Solution X-ray scattering combined with computational modeling reveals multiple conformations of covalently bound ubiquitin on PCNA

Abstract

PCNA ubiquitination in response to DNA damage leads to the recruitment of specialized translesion polymerases to the damage locus. This constitutes one of the initial steps in translesion synthesis (TLS)—a critical pathway for cell survival and for maintenance of genome stability. The recent crystal structure of ubiquitinated PCNA (Ub–PCNA) sheds light on the mode of association between the two proteins but also revealed that paradoxically, the ubiquitin surface engaged in PCNA interactions was the same as the surface implicated in translesion polymerase binding. This finding implied a degree of flexibility inherent in the Ub–PCNA complex that would allow it to transition into a conformation competent to bind the TLS polymerase. To address the issue of segmental flexibility, we combined multiscale computational modeling and small angle X-ray scattering. This combined strategy revealed alternative positions for ubiquitin to reside on the surface of the PCNA homotrimer, distinct from the position identified in the crystal structure. Two mutations originally identified in genetic screens and known to interfere with TLS are positioned directly beneath the bound ubiquitin in the alternative models. These computationally derived positions, in an ensemble with the crystallographic and flexible positions, provided the best fit to the solution scattering, indicating that ubiquitin dynamically associated with PCNA and is capable of transitioning between a few discrete sites on the PCNA surface. The finding of new docking sites and the positional equilibrium of PCNA–Ub occurring in solution provide unexpected insight into previously unexplained biological observations.

Classical DNA polymerases—those involved in normal DNA replication and repair—cannot accommodate many types of DNA damage in the template strand. As a result, replication forks stall when encountering DNA damage, and this is one of the major sources of genome instability. One of the principal pathways for overcoming replication blocks is translesion synthesis (TLS) by one of a variety of specialized TLS polymerases (1–4). These TLS synthesis polymerases are recruited to stalled replication forks where they replace the classical polymerase and carry out DNA synthesis across from the DNA damage. DNA polymerase eta (pol η), for example, replicates efficiently and accurately through thymine dimers and 8-oxoguanine lesions (5, 6). Significant insights into how TLS polymerases accommodate DNA damage have come from structural and kinetic studies of these enzymes over the last decade (2, 4).

Arguably the least understood steps of translesion synthesis are the recruitment of the TLS polymerase to stalled replication forks and the subsequent polymerase switch between the stalled classical polymerase and the TLS polymerase. These steps are governed by the monoubiquitylation of proliferating cell nuclear antigen (PCNA). PCNA is monoubiquitylated on Lys-164 by the Rad6-Rad18 complex in response to DNA damage (7, 8). The attachment of ubiquitin provides an additional binding surface for TLS polymerase, most of which possess ubiquitin-binding motifs (9). In addition, studies using in vitro reconstituted reactions have shown that the polymerase switching event requires the ubiquitylation of PCNA (10).

Some insights into the structural and mechanistic basis of TLS polymerase recruitment have come from a recent X-ray crystal structure of PCNA–Ub (11). The ubiquitin moiety interacted with PCNA on the back face of PCNA, in a previously undescribed position for PCNA-binding proteins. Generally, most PCNA-binding proteins are located on the front face, the side where replication would be occurring. The attachment of ubiquitin to the PCNA did not significantly alter the structure of PCNA, arguing against allosteric models of TLS polymerase recruitment. Instead, it suggests that ubiquitylation of PCNA simply provides an additional binding surface for the TLS polymerases. The X-ray crystal structure, however, poses a problem regarding the mechanism of TLS polymerase recruitment. The surface of the ubiquitin that interacts with ubiquitin-binding motif of TLS polymerases and centered on Leu-8, Ile-44, and Val-70 (12) is buried at the ubiquitin–PCNA interface in the X-ray structure of PCNA–Ub (11). Consequently the binding of the TLS polymerase requires that the conformation of PCNA–Ub change to expose this surface.

To understand the dynamics of PCNA–Ub in solution, we combined small angle X-ray scattering (SAXS) with multiscale molecular modeling. We identified previously undescribed positions for ubiquitin on the side of the PCNA ring. We show that a model of PCNA–Ub that is in dynamic equilibrium between side, back, and flexibly extended positions best matches the experimental SAXS data. We argue that this dynamic range of positions is important for PCNA–Ub function in binding and positioning of TLS polymerases for lesion bypass.

