Two glycosylase families diffusively scan DNA using a wedge residue to probe for and identify oxidatively damaged bases

Significance

Base excision repair, an evolutionarily conserved process responsible for the repair of most endogenous damage, is initiated by DNA glycosylases. Observation of the motion of single molecules of three bacterial glycosylases in two structural families, Fpg, Nei, and Nth, together with mutational analysis, has demonstrated that both families use a wedge residue to scan DNA for damage. Glycosylases pause during diffusion to interrogate bases and upon encountering a damage stop to evert and excise it. Moreover, we have derived a simple chemomechanical simulation that fits our data and is in agreement with ensemble studies.

Abstract

DNA glycosylases are enzymes that perform the initial steps of base excision repair, the principal repair mechanism that identifies and removes endogenous damages that occur in an organism’s DNA. We characterized the motion of single molecules of three bacterial glycosylases that recognize oxidized bases, Fpg, Nei, and Nth, as they scan for damages on tightropes of λ DNA. We find that all three enzymes use a key “wedge residue” to scan for damage because mutation of this residue to an alanine results in faster diffusion. Moreover, all three enzymes bind longer and diffuse more slowly on DNA that contains the damages they recognize and remove. Using a sliding window approach to measure diffusion constants and a simple chemomechanical simulation, we demonstrate that these enzymes diffuse along DNA, pausing momentarily to interrogate random bases, and when a damaged base is recognized, they stop to evert and excise it.

Oxidative DNA damage is produced endogenously during normal cellular metabolism or exogenously by chemical agents and ionizing radiation (1, 2). Oxidatively induced DNA damage resulting from attack by reactive oxygen species accounts for approximately one-half of all DNA base damages (3). Some oxidative base damages, such as thymine glycol, are blocks to DNA polymerases and thus potentially lethal; however, the majority of oxidative base lesions mispair with noncognate bases and are potentially mutagenic (4). Therefore, damaged bases must be repaired to maintain the cell’s genomic integrity. With substantial in vivo steady-state levels of oxidative damages, alkylation damages, and apurinic/apyrimidinic (AP) sites among the nearly six billion normal bases, how DNA repair enzymes locate these damages in the sea of undamaged bases has been the subject of much speculation.

The DNA repair mechanism responsible for the removal of the majority of endogenous DNA damages is the base excision repair (BER) pathway (4–6). The critical first step in BER is carried out by a DNA glycosylase that, fueled only by thermal energy, locates a damaged base and cleaves the N-glycosyl bond, thus removing the base lesion from the sugar phosphate backbone. Glycosylases are small monomeric proteins that are found in all living organisms and can be separated into different families based on substrate specificity and structural motifs. In Escherichia coli, there are three glycosylases, Nth, Fpg, and Nei, that directly remove oxidized DNA bases, and all three have an associated lyase activity that cleaves the DNA backbone. These glycosylases are members of two structural families, the helix–hairpin–helix (HhH) or Nth superfamily and the helix–two turns–helix (H2TH) or Fpg/Nei family (7–9) (Fig. 1A). Interestingly, the HhH superfamily member, endonuclease III (Nth), and Fpg/Nei family member, endonuclease VIII (Nei), primarily catalyze the removal of oxidized pyrimidines, such as 5,6-dihydroxy-5,6-dihydrothymine (Tg), whereas formamidopyrimidine DNA glycosylase (Fpg) primarily removes oxidized purines, including 8-oxo-7,8-dihydroguanine (8-oxoG) and 2,6-diamino-4-hydroxy-5-formamidopyrimidine (FapyG) (4). Although Fpg and Nei have different substrate specificities, their two domain structures that contain the H2TH as well as the zinc finger motifs are essentially superimposable (9–12). Nth also contains two domains, including a [4Fe-4S] cluster and an HhH motif, which create a deep groove that houses the catalytic residues and accommodates the extrahelical base (13–15). Despite differences and/or similarities in structure–function, upon binding the lesion, all glycosylases induce a bend in the DNA that extrudes the damaged base into the substrate binding pocket and is accompanied by the intrahelical insertion of a set of amino acid residues called the intercalation loop. In E. coli Fpg, these residues are Met74, Arg109, and Phe111 (10, 11), whereas the corresponding residues in E. coli Nei are Gln69, Leu70, and Tyr71 (12, 16). In E. coli Nth, Ile79, Leu81, and Gln41 have been shown to play a similar role in lesion extrusion (15). Although crystal structures have shed light on the glycosylase–DNA interactions upon lesion binding, single-molecule biophysical studies may define the mechanism by which glycosylases locate damaged bases while rapidly diffusing along the DNA (17–19).

Fig. 1.

Fpg, Nei, and Nth wedge residues and the single-molecule assay. (A) Fpg [Protein Data Bank (PDB) ID code 2F5O], Nei (PDB ID code 1K3X), and Nth (PDB ID code 1ORN) with respective wedge residues Phe114, Tyr71, and Leu82 highlighted in red based on B. stearothermophilus Fpg, E. coli Nei, and B. stearothermophilus Nth crystal structures. (B) Intrahelically inserted Fpg Phe114 (PDB ID code 2F5O) inducing minor distortion of the interrogated G:C base pair, shown to the Left of the Phe114. (C) Single-molecule assay is constructed through hydrodynamic flow-mediated DNA elongation of λ DNA “tightropes” between 3-μm silica beads. Hexahistidine-tagged glycosylases are imaged through sandwich conjugation with a biotinylated anti-histidine antibody linked to a streptavidin-coated Qdot [conjugation illustration created using crystal structures for Nth (PDB ID code 1ORN), IgG (PDB ID code 1HZH), and DNA (PDB ID code 3BSE) (15, 57, 58)].

