Transient kinetic analysis of oxidative dealkylation by the direct reversal DNA repair enzyme AlkB

AlkB is a bacterial Fe(II)– and 2-oxoglutarate–dependent dioxygenase that repairs a wide range of alkylated nucleobases in DNA and RNA as part of the adaptive response to exogenous nucleic acid–alkylating agents. Although there has been longstanding interest in the structure and specificity of Escherichia coli AlkB and its homologs, difficulties in assaying their repair activities have limited our understanding of their substrate specificities and kinetic mechanisms. Here, we used quantitative kinetic approaches to determine the transient kinetics of recognition and repair of alkylated DNA by AlkB. These experiments revealed that AlkB is a much faster alkylation repair enzyme than previously reported and that it is significantly faster than DNA repair glycosylases that recognize and excise some of the same base lesions. We observed that whereas 1,N6-ethenoadenine can be repaired by AlkB with similar efficiencies in both single- and double-stranded DNA, 1-methyladenine is preferentially repaired in single-stranded DNA. Our results lay the groundwork for future studies of AlkB and its human homologs ALKBH2 and ALKBH3.

DNA bases that are damaged by alkylation may cause mutations that lead to cell death or disease. Pathways for repair of alkylated bases in many organisms include base excision repair (BER) 2 and direct reversal repair (DRR). For example, the adaptive response to alkylation damage in Escherichia coli includes both AlkA, a DNA glycosylase that initiates BER by excising alkylated bases, and AlkB, an oxidative dealkylase that uses ferrous iron, oxygen, and 2-oxoglutarate (2OG) to catalyze DRR of alkylated bases (1).
A broad range of alkylated nucleobases can be repaired by AlkB, including the monoalkyl substrate 1-methyladenine (1mA) and the exocyclic bridged substrate 1,N 6 -ethenoadenine (⑀A) (2). AlkB-catalyzed hydroxylation of these alkyl adducts is followed by spontaneous release of an aldehyde byproduct to restore an undamaged nucleobase (Fig. 1). Although a large number of AlkB substrates are known, information about its substrate specificity is limited. Previous studies of AlkB repair have often been qualitative or semiquantitative in scope. The more quantitative studies among them have surveyed alkylated bases within dissimilar nucleic acid contexts using a wide variety of reaction conditions and assays, and these experimental differences have hindered efforts to combine results across studies into an accurate understanding of AlkB's substrate preferences. The tendency of members of the Fe(II)-and 2OG-dependent dioxygenase superfamily, which includes AlkB, to lose activity due to self-catalyzed oxidative modification further complicates interpretation of previous results (3)(4)(5).
To systematically study the substrate specificity and kinetic mechanism of AlkB, we developed a quantitative DNA glycosylase-coupled assay and used it to perform transient kinetic studies. This approach takes advantage of rapid reaction methods and single-turnover reaction conditions to avoid the problem of dioxygenase inactivation during multiple turnover. The known AlkB substrates 1mA (6,7) and ⑀A (8), embedded in both single-stranded (ss) and double-stranded (ds) DNA oligonucleotides of identical length and sequence, were chosen for direct comparison. Selection of 1mA was guided by the knowledge that E. coli alkB mutants are defective in processing methylated ssDNA (9), and 1mA forms appreciably in ssDNA exposed to alkylating agents (10). The ⑀A oligonucleotides were included in the analysis because AlkB has been shown to be important for repair of a plasmid containing ⑀A damage in E. coli cells (11). In addition, several well-characterized DNA glycosylases excise ⑀A from dsDNA (12,13), so evaluation of oxidative dealkylation of the same lesion by AlkB allows recognition and repair of a common substrate by enzymes in the BER and DRR pathways to be compared.