Results

To resolve questions arising from the crystal structures of yeast PCNA–Ub regarding how Pol η might access the ubiquitin interface, we used both SAXS and computational analyses to characterize the structure of PCNA–Ub in solution.

Small Angle X-ray Scattering in Solution.

To experimentally test the architecture and flexibility of PCNA–Ub, we collected SAXS data on yeast PCNA–Ub in solution (Fig. 1). SAXS combined with crystal structure restraints and conformational analyses can accurately define flexible conformations and ensembles in solution (13–17). We analyzed PCNA–Ub complexes formed by either split-fusion (11) or by chemical cross-linking with PCNA mutant K164C (18) to ensure that the structural results are consistent regardless of the type of linkage between Ub and PCNA. The experimental scattering curves were practically identical for both the split-fusion and chemically cross-linked complexes and followed the overall shape of a scattering curve calculated from the crystal structure of yeast PCNA–Ub (3L10.pdb) (11). Guinier plots were linear, indicating that the samples were not aggregated. The radius of gyration, R_g, is accurately defined by the SAXS experiment independent of sample concentration and contrast and so provides an important constraint for possible solution structures (13). The R_g values based on the Guinier analyses were 35.6 and 36.8 Å, respectively, resembling the 36.05 Å R_g calculated from the crystal structure.

Fig. 1.

SAXS analysis of PCNA–Ub in solution suggests that ubiquitin is not exclusively oriented in the position determined by crystallography. (A) SAXS curves of split-fusion and cross-linked PCNA–Ub overlaid on the calculated curve for PCNA–Ub trimer from the crystal structure 3L10.pdb. (B) Guinier analyses of SAXS data showing relative linearity of sample in Guinier region, indicating lack of aggregation in sample. (C) Ab initio shape predictions calculated from experimental scattering curves of split-fusion or cross-linked PCNA–Ub suggests only one or two ubiquitin positions are extending away from the PCNA ring.

Ab initio shape predictions from the experimental scattering curves of the split-fusion and cross-linked PCNA–Ub complexes revealed a toroidal shape readily identifiable as PCNA. However, only one protrusion corresponding to the crystallographic position for ubiquitin was determined from the ab initio shape reconstruction of the split-fusion PCNA–Ub (Fig. 1C). The shape prediction for the cross-linked PCNA–Ub had a second protrusion that was located in a position not corresponding to the one in the crystal structure. The χ² fit of the SAXS model to the PCNA–Ub crystal structure produced relatively high values, consistent with significant discrepancies between the experimental and computed scattering profiles (Fig. 1A). In fact, the fit was significantly better for a model of PCNA with only one or two ubiquitins per trimer (χ² = 12.32 for split-fusion and χ² = 5.32 for cross-linked). However, a PCNA homotrimer with a single ubiquitin modification was inconsistent with the PAGE analysis showing that the Ub–PCNA linkages were present in 1∶1 ratio for all three PCNA subunits for both the split and cross-linked protein preparations. Thus we postulated that the ubiquitin must occupy positions different from the crystallographically observed position, either at a different position docked against PCNA or in flexible conformation.

We first tested if a model that allowed flexibility in the linker between PCNA and ubiquitin would better fit the split-fusion PCNA–Ub SAXS data. To objectively examine the implications of the SAXS results, we used the CHARMM program implemented in BILBOMD to generate 2,200 models where the three ubiquitins were allowed to move in solution (19). A minimal ensemble search (MES), a program that enables analysis of flexible proteins, was used to identify three conformations that as an ensemble best fits the scattering data. The fit to the split-fusion PCNA–Ub SAXS data was significantly better with the χ² fit decreasing from 23.8 to 7.53. Interestingly, some of the positions were near the PCNA ring, suggesting that there may be alternative positions for ubiquitin to bind to PCNA.

Computational Docking of Ubiquitin on PCNA.

