Structure of the human clamp loader reveals an autoinhibited conformation of a substrate-bound AAA+ switch

Significance

DNA replication and repair depend on AAA+ ATPase protein complexes called clamp loaders that open and load ring-shaped sliding clamps onto DNA. Using cryogenic electron microscopy, we determined the first structure of the human clamp loader (RFC), which is in an autoinhibited conformation while bound to the sliding clamp (PCNA). We assign this to be a reaction intermediate prior to clamp opening and propose a conformational change necessary for activation, leading to a unique paradigm for the AAA+ ATPase mechanism. We examined RFC’s interaction with a PCNA disease variant, which illuminates how this variant maintains tight interactions with some partners. Finally, mapping of cancer mutations onto RFC’s structure suggests stability as a key factor in proper function and human health.

Abstract

DNA replication requires the sliding clamp, a ring-shaped protein complex that encircles DNA, where it acts as an essential cofactor for DNA polymerases and other proteins. The sliding clamp needs to be opened and installed onto DNA by a clamp loader ATPase of the AAA+ family. The human clamp loader replication factor C (RFC) and sliding clamp proliferating cell nuclear antigen (PCNA) are both essential and play critical roles in several diseases. Despite decades of study, no structure of human RFC has been resolved. Here, we report the structure of human RFC bound to PCNA by cryogenic electron microscopy to an overall resolution of ∼3.4 Å. The active sites of RFC are fully bound to adenosine 5′-triphosphate (ATP) analogs, which is expected to induce opening of the sliding clamp. However, we observe the complex in a conformation before PCNA opening, with the clamp loader ATPase modules forming an overtwisted spiral that is incapable of binding DNA or hydrolyzing ATP. The autoinhibited conformation observed here has many similarities to a previous yeast RFC:PCNA crystal structure, suggesting that eukaryotic clamp loaders adopt a similar autoinhibited state early on in clamp loading. Our results point to a “limited change/induced fit” mechanism in which the clamp first opens, followed by DNA binding, inducing opening of the loader to release autoinhibition. The proposed change from an overtwisted to an active conformation reveals an additional regulatory mechanism for AAA+ ATPases. Finally, our structural analysis of disease mutations leads to a mechanistic explanation for the role of RFC in human health.

DNA replication in all cellular life requires sliding clamps, ring-shaped protein complexes that encircle DNA to topologically link numerous factors to DNA. Sliding clamps are necessary for DNA synthesis, because they increase polymerase processivity and speed by orders of magnitude (1–5). Sliding clamps additionally bind and facilitate the function of many other proteins involved in diverse DNA transactions, such as DNA repair, recombination, and chromatin structure (6). The sliding clamp of eukaryotes, proliferating cell nuclear antigen (PCNA), is critical for human health. PCNA’s central role in controlling many cancer pathways makes it a common cancer marker (7). Recently, the genetic disease PCNA-associated DNA repair disorder (PARD) was shown to be caused by a hypomorphic mutation in PCNA that disrupts partner binding (8, 9).

PCNA’s ring shape necessitates active loading onto DNA by the replication factor C (RFC) sliding clamp loader. Clamp loaders are pentameric ATPase machines that can open the sliding clamp and close it around DNA. Clamp loaders are found in all life, although their composition varies across different kingdoms (10). The primary clamp loader in eukaryotes consists of five distinct proteins, RFC1–5. In humans, RFC plays a role in several diseases, such as cancer (11–13), Warsaw breakage syndrome (14), cerebellar ataxia, neuropathy, and vestibular areflexia syndrome (15), Hutchinson–Gilford progeria syndrome (16), and in the replication of some viruses (17–19). It is unknown whether loading by RFC contributes to PARD disease.

Clamp loaders are members of the AAA+ family of ATPases (ATPases associated with various cellular activities), a large protein family that uses the chemical energy of adenosine 5′-triphosphate (ATP) to generate mechanical force (20). Most AAA+ proteins form hexameric motors that use an undulating spiral staircase mechanism to processively translocate a substrate through the motor pore (21–23). Unlike most other AAA+ proteins, clamp loaders do not use ATP hydrolysis as a force-generation step. Instead, the ATP-bound clamp loader forces the sliding clamp ring to open through binding energy alone (24–26). Subsequent binding of primer–template DNA into the central chamber of the clamp loader activates ATP hydrolysis, which results in clamp closure and ejection of the clamp loader (27–33). The sliding clamp is now loaded at a primer–template junction for use by DNA metabolic enzymes, such as DNA polymerases. Thus, the clamp loader is an ATP-dependent protein-remodeling switch (31).

The pentameric clamp loader structure is broadly conserved. The five subunits are named A through E going counterclockwise around the assembly. Each of the five subunits consists of an N-terminal AAA+ ATPase module, followed by an α-helical “collar” domain that serves to oligomerize the complex (Fig. 1A). The Rossman fold and Lid domains that comprise the AAA+ module contain the catalytic residues for ATPase activity. Although most of the catalytic machinery is used in cis, the B, C, D, and E subunits all contain arginine finger residues that are provided in trans to complete the active site of a neighboring subunit.

Fig. 1.

Human clamp loader (hRFC) composition and function. (A) hRFC consists of five different AAA+ ATPase subunits, named A to E. Each subunit consists of an ATPase module and a collar domain. In T4 phage and eukaryotic clamp loaders, the A subunit has a C-terminal extension called the A′ domain that bridges the gap between the A and the E subunit; accordingly, the A′ domain is critical for access to the inner chamber of the clamp loader. The ATPase module has the active site sandwiched between an N-terminal Rossman fold ATPase domain and the Lid domain. (B) The clamp loading reaction begins with binding of ATP to the clamp loader, followed by clamp binding and opening. The clamp loader:clamp complex binds primer–template DNA, which triggers ATP hydrolysis, clamp closure, and clamp loader ejection.