We previously showed (18) that Fpg glycosylase appears to locate damaged bases by inserting a single “wedge” residue into the DNA duplex apparently to test the strength and flexibility of base pairs as well as the sugar pucker (Fig. 1A). Crystal structures of Bacillus stearothermophilus Fpg covalently cross-linked to undamaged DNA (20–22) show Phe114 inserting into the helix of undamaged DNA and buckling the base pair, which causes the 8-oxo carbonyl of 8-oxoG to clash with the DNA backbone. This interaction creates an “extrudogenic” conformation of the DNA backbone, which would not occur with the normal guanine base (20–22). Our work (18) supported this hypothesis by showing that mutation of the E. coli Fpg Phe111 wedge residue to Ala causes a dramatic increase in the overall diffusion constant of the mutant protein on elongated λ DNA compared with its wild-type (WT) counterpart. Here, we show that analogous Ala wedge residue mutations in E. coli Nei, Tyr72, and E. coli Nth, Leu81, also lead to faster diffusive behavior. Thus, intrahelical base interrogation by the wedge residue is most likely a common mechanism for the one-dimensional search by all three glycosylases and thus common to two different structural families. Furthermore, with increasing oxidative damage to the λ DNA substrates, we found a dose-dependent reduction in the mean diffusion constants and a prolonged binding lifetime for all three glycosylases. In addition, on damaged DNA, individual glycosylases were observed switching between faster diffusive behavior and an immobile state, suggesting that glycosylases alternate between rapidly scanning the DNA for damaged bases and stopping at a site once a damaged base is recognized.

Results

The Wedge Residue Modulates the Diffusive Behavior of Fpg, Nei, and Nth.

Recent crystallographic structures (21, 22) and our single-molecule experiments (18) suggest that E. coli Fpg distinguishes between normal and damaged bases by intrahelical insertion of a single Phe111 wedge residue (Fig. 1 A and B). Based on crystallographic structures, E. coli Nei Tyr72 (23) and E. coli Nth Leu81 (15) may serve a similar function. To determine whether E. coli Nei and Nth use these residues for intrahelical base interrogation, we characterized the movement of all three Qdot-labeled WT and mutant glycosylases (Fpg F111A, Nei Y72A, Nth L81A) on λ DNA “tightropes” (Fig. 1C) with high temporal (17 ms) and spatial (15 nm) resolution (Materials and Methods).

Displacement vs. time traces for Nei (Fig. 2A) and its wedge mutant, Nei Y72A (Fig. 2B), which are representative of all glycosylases studied, were characterized by “mean squared displacement” (MSD) analysis (Insets). Linear MSDs are indicative of diffusive motion, the slope of which defines the glycosylase’s overall diffusion constant during a given trajectory (D_traj) (Materials and Methods). Compared with the WT enzymes, D_traj for all three wedge mutant enzymes were increased by an order of magnitude (P < 0.001) (Fig. 2H), whereas the binding lifetimes of the mutants were essentially unchanged compared with WT (Fig. 2H).

Fig. 2.

The diffusive behavior of WT and wedge mutant DNA glycosylases on undamaged λ DNA. Representative displacement vs. time traces for (A) Nei and (B) Nei Y72A. Overall diffusion constants (D_traj) are determined by linear fits to the first 25% of the MSD plots (Insets; error bars indicate SEM). Sliding window diffusion constant D_int for representative traces of (C) Nei and (D) Nei Y72A show the time course of diffusive behavior, showing that the wedge mutant shows overall faster diffusion and an absence of periods of little to no motion. Ordinate tick marks indicate powers of 10. D_int distributions for (E) Fpg, (F) Nei, and (G) Nth WT enzymes (filled symbols) relative to their respective wedge mutants (open symbols) (and see also Fig. S2). (H) Mean overall diffusion constant D_traj of variants (dark bars) of Fpg, Nei, and Nth are significantly faster than the WT versions of the enzymes (light bars; ***P < 0.001). See also Movies S1 and S2.

Interestingly, within a trajectory, glycosylases undergo periods of fast vs. slow diffusive motion (Fig. 2 A and B and Movie S1). To identify these periods, we used a 12-frame (i.e., 800 ms), “sliding window” MSD analysis to characterize a time interval-based diffusion constant (D_int) during a trajectory as shown in Fig. 2 C and D. Based on this analysis, a given glycosylase can exhibit diffusive motions that span three orders of magnitude (i.e., 0.001–0.1 µm²/s), as highlighted by distributions of D_int for all glycosylases (Fig. 2 E–G). For WT Fpg, the D_int distribution is centered at 0.01 µm²/s, whereas the D_int distributions for WT Nei and Nth appear bimodal. For example, WT Nei has a fast population near 0.01 µm²/s and a slower population near 0.001 µm²/s. WT Nth has both fast and slow populations that are shifted to faster diffusive rates. The Fpg, Nei, and Nth wedge mutants display a marked rightward shift toward faster diffusion constants compared with their WT controls (Fig. 2 E–G and Movie S2), which was also observed when characterized by D_traj (Fig. 2H). It is worth noting that the D_int distribution for NeiY72A retains the bimodal appearance seen in WT Nei.

DNA Glycosylases Stop Upon Encountering a DNA Lesion.

To examine the glycosylase motion characteristics on DNA containing damaged bases, λ DNA molecules were treated to produce specific damages recognized by the three glycosylases studied. Fpg primarily removes oxidized purines, whereas Nei and Nth primarily remove oxidized pyrimidines. We generated a substrate for Fpg by treating λ DNA with methylene blue (MB) plus visible light (24–26). Treatment of DNA with MB results not only in the formation of 8-oxoG but primarily its further oxidation products such as spiroiminodihydantoin (27, 28), which are also recognized by Fpg (29). A substrate containing Tg was generated for Nei and Nth using osmium tetroxide (OsO₄) plus heat (30–32) (Materials and Methods). To estimate the number of purine and Tg lesions present in λ DNA, plasmid DNA (pBR322) was treated under the same MB and OsO₄ conditions. The number of damages (n) per pBR322 molecule was determined from the Poisson distribution, n = −ln e, where e is the fraction of supercoiled (form I) compared with circular (form II) and linear (form III) forms of plasmid DNA post glycosylase treatment. From these results (Fig. S1), we estimate that, on average, individual λ DNA molecules treated with low- and high-dose MB have ∼120 and ∼340 damages, respectively (Fig. S1), whereas low- and high-dose OsO₄ treatment resulted in ∼100 and ∼270 damages per λ DNA molecule, respectively (Fig. S1).