The MES analysis outcome suggested that it may be worthwhile to examine the available conformational space for the Ub moiety on PCNA in more systematic ways to identify the discrete binding positions. Fortunately, a computational modeling study of the human PCNA–Ub complex, which was already in progress, provided alternative models for the PCNA–Ub complex and these computational results could be directly incorporated into the experimental SAXS investigation. The computational modeling involved a combination of tethered Brownian dynamics (TBD) (20), protein–protein docking with RosettaDock (a component of the ROSETTA++ software package) (21–25), flexible loop modeling with ModLoop (26, 27), and molecular dynamics (Fig. S1).

This multiscale approach is similar to the relaxed complex scheme (28), with the TBD simulations serving to identify protein–protein conformations as likely candidates for the natively bound complex based on electrostatic and shape complementarity. At the first stage of our protocol, TBD simulations were employed to generate a large ensemble of 6,837 electrostatically and geometrically favorable configurations for one ubiquitin to one PCNA homotrimer (Fig. 2 and Fig. S1).The advantage of TBD is that it allows very extensive sampling of the conformational space for Ub–PCNA in the course of a 34-μs simulation. Subsequently we clustered the conformations using a previously reported clustering algorithm (29) implemented in GROMACS (g_cluster) (30) and calculated centroids for 90 clusters. Because in TBD, PCNA and ubiquitin interact as rigid bodies, it was necessary to apply local docking with RosettaDock to allow for packing of the interacting side chains of Ub and PCNA.

Fig. 2.

Tethered Brownian dynamics simulation shows the range of covalently bound Ub positions on the surface of PCNA. The positions of Ub heavy atoms (C,N,S, and O) in 6,837 frames from a 34-μs TBD simulation were binned and displayed relative to PCNA as a 3D histogram.

With Rosetta Dock, we input in the 90 cluster centroids from TBD and developed 510 refined models of the PCNA/monoubiquitin complex from a local Monte Carlo search using the Rosetta potential energy function. The function includes van der Waals interactions, ee1 implicit solvation model, hydrogen bonding and Coulombic interactions. We examined the largest 90 clusters. One of the clusters corresponded to the one observed in the PCNA–Ub crystal structure (Fig. S2). From the 90 clusters, we took the top three Rosetta-scoring models, which were located along the plane of the ring in the cavity formed between PCNA subunits. We modeled in the linker (Ub residues 72–76) into the structures obtained from clustering the RosettaDock output (29) using Modloop. We then refined the three models using all-atom explicit solvent molecular dynamics with the program NAMD (31, 32) and the AMBER Parm99SB force field (33).

The three models were positioned distinctly from the ubiquitin in the crystal structure and are nestled in a groove of the PCNA ring directly above the PCNA subunit interface (Fig. 3). The models were well-packed and exhibit a high degree of electrostatic and geometric complementarity, well-known factors in determining productive complexation. Approximately half of the interacting residues are conserved between human and yeast, indicating that the MD-identified positions would be similar in the yeast PCNA–Ub. The buried surface area between PCNA and ubiquitin ranged from 730 Å to 1,038 Å, an area similar in size to the 866 Å buried in the yeast PCNA–Ub structure 3L10.pdb. Importantly, in the low resolution SAXS ab initio shape prediction, the ubiquitin occupying these positions would be hidden along the ring.

Fig. 3.

Three positions of ubiquitin derived from multiscale refinement. The three computationally derived positions (blue, orange, red) are shown relative to PCNA (gray) and to the ubiquitin position determined in the crystallographic studies (black). The computationally derived positions cover up biologically important positions (G178 and E113) identified in yeast mutation studies and the J and P loops that are thought to be important structurally for ubiquitin and sumo positioning.

Minimal Ensemble Search to Determine Positions of Ubiquitin on PCNA in Solution.

Because the binding sites identified by computational methods were conserved between human and yeast, we used the positions identified for human PCNA–Ub to reexamine the yeast PCNA–Ub solution data with the MES program. First, we made a set of 30 unique models for every permutation of yeast PCNA–Ub with three ubiquitins per PCNA homotrimer (Fig. 4). In each of the three positions, ubiquitin was placed in the crystallographic position (x, 3L10.pdb) or one of the three computationally identified positions (a,b,c). To allow for potential flexibility in the system, we next generated all possible models where 1, 2, or 3 ubiquitins per PCNA homotrimer were allowed to move flexibly using the CHARMM molecular dynamics program implemented in BILBOMD (19). We then added the models from each BILBOMD run that best fit the SAXS data to the 30 models where the ubiquitin is only in the discrete positions bound to PCNA to generate a final set of 130 models.