Structural studies have revealed critical intermediates for the clamp loading mechanism. Early structures of the Escherichia coli clamp loader revealed the general organization of the complex (34, 35). A subsequent structure of a mutated form of the Saccharomyces cerevisiae clamp loader RFC bound to PCNA showed RFC in a collapsed and overtwisted spiral conformation that is bound to a closed PCNA ring (36). This conformation was initially hypothesized to represent an intermediate toward the end of the clamp loading reaction, with PCNA closed around DNA and still bound to RFC prior to ATP hydrolysis. It has also been hypothesized that this conformation is an artifact of the mutation of the arginine fingers that prevents proper assembly (31, 37). The structure of an off-pathway intermediate of the E. coli clamp loader bound to a primer–template junction confirmed the “notched-screwcap” mode of DNA binding (38). Finally, the T4 phage clamp loader was crystallized with sliding clamp and DNA, revealing how ATP hydrolysis is linked to clamp closure (39).

Despite many years of study, several central questions about the clamp loader mechanism remain unanswered. Primarily, the mechanism of clamp opening is still unknown. It has been postulated that clamp opening occurs in a multistep process, with initial clamp loader binding followed by clamp opening (32, 33). However, the structure of this preopening intermediate state remains unknown. Moreover, it remains unknown how disease mutations perturb clamp loader function, as there is currently no structure of the human RFC complex.

Here we describe a cryogenic electron microscopy (cryo-EM) reconstruction of human RFC (hRFC) bound to PCNA. The structure reveals that PCNA is closed, despite all active sites of hRFC being bound to ATP analogs. The spiral of AAA+ modules is constricted, which prevents opening of the clamp and blocks the DNA-binding region in the central chamber of the clamp loader. We propose that this represents an autoinhibited form of the clamp loader that occurs prior to clamp opening. Our work provides a framework for understanding the clamp loader’s mechanism and function in human health.

Results

Structure Determination of the hRFC:PCNA Complex.

We sought to obtain a structure of human RFC bound to PCNA by single-particle cryo-EM. We purified an hRFC construct with a truncation of the A subunit’s N-terminal region (RFC1∆N555, missing residues 1 to 555). The truncated version expresses in E. coli and results in higher yields of active protein than the full-length construct without sacrificing clamp loading activity (SI Appendix, Fig. S1 A–C and ref. 40). This hRFC construct is similar to the Hutchinson–Gilford progeria syndrome variant, where the A subunit in hRFC is proteolytically truncated to a ∼75-kDa C-terminal fragment, removing the first ∼500 residues. As expected (41, 42), our purified hRFC has highest ATPase activity in the presence of both the sliding clamp and primer–template DNA (SI Appendix, Fig. S1D). For the rest of the paper, we refer to this complex as hRFC.

In order to visualize how the clamp loader interacts with the sliding clamp, we formed a complex of hRFC with PCNA and the slowly hydrolyzing ATP analog ATPγS. Brief cross-linking of the complex with bis(sulfosuccinimidyl)suberate (BS3) was necessary to obtain a quality reconstruction (SI Appendix, Fig. S2A). BS3, which cross-links primary amines with a linker arm of ∼11.4 Å, is often used to obtain high-resolution cryo-EM structures of labile complexes (43, 44). BS3 treatment preserves the structure (45) and mitigates particle preferred orientation (SI Appendix, Figs. S3B and S4A). Mass spectrometry reveals that most cross-links are intramolecular and ∼55% of all cross-links are in flexible loops that are not visible in the reconstruction, with few cross-links connecting RFC to PCNA (SI Appendix, Fig. S2 B–C and Dataset S1).

We determined the cryo-EM structure of hRFC:PCNA to 3.4 Å with single-particle cryo-EM using three-dimensional (3D) classification, contrast transfer function (CTF) refinement, and multibody refinement (SI Appendix, Fig. S3 C and D and S4B). Representative images are shown in SI Appendix, Fig. S3A. The local resolution ranges from ∼2.9 to 4.1 Å, with the highest resolution in the inner chamber of hRFC subunits A, B, and C and the lowest resolution in peripheral loops (SI Appendix, Fig. S4C). The resulting reconstruction was of sufficient quality (SI Appendix, Fig. S4D) to unambiguously assign each of the five chains (RFC1 is the A subunit, RFC2 is B, RFC5 is C, RFC4 is D, and RFC3 is E) and to build an atomic model of hRFC (Figs. 1A and 2 A and C). The final model refines to an overall model-to-map correlation coefficient of 0.85 with good stereochemistry (SI Appendix, Table S1).

Fig. 2.

Architecture of hRFC bound to a closed PCNA ring. (A) The human hRFC:PCNA cryo-EM map is segmented and colored to show each subunit. (B) A cartoon presentation highlights that PCNA is closed. (C) Side view of the atomic model of hRFC:PCNA (PDB 6VVO) and the yeast RFC:PCNA crystal structure (PDB ID code 1SXJ). (D) Back view of the hRFC:PCNA cryo-EM map (Electron Microscopy Data Bank entry 21405) showing the interaction sites of the B and C subunit with PCNA. (E) Top view on the interaction sites of the hRFC A, B, and C subunits with PCNA. (F) Cartoon representation of the interaction sites of RFC and PCNA observed in this structure, with the PCNA and RFC AAA+ ATPase modules flattened onto the page. The A and C subunit contact the IDCL, whereas B is bound at the interface of PCNA subunits I and II. Subunits D and E do not contact PCNA.