Fig. 3A shows a time course, under multiple turnover conditions, of WT Fpg and Fpg F111A glycosylase activity on 8-oxoG embedded in a 35-mer oligodeoxyribonucleotide. Under these conditions, the Fpg wedge mutant glycosylase activity on 8-oxoG is greatly reduced compared with WT Fpg as is the Nth L81A activity on Tg compared with WT Nth (Fig. 3C). Surprisingly, the Nei Y72A wedge mutant showed similar behavior to its WT counterpart (Fig. 3B) on the substrate containing Tg.

Fig. 3.

Gel-based enzymatic activity of WT and variant enzymes. Time courses of product production under multiple-turnover conditions (20 nM substrate, 2 nM enzyme). The substrate for Fpg is 8-oxoguanine lesion opposite C (A), whereas substrates for Nth and Nei contain a thymine glycol lesion opposite A (B and C). Error bars represent SEM.

The presence of damaged bases in λ DNA tightropes resulted in a dose-dependent slowing of D_traj for WT Fpg, Nei, and Nth as the number of damages increased (Fig. 4A, Table S1, and Movie S3). At high dose damage, D_traj for WT glycosylases was reduced by nearly an order of magnitude (P < 0.05) compared with undamaged DNA (Fig. 4A and Table S1). The reduction in D_traj in response to damage is due to the presence of extremely slow diffusive periods within trajectories for all WT glycosylases, as indicated by the emergence of a distinct population of D_int ≤ 0.001 µm²/s within D_int distributions (Fig. S2). It is worth noting that the D_int for this slow diffusing population on damaged DNA is comparable to the apparent diffusion constant (0.0006 ± 0.0002 µm²/s) we determined previously for stationary Qdot-labeled glycosylases (18). Therefore, periods of glycosylase motion with D_int ≤ 0.001 µm²/s will be defined as the enzyme stopping at a given location, presumably at a damaged base. Moreover, the characteristic binding lifetimes of WT glycosylases on damaged DNA increased significantly (P < 0.001) in a dose-dependent manner with increasing damage, going from 1.2 to 3.5 s for Fpg, 1.3 to 3.5 s for Nei, and 1.2 to 2.5 s for Nth (Fig. 4B and Table S1). It is likely that, upon encountering a damage, the WT enzymes are proceeding to catalysis because both the buffer conditions and the enzyme/substrate ratios used in the single-molecule assay are commensurate with the multiple turnover conditions used in our ensemble assays. Moreover, under high damage conditions, we occasionally observe the double-stranded λ DNA breaking apart because substrate lesions that are closely opposed in opposite strands can be cleaved by the lyase activity of the DNA glycosylase and result in double-strand breaks (33–36).

Fig. 4.

Effect of the number of damages in the λ DNA substrate on the diffusion constants and binding lifetimes of Fpg, Nei, and Nth. Undamaged (light gray), low-dose (gray), and high-dose (dark gray) treated λ DNA substrates were used in the single-molecule assay to measure glycosylase diffusion constants. See Fig. S1 for how the number of damages per molecule was determined. (A) Diffusion constants D_traj were calculated from the slope of the first 25% of the MSD plot for WT Fpg, Nei, and Nth and the wedge mutants diffusing on damaged DNA (see Fig. S2 for D_int distributions, Table S1 for specific values and numbers of observations for each condition, and also Movie S3). Ordinate tick marks indicate powers of 10. (B) Characteristic binding lifetimes were calculated from single-exponential fits to binding lifetime histograms for each trajectory that initiates with glycosylase binding and terminates at dissociation from the DNA. Error bars represent SEM. *P < 0.05; ***P < 0.001.

To further dissect the enzyme–substrate interaction, we determined whether or not the enzyme continued its diffusive behavior or dissociated from the DNA after a discernable immobile period on damaged DNA (for example, the period denoted by arrowheads in Fig. 2B). Correlated cleavage studies had shown that the bacterial glycosylases may not dissociate from the DNA after removing a damaged base but continue to scan the DNA until a second damage is located. In our studies at 150 mM KGlu, for 27 trajectories of WT Fpg on damaged DNA that showed both visually distinct immobile and diffusive periods of motion, 64% dissociated from the DNA after stopping (presumably after removal of the damaged base), whereas 36% remained on the DNA and scanned further. These results are similar to the correlated cleavage studies of Fpg acting on oligodeoxyribonucleotides containing 8-oxoG, although the ensemble studies were done at low salt concentrations (37, 38).

To demonstrate that the alteration in glycosylase diffusive behavior on damaged DNA is related to the enzyme’s catalytic activity, we characterized the motion of WT Fpg on high-dose OsO₄-treated λ-DNA compared with its preferred substrate generated by MB treatment (Fig. 5). The oxidative damage produced by OsO₄ treatment (i.e., Tg opposite A) is a poor substrate for Fpg as its catalytic efficiency on this oxidized pyrimidine is 20-fold less than its activity on 8-oxoG and over 100-fold less that of Nei on Tg (39). As expected, the mean overall diffusion constant, D_traj, for WT Fpg on OsO₄-treated λ-DNA (0.002 ± 0.006 μm²/s) was the same as that observed on undamaged DNA (0.003 ± 0.001 μm²/s) (P > 0.4), whereas diffusive motion of Fpg on MB-treated DNA was significantly lower (0.00065 ± 0.00022 μm²/s) (P < 0.001) (Fig. 5A), confirming that the slower D_traj for WT Fpg on MB-treated DNA is indicative of WT Fpg recognizing its appropriate substrate. Although the catalytic activity of Fpg on Tg damage may be low, the enzyme can still recognize Tg; thus, the binding lifetime on OsO₄-treated λ-DNA is prolonged (2.0 ± 0.4 s) (P < 0.001) compared with undamaged DNA (1.2 ± 0.1 s) but not to the same extent as on MB-treated DNA (3.5 ± 0.9 s) (Fig. 5B), which is in keeping with the ensemble studies (39).