Fig. 4.

An MES ensemble of both discrete and flexible positions of ubiquitin relative to PCNA best fit the experimental SAXS data for split-fusion (green) and cross-linked (blue) PCNA–Ub. (A) Schematic showing MES methodology. One hundred thirty PDB models were generated where the three ubiquitins per PCNA homotrimer were placed at the crystallographic (x) position, the MD-identified positions (a,b,c), or the BILBOMD-generated flexible (f). Ensembles of three models were then compared to the experimental SAXS data in FOXS. (B) The scattering curve of the best MES ensemble fits the experimental scattering data better than the crystal structure 3L10.pdb. (C) P(r) plots showing the good fit of the MES ensemble to the experimental data. (D) The three models that as an ensemble best fit the experimental scattering curve are shown in ribbon models. Relative proportion of each position in x, a/b/c, or f shows that ubiquitin adopts both the crystallographic and computationally determined discrete positions as well as being flexible in solution.

The fit to the experimental data improved significantly (Fig. 4). For the split-fusion PCNA–Ub, the χ² fit went from 23.8 to 4.35 for the best ensemble of three conformations. For the cross-link PCNA–Ub, χ² fit went from 6.8 to 1.89. There is one deviation of the MES curve from the experimental curve at high angle; comparison of the electron pair distribution or P(r) plots show that this difference is negligible for the global shape and is likely due to more high resolution differences with the pdb models, such as ordered water packing, which is difficult to model. Based on the representation of each conformation in solution, we calculated the frequency in which the ubiquitin would have populated the crystallographic, computational, or flexible position. In both the split-fusion and the cross-linked PCNA–Ub SAXS studies, the ubiquitin was about 25 to 30% in the crystallographic position, about 40 to 50% in the computationally determined positions, and about 25% to 30% flexible in solution. As a negative control we included in the MES analyses models where ubiquitin was placed in other positions or unmodified PCNA. These other models were not picked up by the MES program as significant contributors to the SAXS data. Although we cannot exclude the possibility that other discrete conformations are also present, SAXS does provide a picture of the range of motions possible for PCNA–Ub. The SAXS analysis supported a model of segmental flexibility, where ubiquitin can occupy, relative to PCNA, both discrete and flexible positions moving from one to the other in dynamic equilibrium.

Discussion

Mechanisms controlling PCNA interactions with translesion polymerases including the role of posttranslational modifications have general and broad-based implications for cell biology. The attachment of ubiquitin to Lys-164 of PCNA is essential for TLS (7, 8), and this posttranslational modification is believed to be necessary for the recruitment of TLS polymerases to stalled replication forks (9, 34). Structural characterization of the conformation(s) of PCNA–Ub is an important step to understanding the structural and mechanistic basis for TLS polymerase recruitment.

Our analysis of PCNA–Ub in solution expands the recent crystallographic analysis of split-fusion PCNA–Ub that showed a single PCNA-binding position for ubiquitin on the back face of the PCNA ring (11). Combining SAXS with multiscale computational modeling we revealed a previously undescribed model for PCNA–Ub, where the position of the ubiquitin moiety is dynamic and can adopt a variety of positions relative to the PCNA ring. Approximately 25 to 30% of the time the ubiquitin moieties are positioned on the back face of the ring in the position indicated by the X-ray crystal structure. About 25 to 30% of the time the ubiquitin moieties are in a flexible position in which the ubiquitin is attached to Lys-164 of PCNA but is not otherwise interacting with the PCNA. About 40 to 50% of the time the ubiquitin moieties are interacting with the side of the PCNA ring at the subunit–subunit interface, as predicted by the Brownian dynamics, docking and molecular dynamics simulations. This model was obtained using two different analogs of PCNA–Ub: a chemically cross-linked one and a split-fusion one. Thus the dynamics and the orientation of the ubiquitin are not impacted by the chemical nature of the ubiquitin–PCNA linkage. Notably, none of these discrete positions are on the front face of the PCNA ring where most PCNA-binding proteins are located near the interdomain linker (35, 36), and thus all of these positions may allow PCNA–Ub to function as a “tool belt” (11, 37). This means that the TLS polymerase can be recruited to the back face or side of the PCNA ring without disrupting ongoing activity on the front face of the PCNA ring.

