A New Protein Architecture for Processing Alkylation Damaged DNA: The Crystal Structure of DNA Glycosylase AlkD

Summary

DNA glycosylases safeguard the genome by locating and excising chemically modified bases from DNA. AlkD is a recently discovered bacterial DNA glycosylase that removes positively charged methylpurines from DNA, and was predicted to adopt a protein fold distinct from other DNA repair proteins. The crystal structure of Bacillus cereus AlkD presented here shows that the protein is composed exclusively of helical HEAT-like repeats, which form a solenoid perfectly shaped to accommodate a DNA duplex on the concave surface. Structural analysis of the variant HEAT repeats in AlkD provides a rationale for how this protein scaffolding motif has been modified to bind DNA. We report 7mG excision and DNA binding activities of AlkD mutants, along with a comparison of alkylpurine DNA glycosylase structures. Together, these data provide important insight into the requirements for alkylation repair within DNA and suggest that AlkD utilizes a novel strategy to manipulate DNA in its search for alkylpurine bases.

Introduction

The integrity of DNA is constantly challenged by chemical attack from endogenous metabolites and environmental agents. Chemical modification of DNA nucleobases by alkylation, oxidation, deamination, or hydrolysis produces mutagenic or cytotoxic lesions that can result in heritable disease, cancer, and cell death reviewed in ¹. To safeguard against DNA damage, all organisms have evolved DNA repair mechanisms to eliminate modified bases from the genome. Base excision repair (BER) is the principal pathway by which small modifications to DNA nucleobases are removed in both prokaryotes and eukaryotes. DNA glycosylases initiate the BER pathway by catalyzing the hydrolysis of the C1′-N glycosylic bond to liberate the modified base from the phosphoribose backbone. The resulting abasic site is further processed by AP endonuclease, phosphodiesterase, DNA polymerase, and DNA ligase functions to restore the DNA to an undamaged state.

DNA glycosylases are specific for a particular type of base damage. The mechanism by which these enzymes initially locate their target substrates is believed to proceed by a processive sliding search for destabilized base pairs along DNA ^{2; 3; 4; 5}. DNA glycosylases read out the identity of the lesion by flipping the damaged nucleotide out of the DNA helix and into a nucleobase binding pocket on the protein, the chemical and physical properties of which are complementary to the correct substrate. All glycosylases examined to date utilize a similar strategy for binding DNA and base flipping despite their structural diversity. There are five structural glycosylase superfamilies reviewed in ⁶, represented by i) T4 pyrimidine dimer specific DNA glycosylase EndoV ⁷; ii) uracil DNA glycosylase UDG, ⁸, iii) bacterial 8-oxoguanine DNA glycosylase MutM/FPG ^{9; 10} ; iv) human alkyladenine DNA glycosylase AAG/MNPG ¹¹; and v) the helix-hairpin-helix (HhH) superfamily ^{12; 13}. A broad range of substrate specificities exist within the HhH superfamily, which includes E. coli endonuclease III EndoIII, ¹⁴, E. coli adenine DNA glycosylase MutY, ¹⁵, human 8-oxoguanine DNA glycosylase hOGG1, ⁹, and several alkylpurine specific DNA glycosylases ^{16; 17; 18}.

DNA alkylation damage from endogenous methyl donors, environmental toxins, or from chemotherapeutic agents produces a chemically diverse spectrum of alkylated nucleobases ^{19; 20; 21}. DNA glycosylases remove N3- and N7-substituted alkylpurines, as well as 1,N⁶-ethenoadenine (εA). The variation in substrate specificities of alkylpurine DNA glycosylases from different organisms has been useful in understanding the mechanisms of base selection and removal. E. coli 3-methyladenine (3mA) DNA glycosylase I TAG, ²² and H. pylori methyladenine DNA glycosylase MagIII, ^{23; 24} are highly specific for 3mA. Thermatoga maritima methylpurine DNA glycosylase II MpgII, ²³ excises 7-methylguanine (7mG) in addition to 3mA. At the other extreme, E. coli 3mA DNA glycosylase II (AlkA), S. cerevisiae MAG and human AAG have a broad, yet well-defined specificity toward 3mA, 7mG, εA, hypoxanthine, and various oxidized purines ^{25; 26; 27; 28; 29; 30; 31}.