Fig. 5.

Mean diffusion constants (D_traj) and binding lifetimes of Fpg on undamaged, methylene blue (MB)-damaged, and OsO₄-damaged λ DNA. Undamaged, MB-damaged, and OsO₄-damaged λ DNA substrates were used in the single-molecule assay to measure the mean diffusion constant (D_traj) and binding lifetime for Fpg. (A) The diffusion constants were calculated from the slope of the first 25% of the MSD plot for each trajectory that initiates with glycosylase binding and terminates at dissociation from the DNA. Ordinate tick marks indicate powers of 10. (B) The characteristic binding lifetime was calculated from single-exponential fits to binding lifetime histograms, where binding events initiate with glycosylase binding and terminate at dissociation from the DNA. Error bars represent SEM. ***P < 0.001; *P < 0.05.

To examine the role of the wedge residue in identifying oxidatively damaged bases, we measured the diffusive behavior of the wedge mutant enzymes on damaged DNA substrates. Although all of the wedge mutants have significantly higher (P < 0.05) D_traj than their WT counterparts on the same substrate (Fig. 4A and Table S1), all three enzymes demonstrate an unexpected decrease in D_traj on damaged DNA. This is most pronounced with Nth L81A, where the decrease in D_traj is clearly dose dependent. As with the WT enzymes on damaged DNA, the slower D_traj for the wedge mutant enzymes is due to the emergence of periods of slow diffusive behavior as seen in the D_int histograms (Fig. S2). However, unlike their WT counterparts, the binding lifetimes of the Fpg F111A and Nth L81A demonstrated no clear response to the damage levels of the substrate (Fig. 4B and Table S1). Interestingly, the binding lifetimes of Nei Y72A showed a dose-dependent increase with increasing damage, from 0.9 s on undamaged DNA to 3.0 s at high damage, almost identical to the behavior of WT Nei (Fig. 4B and Table S1).

Discussion

Given the high rate of oxidative damage that occurs normally in DNA, the glycosylase search must be rapid and efficient to maintain the cell’s genomic stability. Furthermore, all DNA glycosylases studied to date require the damaged base to be extruded extrahelically into the enzyme’s substrate binding pocket. However, everting each base for inspection is thermodynamically untenable (17). Some DNA glycosylases locate damaged bases by capturing randomly extruded bases, in the case of uracil glycosylase, a transiently extruded uracil residue (40). However, transient lesion extrusion of 8-oxoG, which is a substrate for Fpg, is thought to be infrequent (17). Alternatively, we (18) and others (5, 20–22, 40) have postulated that glycosylases may locate damages by recognizing them in situ inside the DNA helix. Here, we characterize the movement of individual Qdot-labeled glycosylases as they diffusively search for oxidatively damaged bases along λ DNA tightropes and show that two structural families of glycosylases rely on a wedge residue, one of three residues that comprise the so-called intercalation loop, to probe for DNA damage. To support this, mutants of Fpg, Nei, and Nth, where the respective wedge residues have been mutated to alanines, exhibit faster diffusive movements on undamaged DNA tightropes, presumably due to an ineffective interrogation process. Molecular-dynamics simulations based on crystallographic data support wedge residue insertion into the DNA helix as part of the glycosylase search mechanism, given that B. stearothermophilus Fpg cross-linked to undamaged DNA structures show the phenylalanine wedge inserted adjacent to the base opposite to the potential lesion (20, 22). Both WT and mutant glycosylases exhibit slower diffusive motion on oxidatively damaged DNA, which we attribute to the enzyme stopping at a damaged base. Crystal structures of Fpg with either 8-oxoG (10, 16) or FapyG (41) show that the lesion is stabilized in the binding pocket by the αF-β9/10 loop. However, FapyG is recognized even when this loop is deleted (42). There are currently no structures of bacterial Nei or Nth with lesions in the substrate binding pocket, but the structure of a viral ortholog of human NEIL1, MvNei, has been solved with Tg or 5-hydroxyuracil in the binding site pocket (43). In these structures, there are no enzyme–DNA interactions that allow discrimination between the lesion and the normal base. Thus, both the crystallographic data and the single-molecule results presented here support the idea that the lesion is recognized in the DNA helix by the wedge residue before base extrusion into the substrate binding pocket of the glycosylase.

Types of Diffusive Motion Exhibited by DNA Glycosylases.

The glycosylase search is powered only by thermal energy; thus, it is not surprising that glycosylase motion on DNA tightropes is diffusive in nature. Interestingly, the diffusive rates for WT Fpg, Nei, and Nth can span three orders of magnitude on undamaged DNA, occurring even within a single DNA–glycosylase interaction (i.e., trajectory) (Fig. 2C). This broad range of diffusive behavior may reflect multiple underlying processes involved in the glycosylase’s search and subsequent catalytic activity. To address this, we take advantage of alterations to glycosylase diffusive behavior in response to the wedge residue mutation and to oxidative DNA damage.

The broad D_int distributions observed for all WT glycosylases on undamaged DNA appear multimodal and described by three regimes, having characteristic D_int that differ by one to two orders of magnitude (i.e., D_int of ∼0.001 μm²/s, ∼0.01 μm²/s, and ∼0.1 μm²/s). For the regime where glycosylases diffuse at D_int ∼ 0.001μm²/s, this apparent motion cannot be distinguished from a glycosylase that is truly stopped at a given location. Our inability to define such stationary enzymes arises from the positional noise introduced by the Brownian motion of the DNA tightrope, the imaging system, and tracking algorithms, giving an apparent diffusion constant of 0.0006 ± 0.0002 µm²/s for a control stationary Qdot (18). Thus, enzymes that truly stop at a location will have D_int ≤ 0.001 μm²/s. A significant portion of the D_int distributions for WT glycosylases on undamaged DNA fall within this regime, and it appears that trajectories are punctuated by the enzyme stopping at sites for at least 800 ms, based on the sliding window MSD analysis (Fig. 2 C and D). Therefore, we cannot rule out the possibility that, on undamaged DNA, the glycosylase stops at preexisting lesions (such as AP sites) or some undefined aspect of the DNA’s physical landscape.