Flexibility of the ubiquitin moiety on PCNA–Ub is supported by a recent finding that PCNA chemically monoubiquitylated at Lys-164, Lys-127, Lys-107, or Arg-44 were indistinguishable in recruiting TLS polymerase pol η to exchange with classical pol δ in an in vitro polymerase exchange reaction (18). This observation suggests that ubiquitin moieties attached at distinct positions can achieve whatever conformations are necessary to form functional complexes with pol η and that conformations are not determined by the linker or linker position.

SAXS support of a discrete position of ubiquitin on the back face of PCNA was expected given the X-ray crystal structure of PCNA–Ub. Finding a second set of discrete positions of ubiquitin on the side of the PCNA ring, however, was surprising. The importance of the side position for TLS is supported by experimental results on PCNA mutant proteins. E113G and G178S, two key amino acid substitutions in PCNA that disrupt TLS, are located at the subunit-subunit interface and would directly contact the ubiquitin moiety in the side position (Fig. 3). The E113G blocks the functional interactions between PCNA and TLS polymerases but does not block PCNA ubiquitination in the presence of DNA damage (38). Similarly, the G178S substitution, which is directly across the subunit–subunit interface from E113G, also blocks functional interactions with TLS polymerases (39, 40). Neither amino acid substitution interferes with normal DNA replication and repair. Moreover, the position of ubiquitin on the side of the PCNA ring is reminiscent of the position of the C-terminal domain (the PAD or “little finger” domain) of the Escherichia coli TLS polymerase pol IV bound to the beta sliding clamp (41). Pertinent to the role of PCNA–Ub in other pathways as well, the side positions are adjacent to conserved patches in PCNA, suggesting sites where interacting proteins could bind to both PCNA and ubiquitin (Fig. S3).

In general, the spatial organization of proteins on PCNA and DNA needed for efficient replication and repair is a critical aspect of cell biology. Spatial control is essential for managing of DNA ends while avoiding off pathway activities that would otherwise cause toxicity and mutations. In this context, these results provide insights into how PCNA posttranslational modification provides both specificity and flexibility during translesion synthesis in response to DNA damage. Each of the three PCNA–Ub interaction sites identified here (flexibly, back, side) likely plays a different role in TLS (Fig. 5). In the flexible position, for example, the hydrophobic surface of the ubiquitin is free to bind to the TLS polymerase. As a result, this position likely plays a key role in initially binding to TLS polymerase. Once the polymerase is bound, the ubiquitin moiety could remain in the flexible position or move to either the back or the side position. Of course, when the polymerase is bound, the ubiquitin in the back or side positions would likely have to reorient or rotate slightly to accommodate the polymerase–ubiquitin interaction. Besides the canonical hydrophobic patch of ubiquitin centered on Ile44, TLS polymerase such as Pol η likely binds ubiquitin at other surface patch(es) (42). This mode of interaction may allow Pol η to access the ubiquitin moiety without fully dislodging the ubiquitin from the back or side positions on PCNA. The back position could be the position of the ubiquitin when the PCNA–Ub is functioning as a tool belt. In this position, the TLS polymerase could be held in reserve away from ongoing activity on the front face of the PCNA ring until needed. The side position could be the position of the ubiquitin when the TLS polymerase is engaged in DNA synthesis on the primer template in the front of the PCNA ring. Further understanding of these issues awaits structures of PCNA–Ub bound to TLS polymerases.

Fig. 5.

Possible biological significance of the different ubiquitin positions relative to PCNA. The observation that covalently attached ubiquitin can dynamically occupy different positions on PCNA allows a tool belt model for TLS polymerase binding and functions.

On a more general level, this work illustrates how multidomain or covalently modified proteins can dynamically adopt multiple discrete conformations and have segmental flexibility. These systems are difficult to study structurally and often depend on the ability to capture discrete conformations in different crystals. Here, our study that used multiscale computational analysis to identify discrete positions and SAXS for structural analysis in solution has revealed the range of motions possible for ubiquitin linked to PCNA and a resulting segmental flexibility for the PCNA–Ub complex suitable to regulate recruitment and coordinated actions of TLS polymerase.