Several alkylpurine DNA glycosylases have recently been identified in the genomic sequences of gram-positive bacteria and lower eukaryotes Entamoeba histolytica and Dictyostelium discoideum ^{32; 33}. Bacillus cereus AlkC and AlkD excise 3mA and 7mG, but not εA, and thus have an intermediate specificity similar to MpgII ^{23; 32}. These enzymes share no sequence homology with any other glycosylase, and they have been predicted to represent a new structural class of glycosylase enzymes ³³. We present here the crystal structure of B. cereus AlkD at 2.1 Å resolution together with mutational analyses of 7mG excision and DNA binding. AlkD is composed of a tandem array of helical repeats reminiscent of HEAT motifs, which are known to facilitate protein-protein interactions and have not yet been associated with DNA binding or catalytic activity. A comprehensive structural analysis of the variant HEAT repeats in AlkD helps to explain how this protein interaction motif has been adapted to bind DNA. Importantly, the complementarity between AlkD's concave surface and the B-DNA backbone in our docking model raises the possibility that the enzyme uses a different strategy to engage DNA and access the damaged base.

Results

The structure of AlkD

The crystal structure of AlkD from B. cereus was determined using experimental phases from a single-wavelength anomalous dispersion (SAD) experiment from a crystal derivatized with platinum tetrachloroplatinate(II) (Table 1). A crystallographic model that consists of one AlkD molecule (residues 1-225) in the asymmetric unit was built into 2.0-Å Pt-SAD electron density (Figure 1A). To avoid the possibility of structural artefacts resulting from platinum binding to the protein, the original SAD model was refined against native diffraction data extending to 2.1 Å resolution, resulting in a crystallographic residual (R_cryst) of 18.9% and an R_free value of 22.9% (Table 1).

Table 1.

Data collection, phasing and refinement statistics

	Native	K2PtCl4
Data collection
Space group	P43	P43
Cell dimensions
a, b, c (Å)	77.9 77.9 55.1	78.8 78.8 55.3
α, β ,γ (°)	90.0 90.0 90.0	90.0 90.0 90.0
Wavelength	1.0000	1.0718
Resolution (Å)	50-2.08 (2.15-2.08)	50-1.85 (1.92-1.85)
Rsym or Rmerge	0.046 (0.531)	0.052 (0.321)
I / σI	26.9 (3.4)	50.1 (3.1)
Completeness (%)	99.1 (95.3)	97.4 (84.3)
Redundancy	7.1 (6.2)	7.2 (4.8)
Refinement
Resolution (Å)	50-2.10 (2.21-2.10)
No. reflections	18356 (2605)
R work	0.188 (0.283)
R free	0.227 (0.313)
No. atoms
Protein	1893
Ion / Ligand	0
Water	81
B-factors
Protein	52.7
Water	55.2
R.m.s. deviations
Bond lengths (Å)	0.018
Bond angles (°)	1.571

Data in parentheses refer to the highest resolution shells

Figure 1.

Structure of B. cereus AlkD. (A) Stereo-view of helices αG and αF from the refined crystallographic model superimposed on a 2.0-Å Pt-SAD experimental electron density map (contoured at 1σ). (B) AlkD forms a solenoid structure with a left-handed superhelical twist. Residues forming an aromatic cleft on the concave surface are shown as yellow sticks. (C) Solvent accessible surface representation of AlkD in the same orientation as panel B, colored according to electrostatic potential (red negative, blue positive, −7 to +7 k_BT). Potentials in Figures 1 and 5 were calculated with the program DelPhi ⁶⁴. All molecular images were rendered using PyMOL (http://www.pymol.org).