For WT glycosylases with D_int ∼ 0.01 μm²/s, this rate of diffusion is well above our motion detection limits, and thus the enzyme is undergoing true diffusive motion. Given that glycosylases are conjugated to ∼20-nm Qdots, the viscous drag of this label will reduce the diffusive motion of the enzyme on DNA. In fact, Blainey et al. (19) demonstrated that glycosylases conjugated to fluorescent labels with increasing Stokes radii diffused along DNA molecules with diffusion constants that were consistent with the glycosylase tracking the DNA’s helical contour (i.e., rotating along the DNA), most probably in the minor groove where they bind (7). By this analysis, our Qdot labeling strategy gives an apparent one-dimensional diffusion constant of ∼0.05 μm²/s (18) as the glycosylase diffuses along the helical path of the DNA. Therefore, we attribute D_int ∼ 0.01 μm²/s to diffusive motion that has a significant rotational component along the DNA backbone. Finally, for enzymes, most notably Nth L81A and to a lesser extent WT Nth, that can diffuse at D_int ≥ 0.1 μm²/s, this rate is nearly an order of magnitude faster than the predicted free rotational diffusion, implying that the enzyme translates linearly along the DNA at least part of the time. Although globally, Nth has a similar number of electrostatic interactions with DNA as Fpg and Nei, the leucine wedge residue is lacking the aromatic ring present in phenylalanine (Fpg) and tyrosine (Nei) and thus, once inserted, would not form as strong stacking interactions with the adjacent DNA bases (11, 12, 15, 23).

If one or more of these diffusive regimes reflects the glycosylase probing for damaged bases through the use of its wedge residue, then mutating the wedge residue should help identify the diffusive behavior most dependent on wedge residue function. For all three wedge residue mutants, D_int distributions were shifted by at least an order of magnitude to higher rates of diffusion (Fig. 2 E–G). The most notable effect of the wedge residue mutations was the significant reduction in the population of D_int associated with the slowest diffusive regime, i.e., periods where the glycosylase pauses and inserts the WT wedge residue presumably to spot check and interrogate regions of the DNA helix for damage. With the wedge mutants now only inserting an alanine, which results in overall faster diffusive behavior, we propose, as we have previously (18), and as have others using crystallographic (20–22) and stopped-flow kinetics approaches (44–47) that base interrogation relies on the insertion of the wedge residue into the DNA helix.

Linking Glycosylase Motion to the Enzyme’s Biochemical Activity.

An important aspect of glycosylase function is the enzyme’s capacity to link its diffusive motion along DNA to its enzymatic activity, specifically, wedge insertion and base recognition, eversion, and excision. Can alterations to the enzyme’s diffusive behavior observed on DNA tightropes that result from wedge residue mutation and the presence of oxidative damage in the DNA help tie glycosylase diffusive behavior to its underlying catalytic activity? To address this, we developed a simple two-state mechanical model (i.e., diffusive and immobile states) that links the observed motion on DNA tightropes to the enzyme’s underlying catalytic states and kinetics (see SI Materials and Methods for details of stochastic simulation and model results). Specifically, we assumed that, during the bound lifetime of a glycosylase’s encounter with DNA, it searches for damaged bases by free rotational diffusion along the DNA helix at 0.05 μm²/s (18), stopping frequently to interrogate the DNA (k_interrogate > 2,000 s⁻¹) by insertion of the wedge residue (Fig. 6A). If the base is undamaged, the enzyme rapidly (k_−und > 500 s⁻¹) resumes scanning. However, if a damaged base is encountered, which experimentally would occur every ∼280 nm on low-dose and every ∼100 nm on high-dose damaged DNA, then the enzyme will stop at the damaged site. Based on biochemical evidence (44, 45, 48), even though the damaged base is recognized and the enzyme will normally commit to catalysis, there is the potential that the enzyme will resume scanning without catalysis (k_−dam < 1 s⁻¹). If catalysis occurs, then the enzyme may remain at the damage site for the remainder of its bound lifetime (up to 3 s), although we do observe some enzymes to either dissociate from the damage or remain on the DNA and continue to scan (Results). This assumption is based on biochemical data, where catalysis results in a long-lived immobile state having a lifetime greater than 1 s, although Nei is about 1 s (39). Finally, to account for the observation that glycosylases apparently stop on undamaged DNA (i.e., D_int ∼ 0.001 μm²/s) (Fig. 2C), we assumed a basal level of nonspecific lesion sites in undamaged DNA, 1 every 553 nm. Simulated trajectories (Fig. 6B) were then analyzed as were the experimental data (Fig. 6 C and D) for all WT enzymes on undamaged and damaged DNA as well as for the wedge mutants (Fig. 6 E and F and Figs. S3–S8). The model has only three tunable rates associated with wedge residue insertion and interrogation, k_interrogate, and resumption of scanning depending on whether the interrogated base is undamaged, k_−und, or damaged, k_−dam. These rates were optimized to give the best global fit for D_int distributions for each enzyme across all levels of DNA damage (SI Materials and Methods). The model results compare favorably to the experimental data (Fig. 6 D–F and Figs. S3–S8).

Fig. 6.

Kinetic scheme used for chemomechanical simulation and agreement between simulation and experimental D_int distributions. (A) Kinetic scheme showing transitions between the “diffusing state” (Top) and different immobile states as the enzyme encounters, interrogates, and excises damaged bases in DNA. (B) Simulated displacement vs. time trace for a single molecule of Nei. In the simulation, the enzyme recognizes a damaged base at t = 2.2 s and remains immobile for the remainder of the simulation. (C) Sliding window diffusion constant D_int for representative traces show the time course of diffusive behavior. Ordinate tick marks indicate powers of 10. (D) Overall diffusion constants (D_traj) are shown for simulations of Nei and Nei Y72A on undamaged, low-dose damage, and high-dose damage. Agreement is shown between simulated (lines) and experimental (points) D_int distributions for (E) Nei and (F) Nei Y72A on undamaged (light gray), low-dose damage (gray), and high-dose damage (dark gray) substrates. To simulate increasing damage, only the rate at which the enzyme encounters damaged bases was altered. See SI Materials and Methods for the details of the chemomechanical simulation and Figs. S3–S5 for the best-fit model describing the WT glycosylases on undamaged and damaged DNA and Figs. S6–S8 for the wedge mutant glycosylases on undamaged and damaged DNA.

