Main

Error-free genome duplication is essential for precise genetic inheritance and for protection against genomic instability—a driver of oncogenic transformation and cancer progression10,11. In eukaryotes, such precision is maintained by two fundamental controls: the number of active replication origins and the velocity of active replisomes12,13. The DNA replication program is established in G1 by the licensing of an excess pool of origins, marked by the loading of inactive minichromosome maintenance (MCM) helicases 2–7 onto chromatin. During S phase, only around 10% of these origins are activated to form active replisomes, in which MCM2–MCM7 assemble with CDC45 and GINS1–GINS4 (Go-Ichi-Ni-San 1–4 subunits) to form the CMG helicase, together with DNA polymerases and additional regulatory proteins14. The S-phase checkpoint, governed by the ataxia telangiectasia and RAD3-related (ATR) pathway, coordinates origin firing with ongoing synthesis in cells both in unperturbed S phase and under replication stress, thereby preventing premature or excessive origin activation and untimely mitotic entry1,15. Whereas insufficient origin activation causes DNA under-replication16, excessive origin firing exhausts replication resources and destabilizes the genome2,3. Thus, we hypothesize that the S-phase checkpoint operates within boundaries set by as-yet-unidentified replisome dynamics that define a global replication capacity and prevent replication from exceeding this threshold.

Unscheduled origin activation affects DNA replication

To test this hypothesis, we examined whether rapid origin firing would deplete core replisome components and thereby reveal the key rate-limiting steps of global genome replication. Therefore, we applied brief ATR inhibition to induce rapid and aberrant cyclin-dependent kinase (CDK) activation in a cell line expressing a TagRFP chromobody targeting endogenous PCNA, a proxy for active replisomes and local replication dynamics17 (Fig. 1a). Using quantitative image-based cytometry (QIBC)2,18, we first confirmed that short-term inhibition of ATR (40 min) robustly activates CDKs during S phase, as indicated by phosphorylation of FOXM1—an established CDK1 and CDK2 (hereafter, CDK1/2) target1—without triggering a DNA damage response, as evidenced by the absence of pan-nuclear γH2AX (phosphorylated histone H2AX) (Fig. 1a, Extended Data Fig. 1a and Supplementary Fig. 1a,b). As anticipated19, such ATR inhibition led to a pronounced increase in active fork density (Extended Data Fig. 1b) and a corresponding reduction in replication fork speed (Extended Data Fig. 1c). However, despite increased CDK activity and abrupt activation of new origins, we did not observe an increase in PCNA chromobody foci after ATR inhibition, as assessed through QIBC of a fixed cell population and live-cell tracking (Fig. 1a and Supplementary Fig. 1c).

Fig. 1: Excess replication origins drive the depletion of chromatin-bound PCNA and lagging-strand activities.

a, Left, workflow using U2OS cells stably expressing the PCNA TagRFP chromobody. Scale bar, 10 µm. ATRi, ATR inhibitor; IF, immunofluorescence; pFOXM1, FOXM1 phosphorylated at Thr600. Middle, QIBC of pFOXM1 in cells treated with dimethyl sulfoxide (DMSO) or exposed to ATRi for the indicated durations. Nuclear DNA was counterstained with DAPI (4′,6-diamidino-2-phenylindole), and total DAPI intensity was used to infer DNA content (2n, G1; 4n, G2). n > 8,000 cells per condition. The colour gradient denotes mean PCNA TagRFP chromobody intensity in individual nuclei. a.u., arbitrary units. Right, fold change in CDK1/2 activity (pFOXM1 intensity) and number of PCNA chromobody spots, normalized to the DMSO-treated condition. b, Left, workflow. Middle, three-dimensional confocal images of endogenously tagged CDC45–GFP U2OS cells immunostained for PCNA. Scale bars, 10 µm. Right, intensity profiles from the insets for CDC45–GFP (green) and PCNA (red) within replication foci. c,d, QIBC of CDC45, PCNA (c) and DNA ligase 1 (d) chromatin binding (n > 5,000 cells). Pink dotted lines in the scatter plots denote approximate upper limits of chromatin-bound protein levels. Fold changes in S-phase-specific chromatin association are shown. e,f, QIBC of RPA1 and γH2AX (e), and of RPA1 and PCNA (f), in U2OS cells under the indicated conditions (n > 8,000 cells). The dotted line marks the onset of replication catastrophe. ssDNA, single-stranded DNA. g, Fold change on chromatin for the indicated proteins during ATRi treatment relative to DMSO (from data in Supplementary Fig. 2d,e). h, Western blot of chromatin from U2OS cells treated with the indicated short interfering RNAs (siRNAs) for 48 h. siATAD5, ATAD5 siRNAl; siRFC1, RFC1 siRNA. i, QIBC of chromatin-bound RPA1 and γH2AX in control or ATAD5-depleted U2OS cells treated with hydroxyurea (HU) or ATRi. The colour gradient reflects the PCNA chromatin-bound intensity. The dotted line marks the onset of replication catastrophe; the red boxes indicate cells undergoing catastrophe (n > 10,000 cells). The QIBC plots shown are representative of at least two independent biological replicates. The camera icon in a was created with BioRender.com. Somyajit, K. (2025) https://BioRender.com/ergwc0j.

