (Molecular Probes, A-11007, Thermo Fisher Scientific); Alexa Fluor 350-conjugated goat anti-mouse IgG (H+L) (Molecular Probes, A-11045, Thermo Fisher Scientific); Alexa Fluor 488-conjugated goat anti-mouse IgG1 (Molecular Probes, A-21121, Thermo Fisher Scientific). They were all used at 1/2000 dilution.

To prepare chromatin-enriched fractions (Fig. S2E), cells were lysed in cytoskeleton (CSK) buffer [10 mM HEPES-KOH (pH 7.4), 100 mM NaCl, 3 mM MgCl₂, 300 mM sucrose] containing 0.2% Triton X-100, 1× cOmplete protease inhibitor cocktail EDTA-free (Roche, 04693159001), and 1× HALT protease and phosphatase inhibitor (Thermo Fisher Scientific, 78446) for 10 min on ice. Lysed cells were then centrifuged for 3 min at 2000 g. The pellet was washed once with CSK buffer, centrifuged for 4 min at 2000 g and resuspended in CSK buffer containing 10 µl/ml Benzonase (Millipore) for 30 min on ice. Samples were boiled in 1× Laemmli sample buffer for 10 min, and 5% β-mercaptoethanol was added.

To prepare chromatin-enriched fractions for analysis of p21 (Fig. 7A), CDT1 (Fig. S2C, left), CDC6 and ORC1 (Fig. S2C, middle), cells were lysed in low-salt extraction (LSE) buffer [10 mM potassium phosphate buffer (pH 7.4), 10 mM NaCl, 5 mM MgCl₂] containing 0.1% Igepal CA-630 and 1 mM PMSF for 5 min on ice. Lysed cells were then centrifuged (10 min at 12,000 g), and the pellet was washed once with LSE buffer. The pellet was resuspended and boiled in 1× Laemmli sample buffer for 10 min, and 5% β-mercaptoethanol was added.

Protein concentrations in whole-cell extracts (WCEs) were determined using the Bio-Rad RC DC Protein assay kit. For western blots, an equal amount of total protein was loaded on each WCE lane, and loading for the corresponding chromatin fractions was calculated based on cell equivalency. Equal loading was further confirmed by examining total protein using Mini-PROTEAN stain-free gels (Bio-Rad).

For analysis of nascent DNA on DNA fibres, cells were pulse labelled sequentially with CldU (50 µM; Sigma-Aldrich, C6891) and IdU (250 µM; Sigma-Aldrich, I7125) for 20 min each. Cells were then collected, and DNA combing carried using a FiberComb instrument (Genomic Vision, Bagneux, France) according to the manufacturer's instructions. Detection of CldU and IdU was as previously described (Garzon et al., 2019). Images were acquired on a Zeiss Axio Imager M2 microscope and 63×/NA1.4 objective equipped with ORCA-Flash 4.0LT CMOS camera (Hamamatsu Photonics). Images were analysed as previously described (Garzon et al., 2019). For inter-origin distance measurements, 1 µm was converted to 2 kb based on a predetermined value (Bensimon et al., 1994, 1995).

Cell cycle analysis of cells stained with DAPI was performed as described previously (Hiraga et al., 2017; Watts et al., 2020). EdU labelling and its detection by flow cytometry have been previously described (Hiraga et al., 2017). Cells were extracted before fixation with low-salt extraction buffer [0.1% Igepal CA-630, 10 mM NaCl, 5 mM MgCl₂, 0.1 mM PMSF, 10 mM potassium phosphate buffer (pH 7.4)] unless otherwise noted. Detection and analysis of chromatin-bound proteins by flow cytometry were performed as previously described (Hiraga et al., 2017) with multiplexing as described below. For analysis of total proteins, cells were fixed with formaldehyde prior to permeabilisation. Data were acquired on Becton Dickinson LSRII or Fortessa flow cytometers with FACSDiva software (Becton Dickinson) and were analysed using FlowJo software version 10.4.2 (FlowJo LLC., Ashland, OR, USA).

We found that apparent MCM levels per cell are very sensitive to the number of cells analysed (i.e. the ratio of cells to antibody during immunostaining), causing tube-to-tube variations. To avoid this issue, we adopted a ‘multiplexing’ strategy. In brief, before immunostaining, samples were differentially labelled with CellTrace Yellow (Molecular Probes, Thermo Fisher Scientific) at a concentration unique to each sample (between 0 µM and 0.5 µM final concentration). Differentially stained samples were then mixed and were immunostained in a single tube to eliminate tube-to-tube variations. After data acquisition by flow cytometry, cell populations were separated based on their CellTrace Yellow signal levels. We confirmed that the CellTrace Yellow signal does not affect the quantification of Alexa Fluor 488 and Alexa Fluor 647 signals.

For visualisation of chromatin DNA within the nucleus, cells were grown on Ibidi chambered slides (ibi-treated; obtained from Thistle Scientific Ltd., Glasgow, UK). Cells were washed with phosphate-buffered saline (PBS), and fixed with neutral buffered 4% formaldehyde (Sigma) for 15 min at room temperature, then permeabilised with 0.1% Triton X-100 in PBS for 15 min at room temperature. After washing cells three times with PBS containing 0.1% Igepal CA-630, DNA was stained with 0.25 µg/ml DAPI for 30 min. Cells were finally washed and mounted in Ibidi mounting medium. Eleven z-section images were acquired at 170 nm intervals on a Zeiss LSM-880/AiryScan microscope with 63×/NA 1.3 objective (with Zen Black software; Zeiss). The middle section of the z-stacks was assigned as the plane where each nucleus has the largest x–y projection. After AiryScan processing (with the automatic 3D AiryScan processing condition), the middle section was used for analysis. The areas of DAPI-high and DAPI-low regions were determined in an unbiased manner by a custom pipeline utilising Minimal Cross-Entropy on CellProfiler 3.19 (McQuin et al., 2018).