The hRFC:PCNA Complex Is Closed.

Our reconstruction shows hRFC bound to a closed PCNA ring (Fig. 2 A–C). PCNA is in a planar, undistorted conformation, with only localized conformational changes where the ring is contacted by hRFC (overall C_α rmsd is 0.98 Å; SI Appendix, Fig. S5A). The closed PCNA ring was unexpected because clamp loaders typically open the sliding clamp when bound to ATP or ATP analogs (46).

Only the A, B, and C subunits of hRFC contact the sliding clamp; the D and E subunits are lifted off the PCNA surface (Fig. 2 D–F). The A and C subunits bind PCNA’s interdomain connecting loop (IDCL), the primary binding site for PCNA interaction partners (6). The interaction between the A subunit of hRFC and subunit I of PCNA is the most extensive, burying ∼2,000 Ų of surface area, primarily through hydrophobic interactions. The A subunit interacts with PCNA using a classic PIP-box motif that is similar in sequence and structure to the p21 protein, a high-affinity PCNA partner (47) (SI Appendix, Fig. S5A and S6). This interaction induces a conformational change in the IDCL, which we refer to as “high-affinity” PCNA conformation. The interaction between hRFC-C and the IDCL of PCNA subunit II is less extensive (∼1,200 Ų of surface area buried) and does not induce the “high-affinity” PCNA conformation (SI Appendix, Fig. S5A). Similar to hRFC-A, hRFC-C inserts an aromatic residue into the PIP-box binding cleft of PCNA to serve as the anchor (SI Appendix, Fig. S5B). The RFC-B subunit binds the top of the interface between PCNA subunits I and II (Fig. 2 D–F). The hRFC-B:PCNA interaction is much less substantial than the other two sites, with only ∼700 Ų of buried surface area, primarily mediated by salt bridges. Therefore, the three contacts are not equal, with the hRFC-A interactions the strongest and the hRFC-B the weakest.

The AAA+ Modules Form an Inactive, Asymmetric Spiral.

All five subunits contain a nucleotide bound at the interfaces between hRFC subunits. The map is most consistent with ATPγS bound in the A, B, C, and D subunits; the density in the E subunit is consistent with adenosine 5′-diphosphate (ADP), with no density at the γ-phosphate position (Fig. 3A). The E subunit is inactive due to absence of several catalytic residues and an altered P-loop that prevents binding of ATP (SI Appendix, Fig. S7A). Therefore, the presence of ADP is not due to phosphorothioate hydrolysis in the E site but instead is likely copurified with the complex or from the presence of ADP contaminants typically found in ATPγS preparations. ADP binding in the E subunit is dispensable for the ATPase or replication activity of hRFC (48, 49), yet the yeast and T4 phage clamp loaders also retain ADP in the inactive E subunit (36, 39). Therefore, we hypothesize that ADP binding is a conserved function of the E subunit and likely plays a more structural role to stabilize this subunit. Related AAA+ machines such as the origin recognition complex are thought to use nucleotide binding in a similar role (50).

Fig. 3.

Architecture of the AAA-ATP domains and nucleotide binding. (A)Top view on the AAA+ domains of hRFC. All active sites (A, B, C, and D subunits) contain ATPγS. ADP is bound in the inactive E subunit. The nucleotide density at each site is shown in orange mesh. (B) Detailed view of all four active sites. The arginine fingers are distant in the B, C, and D ATPase sites, rendering them inactive. (C) Side and top views of the AAA+ spirals of the DNA-bound T4 phage clamp loader (PDB ID code 3U60, Top) and hRFC. The rotation axes that relate the A to B, the B to C, the C to D, and the D to E subunits are shown in red, green, blue, and magenta, respectively. The rotation axes of T4:DNA complex are coincident with each other and the central axis of DNA; in contrast, the axes of hRFC are severely skewed. (D) The AAA+ spiral of hRFC is overtwisted relative to the active, DNA-bound form.

The B, C, and D active sites are disrupted due to loss of interactions with the trans-acting arginine fingers (Fig. 3B). The B–C interface is the least disrupted but is still too expanded for the arginine finger to contact the γ-phosphorothioate (∼11 Å for Arg168 of hRFC-C). The D subunit is swung out, disrupting arginine finger interactions at both the C and D active sites (∼17 Å for Arg193 of hRFC-D; ∼8 Å for Arg178 of hRFC-E) (Fig. 3B). The interface between the A and B subunits is tightest and similar to that found in active conformations of other clamp loaders (38, 39) (Fig. 3 A and B). However, this active site is unnecessary for clamp loading in all loaders tested (37, 51, 52), so the functional relevance of this conformation is unclear. Therefore, all ATPase sites required for clamp loading activity are in an inactive conformation.

The hRFC AAA+ spiral is asymmetric and overtwisted, which sterically hinders DNA binding (Fig. 4A). The axes of rotation relating adjacent AAA+ domains are not coincident with each other, indicating that the spiral in hRFC lacks helical symmetry (Fig. 3C). In contrast, the structures of T4 and E. coli clamp loaders bound to DNA show a symmetric spiral of AAA+ domains around DNA (Fig. 3C) (38, 39). The hRFC spiral observed here has a reduced helical radius, although the pitch is similar to that of the DNA-bound spirals. The D and E subunits of hRFC are particularly overtwisted such that they fill the DNA binding region (Figs. 3D and 4A). Moreover, an intramolecular interaction between the A′ domain and the A subunit AAA+ module prevents DNA access to the central chamber of hRFC. The interaction between the two domains is rather weak with little buried surface area (∼250 Ų) and is therefore unlikely to be the main driving force for overtwisting the AAA+ spiral. The overtwisted state of the AAA+ spiral has important ramifications for how hRFC opens PCNA and binds DNA (Discussion).