AlkD adopts a single α-helical domain. Twelve of the fourteen helices (αA-αN) pair in an antiparallel fashion to form 6 tandemly repeated α-α motifs (αA/αC, αD/αE, αF/αG, αH/αI, αJ/αK, and αL/αM) (Figures 1B and 3). The helical repeats are stacked into a superhelical solenoid in which helices B, C, E, G, I, K and M form a concave surface with an aromatic cleft at its center (Figure 1B). Residues within this cleft were shown previously to impair base excision activity of AlkD ³³. The concave surface of the protein is strikingly electropositive (Figure 1C), a feature that is distinct from other helical repeat proteins and one that likely facilitates binding to DNA (discussed below). A structural homology search of the PDB using the DALI server ³⁴ indicated that AlkD is most similar in tertiary structure to Enterococcus faecalis EF3068 (PDB ID 2B6C) and B. cereus BC3264 (PDB ID 1T06), two hypothetical proteins of unknown function determined from the Midwest Center for Structural Genomics (Figure S1, Table S1). EF3068 shares 35% sequence identity with AlkD (Figure S2) and was used in a previous study as the basis for a homology model of the Bacillus enzyme ³³. The structures of AlkD and EF3068 are similar in overall fold (r.m.s.d. of 1.38 Å for all backbone atoms), with the most notable differences in helices αA/αC at the N-terminus (Figure S1).

Figure 3.

AlkD is composed of six variant HEAT motifs. (A) Cylinder representation of AlkD with HEAT-like tandem repeats colored independently. Residues lining the putative active site are shown as ball-and-stick. (B) Structures of individual AlkD α-α pairs (AC, DE, FG, HI, JK, LM) are compared to canonical HEAT repeat 10 (XY) from protein phosphatase 2A PR65/A subunit, PDB ID 1B3U ⁴⁷. (C) Structure-based sequence alignment of HEAT motifs in panel B, with AlkD residues lining the aromatic cleft in boldface and Asp113/Arg148 colored red. Positively charged residues contributing to the electropositive concave surface are highlighted yellow. Interdigitating residues important for the relative orientation of paired α-helices are shaded grey.

Putative active site

One striking feature of the AlkD structure is the apparent similarity in the chemical environment of the aromatic cleft to the nucleobase binding pockets of other alkylpurine DNA glycosylases. AAG, AlkA, TAG, and MagIII all feature electron-rich, aromatic pockets that accommodate the extrahelical alkylpurine base through both shape complementarity and π-π stacking interactions ^{18; 35; 36; 37; 38}. With the exception of TAG, which exclusively excises 3mA and 3mG, the alkylpurine binding pockets also feature a conserved acidic residue that is essential for base excision activity.

AlkD contains a cluster of aromatic residues (Trp109, Trp187, Phe179, Phe180, and Tyr27) that form a shallow cleft at the center of the concave surface (Figure 2A). At the back of this cleft, Asp113 and Arg148 form an electrostatic interaction and were observed in the crystal structure to hydrogen bond to several well-ordered water molecules (Figure 2A). The relative positions of these side chains are similar to the aromatic and catalytic Asp238 residues inside the shallow nucleobase binding pocket of AlkA (Figure 2B) ^{16; 37}. Asp113, Arg148 and Trp109 are invariant in sequence among AlkD homologs (Figure S2) and were identified in a previous study to be important for release of methylated bases from N-methyl N-nitrosourea (MNU)-treated genomic DNA and for growth of E. coli tag alkA mutant cells in the presence of methylmethane sulfonate (MMS) ³³. Because wild-type AlkD excises several alkylated nucleobases produced from MNU or MMS treatment ³², we tested the effect of AlkD mutants on the specific excision of 7mG from a defined oligonucleotide substrate. As shown in Figures 2C and 2D, Asp113Asn and Arg148Ala each resulted in a dramatic (100-fold) decrease in the single-turnover rate of 7mG excision relative to the wild-type enzyme (Table 2), indicating that Asp113 and Arg148 play a specific role in 7mG excision by AlkD. These results show that AlkD provides catalytic assistance to liberate 7mG from DNA, and implicate the aromatic cleft as the active site of the enzyme.

Figure 2.

The putative active site of AlkD. (A) Close-up of the concave cleft shows aromatic and charged residues implicated in nucleobase and DNA binding. Water molecules are shown as red spheres and hydrogen bonds are depicted as black dotted lines. (B) The electron-rich active site of AlkA, showing the conserved Asp238 essential for base excision. (C) 7-methylguanine (7mG) excision by AlkD. Denaturing polyacrylamide gel showing the disappearance of 7mGDNA substrate (S) and appearance of alkaline-cleaved abasic-DNA product (P) as a result of reaction with no enzyme, wild-type AlkD, Asp113Asn, and Arg148Ala mutants for increasing amounts of time. (D) Quantitation of the data in panel C showing the inactivity of the Asp113Asn (squares) and Arg148Ala (triangles) mutants as compared to wild-type AlkD (closed circles). Non-enzymatic 7mG hydrolysis is shown as crosses and dotted line curve fit. Error bars represent the standard deviation from the average of three independent experiments.