On undamaged DNA, the model predicts a significant peak in the D_int distributions at ∼0.01 μm²/s as observed experimentally (Nei, Fig. 6E; Fpg and Nth, Figs. S3–S8). Based on the model, this D_int reflects the WT glycosylase rotationally scanning the DNA at 0.05 μm²/s but stopping frequently (k_interrogate ≥ 2,000 s⁻¹) for ∼0.2–2 ms (i.e., 1/k_−und) at random sites as the enzyme inserts the wedge residue and interrogates its candidate base. Interestingly, if the enzyme were to interrogate every base or even every G, the model predicts an overall diffusion constant on undamaged DNA of 0.0003 μm²/s, two orders of magnitude slower than we observe experimentally. Although not assumed in our model, it is possible that the enzyme does go on detailed interrogation sprees at random times, leading to the <0.001 μm²/s diffusion constants observed on undamaged DNA.

With regards to damaged base recognition, our single-molecule measurements suggest that the motion of all three WT glycosylases is sensitive to the presence of oxidative damage in the DNA substrate, as evidenced by a dose-dependent decrease in D_traj that is coupled to a dose-dependent increase in binding lifetimes (Fig. 4 A and B and Table S1). Based on our modeling, the slower D_traj with increasing damage (Fig. 6D) is due simply to glycosylases having a higher probability of encountering a damaged base upon wedge residue insertion, and once recognized, the enzyme effectively stops at the site (i.e., model predicted D_int ≤ 0.001 μm²/s) while undergoing catalysis. This accounts for the observed “shoulder” that emerges in the left-hand side of D_int distributions of all WT enzymes (Fig. S2; Nei, Fig. 6E; Fpg and Nth, Figs. S3–S8).

The simulation also provides a platform to interpret the diffusive properties of the wedge residue mutants. Best-fit parameters from the simulations indicate that the wedge mutant enzymes continue to interrogate bases at a high rate, but are quicker to resume scanning upon interrogation of an undamaged base (k_−und ≥ 2,000 s⁻¹). Furthermore, upon interrogation of a damaged base (either nonspecific or induced), wedge mutant enzymes are also prone to release the base and resume scanning (k_−dam ∼ 1–2 s⁻¹), unlike the WT enzymes. These increased release rates are sufficient to predict the 5- to 10-fold increase in the fast diffusive (D_int) population observed for all three wedge mutants (Fig. S2; Nei, Fig. 6F; Fpg and Nth, Figs. S3–S8). In effect, this model predicts the D_int distributions to be closer to the unimpeded D_int of 0.05 μm²/s. The alanine wedge still allows the mutants to retain their ability to “sense” damage, indicated by the reduction in D_traj of all three wedge mutant enzymes in response to damage (Figs. 4A and 6D and Table S1). Initially, this was surprising to us given that the biochemical data indicated that the glycosylase activities of both the Fpg and Nth wedge mutants were substantially reduced (Fig. 3) despite the fact that their catalytic residues were intact. However, prior ensemble studies with the E. coli MutY glycosylase wedge mutant proteins containing different amino acid substitutions showed a number of distinct effects on enzymatic activity as well as on binding, illustrating the multiple roles in the catalytic process undertaken by the wedge amino acid (49). As stated above, we are proposing that the mutant wedge residue still inserts into the DNA helix and may play a role in recognition of a damaged base. This process slows the enzyme’s effective diffusive motion. However, as the stability of the “interrogation complex” is reduced in the wedge mutants (i.e., faster k_−und and k_−dam), the slowing due to these pause events is attenuated, resulting in faster motion for the mutant enzymes. Furthermore, the reduced size of the alanine wedge residue compromises the ability of the wedge mutants to evert a damaged base into the extrahelical active site thus compromising catalysis. In support of this hypothesis, the diffusive motion of Fpg F111A on damaged DNA is slowed but the binding lifetime of Fpg F111A is not affected by damage. Also, a recent structure of the B. stearothermophilus Fpg alanine wedge mutant (50) on damaged DNA containing 8-oxoG shows the alanine wedge residue side chain inserted. However, instead of causing buckling of the target base pair, the alanine’s wedge insertion results in an anti→syn rotation about the glycosidic bond, disrupting hydrogen bonding with the complementary C and 8-oxoG extrusion into the substrate binding pocket of Fpg does not occur. One caveat is that, in the structure, the DNA is already bent, an event that purportedly extrudes the damaged base through the major groove into the active site pocket (44, 45). Like Fpg F111A, Nei Y72A slowed in response to damage also not in a dose-responsive manner (Fig. 4A and Table S1). If Nei Y72A stalls long enough, however, catalysis may ensue because Nei Y72A retains nearly WT catalytic activity (Fig. 3). These events are likely facilitated by the remaining two members of the three intercalation residues being on the same structural loop, which is not true for Fpg and Nth. This idea is further supported by the increased binding lifetime of Nei Y72A in response to damage that is comparable to WT (Fig. 4 and Table S1). Explaining the highly significant dose-dependent reduction in D_traj of Nth L81A is much more difficult. It is clear that the wedge mutant is sensing damage; however, catalysis does not ensue because Nth L81A retains almost no glycosylase activity, and unlike Nei Y72A, which retains glycosylase activity and shows an increase in binding lifetime in response to damage, the binding lifetime of Nth L81A is unaffected by damage. However, despite the variation in the responses of the wedge mutants to damage, the model adequately predicts their D_int distributions; not by preventing wedge residue insertion but by increasing the rate of wedge residue removal following interrogation by as much as fourfold whether the base is damaged or not (Figs. S6–S8).