Independent to the PCNA chromobody-specific readout, anti-PCNA immunostaining yielded identical results (Fig. 1b,c and Supplementary Fig. 1d). ATR inhibition led to a robust focal chromatin accumulation of early replisome components—including CDC45 of the CMG complex, the replicative DNA polymerases and the replication progression complex (RPC) factors Timeless and Claspin (Extended Data Fig. 1d). However, chromatin-bound PCNA showed only a modest (1.2-fold) increase, and did not scale with CDC45, which rose by 3.6-fold (Fig. 1b,c).

PCNA, a highly abundant homotrimeric sliding clamp, acts as a structural scaffold for continuous leading-strand synthesis by POLε20, and coordinates the dynamic interplay among POLδ, FEN1 and DNA ligase 1 for the synthesis and maturation of short Okazaki fragments (OkFs)21,22. Notably, although core and leading-strand-specific factors were enriched on S-phase chromatin (Extended Data Fig. 1d), further analysis revealed that lagging-strand processes that depend on PCNA—such as those involving FEN1 and DNA ligase 1—did not increase and were instead mildly diminished on chromatin during excess origin activation (Fig. 1d and Extended Data Fig. 1e). PCNA also orchestrates epigenome maintenance by recruiting DNA methyltransferase 1 (DNMT1) to post-replicative nascent DNA23. Of note, the depletion of PCNA and lagging-strand factors from the chromatin pool was accompanied by a failure to load additional DNMT1 onto S-phase chromatin after ATR inhibition (Extended Data Fig. 1f), further highlighting a unique chromatin-level paucity of PCNA induced by unscheduled origin firing.

The functional exhaustion of OkF processing after ATR inhibition was confirmed using proximity ligation assay (PLA)–QIBC, which revealed reduced interactions between PCNA and DNA ligase 1 (Extended Data Fig. 1g), despite increased Timeless–RPA2 proximity in new replisomes (Extended Data Fig. 1g, right). Moreover, these results were corroborated by chromatin fractionation immunoblotting (Extended Data Fig. 1h,i) and QIBC across two additional cell lines (Extended Data Fig. 1j). Finally, the findings obtained using an ATR inhibitor were recapitulated by both WEE1 inhibition and Claspin depletion, both of which enhance CDK activity and are known to promote aberrant origin firing (Supplementary Fig. 2a–c).

Unligated OkFs are processed through a non-canonical pathway, which is mediated by PARP124. This is evidenced by the accumulation of nascent, S-phase-specific polyADP-ribosylation (PAR) chains after short-term inhibition of poly(ADP-ribose) glycohydrolase (PARG)24 (Extended Data Fig. 2a). Consistent with our observation that PCNA lagging-strand activities are depleted by excessive origin firing (Fig. 1c,d), even under normal replication conditions, inhibition of PARG triggered a rapid accumulation of PAR chains specifically in S phase (Extended Data Fig. 2b). This effect intensified with acute inhibition of ATR, but was completely suppressed by blocking dormant origin firing (Extended Data Fig. 2b), indicating that unresolved lagging-strand intermediates after excessive origin firing drive the early activation of PARP1, before fork collapse (40 min of ATR inhibition). The essentiality of this non-canonical pathway is underscored by the fact that PARP1 inhibition and ATR or WEE1 inhibition are synthetic lethal (Extended Data Fig. 2c,d).