Table 2.

7-Methylguanine excision activities for wild-type and mutants of AlkD

	kcat (10−3 min−1)a	Relative activity	Rate enhancementb
WT	171.8 ± 11.1	(1.0)	231.6
D113N	1.8 ± 0.5	0.01	2.5
R148A	1.8 ± 0.3	0.01	2.5

^aFirst-order single-turnover rate constants for 7mG excision from a 25mer oligonucleotide duplex containing a 7mG·C base pair. Values represent the averages and standard deviations from three experiments.

^bRate enhancements are based on the fold increase above a non-enzymatic control (k_non = 0.7 × 10⁻³ min⁻¹).

The variant HEAT motif

The list of structurally homologous proteins from the DALI analysis was exclusive to proteins that contain tandem helical repeats (Table S1). Apart from EF3068 and BC3264, AlkD is most similar in structure to the PR65/A subunit of protein phosphatase 2A (PP2A) ^{39; 40} and to the SCF ubiquitin ligase regulatory subunit Cand1 ⁴¹, both of which contain Huntington/Elongation/A subunit/Target of rapamycin (HEAT) repeats ⁴². HEAT motifs are ~45-amino-acid sequences arranged into two antiparallel α-helices, which are packed together by a conserved hydrophobic interface and are tandemly repeated to form superhelical α-structures ⁴². The α-α pairs stack in a parallel arrangement to form a solenoid in which the C-terminal helix of each repeat lines the inside groove of the superhelix. The solenoid structure is typically formed from no less than 14 HEAT motifs and provides a molecular scaffold that facilitates protein-protein interactions. Out of 393 structurally similar proteins identified from the PDB search, only three were eukaryotic nucleic acid-binding proteins, including transcription factor Rcd-1 (Z-score = 5.9) ⁴³ and the RNA-binding proteins Pumilio1 (Z-score = 4.8) ⁴⁴ and Ro autoantigen (Z-score = 3.5) ⁴⁵. Ro contains HEAT repeats that are quite divergent from the canonical motif, while Rcd-1 and Pumilio-1 are comprised entirely of armadillo and pumilio repeats, respectively ^{42; 46}. Thus, AlkD represents the first example of a DNA-binding HEAT protein.

In contrast to the regular, repeating α-α structures found in archetypical HEAT proteins, the AlkD repeats exhibit greater structural variation despite the conservation of hydrophobic residues that stabilize the interface between helices (Figure 3). The most obvious outlier is αA/αC, which has the opposite handedness from the other repeats, as well as an inserted helix αB that contributes considerable surface area to the concave face of the protein (Figure 3A). The r.m.s.d. among the remaining 5 repeats αD/αE-αL/αM in AlkD is 2.48 Å, as compared with 1.76 Å among five consecutive HEAT repeats in PP2A ⁴⁷. AlkD's helices are on average one turn shorter in length and, with the exception of αF/αG, do not contain the intrahelical kink characteristic of HEAT repeats (Figure 3B). Thus, the helical repeats that comprise AlkD represent a structural variant of the canonical HEAT motif and thus give rise to a unique architecture.

The most striking feature of AlkD's variant HEAT repeats is the abundance of basic residues along the C-terminal helix (Figure 3C). The positions and spacing of these basic residues are conserved among the repeats, and consequently, the concave surface of the protein is highly electropositive (Figure 1C). This positive charge along the concave surface is conserved among AlkD orthologs (Figures S1, S2), but is not observed in other helical repeat proteins (Figure S3), which faithfully utilize this surface to bind polypeptides ^{40; 41; 48; 49}. Thus, it seems that AlkD's variant HEAT repeats have evolved this additional electrostatic feature in order to stabilize the negatively charged DNA backbone at the molecular binding interface.