Metal binding to purified AlkB
Recombinant E. coli AlkB was purified without metal and stored in 1 mM EDTA to ensure protein stability. To evaluate the quality of the purified enzyme, we compared the total concentration of AlkB, as determined by its absorbance at 280 nm and its predicted extinction coefficient, with the total concentration of metal that it binds, using ICP-MS to perform quantitative metal analysis. Fe 2ϩ binds stoichiometrically in the AlkB active site and is required for oxidative repair of DNA by AlkB, but it is easily oxidized and can cause oxidative damage to AlkB in aerobic conditions. In the absence of Fe 2ϩ , the metal-binding site of AlkB is able to bind to other metals, including Zn 2ϩ (14 -16). We performed parallel dialysis on AlkB samples that contained an excess of Zn 2ϩ and on AlkB samples without added metal. At equilibrium, each sample type is expected to contain the same amount of Zn 2ϩ , corresponding to the concentration of AlkB that is capable of binding a metal ion in its active site. The molar ratio of zinc to AlkB was 0.92 Ϯ 0.02 for extensively dialyzed samples that started with AlkB and an excess of Zn 2ϩ and 0.86 Ϯ 0.06 for samples that started with AlkB only (Table S1). This result indicates that ϳ90% of the purified AlkB is competent to bind metal and is therefore expected to be enzymatically active.

DNA glycosylase-coupled assay for direct DNA repair catalyzed by AlkB
Some organisms that lack an AlkB homolog have DNA glycosylases with expanded substrate repertoires that are able to initiate BER of known AlkB substrates (17)(18)(19)(20), and we expected that such glycosylases would be useful for assaying AlkB and its homologs. The alkyladenine DNA glycosylase from Bacillus subtilis (bAAG) (21) excises both the monoalkyl lesion 1mA and the exocyclic bridged lesion ⑀A from DNA (Fig. S1); we therefore developed a DNA glycosylase-coupled assay for AlkB using bAAG ( Fig. 2A). After AlkB repair reactions of fluorescein-labeled oligonucleotide substrates (Table S2) are quenched, bAAG is added to excise any unrepaired base lesions. The resultant abasic sites are cleaved by heating in sodium hydroxide, generating shortened strands that resolve from full-length AlkB-repaired products on a denaturing polyacrylamide gel (Fig. 2B). The DNA glycosylase-coupled assay is suitable for any AlkB substrate for which there is a compatible glycosylase, and a similar assay that used E. coli MUG and human AAG has been applied to ethenobase-containing DNA (22).

Single-turnover reactions of AlkB
Loss of catalytic activity for 2OG oxygenases is commonly observed and has been attributed to nonproductive reactions with oxygen, including auto-oxidation of active site residues and oxidation of the active-site ferrous iron to a higher oxidation state (3)(4)(5). To avoid this problem, we initially applied the repair assay to single-turnover reactions of AlkB. A single turnover requires less time than multiple turnovers, and the DNA concentration remains below the AlkB concentration in singleturnover reactions ([AlkB] Ͼ [DNA]), minimizing time-dependent and potential DNA-dependent enzyme inactivation.
AlkB was purified without iron to ensure protein stability; therefore, single-turnover repair reactions required the addition of the iron cofactor as well as the 2OG and DNA substrates. We compared two different orders of addition to assess whether or not the timing of the DNA substrate addition mattered (Fig. 3A). Representative time courses for repair of ⑀A-25mer (ssDNA) by an excess of AlkB are shown in Fig. 3B and provide examples of the quality of kinetic data that can be obtained from the DNA glycosylase-coupled assay. Repair of ⑀A-25mer is noticeably faster at all concentrations of AlkB ( Fig.  3C) when AlkB is combined with iron and 2OG before DNA substrate is added (OA2). Repair of ⑀A⅐T-25mer (dsDNA), 1mA-25mer (ssDNA), and 1mA⅐T-25mer (dsDNA) is likewise faster for OA2 than for OA1 at all concentrations of AlkB ( Fig.  S2 and Table S3). The inhibitory effect of adding DNA substrate, iron, and 2OG simultaneously to AlkB reactions may be caused by nonproductive binding of DNA substrate to apo-AlkB (E apo ), and it suggests that assembly of the AlkB active site is ordered (Fig. 3D).
We focused on the more favorable OA2 conditions for comparison of single-turnover repair of the DNA substrates ⑀A-25mer, ⑀A⅐T-25mer, 1mA-25mer, and 1mA⅐T-25mer by AlkB. This is likely to be the biologically relevant pathway, considering the physiological concentrations of Fe 2ϩ and 2OG and the relative rarity of alkylated bases. The k obs values determined from repair time courses for all four substrates show a hyperbolic dependence on AlkB concentration (Fig. 4), and their