Fig. 4.

The E-plug blocks the DNA binding chamber. (A) The hRFC AAA+ spiral (top view) is incompatible with DNA binding, primarily through clashes with the D and E subunits. DNA is superposed from the structure of T4 phage clamp loader bound to DNA (PDB ID code 3U60). (B) The “E-plug” is a β-hairpin that extends into the DNA binding chamber. The tip of the E-plug contains conserved basic residues.

The conformation of the hRFC:PCNA complex we observe here is very similar to the crystal structure of the mutated yeast RFC:PCNA complex (Fig. 2C and SI Appendix, Fig. S7 B and C) (36). Both structures contain a closed PCNA ring contacted by only the A, B, and C RFC subunits, with similar buried surface between the two proteins (4,600 vs. 5,200 Ų buried surface area for human and yeast RFC:PCNA complexes, respectively). The AAA+ spirals of hRFC and yRFC adopt similar conformations, with only a small difference in position of the D subunit (SI Appendix, Fig. S7C).

The E-Plug Region Fills the DNA Binding Chamber.

We observe a β-hairpin of the hRFC E subunit that blocks the DNA binding region (Fig. 4 A and B). This β-hairpin extends the β-sheet in the Rossman fold domain of the AAA+ module, forming a mixed parallel/antiparallel β-sheet. To our knowledge, this type of topology is not found in other AAA+ proteins. We call this β-hairpin the “E-plug” because it completely blocks the DNA binding region by interacting with the Rossman fold of the A subunit and possibly the B subunit (Fig. 4B and SI Appendix, Fig. S8B). The E-plug was not modeled in the yeast RFC crystal structure due to weak density (36). However, sequence alignments reveal that the E-plug feature is conserved in RFC-E subunits throughout eukaryotes (SI Appendix, Fig. S8A). Therefore, in the observed conformation, the E-plug is playing an inhibitory role by blocking the gate for DNA access.

The tip of E-plug β-hairpin is particularly critical, as it latches the E subunit onto the AAA+ domain of the A subunit. Metazoans have conserved basic residues (Lys79 to 81), and hRFC has two phosphorylation site (Thr75 and Thr76) near the tip of the β-hairpin (Fig. 4B and SI Appendix, Fig. S8A) (53). Phosphorylation of Thr76 is predicted to be important for hRFC function based on a machine-learning approach (53). We hypothesized that the three basic residues are important for maintaining the autoinhibited, closed state. To investigate the role of these residues in RFC function, we constructed two hRFC variants: one in which the three lysines at the E-plug tip were simultaneously mutated to alanine (K79A, K80A, K81A, or the 3K → 3A mutant) and the phosphomimetic mutant Thr76Asp (T76D). We find that the ATPase activity of hRFC-T76D is nearly the same as wild type (WT)-hRFC and hRFC-3K → 3A exhibits a modest reduction (SI Appendix, Fig. S8C). Furthermore, the DNA dependence of ATPase activity reveals that both variants bind DNA with equivalent affinity as WT-hRFC (SI Appendix, Fig. S8 C and D). These results indicate that the residues at the E-plug tip are not critical for DNA binding affinity and potentially only play a limited role in controlling ATPase activity. Therefore, although the E-plug must move to open the gate for DNA access, the role of the conserved tip residues remain unclear.

Investigating Disease Mutations in hRFC and PCNA.

To investigate how RFC mutations commonly found in cancer could affect loading of PCNA, we mapped somatic cell cancer mutations onto the hRFC:PCNA structure using the Catalogue of Somatic Mutations in Cancer (COSMIC) database (Fig. 5 and SI Appendix, Fig. S9 A and B). To enrich for putative driver mutations, we focused on the most common mutation sites (analyzed sites have missense and nonsense mutations from three or more patient isolates; Dataset S2). Furthermore, we introduced each missense mutation in silico to analyze the molecular consequence of each mutation (SI Appendix, Fig.S10) and used the computational tool Rhapsody to predict their pathogenicity (54).

Fig. 5.

hRFC in disease. (A) Cancer mutations mapped onto primary sequence of hRFC. Triangle size is scaled by the number of hits at that site and colored according to the mutation type. Gray regions are not visible in the cryo-EM structure. (B) hRFC ATPase activity dependence on WT-PCNA and S228I-PCNA. S228I-PCNA has a minor effect on the ATPase hydrolysis rate of hRFC. (C) Rate of hRFC ATP hydrolysis as a function of WT-PCNA and S228I-PCNA concentration. S228I-PCNA has a mild enhancement in hRFC affinity.

Rhapsody predicts that each of the five RFC subunits contains at least one deleterious missense mutation (SI Appendix, Fig. S9 B and C). Surprisingly, cancer mutations are significantly more prevalent in the collar domain than in the AAA+ spiral or the sliding clamp (P = 0.006) (Fig. 5A and SI Appendix, Fig. S9A). Additionally, the unstructured N-terminal and C-terminal regions of the A subunit have particularly low cancer mutation hit rates, suggesting that mutations in these regions are less important for driving cancer. Taken together, we find that the collar region is a hot spot for cancer mutations.