A DNA binding model

Despite their structural divergence, the five existing superfamilies of DNA glycosylases ^{6; 50} use a common mode of binding DNA and gaining access to the lesion via a base flipping mechanism (Figure 4). Structures of these enzymes in complex with DNA show that they all use a positively charged surface to help anchor the DNA against the enzyme. The DNA duplex is highly kinked to allow for extrusion of the damaged base into a sterically constrained active site pocket that lies adjacent to the positive surface (Figure 4). The distorted conformation of the DNA is stabilized by a pair of side chains that intercalate into the DNA helix at the site of the lesion.

Figure 4.

The six structural superfamilies of DNA glycosylases. Representative crystal structures shown are B. cereus AlkD; E. coli 3-methyladenine DNA glycosylase II, AlkA 1DIZ, ³⁷; human methylpurine DNA glycosylase, AAG 1EWN, ³⁶; human uracil-DNA glycosylase, UDG 1EMH, ⁶⁵; Bacillus stearothermophilus 8-oxoguanine DNA glycosylase, MutM 1L1T, ⁶⁶; and T4 pyrimidine dimer DNA glycosylase, EndoV 1VAS, ⁶⁷. Proteins are colored according to secondary structure, and lesion-containing DNA is shown as blue sticks.

The solenoid structure of AlkD is distinctly different from the other DNA glycosylases (Figure 4). AlkD is the only glycosylase to form a linear, cylindrical groove along the entire width of the protein. Consequently, the environment around the putative active site is much less constrained than other DNA glycosylase architectures. To investigate how such a unique glycosylase architecture might engage DNA, we constructed a model of DNA docked onto the structure of AlkD (Figure 5). The lack of structural similarity between AlkD and the other glycosylases precluded us from superimposing previous glycosylase-DNA complexes onto AlkD. Instead, we took advantage of the fact that the dimensions across the concave opening are ~24 Å, which can easily accommodate a B-DNA duplex. Canonical B-DNA (PDB ID 1BNA) was manually docked onto the concave surface with the minor groove of the DNA positioned toward the putative active site, consistent with structures of other glycosylase/DNA complexes. Importantly, the linear DNA model formed favorable van der Waals and electrostatic interactions along the entire groove of the protein with no steric collisions (Figure 5).

Figure 5.

Theoretical model of DNA bound to AlkD. B-DNA from PDB entry 1BNA manually docked onto the AlkD crystal structure shows that the DNA need not adopt a distorted conformation to access the concave surface and putative active site of the protein. The molecular surface of the protein is colored according to electrostatic potential (red negative, blue positive, −7 to +7 k_BT). Asp113 and Arg148 are labeled D and R, respectively.

The DNA docking model was used as a guide for mutational and chemical modification experiments designed to probe the nature of DNA binding to AlkD. The DNA backbone in our docking model was in close proximity to the catalytically important Asp113 and Arg148 residues (Figure 5). The effects of these residues on DNA binding were quantified using an in vitro fluorescence anisotropy assay (Table 3, Figure S4). Substitution of Asp113 with asparagine increased the affinity of AlkD for DNA containing either 7mG substrate or a tetrahydrofuran (THF) abasic product analog (Table 3). Likewise, an Arg148Ala mutation resulted in a modest but significant decrease in DNA binding. These results are consistent with an electrostatic interaction between Asp113-Arg148 and the DNA backbone. To further investigate this interaction, we measured binding of wild-type AlkD to positively-charged 1-azaribose (C1′→N) and pyrrolidine (O4′→N) abasic sites, which are potent inhibitors of many DNA glycosylases and were designed to mimic the positive charge of the proposed transition state for hydrolysis of the glycosidic bond ^{37; 51; 52; 53}. The affinity of AlkA for pyrrolidine-DNA is three orders of magnitude greater than for THF-DNA, and AlkA binds to both pyrrolidine and 1-azaribose ~15-fold tighter than to 7mG-DNA ⁵⁴. It was therefore surprising that wild-type AlkD bound with the same affinity to all DNAs tested (Table 3), suggesting that AlkD engages DNA in the active site differently than AlkA. Taken together, the docking model and mutational data suggest that DNA need not obtain a distorted, extrahelical conformation in order to bind to the protein active site surface.

Table 3.