Premature origin firing and DNA replication stress have a catastrophic effect on the replicating genome, collectively known as ‘replication catastrophe’25. Exhaustion of the available pool of genome-protective RPA protein marks the onset of such terminal replication collapse, especially when checkpoint failure coincides with DNA replication stress induced by hydroxyurea-triggered nucleotide depletion2 (Fig. 1e). In light of our new findings, we compared the rate of RPA exhaustion with the chromatin paucity of PCNA. Notably, chromatin-bound PCNA and DNA ligase 1 were already saturated—or declined—before RPA exhaustion under hydroxyurea plus ATR inhibition (Fig. 1f and Extended Data Fig. 2e), suggesting that a loss of core replisome activity precedes RPA depletion in driving replication catastrophe.

Together, these results reinforce the notion that PCNA and lagging-strand processes undergo functional exhaustion during unperturbed origin activation and become further depleted upon unscheduled origin firing.

Mechanism of chromatin-level PCNA depletion

To delve deeper, we combined mass spectrometry (MS) with TurboID-based biotinylation of PCNA-proximal proteins (Extended Data Fig. 2f–h) to identify factors that might influence its rate-limiting mechanisms. We focused on PCNA loaders, including canonical replication factor C 1–5 (RFC1–RFC5) and the CTF18-RFC variant—which encircle PCNA homotrimers on the lagging and on the leading strands, respectively4,26—and PCNA-associated factor 15 (PAF15, also known as PCLAF). PAF15 contains a high-affinity PCNA-interacting peptide (PIP) motif and has been implicated in restraining error-prone DNA polymerases during DNA damage5,6,7,8,27, as well as regulating DNMT1 chromatin association through dual mono-ubiquitination of Lys15 and Lys24 by the E3 ligase UHRF128. However, the direct role of PAF15 in unperturbed DNA replication remains unknown.

Mapping the dynamic range of origin activation after ATR inhibition, QIBC analysis revealed that PCNA—and its associated factor DNA ligase 1—accumulate on chromatin by no more than 1.3-fold, even as new origins continue to fire over a 60–90-min window, as evidenced by increasing levels of chromatin-bound Timeless and RPA (Fig. 1g and Supplementary Fig. 2d). Whereas the leading-strand PCNA loader CTF18 accumulated on replicating chromatin, RFC1—of the canonical RFC complex—showed only limited recruitment, revealing a bottleneck in PCNA loading during excessive origin activation, consistent with recent reports of RFC1 depletion under replication stress and checkpoint loss29,30. Notably, PAF15 showed chromatin accumulation closely mirroring that of PCNA after ATR inhibition (Fig. 1g and Supplementary Fig. 2e), raising the possibility that its natural depletion exposes a rate-limiting control point for PCNA function after chromatin loading, particularly in the context of strand-specific DNA replication dynamics.

In line with this, depletion of the PCNA unloader ATAD531, which stabilizes PCNA, OkF processing factors and PAF15 on chromatin (Fig. 1h and Extended Data Fig. 2i–k), rescued replication catastrophe by preventing the loss of PCNA and DNA ligase 1 during excess origin firing (Fig. 1i and Extended Data Fig. 2l). These findings suggest that continuous PCNA unloading under excessive origin activation leads to chromatin depletion of PCNA, whereas exhaustion of key regulators such as PAF15 imposes a rate-limiting control.

PAF15 controls PCNA during lagging-strand maturation

We next examined whether chromatin-bound PCNA dynamics depend on the natural paucity of soluble PCNA-associated factors such as PAF15. Quantitative, cell-cycle-resolved comparison of total and chromatin-bound pools showed that whereas PCNA and its effectors—including DNA ligase 1, DNA polymerases and CTF18—remained abundant, RFC1 exhibited notable paucity in total levels29,30 (Fig. 2a, Extended Data Fig. 3a–e and Supplementary Fig. 3a). PAF15 was almost entirely chromatin-bound, with negligible soluble excess during normal origin firing (Fig. 2b), and this was consistent across cell types, antibodies and synchronization assays (Extended Data Fig. 3f–i).