Materials and Methods

SAXS Analysis of Split-Fusion and Cross-Linked PCNA–Ub.

Split-fusion and cross-linked yeast PCNA–Ub was purified as before (11, 18). SAXS data of PCNA–Ub were collected at the SIBYLS 12.3.1 beamline at the Advanced Light Source, LBNL (17, 43). Scattering measurements were performed on 20-μL samples at 15 °C loaded into a helium-purged sample chamber, 1.5 m from the Mar165 detector. Prior to data collection, PCNA–Ub were purified by gel filtration. Split-fusion PCNA–Ub was purified on a 24 mL Superdex200 column equilibrated in 20 mM Tris pH 7.5, 50 mM KGlutamate, 5 mM DTT, and 5% glycerol. Cross-linked PCNA–Ub was purified on a 24 mL Superose6 column (GE Healthcare) equilibrated in 20 mM Tris pH 7.5, 150 mM NaCl, 5% glycerol. Data were collected on both the original gel filtration fractions and samples concentrated approximately 2 × –8× from individual fractions. Fractions prior to the void volume and concentrator eluates were used for buffer subtraction. Sequential exposures (0.5, 0.5, 5, and 0.5 s) were taken at 12 keV. Although scattering of the split-fusion PCNA–Ub showed no differences between sequential exposures, the scattering of the cross-linked PCNA–Ub showed a decrease in slope, indicative that the molecules in solution were becoming smaller. It is likely due to X-ray irradiation breaking the disulfide bond. The first and second 0.5 s exposures overlaid, so the first exposure was assumed to have only minimal damage. The best data, based on signal-to-noise and Guinier, were collected on split-fusion PCNA–Ub (4.1 mg/mL, 5 s exposure) and cross-linked PCNA–Ub (9.5 mg/mL, 0.5 s exposure). Data was analyzed using PRIMUS. The Porod exponent was determined from a linear regression analysis (I vs Q) of the top of the first peak in the Porod–Debye plot (q⁴ ∗ I(q) vs. q⁴) of the scattering data, implemented in OPTIMUS. The FOXS/MES server was used for rapid comparison of experimental scattering curves with individual or ensembles of pdb models (23). We applied BILBOMD (19) to generate models of PCNA–Ub where one, two, or three ubiquitins were moved as rigid bodies relative to the remainder of PCNA–Ub. The other fixed ubiquitins were either in the crystal structure position 1 or one of three positions identified in the computational modeling. The range of R_g was set between 30 and 50 Å. For each set of positions, 2,200 models were generated.

Computational Analyses.

To model the human PCNA–Ub complex we employed a multiscale computational strategy concisely summarized in Fig. S1. Details of the analysis are given in Supplemental Methods. The modeling protocol involved the following consecutive stages: (i) TBD (20); (ii) protein–protein docking with RosettaDock (21–25) (a component of the ROSETTA++ software package); (iii) flexible loop modeling with ModLoop (26, 27); and (iv) molecular dynamics. The results of each intermediate stage were clustered based on pairwise distance rmsd (29, 30) with the centroids of the clusters subsequently used as input for the next stage. The clustering results from TBD (29) were used as a starting point for extensive local docking searches with RosettaDock. The major advantage of TBD is the exhaustive sampling of different orientations of Ub on PCNA allowable by the length of the flexible linker connecting the two proteins (Ub residues 72–76). This advantage is partially offset by the limitation that both Ub and PCNA are modeled as rigid bodies. The limitation is partially relaxed at the second stage of our protocol (local docking) by allowing side chain repacking and optimization. The protein backbone is still kept fixed in order to reduce the computational demands as compared to a fully flexible search. The six-residue linker between Ub and PCNA was omitted at the docking stage. Therefore, in stage three it had to be reintroduced through a loop modeling procedure (Modloop) (25–27). Finally, in the fourth stage (all-atom explicit solvent molecular dynamics), all of the above-mentioned restrictions were removed and a fully flexible conformational search is carried out to produce final refined models from each cluster. These multiple trajectory MD runs (each with approximately 25-ns duration) required the use of extensive supercomputing resources provided by the Teragrid and the Oak Ridge Leadership Computing Facility (OLCF).