DNA binding activities for wild-type and mutants of AlkD

	Wild-type		D113N		R148A
	Kd (μM)	Relative affinity	Kd (μM)	Relative affinity	Kd (μM)	Relative affinity
G	2.0 ± 0.5		n.d.		n.d.
7mG	1.8 ± 0.5	(1.0)	0.5 ± 0.2	3.5	3.8 ± 0.6	0.5
THF	3.1 ± 0.3	(1.0)	1.3 ± 0.3	2.4	6.2 ± 1.3	0.5
Pyr	3.4 ± 0.1		n.d.		n.d.
Aza	1.8 ± 0.3		n.d.		n.d.

Dissociation constants (K_d) for a 25mer oligonucleotide duplex containing the specified modification paired with cytosine were measured by fluorescence anisotropy as described in Experimental Procedures. n.d., not determined

Discussion

The crystal structure of B. cereus alkylpurine DNA glycosylase AlkD reveals variant HEAT-like motifs that form an electropositive concave surface necessary to engage DNA. At the heart of the concave surface lies a shallow cleft of aromatic and charged residues, which we verified to be important for binding the DNA backbone and for 7mG excision. These catalytically essential residues and the unique DNA binding HEAT repeat architecture were previously predicted from an AlkD homology model ³³, which was constructed from an unpublished structure (EF3068) determined by the Midwest Structural Genomics Center as part of the Protein Structure Initiative ⁵⁵. In the present study, we extend these results by a structural and mutagenic analysis of nucleic acid binding by the variant HEAT motifs in AlkD.

The AlkD solenoid is a unique DNA glycosylase fold, and therefore represents a sixth structural superfamily for these DNA repair enzymes (Figure 4). Perhaps the most unique feature of the AlkD glycosylase is its concave surface, which forms a groove perfectly complementary in shape and charge to a linear B-DNA duplex. Our DNA docking model shows an excellent fit between AlkD and an undistorted B-DNA molecule. As a comparison, we superimposed the kinked DNA from the AlkA/DNA complex ³⁷ onto our structure, as has been done previously for the theoretical homology model ³³. The protein surface occupied by the highly bent DNA was not as extensive as the undistorted B-DNA contact surface, suggesting that AlkD might bind DNA differently than other glycosylases. In order to stabilize extrahelical nucleobases and a large helical distortion in the DNA duplex, DNA glycosylases typically intercalate a pair of side chains into the DNA base stack. Interestingly, neither linear nor kinked DNA-AlkD models revealed any obvious candidate side chains that might penetrate the stacked bases in order to stabilize a flipped nucleotide. We cannot exclude the possibility that AlkD might adjust its conformation in order to engage a flipped-out nucleotide, although such a DNA-induced protein conformational change has not been observed in other glycosylases. Nevertheless, the open architecture and the apparent lack of DNA intercalating residues strongly suggests that AlkD interrogates damaged DNA differently than other glycosylases. Our model illustrates how AlkD might utilize extensive contacts with the minor groove of both DNA strands in order to sense a subtle distortion in 3mA·T and 7mG·C base pairs. This binding regime, although speculative, illustrates that this novel glycosylase architecture is well suited to scan DNA in search for a lesion without distorting the conformation of the double helix.

The distinct substrate specificity of AlkD for only positively charged alkylpurines might be explained by the Asp113-Arg148 salt bridge within the putative active site. Because it participates in an electrostatic interaction, Asp113 is not analogous to the conserved aspartate or glutamate residues found in other glycosylases (e.g., Asp238 in AlkA). The crystal structure of AlkA in complex with DNA containing 1-azaribose revealed a direct interaction between the extrahelical positively charged abasic site and Asp238 ³⁷. The observation that neither 1-azaribose nor pyrrolidine enhances DNA binding by AlkD (Table 3) suggests that AlkA and AlkD use different mechanisms to catalyze nucleobase excision. Acidic residues within glycosylase active sites have been suggested to catalyze base hydrolysis by activating a water or protein nucleophile for attack of the C1' anomeric carbon ^{13; 56}, or by directly stabilizing a carbocation intermediate formed during cleavage of the glycosylic bond ³⁷. However, the formal positive charges of 3mA and 7mG nucleotides make these bases favorable to hydrolysis even in the absence of direct activation by a catalytic residue. Mutation of Arg148 reduced activity by two orders of magnitude, suggesting that Asp113 alone is not sufficient for catalysis, or that the salt bridge stabilizes the local conformation of the active site. Indeed, our DNA docking model predicts that the precise orientations of the Asp113 and Arg148 side chains with respect to the lesion are critical. In this model, the phosphoribose backbones of both DNA strands are abutted against the Asp113-Arg148 pair, leaving no room for a flipped nucleotide (Figure 5). This supports the scanning mechanism described above, and raises the possibility that this aromatic cleft is not an extrahelical nucleobase binding pocket. In support of this, binding of free alkylpurine bases to AlkD could not be detected in solution by intrinsic tryptophan fluorescence quenching or 7mG excision inhibition assays, nor did we observe electron density corresponding to 3mA or 7mG bases that were soaked into AlkD crystals at 10-fold molar excess of free base (data not shown). Based on the instability of the 3mA and 7mG bases and the proximity of Asp113-Arg148 to the DNA backbone in our model, AlkD should be able to gain access to the glycosylic bond for catalysis without flipping the base into the aromatic cleft.