Fig. 2: PAF15 is a dosage-limited, lagging-strand-specific PCNA factor.

a,b, QIBC-based side-by-side imaging and quantification of total and chromatin-bound pools of PCNA (a) and PAF15 (b) in U2OS cells, together with their distribution across cell-cycle phases. Cell-cycle staging was determined using EdU incorporation and DAPI intensity. n > 9,000 cells analysed per condition. Pink dotted lines in the scatter plots denote the approximate maximum abundance of PCNA or PAF15. c, QIBC analysis showing the linear correlation between chromatin-bound PCNA and PAF15. More than 4,000 cells were quantified across two independent experiments. Cell-cycle phases are colour-coded: G1 (grey), S (purple) and G2 (black). d, Confocal and stimulated emission depletion (STED) microscopy of U2OS cells stained for PAF15 and labelled for active DNA synthesis using EdU. Scale bars, 5 µm. e, Representative images (top) and QIBC-based quantification (bottom) of PLA focus sum intensity for the indicated antibody pairs: PAF15–POLε1, PAF15–POLδ1, PAF15–FEN1 and PAF15–DNA ligase 1. PLA signals are plotted across total DNA content to visualize cell-cycle-stage-dependent interactions. n > 8,000 cells were analysed per condition. Scale bars, 10 µm. f, Example genomic locus showing replication fork directionality (RFD; grey) local PAF15 strand partitioning (green) and initiation zones (black) in E14 mouse embryonic stem cells. g, Average of PAF15 strand partitioning (green) and RFD (grey) around replication initiation zones. h, Immunoprecipitation of Flag-tagged PAF15 in naive U2OS cells and U2OS cells constitutively expressing PAF15-MYC-3×Flag. The indicated proteins from input and Flag immunoprecipitation lysates were analysed by western blot. i, Model depicting the presence of PAF15 at the lagging strand only, and its exclusion from the leading strand. Each QIBC plot shown is representative of at least two independent biological replicates. The model in** i** was created with BioRender.com. Somyajit, K. (2025) https://BioRender.com/yh2ia9e.

Despite its low abundance, PAF15 constitutes a bona-fide component of the active replisome (Fig. 2c,d and Extended Data Fig. 3j,k). MS-based interactome analysis of chromatin-fractionated PAF15 suggested that PCNA is the main partner responsible for recruiting PAF15 to the replisome (Supplementary Fig. 3b). Consistently, the highly conserved PIP motif in PAF15 is essential for its localization to replicating chromatin (Extended Data Fig. 4a,b). By contrast, blocking APC/C (anaphase-promoting complex, also known as the cyclosome)- and CDH1-mediated degradation of PAF15 via its KEN-box5 did not increase PAF15 chromatin loading (Extended Data Fig. 4a,b), suggesting that PAF15 assembly occurs downstream of origin firing, probably with additional regulation.

Because PCNA is essential for processive DNA replication on both strands, we next investigated which PCNA pool recruits PAF15 at the replisome. Using PLA to capture spatially proximal protein–protein interactions32, we found that PAF15 specifically interacts with lagging-strand components (POLδ, FEN1 and DNA ligase 1) during S phase, but is excluded from leading-strand factors such as POLε (Fig. 2e). To independently assess the partitioning of PAF15 between the leading and lagging strands at replication forks, we used sister chromatid after replication sequencing (SCAR-seq)33 and OkF sequencing (OK-seq)34,35 in mouse embryonic stem (ES) cells. Strand-resolved analyses revealed a pronounced enrichment of PAF15 in lagging-strand regions (Fig. 2f,g), inversely correlated with replication fork directionality (RFD) (Fig. 2f,g and Extended Data Fig. 4c–e). Co-immunoprecipitation analyses revealed the same pattern, showing that PAF15 associates with PCNA in a complex with lagging-strand factors36, but not with the leading-strand polymerase POLε (Fig. 2h). These results show that PAF15 is associated predominantly with the lagging arms of replication forks (Fig. 2i). Although perhaps unexpected, this enrichment aligns with quantitative MS showing that the levels of PCNA exceed those of PAF15 by ninefold in total and by twelvefold on chromatin (Extended Data Fig. 4f), indicating that PAF15 is selectively recruited to a subset of PCNA that is engaged in lagging-strand synthesis.

PCNA, as a homotrimer, coordinates POLδ, FEN1 and DNA ligase 1 by functioning as a dynamic ‘toolbelt’4,8,21,37. However, how lagging-strand factors avoid steric conflict during hand-off remains unclear. Structural alignment of the PCNA–POLδ–FEN1–DNA cryo-electron microscopy (cryo-EM) complex with the PCNA–PAF15–DNA crystal structure reveals severe clashes between PAF15 and POLδ–FEN1, indicating that PAF15 cannot coexist with these PCNA-bound enzymes (Extended Data Fig. 4g). By contrast, the sequential occupancy of PCNA subunits by POLδ, FEN1 and DNA ligase 1 is compatible with one or two PAF15 molecules (Fig. 2i, Extended Data Fig. 4g,h and Supplementary Fig. 4), suggesting that PAF15 anchors PCNA and coordinates orderly factor exchange to prevent steric interference. This model requires future structural validation.