Conclusions

In summary, by characterizing the diffusive behavior of individual WT and wedge mutant glycosylases on DNA tightropes coupled with kinetic modeling of the enzyme/DNA interaction, we have provided insight into the molecular basis for glycosylase search and base interrogation. This search process must be robust and efficient so that glycosylases can meet the constant challenge presented by oxidative DNA damage. For Fpg, Nei, and Nth, base interrogation and recognition begin by insertion of the wedge residue, and thus damage recognition occurs before base extrusion into the enzyme’s substrate binding pocket in keeping with our previous results (18) as well as recent crystallographic studies (20–22, 51) and ensemble kinetics analysis (44). Although our modeling suggests that glycosylases only “spot check” for damage through wedge residue insertion, diffusive motion is highly redundant, meaning that the glycosylase revisits individual bases many times during a trajectory. Based on D_traj and the lifetime of a single enzyme/DNA for the WT enzymes under our experimental conditions (Fig. 2H), we estimate that the mean length of DNA scanned is 450–600 bp. Given the model’s predicted high rates of wedge residue insertion (∼2,000 s⁻¹) and removal (∼500 s⁻¹), simulations indicate that most (60–75%) of these bases will be interrogated at some point during a single enzyme’s encounter with DNA. In our studies with damage-containing DNA, at most, a single damaged base exists within this 450- to 600-bp length of DNA scanned in one encounter. It is estimated that there are 400 molecules of each of these glycosylases present in an E. coli cell (52). Given this complement and correcting for the physiological absence of the quantum dot label, it would take each glycosylase pool about 4 min to scan almost 90% of both strands of the 4.6 million base pair genome and 10 min to scan almost 100%. We conclude that, by applying the experimental and analytical approaches described here to mammalian DNA glycosylases, further insight into the fundamentals of base excision repair in cells should be forthcoming.

Materials and Methods

Glycosylase Preparation.

C-terminal hexahistidine tag addition was accomplished through ligation of E. coli Fpg, Nei, and Nth cDNA into pET22b. The intrahelical interrogation residue mutants Fpg F111A, Nei Y72A, and Nth L81A were prepared using a site-directed mutagenesis kit (Agilent Technologies). Protein expression and purification were done as previously described (53). The active fraction of each enzyme was measured using our previously described fluorescence-based high-throughput molecular accessibility method using single-molecule buffer conditions (50 mM Tris⋅HCl, pH 8.1, 150 mM KGlu, 1 mM DTT, and 1 mg/mL BSA) (54).

Multiple Turnover Assay.

The glycosylase assays were carried out with 20 nM damage-containing substrate and 2 nM active protein (Nei WT, Nei Y72A, Nth WT, Nth L81A, Fpg, and Fpg F111A) in 50 mM Tris, pH 8.1, 150 mM potassium glutamate, 1 mM DTT, and 1 mg/mL BSA in a total volume of 100 μL at 37 °C. The damage-containing sequence [dTGTCAATAGCAAG(X)GGAGAAGTCAATCGTGAGTCT], where “X” indicates the damaged base (8-oxoG or Tg), was end-labeled with T4 polynucleotide kinase and [γ-³²P]ATP. The labeled oligodeoxyribonucleotide was purified and annealed to its complement to make the 8-oxoG:C or Tg:A substrate. A 10-μL aliquot was removed at 0.25, 0.5, 1, 2, 5, 15, 30, and 60 min, and added to 2 μL of 2 M NaOH. The samples were heated at 95 °C for 5 min, cooled to room temperature, and 10 μL of formamide stop solution [98% (wt/vol) formamide, 10 mM EDTA, 0.1% bromophenol blue, and 0.1% xylene cyanol] was added. The samples were separated on a 12% (wt/vol) polyacrylamide sequencing gel, transferred to Whatman 3MM paper, dried, exposed to a phosphor imager screen, and imaged with an isotope imaging system (Molecular Imaging System; Bio-Rad).

Single-Molecule Substrates.

Single-molecule experiments were performed using λ DNA (New England BioLabs). λ DNA containing Tg or oxidized purine was generated by exposure to osmium tetroxide (OsO₄) plus heat and MB plus visible light, respectively, as previously described (24–26, 30–32). Briefly, 100 μg/mL λ DNA was treated with 0.2 and 1 μg/mL MB and then illuminated for 60 min with one 100-W lamp (Reveal; 100 W; 1,352 Lumens; A-19 Shape; General Electric) that was placed at a distance of 20 cm from the surface of a thinly coated 10-cm plate containing the DNA solution placed on ice. DNA was then ethanol precipitated, dried, and resuspended in TE buffer [10 mM Tris⋅HCl (pH 8.0) and 1 mM EDTA] (26). In addition, 50 μg/mL λ DNA was treated with 0.001 and 0.1 mM OsO₄ and incubated at 60 °C for 5 and 2.5 min, respectively, and ethanol precipitated and resuspended in TE buffer (32). Plasmid relaxation assays were then used to determine the approximate number of damages produced by either MB or OsO₄ treatment to λ DNA. Briefly, plasmid DNA (pBR322) was treated with MB or OsO₄ under the same conditions as λ DNA detailed above, treated with glycosylase in enzyme excess for 30 min at 37 °C, and run on a 1% agarose gel for 30 min. By measuring the relative amounts of supercoiled (form I), circular (form II), and linear (form III) forms of plasmid DNA post glycosylase treatment, the number of damages (n) per pBR322 molecule was determined from the Poisson distribution, n = −ln e, where e is the fraction of form I. Fig. S1 shows the analysis for determining the number of damages per λ DNA molecule.

Single-Molecule Assay.