DNA repair kinetics of AlkB
k max , K1 ⁄ 2 , and k max /K1 ⁄ 2 values are summarized in Table 1. Our results are consistent with the prevailing view that 1mA in ssDNA is a preferred substrate of AlkB (Table S4), and our single-turnover methodology enables us to report this preference accurately and quantitatively; comparison of the substrate specificity constants k max /K1 ⁄ 2 reveals that AlkB prefers 1mA in ssDNA over 1mA in dsDNA by 18-fold, over ⑀A in ssDNA by 160-fold, and over ⑀A in dsDNA by 110-fold. The k max /K1 ⁄ 2 values for ⑀A-25mer and ⑀A⅐T-25mer are within 2-fold of one another, indicating that DNA strand context is less important for repair of ⑀A than for repair of 1mA by AlkB. The similar maximal repair rate constants of ϳ1 s Ϫ1 for the 1mA and ⑀A lesions, irrespective of the DNA strand context (k max ; Table 1), show that repair of ⑀A by AlkB is limited by recognition, not rate of repair.

Multiple-turnover reactions of AlkB are poorly behaved
Despite concerns about possible confounding nonproductive reactions with oxygen (3-5), we next applied the repair assay to multiple-turnover reactions of AlkB, to facilitate more direct comparisons with previously published data (Table S4). A, a fluorescein-labeled oligonucleotide substrate containing an alkylated base lesion (A*) is incubated with AlkB and cosubstrates. The repair reaction is quenched with EDTA and sodium hydroxide and then neutralized. Unrepaired base lesions (right pathway) are excised by a DNA glycosylase, yielding abasic sites (ap) that are cleaved when heated in sodium hydroxide to generate shortened strands that resolve from the stable repaired product (left pathway) on a denaturing polyacrylamide gel. For repair reactions of ssDNA substrates, an excess of the complementary strand was included in the quench solution to improve subsequent DNA glycosylase activity. B, a gel image shows the mobility of species present after ⑀A-25mer is incubated with the indicated components. After a complete dealkylation reaction (far right lane), nearly all of the ⑀A base has been converted to A, which is not excised by bAAG, resulting in a repaired product that runs as an intact 25mer. . Each reaction progress curve was fit by a single exponential, giving observed rate constants (k obs ) of 0.069 s Ϫ1 for OA1 and 0.32 s Ϫ1 for OA2. C, the dependence of single-turnover repair rate constants for ⑀A-25mer on the AlkB concentration differs for OA1 (open circles) and OA2 (closed circles) but is hyperbolic for both orders of addition, and k max and K1 ⁄2 values for repair are reported in Table S2. The average Ϯ S.D. of 2-4 reactions is shown. D, model for nonproductive binding of alkylated DNA substrate (DNA*) to apo-AlkB (E apo ) that may account for the inhibitory effect of OA1 on repair reactions.

DNA repair kinetics of AlkB
However, inspection of multiple-turnover time courses for repair of 1mA⅐T-25mer by AlkB revealed that the rate of product formation decreases rapidly (Fig. 5A) instead of staying constant during the first 10% of substrate depletion as expected for a well-behaved enzymatic reaction. This drop-off in 1mA repair during multiple turnover can be clearly seen by plotting the different instantaneous v/[E] values (calculated from three consecutive time points; Fig. 5B); a similar drop-off is observed for ⑀A repair during multiple turnover (Fig. S3). To find out if instability of AlkB in our reaction conditions could account for the repair slowdown, we preincubated the enzyme at 37°C for 90 min in reaction buffer only, or in reaction buffer plus all other reaction components, prior to adding DNA substrate to initiate multiple-turnover repair (Fig. 5C). There is no difference between the trajectories of the original reaction (Fig. 5C, filled A, dependence of single-turnover repair rate constants for ⑀A-25mer and ⑀A⅐T-25mer on the AlkB concentration. B, dependence of single-turnover repair rate constants for 1mA-25mer and 1mA⅐T-25mer on the AlkB concentration. The single-turnover repair rate constants for each substrate exhibited a hyperbolic dependence on AlkB concentration, and k max and K1 ⁄2 values for repair of the four substrates are reported in Table 1. OA2 conditions were used, and the average Ϯ S.D. of 2-4 reactions is shown.