Our structure also sheds light on a rare genetic disease, PARD, which is caused by a hypomorphic mutation in PCNA that converts serine 228 to isoleucine (9). We had previously shown that the S228I mutation deforms the IDCL of PCNA, thereby weakening the affinity of some partners (8). PCNA loading is the primary means of regulating PCNA function, which raises a critical question: Does the S228I mutation perturb PCNA loading? Structural comparisons suggest that the interaction between hRFC and S228I-PCNA is disfavored (SI Appendix, Fig. S5B). To test this hypothesis, we measured hRFC ATPase activity as a function of PCNA concentration, which gives an estimate of the affinity for PCNA (Fig. 5 B and C). We find that the S228I-PCNA mutation has no effect on maximal activity, (V_{max, WT} = 0.16 ± 0.01 µM/s, V_{max, S228I} = 0.15 ± 0.01 µM/s) and only a mild stimulatory effect on affinity (apparent K_d of 200 ± 28 nM and 120 ± 12 nM for WT and S228I, respectively). In contrast to expectations, these results suggest that the disease mutation causes no substantial defect on PCNA loading.

Discussion

Implications for Clamp Loading.

We were expecting to obtain a structure of hRFC bound to an open sliding clamp, because previous fluorescence resonance energy transfer (FRET) experiments indicate that PCNA is rapidly opened after initial binding to RFC (40, 55, 56). However, we did not observe a significant number of particles with PCNA in the open form. Subclassification revealed two-dimenstional (2D) class averages that likely represent RFC:PCNA in an open state (SI Appendix, Fig. S3B), but a lack of side views prevented us from obtaining a high-resolution reconstruction. Furthermore, our cross-linking experiments indicate cross-linking between Lys144 of the RFC-E subunit and Lys254 of PCNA. The two residues are buried in the closed hRFC:PCNA conformation. Based on comparison with the open state of the T4 loader:clamp complex (39), interaction between those two residues probably occurs once PCNA is opened and therefore likely indicates the presence of the open hRFC:PCNA complex. However, we cannot rule out that this cross-link occurs in an intermolecular interaction between two different hRFC:PCNA complexes. Taken together, the open hRFC:PCNA complex is likely populated in solution but not suitable for high-resolution structure determination. We hypothesize that the open hRFC:PCNA complex interacts with the air–water interface, likely through exposed hydrophobic residues in the open clamp, which may result in aggregation or denaturation (57, 58).

The conformation of the hRFC:PCNA complex captured here is very similar to the crystal structure of a mutated form of the yeast RFC:PCNA complex (36). This similarity is surprising, as it had been presumed that mutation of the arginine finger residues caused this “overtwisted” state (31, 59). The structural similarity observed here using a functional version of the hRFC complex suggests that this state is actually well-populated in solution and is conserved across ∼1 billion y of evolution since the S. cerevisiae and Homo sapiens lineages diverged (60). If so, what intermediate in clamp loading does it represent?

RFC initially binds PCNA without ring opening (56, 61); we propose that the conformation captured here by cryo-EM is the first-encounter complex between RFC and PCNA (Fig. 6). All four of the ATPase sites are occupied by ATP analogs, indicating that no ATP hydrolysis is necessary for formation of this state. Because ATP hydrolysis occurs at the end of clamp loading (10), our structure must represent an encounter complex at the beginning of the clamp loading cycle, before clamp opening.

Fig. 6.

Model for PCNA opening. In the limited change model, PCNA is opened without RFC undergoing a large conformational change. Contact with primer–template DNA induces a widening of RFC’s spiral to allow for binding into RFC’s central chamber. Alternatively, in the crab-claw model, RFC undergoes a large conformational change concurrent with clamp opening to open up the DNA binding chamber.

Our structure indicates that hRFC can adopt an autoinhibited conformation. The AAA+ spiral is distorted and overtwisted, which disrupts the interfacial contacts necessary for ATPase activity and blocks the inner DNA-binding chamber. Access to the DNA-binding region is further blocked by the E-plug, which reaches across the central chamber of the clamp loader to make extensive contacts with the Rossman fold of the A subunit (SI Appendix, Fig. S8B). Typically, the single-stranded region of template DNA extrudes from the gate between the A′ domain and the AAA+ domain (38, 39). This interface is closed in our structure. Therefore, this state can neither bind DNA nor hydrolyze ATP. Only binding of both clamp and DNA places the clamp loader into an active ATPase conformation (39). This autoinhibited conformation may limit wasteful hydrolysis of ATP by maintaining this inactive conformation.

Clamps across the domains of life display different clamp-opening dynamics. While the T4 bacteriophage clamp is open in solution, PCNA and the E. coli clamp only open upon binding to the clamp loader (37, 62, 63). The E. coli clamp loader and RFC are both reported to form transient encounter complexes with the closed clamp prior to clamp opening, with fivefold slower clamp opening rates for PCNA than for the E. coli clamp (56). The more prominent presence of the inactive encounter complex may be characteristic for eukaryotes; this state may facilitate clamp loader regulation through posttranslational modification or binding of accessory factors. PCNA sequestered with RFC in the encounter complex may be easily activated for timely, controlled loading events, while preventing nonspecific loading and overaccumulation on DNA, which causes genomic instability (64). Further research is necessary to test this hypothesis.

We identified the E-plug as a structural element of the RFC-E subunit that blocks the DNA binding chamber. Based on sequence analysis, the E-plug is conserved in all known eukaryotic RFC-E subunits, yet we know of no analogous feature in other AAA+ proteins. The E-plug blocks DNA binding in the conformation observed here, and we hypothesize that the E-plug needs to be retracted from an inhibitory position to allow DNA binding.