PAF15 doubly monoubiquitinated by UHRF1 at Lys15 and Lys24 mediates DNMT1 recruitment during replication28 and triggers the degradation of PAF15 after DNA damage6. Given that PAF15 is localized preferentially to the lagging strand, we tested whether this modification reflects strand-specific engagement. Acute depletion of LIG1 or RFC1 markedly reduced PAF15 ubiquitination (Extended Data Fig. 5a), and Lig1 knockout (KO) mouse ES cells showed a complete loss of dual mono-ubiquitination (Extended Data Fig. 5b), confirming that both PAF15 and its ubiquitination are functionally coupled to lagging-strand synthesis. As previously shown, PAF15 ubiquitination is essential for DNMT1 loading28; accordingly, loss of this modification in UHRF1-depleted or *PAF15-*KO cells impaired DNMT1 recruitment during S phase, especially under treatment with decitabine (also known as 5-aza-2′-deoxycytidine) (Extended Data Fig. 5c). Conversely, loss of DNMT1 destabilized chromatin-bound PAF15, whereas depletion of UHRF1 did not affect the loading of either PAF15 or PCNA (Extended Data Fig. 5d). These findings suggest that PAF15 has a mono-ubiquitination-independent role in its association with PCNA, prompting us to investigate how PAF15 itself shapes PCNA dynamics on replicating chromatin.

Notably, *PAF15-*KO cells exhibited reduced levels of chromatin-bound PCNA—a phenotype validated across several cell types and methods of genetic ablation (Fig. 3a, Extended Data Fig. 5e,f and Supplementary Fig. 5a). To assess PCNA stability after rapid loss of PAF15, we C-terminally tagged endogenous PAF15 with AID–GFP. The tagged protein did not fully recapitulate native PAF15, and showed impaired chromatin loading, probably because the green fluorescent protein (GFP) tag disrupts the structure of PAF15 and its interaction with PCNA (Supplementary Fig. 5b). Nevertheless, this resulted in hypomorphic PAF15 cells with markedly reduced PCNA chromatin stability, which was worsened by auxin-induced PAF15 degradation (Supplementary Fig. 5b).

Fig. 3: Loss of PAF15 compromises OkF maturation and mounts non-lethal DNA replication stress.

a, Western blot of the indicated proteins from chromatin-purified extracts of parental and *PAF15-*KO U2OS cells. b, Left, schematic showing the dissociation of OkFs from nascent DNA (EdU) after alkali treatment. Middle, Rand plot from QIBC of parental (grey) and *PAF15-*KO (blue) cells with or without alkali. Right, box plot of CldU mean intensity (ssDNA) in both cell lines after alkali treatment. n > 10,000 cells analysed per condition. c, Top, DNA-fibre labelling protocol with in situ S1-nuclease digestion. Bottom, representative DNA fibres with or without S1 nuclease. Scale bars, 20 µm. Right, quantification of total nascent DNA-tract length in parental and *PAF15-*KO cells with or without S1 digestion (n = 200 fibres). d, Left, QIBC of 53BP1 nuclear bodies in the indicated cells, stratified by cell-cycle phases (n > 5,000 cells per condition). Right, distribution of 53BP1 bodies across G1, S and G2 phases in parental and *PAF15-*KO cells. e, QIBC analysis of mono- or poly-ADP-ribosylated (MAR/PAR) chain formation in *PAF15-*KO cells with ATR inhibition (n > 9,000 cells per condition). S-phase quantification (5,000 cells) is shown. f, Quantification of nascent DNA-tract length in parental and PAF15-KO cells treated with PARP inhibitor (PARPi; 500 nM) and digested with S1 nuclease (n = 200 fibres per condition). g, QIBC of RAD51 foci in parental and *PAF15-*KO cells with or without PARP inhibition (n > 9,000 cells per condition), shown across G1, S and G2 phases. h, Quantification of micronuclei in cells treated with PARPi (500 nM, 24 h), based on 500 nuclei per condition. Data are mean ± s.d.; n = 5 biological replicates. P values calculated by one-way ANOVA with Tukey’s test. Each DNA fibre and QIBC plot shown is representative of at least two independent biological replicates.