AlkD is the first structural example of a HEAT protein that contains catalytic activity. Based on the adaptability of the HEAT motif to bind protein and DNA, other examples are likely. Of note is deoxyhypusine hydroxylase (DOHH), an iron(II)-containing enzyme which catalyzes the final step in postranslational hypusinylation of eukaryotic initiation factor 5A (eIF5A). A theoretical model of DOHH predicts the protein to form a pair of HEAT domains similar in structure to AlkD ⁵⁷. Mutation of glutamate residues lining the concave surfaces of the protein impairs binding to the positively-charged deoxyhypusine-eIF5A substrate ⁵⁸. The apparent similarity between catalytic residues within AlkD and DOHH suggest that HEAT proteins might have evolved a common feature to couple substrate binding and catalysis.

Materials and Methods

AlkD purification and crystallization

The AlkD gene was PCR amplified from Bacillus cereus genomic DNA (ATCC 14579) and cloned into a pET27 (Novagen) derived expression vector (pBG103, Vanderbilt Center for Structural Biology) that produces a cleavable N-terminal His₆-SUMO-fusion protein. E. coli HMS174 cells transformed with the AlkD/pBG103 plasmid were propagated in LB media and protein was overexpressed for 3 h at 37°C upon addition of 0.5 mM IPTG. Cells were harvested in 50 mM Tris-HCl (pH 7.5), 500 mM NaCl, and 10% glycerol and lysed with an Avestin Emulsifier C3 homogenizer operating at ~20000 psi. AlkD-fusion protein was purified using Ni-NTA (Qiagen) affinity chromatography, followed by cleavage of the His₆-SUMO tag. AlkD was further purified by heparin affinity and gel filtration chromatography to >99% homogeneity. Protein was concentrated to 12.5 mg/ml and stored in 20 mM Bis-Tris Propane, 100 mM NaCl, and 0.1 mM EDTA. Mutant proteins were prepared by site-directed mutagenesis using a Quik-Change Kit (Stratagene) and purified in the same manner as wild-type AlkD. Structural integrity of mutant proteins was verified by circular dichroism spectroscopy.

Crystals of unliganded AlkD were grown at 21°C by sitting-drop vapor diffusion. Drops containing equal volumes of protein (14.5 mg/ml) and reservoir (85 mM HEPES pH 7.5, 15% PEG 4000, 17% glycerol) were equilibrated against the reservoir. Single crystals with dimensions 0.2 × 0.2 × 2.0 mm³ grew in 4 days and were flash frozen in a liquid nitrogen stream prior to X-ray data collection. Derivative crystals were prepared by soaking AlkD crystals in 1 mM K₂PtCl₄ for 72 hrs at 21°C.

X-ray Data Collection, Phasing, and Structure Refinement

X-ray diffraction data (Table 1) were collected at beamlines 21-ID (native) and 22-ID (derivative) at the Advanced Photon Source (Argonne, IL) and processed using the HKL2000 package ⁵⁹. AlkD crystals belong to space group P4₃ and contain one molecule in the asymmetric unit.