We speculated that mutations at the tip of the E-plug will disrupt interaction with the A subunit and therefore stimulate opening of RFC’s overtwisted spiral to allow for DNA binding. However, mutation of these residues did not result in a change of DNA binding affinity. This result could indicate that the tip residues are not important for mediating autoinhibition. Alternatively, the residues could be playing a dual role: inhibiting DNA binding in the overtwisted state but supporting DNA binding in the open state. In support of this hypothesis, the residues at the tip of the E-plug are conserved as positively charged, hinting at interactions with the phosphate backbone of DNA (SI Appendix, Fig. S8 A and B and Fig. 4B). Moreover, mutation of the tip residues resulted in a slight reduction of ATPase activity, indicating a possible role of the E-plug as DNA sensor for ATPase activation (SI Appendix, Fig. S8 C and D). Hence, the E-plug likely has a dual role, where it inhibits DNA binding and ATP hydrolysis in the overtwisted state but acts stimulating once DNA is bound. Structures of RFC:PCNA:DNA ternary complexes will address this issue.

The overtwist in the AAA+ spiral occurs primarily at the D subunit, which deviates from the active, symmetric conformation more than any other subunit (Fig. 3). The placement of the D subunit’s AAA+ module disrupts two different interfaces: C–D and D–E. Therefore, we propose that the D subunit plays a key role in mediating autoinhibition. In support of this hypothesis, kinetic studies indicate that the C and D subunit’s ATPase activity dictates clamp loader function (59). Taken together, the D subunit’s role is particularly “pivotal” for RFC function: The D subunit is an essential component of the RFC machine, and it acts as a pivot point for AAA+ motion.

Based on our structure, we hypothesize that the clamp loader undergoes a conformational change that opens the gate between A′ and the AAA+ module in the A subunit to allow primer–template DNA binding within the RFC inner chamber. Previous FRET experiments of the E. coli clamp loader indicated that addition of ATP and/or the sliding clamp does not trigger opening of the gate region (32). This study led to the limited change model, which suggests that the ATP-bound clamp loader is already in a conformation competent to open the sliding clamp. Based on our data, we now modify this model to include the presence of this autoinhibited state. In this limited change/induced fit model, clamp binding and opening occurs with minimal change in clamp loader conformation. Subsequent binding of primer–template DNA within the open clamp induces a conformational change in the loader that opens the central chamber so that DNA can be productively bound by both loader and clamp (Fig. 6). In support of this hypothesis, molecular dynamics simulations of PCNA opening with yeast RFC suggested that clamp opening is possible with limited conformational change in RFC (65). Another possible mechanism, the crab-claw model, posits that PCNA opening triggers a concomitant conformational change within RFC that opens the central chamber for DNA binding. However, the crab-claw mechanism does not agree with the previous fluorescence studies (32). Future studies will differentiate between the two potential mechanisms for clamp opening.

How does the autoinhibited state we see with RFC compare with those of other AAA+ machines? Recent structural work has proposed that many AAA+ machines are processive motors that function using an undulating spiral staircase mechanism (21–23). However, our work here and elsewhere indicates that clamp loaders use a different mechanism (31, 39). The inactive state reported here is very distinct from the active, DNA-bound conformation (38, 39). The helical radius of the AAA+ spiral must expand to accept DNA, indicating that the primary mechanism of clamp loader activation is controlled by the shape of the AAA+ spiral. Other AAA+ switches such as the origin recognition complex show autoinhibition mechanisms wherein the spiral is disrupted, although the disruption is not due to overtwisting (50). In contrast, in the “spiral staircase” view of processive AAA+ motors, the helical radius is fairly uniform during function (21). The overtwist of the AAA+ spiral may be due to the fact that RFC is pentameric and not hexameric. The lack of the sixth subunit may afford the space for the helical radius to significantly change, particularly in the absence of DNA. This regulatory mechanism may not be readily available for hexameric AAA+ motors because the sixth subunit creates a steric block. We speculate that the distinct mechanism of pentameric clamp loaders affords new modes of regulation.

Implications for Cancer and the Rare Disease PARD.

We identified the collar as a hot spot region for cancer mutations, many that are predicted to be deleterious (Fig. 5A and SI Appendix, Fig. S9 A–C). The collar is critical for clamp loader assembly, so it stands to reason that collar mutations could disrupt oligomerization of hRFC (Dataset S2 and SI Appendix, Fig. S10). We propose that improperly assembled clamp loaders are particularly detrimental to the cell for two reasons: 1) inducing premature unloading (66) and 2) altering the relative populations of hRFC and the alternative clamp loaders, which are important for genome integrity (67).

Our work also reveals insights into PARD, a rare disease caused by a hypomorphic mutation of PCNA that results in misregulation of the sliding clamp (8, 9, 68). Because clamp loading is the primary means of PCNA regulation (6, 31), understanding the mechanistic impacts of the PARD S228I mutation on clamp loading is of prime importance. We expected that loading of S228I-PCNA would be less efficient because our structure predicts that the interaction between hRFC and S228I-PCNA would be disfavored (SI Appendix, Fig. S5B). However, we observe that the S228I mutation does not reduce hRFC ATPase activity or binding affinity (Fig. 5 B and C), indicating that loading of S228I-PCNA is unlikely to be a driver of PARD disease. PARD effects are likely downstream of loading.

Our results provide insight into how S228I-PCNA interacts with partners. S228I-PCNA has a dramatic loss in affinity for partners that bind PCNA with moderate to low affinity, such as FEN1 and RNaseH2 (8, 9, 68). We find that tight-binding partners, such as hRFC and p21, maintain their binding affinity for S228I-PCNA. Therefore, we propose that absolute binding affinity determines whether the partner loses affinity for the disease mutant. In other words, we predict a correlation between the ∆G_binding and ∆∆G_WT-S228I. Furthermore, we find that RFC-A binds PCNA using a PIP-box motif that closely resembles that of p21 (SI Appendix, Fig. S6). Both p21 and hRFC-A have a tyrosine residue at the second conserved aromatic position of the PIP-box (Tyr151 of p21 and Tyr703 of A), unlike most partners that have a phenylalanine at this position. Tyrosine at the second aromatic position of the PIP-box increases affinity (69). Therefore, we hypothesize that the hydroxyl group of Tyr703-A reorients the IDCL of S228I-PCNA into the tight binding configuration, as Tyr151 does for p21 (8, 69).