We next asked whether the marked loss of chromatin-bound PCNA in PAF15-deficient cells alters replisome architecture. MS analysis of PCNA–TurboID showed that PAF15 deletion selectively disrupts PCNA’s interactions with lagging-strand factors, whereas other replisome-associated contacts remain intact (Extended Data Fig. 5g). Chromatin fractionation confirmed that although PCNA is destabilized, the core replisome is preserved (Fig. 3a). Instead, OkF proteins such as FEN1 and DNA ligase 1 were selectively reduced (Fig. 3a). This was further validated by PLA–QIBC across thousands of S-phase cells, which revealed diminished PCNA–FEN1 and PCNA–DNA ligase 1 interactions (Extended Data Fig. 5h). Ectopic expression of PAF15 increased the levels of chromatin-bound PAF15 and co-enriched PCNA and DNA ligase 1 (Extended Data Fig. 5i), suggesting that PAF15 has a direct role in fostering lagging-strand factors.

Consistent with the notion that PAF15 shapes OkF maturation, single-molecule fibre assays showed that PAF15 loss increased fork speed but caused fork asymmetry (Extended Data Fig. 6a–c), a hallmark of defective OkF maturation38. Such impaired OkF maturation was further evidenced by alkali-sensitive, unligated OkFs (Fig. 3b) and S1-sensitive nascent DNA gaps (Fig. 3c). Moreover, PAF15-deficient cells showed the formation of 53BP1 foci (Fig. 3d and Extended Data Fig. 6d,e) and of micronuclei (Extended Data Fig. 6f), both of which are indicative of replication-associated DNA damage. Under these conditions—both during unperturbed replication and after checkpoint failure—PAF15-deficient cells relied on the non-canonical OkF maturation pathway, as evidenced by a marked increase in S-phase-specific ADP-ribosylation (Fig. 3e). Consequently, under PARP inhibition, loss of PAF15 caused extensive S1-sensitive daughter-strand gaps (Fig. 3f), RAD51-mediated post-replicative repair39 (Fig. 3g) and ultimately lethal replication stress (Fig. 3h and Extended Data Fig. 6g).

Altogether, these results show that PAF15 stabilizes PCNA on chromatin to enable canonical OkF maturation and suppress replication stress. Its loss—or its natural exhaustion during excessive origin firing—disrupts lagging-strand processing, forcing cells to rely on PARP1-dependent repair pathways.

Limited levels of PAF15 sustain PCNA stability

Next, we sought to uncover the mechanistic basis by which PAF15 physically and functionally safeguards chromatin-bound PCNA and regulates lagging-strand processing. PAF15 is an intrinsically disordered protein. Crystallographic and nuclear-magnetic-resonance studies of its extended PIP-box suggest that on binding to PCNA, PAF15 also accesses the DNA-encircling channel of PCNA8,9 (Fig. 4a), although the functional importance of this remains unclear. To investigate, we used AlphaFold modelling40 to analyse full-length PAF15 in complex with the PCNA homotrimer on DNA; this suggested that PAF15 acts as a molecular wedge that completely traverses the PCNA ring, exhibiting interactions that are distinct from those of other PIP-box proteins (Fig. 4a, Extended Data Fig. 6h and Supplementary Fig. 6).

Fig. 4: PAF15 stabilizes PCNA and lagging-strand factors on chromatin, shielding them from unscheduled ATAD5-mediated unloading.