For visualisation of EdU incorporation and immunofluorescence detection of γ-H2AX, cells were grown and fixed as above, and kept in 70% ethanol. Cells were then washed with PBS, permeabilised with 0.5% Triton X-100 in PBS, and incorporated EdU was visualised using an Alexa Fluor 488 EdU imaging kit (Molecular Probes C10337) according to the manufacturer's instructions, followed by indirect immunofluorescence staining of γ-H2AX. Antibodies used were p-histone H2A.X S139 (20E3) rabbit monoclonal antibody (Cell Signaling Technology, #9718) and Alexa Fluor 647-conjugated anti-rabbit IgG (Abcam, ab150063). Z-stack images were acquired at 250 nm intervals to cover entire nuclei in the field. After AiryScan processing, maximum intensity z-projection images were created for downstream analysis using ImageJ (NIH, Bethesda, MD, USA). Detection of cells with diffuse γ-H2AX or γ-H2AX foci were carried out by using a custom CellProfiler pipeline. For the data presented in Fig. 7C, a Zeiss Axio Observer Z1 with Plan-Apochromat 63x/1.40 objective was used. Images were taken with ORCA-Flash4.0 V3 CMOS camera (Hamamatsu Photonics).

Homo sapiens (human) genome assembly GRCh38 (hg38) from Genome Reference Consortium was used throughout the analysis. Single-cell replication timing of siControl and siSAF-A hTERT-RPE1 cells were analysed as described previously (Miura et al., 2020; Takahashi et al., 2019). In brief, single cells were sorted into a 96-well plate by flow cytometry (Sony SH800), and an NGS library was prepared from each single cell as previously described (Miura et al., 2020; Takahashi et al., 2019). Copy number analysis of human genomic segments were performed as described (Miura et al., 2020; Takahashi et al., 2019). Replication timing boundaries were defined as corresponding to the transition point for binarised replication values (from −1 to 1 or from 1 to −1) of 100 siControl mid-S-phase cells. Replication timing changes (RT changes) between siControl and siSAF-A cells were calculated by comparing the average replication timing of single siControl and siSAF-A cells. The ‘−log₁₀P’ values were calculated by comparing the distribution of single-cell replication timing of 100-kb segments between siControl and siSAF-A cells using a two-tailed unpaired t-test. The ‘−log₁₀P’ peaks were defined as those with ‘−log₁₀P’ values above 3, which corresponds to P-values below 0.0001. Bedtools (version 2.30.0) was used for additional data analysis (Quinlan and Hall, 2010). Bedtools intersect was used to identify intersections between different data tracks (e.g. −log₁₀P peaks and RT boundaries).

Bedtools fisher was used to perform Fisher's exact test to evaluate the co-occurrence between −log₁₀P peaks and RT boundaries (Quinlan and Hall, 2010). A one-tailed test was performed, with the null hypothesis that occurrences of −log₁₀P peaks have no correlation with RT boundaries, and the alternative hypothesis that occurrences of −log₁₀P peaks have positive correlation with RT boundaries.

t-SNE clustering analysis of replication timing profile data was performed using Rtsne R library (see https://github.com/jkrijthe/Rtsne), with R version 3.6.1. For t-SNE analysis, all the 40-kb chromosome segments with measurements from all the single cells (i.e. no missing values) were used. The number of 40-kb chromosome segments used was 71,979, which covers 93% of the genome. The following parameters were supplied to Rtsne; perplexity=19, PCA-scaling=T.

Chromatin A/B compartments in hTERT-RPE1 were determined as described previously (Miura et al., 2018), using a previously published Hi-C dataset in 100-kb bins (Darrow et al., 2016). The original genome coordinates in genome assembly GRCh37 (hg19) to GRch38 (hg38) were converted using UCSC liftOver tool (Hinrichs et al., 2006) and remapped using BEDOPS bedmap (with a weighted average option) (Neph et al., 2012), so that both datasets could be compared in a common genomic segmentation system. For the analysis presented in Table 1, A/B compartment and replication timing in siControl cells was compared at each 100-kb segment of the genome.

GraphPad Prism (version 7; GraphPad Software, San Diego, CA, USA) and R Studio (version 1.4.1717 with R 4.1.0) were used for statistical analysis of experimental data and creating graphs. Student's t-test was used for data with normal distributions as indicated in figure legends, and Mann–Whitney–Wilcoxon test was used for non-normal data (in Fig. 1G, Fig. 3C and Fig. 7C). Two-tailed Fisher's exact test was used in Fig. 6D and Table 1. Specific conditions used are stated either in the main text or in the figure legends when necessary. Beanplot R package (version 1.2) was used for creating two-sided bean plots (Kampstra, 2008).

Information for SAF-A expression was obtained at The Cancer Genome Atlas (TCGA) Research Network (https://www.cancer.gov/tcga). We thank Dr Ryu-suke Nozawa for help in the early stage of the project, and Professor Julian Blow for advice on the 3D licensing assay. Thanks to the staff of the Iain Fraser Cytometry Centre, and Microscopy and Histology facility at the University of Aberdeen.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.258991.