The DNA “tightrope” single-molecule assay has been previously described and is effective for imaging single nucleotide excision repair enzymes UvrA and UvrB (55) as well as BER enzymes Fpg, Nei, and Nth undergoing one-dimensional diffusion along single elongated λ DNA molecules (18). Briefly, a flow-through microscope slide chamber was designed to allow nonspecific attachment of hydrodynamic flow-elongated λ DNA molecules between poly-l-lysine coated 3-μm silica beads (PolyK-beads) (Polysciences). PolyK-beads were prepared by overnight shaking at room temperature in a solution containing 0.35 μg/μL poly-l-lysine (Sigma-Aldrich). Fifty microliters of PolyK-beads were then washed once in 1 mL of water followed by centrifugation and resuspension in single-molecule elongation buffer [50 mM Tris⋅HCl (pH 8.0), 1 mM DTT, and 150 mM potassium glutamate (KGlu)]. One hundred fifty microliters of the PolyK-bead-containing solution was then added all at once to the flow cell, leaving a monodispersed field of beads. The flow cell was then cleared of weakly attached PolyK-beads through the addition of three 1-mL washes, where ∼200 μL of elongation buffer was flowed back and forth during each 1-mL wash. λ DNA tightropes were elongated by flowing 200 µL of 31.7 pM λ-DNA in elongation buffer [50 mM Tris⋅HCl (pH 8.0), 1 mM DTT, and 150 mM KGlu] into the chamber followed by cycles of 100-µL infusion and withdrawal at a flow rate of 500 µL/min. λ DNA tightropes were stained with 5 nM YOYO-1 dye (Invitrogen) for 2 min in glycosylase buffer [50 mM Tris⋅HCl (pH 8.0), 1 mM DTT, 1 mg/mL BSA, and 150 mM KGlu] and washed with 300 µL to remove unintercalated dye. Glycosylases were imaged as previously described (18). Briefly, streptavidin-coated Qdots (λ = 655 nm; Invitrogen) were conjugated to a biotin-conjugated anti-histidine antibody (Penta-His; Qiagen) by a 15-min incubation on ice at a molar ratio of 1 Qdot to 5 Penta-His antibodies. Glycosylases were then conjugated to the Qdot-Penta-His moiety via the hexahistidine tag while incubating on ice for 15 min. The final molar ratio of 1 Qdot to 5 Penta-His antibodies to 1 active glycosylase was used to maximize the probability of a 1:1 molar ratio of Qdot and glycosylase. It was determined that the effective molar ratio of 5.7 Qdot to 1 active glycosylase yielded a majority of Qdots without glycosylases with the remaining Qdots bound to a single glycosylase.

Last, only glycosylases that were observed to bind to, diffuse along, and then dissociate from single λ DNA molecules were characterized in this study. The majority (∼90%) of glycosylases were observed to bind and release from the DNA within a single image stack. All experiments were done at ambient room temperature.

Image Acquisition.

Using a custom-built total internal reflectance fluorescence microscope (Nikon TE2000 inverted) with through-the-objective (PlanApo 100×, 1.49 N.A.) excitation light from a 488-nm 50-mW argon-ion laser (Spectra-Physics), glycosylase–DNA interactions were observed as previously described (18). Excitation light was adjusted to a subcritical angle, yielding an obliquely angled illumination ray by defocusing the beam at the edge of the objective’s back aperture. Simultaneous dual-color imaging of the YOYO-1–stained λ-DNA and Qdot-labeled glycosylase was accomplished by passing emitted light through a beam splitter (Optical Insights). Split image fields were captured using an intensified CCD camera (XR Mega-S30 running Piper Control, version 2.3.14, software; Standard Photonics) where typically 4,000 images (2 × 2 pixel binning, 117 nm per pixel) were captured at 15–60 frames per s.

Qdot-labeled glycosylases interacting with single λ DNA molecules were tracked both parallel (x axis) and perpendicular (y axis) to the long axis of the DNA using ImageJ, version 1.45 (National Institutes of Health, Bethesda, MD) with the SpotTracker 2D plug-in (56). The microscope spatial resolution was ∼15 nm (x and y axis SD of positional error), which was calculated by tracking immobile Qdots on a glass surface for 1,000 frames. Glycosylase motion along the x axis was attributed to the glycosylases’ one-dimensional search, whereas motion in the y axis was attributed to thermal fluctuations of the taut λ DNA molecules. The y axis positional error of ∼21 nm was calculated from the SD of 20 1,000 frame trajectories of elongated λ DNA molecules, where the mean distance between attachment points was 6.7 μm. Glycosylase movement along the x axis was found to be orders of magnitude greater than the positional error for the y axis, indicating that the thermal fluctuations of elongated λ DNA molecules did not significantly affect measurements of glycosylase motion.

Data Analysis.

Diffusion constants (D) were calculated by characterizing the displacement vs. time trajectories using MSD analysis (18, 55). MSDs were defined as follows:

$$\mathrm{MSD}=2D\left(n\mathrm{\Delta }t\right),$$ [1]

for 1 < n < N/4, where

$$\mathrm{MSD}\left(n\mathrm{\Delta }t\right)=\frac{1}{N-n}{\displaystyle \sum _{i=1}^{N-n}{\left(x_{i+n}-x_i\right)}^2},$$ [2]

where Δt is the time interval between frames for the trajectory (typically 66 ms), and x_i is the position of the glycosylase at time i. We used two different approaches to characterize diffusion constants. A diffusion constant for each complete trajectory, D_traj, was calculated from the slope of the MSD of the first 25% of the MSD vs. nΔt plot when plotted on linear axes, where N is the total number of frames in the trajectory. Diffusion constants over a short interval of time, D_init, were calculated by determining an MSD within a 12-image frame-wide sliding window, typically 800 ms in length. In this approach, multiple D_init for each trajectory are calculated by sliding the window over one frame at a time. As diffusion constants are well described by lognormal distributions (or combinations thereof), all reported mean diffusion coefficients indicate geometric means.

Enzyme binding lifetimes were calculated by fitting lifetime histograms with a single exponential decay of the following form:

$$P\left(t\right)=P_0{e}^{-t/\tau },$$ [3]

where P(t) is the number of trajectories bound at time t, P₀ is the total number of trajectories, and τ is the “characteristic” binding lifetime.

Trajectories for Fpg, Nei, Nth, and Fpg F111A on undamaged DNA are from experiments reported in ref. 18, but have been reanalyzed as indicated above and are included for comparison. Statistical significance was established using a t test on log-transformed data for diffusion constants, and the log-rank test for binding lifetime data.