a, Comparison between the PCNA–PAF15 crystal structure (Protein Data Bank (PDB): 6EHT, grey) and the AlphaFold 3-predicted model (pink). One PCNA and one PAF15 molecule are removed for clarity. PAF15 is coloured by per-residue confidence score (predicted local distance difference test; pLDDT). b, APBS electrostatic analysis of the modelled PCNA–PAF15–DNA complex containing three PCNA molecules (one removed), two PAF15 molecules and 19-nucleotide DNA. Each PAF15 contributes four positive charges towards the PCNA inner ring. Side-view cartoons and electrostatic surface maps are shown. c, Western blot of chromatin fractions from PAF15-KO U2OS cells treated with control siRNA (siControl) or ATAD5 siRNA (siATAD5). d, FRAP analysis of PCNA–GFP mobility in U2OS cells depleted of PAF15, ATAD5 or both. Cells received a single bleach pulse followed by real-time recovery imaging. Curves represent mean ± s.d. (n = 10 S-phase cells with similar focal PCNA organization). e, Representative PLA images (PCNA–DNA ligase 1) and QIBC quantification of PLA sum intensity in parental and *PAF15-*KO cells treated with control or ATAD5 siRNA (n > 8,000 cells per condition). Scale bar, 10 µm. f, Schematic of DNA-fibre assay and quantification of nascent DNA-tract lengths in parental and *PAF15-*KO cells treated with control or ATAD5 siRNA and digested with S1 nuclease (n = 200 fibres). g, Schematic of PAF15 function during sequential binding of PCNA to POLδ, FEN1 and DNA ligase 1, protecting lagging-strand factors from ATAD5. h, Western blot of PCNA and PAF15 in chromatin and whole-cell lysate (WCL) from parental or *PAF15-*KO U2OS cells with doxycycline (Dox)-inducible PAF15 expression. i, Relative plating efficiency of PAF15-overexpressing cells (mean ± s.d.; n = 3). WT, wild type. j, Replication fork speed in *PAF15-*KO U2OS cells with induced overexpression of PAF15 WT or PIP motif mutant (PIP*). n = 200 fibres. P values calculated by one-way ANOVA with Tukey’s test. The DNA-fibre and QIBC plots shown are representative of at least two independent biological replicates. The schematic in g was created with BioRender.com. Somyajit, K. (2025) https://BioRender.com/yrkrtqc.

APBS (Adaptive Poisson-Boltzmann Solver) software analysis of the modelled PCNA–PAF15–DNA complex revealed that each PAF15 molecule contributes with positively charged residues to both the inner PCNA ring and its N terminus, which is likely to stabilize its interaction with DNA (Fig. 4b). Given that purified human PCNA exhibits weak affinity for DNA in vitro and high rates of diffusion41, we hypothesized that, owing to its unique interaction within the PCNA ring, PAF15 might enhance PCNA–DNA contact, and that the loss of this could explain the reduced levels of PCNA on chromatin.

Indeed, salt sensitivity assays on unperturbed replicating chromatin revealed that the loss of PAF15 rendered chromatin-bound PCNA more susceptible to increasing ionic strength (Extended Data Fig. 6i), indicating a reduced affinity of PCNA for DNA during replication. PCNA is topologically locked onto DNA after loading by the RFC clamp loader and is actively unloaded only by the ATAD5–RFC complex (Elg1–RFC in yeast)31. Thus, the reduced chromatin retention of PCNA observed in PAF15-deficient cells suggests that, in the absence of PAF15, PCNA becomes increasingly vulnerable to premature unloading by this pathway. ATAD5 depletion not only stabilized PCNA on chromatin in PAF15-deficient cells, but also reinforced our conclusion that PAF15 participates in lagging-strand processing (Fig. 4c and Extended Data Fig. 7a). Under these conditions, all tested OkF processing factors were selectively enriched on chromatin, whereas the leading-strand polymerase POLε1 remained unchanged (Fig. 4c). This was further corroborated by fluorescence recovery after photobleaching (FRAP), which showed that although depletion of PAF15 markedly increased the exchange rate of PCNA and shortened its chromatin residence time, co-depletion of ATAD5 and PAF15 restored PCNA immobility during S phase, mirroring the effect of ATAD5 depletion alone (Fig. 4d). This restoration of PCNA stability was accompanied by the functional recovery of lagging-strand processing and genome integrity, as evidenced by the re-establishment of PCNA–DNA ligase 1 interactions, detected by PLA–QIBC (Fig. 4e), a significant reduction in S1-sensitive daughter-strand gaps (Fig. 4f) and diminished 53BP1 foci in PAF15-deficient cells (Extended Data Fig. 7b). Of note, cells with co-depletion of PAF15 and ATAD5 maintained active DNA synthesis and proceeded into mitosis (Extended Data Fig. 7c), indicating partial preservation of replication dynamics and cell-cycle continuity.

These results establish PAF15 as a crucial lagging-strand-specific factor that maintains PCNA stability on replicating chromatin and safeguards lagging-strand-associated proteins from premature unloading by the ATAD5 complex (Fig. 4g).

PAF15 bind