Experimental X-ray phases were obtained from a single-wavelength anomalous dispersion (SAD) experiment using a crystal soaked with K₂PtCl₄. Diffraction data extending to 1.85 Å (Table 1) were collected at the energy corresponding to the platinum absorption peak (11.567 keV). Significant anomalous differences between |F⁺| and |F⁻| amplitudes extended to 2 Å. The program SHARP ⁶⁰ was used to locate and refine the positions of two platinum sites and to carry out phase calculation and density modification. A continuous protein chain corresponding to residues 1-225 was built into the resulting 2.0-Å electron density map. Two non-native N-terminal residues (Val −1, Pro 0) resulting from cleavage of the His₆-SUMO tag were readily identified in the experimental electron density and were included in the final model, while the C-terminal 12 residues (226-237) were unobserved.

The model was refined against both native and derivative amplitudes using experimental Pt-SAD phases and a maximum likelihood target as implemented in REFMAC 5.4 ⁶¹. Improvements to the model were guided by manual inspection of sigma-weighted 2mFo-Dfc and mFo-DFc electron density maps, and were judged successful by a decrease in R_free during refinement. Translation/libration/screw-rotation (TLS) refinement in REFMAC was used to model anisotropic motion of the protein. Refinement of the Pt-derivative model resulted in R_cryst and R_free values of 20.3% and 21.4%, which are higher than expected at 1.85 Å resolution. Efforts to resolve regions in the model corresponding to a partially chlorinated Pt(II) ion and a PEG 4000 molecule inside the putative active site and a second Pt(II) ion bound to Met 1 using both experimental and refined electron density were unsuccessful. In order to avoid any potential structural perturbation of the putative active site from Pt or PEG 4000, the native model was used in all structural analysis despite the lower resolution of the native data (Table 1). The r.m.s.d. between the native and derivative models is 0.72 Å for all atoms. The native AlkD model was validated using PROCHECK ⁶², and has been deposited along with structure factors in the Protein Data Bank under accession code 3BVS.

Glycosylase Activity Assay

AlkD glycosylase activity was measured by alkaline cleavage of the abasic DNA product of 7mG excision from a 25mer oligonucleotide duplex containing a 7mG·C base pair. 7mG was incorporated into DNA duplexes enzymatically using the previously described method ⁶³. Briefly, the primer oligonucleotide 5′- GACCACTACACC was ³²P-labeled at the 5′-end, annealed to a 3-fold excess of the complementary strand (5′-GTTGTTAGGAAACGGTGTAGTGGTC), and extended using DNA polymerase I Klenow fragment (New England Biolabs) in the presence of deoxy-7-methylguanosine 5′-triphosphate (d7mGTP, Sigma), dCTP, dTTP, and dATP. In a 10 μl glycosylase reaction, 2 nM radiolabeled DNA duplex was incubated with 2 μM AlkD in 50 mM HEPES pH 7.5, 100 mM KCl, 10 mM DTT, and 2 mM EDTA. The reaction was stopped at various times by addition of 0.2 N NaOH, and heated at 70°C for 10 min. The 12mer product and remaining 25mer substrate DNA strands were separated by denaturing 15% polyacrylamide gel electrophoresis in 7M urea and quantitated by autoradiography.

DNA Binding Assay

DNA binding of wild-type and mutant AlkD was measured by the change in fluorescence anisotropy as protein was added to an oligonucleotide duplex that contained a site-specific modification (X) in the middle of one strand [d(GACCACTACACCXTTTCCTAACAAC)] and a 6-carboxyfluorescein on the 3'-end of the complementary strand [d(GTTGTTAGGAAACGGTGTAGTGGTC)-FAM]. Oligonucleotides containing abasic sites were chemically synthesized, and those containing 7mG were prepared as above but without the ³²P-label. Increasing concentrations of protein (0-30 uM) were added to a 50 nM DNA in 20 mM Bis-Tris Propane pH 6.5, 100 mM NaCl, 2 mM DTT, and 0.1 mM EDTA. Polarized fluorescence intensities using excitation and emission wavelengths of 485 and 538 were measured at ambient temperature using a SpectraMax M5 microplate reader (Molecular Devices). Measurements were recorded within 2 minutes of adding wild-type AlkD to DNA to ensure that no more than 50% of the 7mG-DNA substrate was converted to abasic-DNA product. Dissociation constants were derived by fitting a two-state binding model to data from three independent experiments.