Materials and Methods

Protein Expression and Purification.

The E-plug mutants (RFC3-3KtoA and RFC3-T76D) were cloned with site-directed mutagenesis of p36-p37-p38-p40-pET-Duet-1 (70). hRFC was overexpressed and purified following the protocols of refs. 40, 71 with minor modifications. p36-p37-p38-p40-pET-Duet-1 and pCDF-1b-RFC140∆N555 plasmids were cotransformed into BL21(DE3) E. coli cells (Millipore). After preculture, transformants were grown in 6 L of prewarmed terrific broth medium supplemented with 50 μg/mL streptomycin and 100 μg/mL ampicillin at 37 °C and induced with isopropyl β-d-1-thiogalactopyranoside at an optical density of 0.8. Protein expression was continued at 18 °C for 15 h.

Cells were harvested by centrifugation at 7,277 × g for 20 min, resuspended in 200 mL of 20 mM Hepes-KOH, pH 7.4 with 200 mM NaCl, and pelleted by centrifugation at 4,000 × g for 20 min. The pellet was resuspended in 15 mL lysis buffer per one-cell optical density. The lysis buffer contained 20 mM Hepes KOH, pH 7.4, 180 mM NaCl, 2 mM EDTA, 5% glycerol (wt/vol), 0.01% Nonidet P-40 (vol/vol), 2 mM dithiothreitol (DTT), and Roche protease inhibitor mix. Cells were lysed using a cell disruptor, pelleted and the supernatant was filtered.

After filtration, the supernatant was applied to a 25 mL HiTrap SP HP column (GE Healthcare). The column was washed with three column volumes of SP column buffer (25 mM Hepes-KOH, pH 7.4, 1,180 mM NaCl, 0.1 mM EDTA, 5% glycerol [wt/vol], 0.01% Nonidet P-40 [vol/vol], and 2 mM DTT) and developed with a seven-column volume linear NaCl gradient (200 to 1,000 mM). Fractions containing all RFC subunits were diluted with 5% glycerol (wt/vol), 0.01% Nonidet P-40 (vol/vol), and 50 mM KPO₄ buffer, pH 7.5, to a salt concentration of 100 mM NaCl and loaded onto a 5-mL Bio-Scale TM Mini CHT Type II column (Bio-Rad) equilibrated with CHT column buffer (5% glycerol [wt/vol], 0.01% Nonidet P-40 [vol/vol], 50 mM KPO₄ buffer, pH 7.5, and 100 mM NaCl). After a 2.5 column volume (CV) wash, the protein was eluted with a stepwise gradient of 140 mM (3 CV), 185 mM (2.5 CV), 230 mM (2.5 CV), 275 mM (2.5 CV), 350 mM (2.5 CV), and 500 mM (2.5 CV) KPO₄ buffer, pH 7.5. Peak fractions of hRFC were pooled and concentrated to 15 to 20 mg/mL into a buffer containing 25 mM Hepes-KOH, pH 7.4, 15% glycerol (wt/vol), 0.01% Nonidet P-40, 300 mM NaCl, and 2 mM DTT for storage. The protein was further purified by gel filtration with a Superose 6 10/300 GL column (GE Healthcare). hPCNA was expressed and purified as described in ref. 8.

ATPase Enzymatic Assays.

hRFC was incubated at room temperature with a master mix (3U/mL pyruvate kinase, 3 U/mL lactate dehydrogenase, 1 mM ATP, 670 μM phosphoenol pyruvate, 170 μM NADH, 50 mM Tris (pH 7.5), 500 μM TCEP, 5 mM MgCl₂, and 200 mM potassium glutamate) with 1 µM PCNA and varying concentrations of annealed oligonucleotides. The annealed DNA has a 10-base 5′ overhanging end. The template strand sequence was 5′-TTT​TTT​TTT​TTA​TGT​ACT​CGT​AGT​GTC​TGC-3′ and the primer strand sequence 5′-GCA​GAC​ACT​ACG​AGT​ACA​TA–3′ with a recessed 3′-end. ATPase activity of hRFC was measured in a 96-well format with a PerkinElmer Victor3 1420 multichannel counter using an excitation filter centered at 355 nm, with a band pass of 400 nm to detect NADH oxidized to NAD+. Initial rates were obtained from a linear fit of the initial slopes. Rates were plotted as a function of primer–template DNA concentration and the data were fit using a hyperbolic equation:

$$v=V_{max}\ast \left[DNA\right]/\left(K_{d,app}+\left[DNA\right]\right)+z,$$

where V_max is the maximum enzyme velocity, K_d,app is the substrate concentration needed to get a half-maximum enzyme velocity, and z is the velocity with no DNA (GraphPad Prism).

Cross-Linking and Mass Spectrometry.

hRFC was cross-linked with BS3 (Thermo Scientific Pierce). For cross-linking, hRFC and hPCNA were mixed in a in a ratio of 1/1.3, respectively, and buffer-exchanged using an Amicon Ultra 0.5-mL centrifugal concentrator (Millipore) into buffer containing 1 mM TCEP, 200 mM NaCl, 50 mM Hepes-NaOH, pH 7.5, and 4 mM MgCl₂. The protein was diluted to 3 µM and after the addition of 1 mM ATPγS and a wait time of 3 min, 1 mM of BS3 was added for cross-linking. The sample was incubated for 15 min at room temperature and subsequently neutralized with Tris⋅HCl. The cross-linked complex was reduced, alkylated, and loaded onto a sodium dodecyl sulfate polyacrylamide gel electrophoresis gel for enrichment. The gel >150 kDa was excised, destained, and subjected to trypsin digestion. The resulting peptides were extracted and desalted as previously described (72) and analyzed with liquid chromatography mass spectrometry coupled to a ThermoFisher Scientific Q Exactive mass spectrometer operated in data-dependent mode selecting only precursors of 3. The data were searched against the UniProt human database, using Byonic and XlinkX within Proteome Discoverer 2.3.

EM.

Negative-staining EM.

Purified hRFC was diluted to a concentration of 100 nM and applied on carbon-coated 400-mesh grids. After blotting, the grids were washed twice with 50 mM Hepes, pH 7.5, and stained with 1% uranyl acetate. Data were collected on a 120-kV Philips CM-120 microscope fitted with a Gatan Orius SC1000 detector.

Cryo-EM sample preparation.

Quantifoil R 2/2 (first dataset) and quantifoil R 0.6/1 (second dataset, Electron Microscopy Sciences) grids were washed with ethyl acetate and glow discharged using a Pelco easiGlow (Pelco) for 60 s at 25 mA (negative polarity). Three microliters of hRFC was applied to a grid at 10 °C and 95% humidity in a Vitrobot Mark IV (FEI). Samples were blotted at a force of 5 for 5 s after a 2-s wait and plunged into liquid ethane.

Cryo-EM data collection.

Grids of hRFC were imaged on a Titan Krios operated at 300 kV. Images were collected on a K3 Summit detector in superresolution counting mode at a magnification of 81,000×, with a pixel size of 0.53 Å. The data were collected in two sessions using the multihole/multishot strategy with SerialEM (73) and beam-image shift. During the first session, 3,695 micrographs were collected with a target defocus range of −1.2 to −2.3 with a total exposure of 40.3 e-/Å2 per micrograph. During the second session, 7,840 micrographs were collected with a target defocus range of −1.2 to −2.6 with a total exposure of 45.0 e-/Å2 per micrograph.

Data processing.

The Align Frames module in IMOD (74) was used to align micrograph frames with 2× binning, resulting in a pixel size of 1.06 Å per pixel. Initial CTF estimation and particle picking was performed using cisTEM (75, 76). Particles were picked in three different batches with a characteristic radius of 50 Å and a maximum radius of 100 Å. Particles were then extracted with a largest dimension of 170 Å and a box size of 240 pixels and subjected to 2D classification into 50 classes. Particles from classes with well-defined features were extracted for processing in Relion. To extract more particles that represent less-well-defined side views, the rest of the dataset was subjected to one more round of classification, and more particles that resembled hRFC:PCNA complexes were extracted.

For all further processing steps, all particle stacks were combined yielding 2,933,726 putative particles (SI Appendix, Fig. S3). The coordinates and micrographs were imported into Relion 3.0.2 (77) and CTF parameters were reestimated with Gctf1.06 (78). Particles were binned to 2.12 Å per pixel for the first round of 3D classification. As reference for 3D classification, a refined 3D reconstruction of hRFC bound to PCNA from a dataset collected on a 200-kV Talos Arctica equipped with a K3 Summit detector was down-filtered to 50 Å (SI Appendix, Fig. S3C).

The binned particles were classified into three classes. The 561,558 particles contributing to the best 3D class with well-defined features were extracted without binning. A final round of 3D classification with local angular search helped to obtain a more homogeneous particle stack. These particles were refined to a resolution of 3.5 Å. We next performed CTF refinement, which yielded a structure with a reported resolution of 3.4 Å. In this reconstruction, however, areas within PCNA appeared to suffer from flexibility and misalignment. Suspecting that this was due to motions of PCNA relative to hRFC, we next performed multibody refinement (79). The best result was obtained with a mask that included the three PCNA-bound and ATPase domains of subunits A through C as well as PCNA (mask A). Mask B included the collar domain of subunits A through C and complete subunits D and E. After multibody refinement, the density for PCNA and some loops in the A′ domain improved. The reported resolutions for mask A and mask B are ∼3.3 Å and ∼3.4 Å, respectively, by the Fourier shell correlation (FSC) gold-standard 0.143 criterion (and are ∼3.7 Å and ∼3.8 Å by the FSC 0.5 criterion). The reconstructions were individually sharpened with Relion postprocess and a composite map was generated with the “vop max” function in UCSF Chimera. The composite map was used for atomic model building and refinement. The local resolution was estimated with Relion LocalRes.

Model building.

A homology model of the hRFC:PCNA was generated with the structure of yeast RFC (Protein Data Bank [PDB] ID code 1SXJ) using SWISS-MODEL (80) (sequence identity between yeast and human subunits: A subunit 24%, B subunit 47%, C subunit 47%, D subunit 53%, and E subunit 42%). For PCNA, the human PCNA structure (47) was used for rigid body fitting. The different subunits were split into individual globular domains that were fit as rigid bodies into the density using UCSF Chimera (81). The fitted model was adjusted in Coot and real-space-refined in Phenix with simulated annealing with rotamer, Ramachandran and secondary structure restraints (82, 83)). The refined model was readjusted in Coot and further refined using the same settings as listed above, but without simulated annealing. UCSF Chimera and Pymol were used for figure generation (81, 84).

Data Availability.

The reported cryo-EM map and atomic coordinates have been deposited in the Electron Microscopy Data Bank (https://www.ebi.ac.uk/pdbe/emdb/, entry number 21405) and the Protein Data Bank (https://www.rcsb.org/, ID code 6VVO).