DNA repair kinetics of AlkB
circles) and the reactions in which AlkB was preincubated (Fig.  5C, open symbols), indicating that AlkB is stable for at least 90 min in our reaction conditions in the absence of DNA substrate. It is thus the addition of DNA substrate that causes the catalytic activity of AlkB to decrease in a time-dependent manner. Many nonheme iron enzymes are known to undergo autocatalyzed oxidative modifications (3)(4)(5), which are postulated to result from side reactions of their highly activated Fe(IV)-oxo intermediates (23). Indeed, AlkB has been reported to self-hydroxylate Trp-178 (24), and hydroxylation and carbonylation of Leu-118 have also been detected (25). Such modifications may occur more readily in the presence of DNA, due to conformational changes in the AlkB active site that occur upon binding of DNA, and inactivating modifications could explain the observed slowdown in AlkB's repair rate during multiple turnover ( Fig. 5 and Fig. S3).

Multiple-turnover competition assays to determine relative k cat /K m values for AlkB
After finding conventional multiple turnover reactions of AlkB to be poorly behaved, we turned to an alternative multiple-turnover method to measure substrate specificity. The relative preference of AlkB for dealkylation of 1mA and ⑀A in dsDNA was measured by placing 1mA⅐T-25mer and ⑀A⅐T-19mer substrates (Table S2) in direct competition in the same AlkB reaction mixture (Fig. 6). Following DNA glycosylase treatment and hydrolysis in our assay, the cleaved 1mA substrate (12mer) and its AlkB-repaired product (25mer) and the cleaved ⑀A substrate (9mer) and its AlkBrepaired product (19mer) are all different sizes and can be resolved on a gel.
Repair of 1mA outcompetes repair of ⑀A when equal concentrations of 1mA-and ⑀A-containing dsDNA substrates are present in the same AlkB reaction mixture (Fig. 6B), corresponding to a relative k cat /K m value of 9.3 Ϯ 0.9 for 1mA⅐T-25mer with respect to ⑀A⅐T-19mer (see "Experimental procedures"). A slight length preference was observed when we controlled for the effect of DNA length on AlkB activity by performing a competition assay between ⑀A incorporated into 25mer and 19mer duplexes (Fig. 6C), which revealed a relative k cat /K m value of 1.3 Ϯ 0.1 for ⑀A⅐T-25mer with respect to ⑀A⅐T-19mer.
The ratio of the two relative k cat /K m values for the 25mers with respect to ⑀A⅐T-19mer gives a relative k cat /K m value of 7.2 Ϯ 0.9 for 1mA⅐T-25mer with respect to ⑀A⅐T-25mer; this is the same, within error, as the relative k max /K1 ⁄ 2 value of 6.0 Ϯ 3.0 for 1mA⅐T-25mer with respect to ⑀A⅐T-25mer ( Table 1; the ratio of their k max /K1 ⁄ 2 values). Therefore, results from multipleturnover competition between two alkylated substrates and from single-turnover repair reactions performed for each substrate individually are in agreement and reveal that AlkB favors 1mA lesions over ⑀A lesions in dsDNA by a factor of 7. This consistency validates the use of the single turnover specificity constant k max /K1 ⁄ 2 to evaluate the substrate specificity of AlkB, as an alternative to multiple-turnover measurements that are adversely affected by AlkB inactivation. Notably, single-turnover studies of substrate specificity and mechanism are widely used for another class of DNA repair enzymes, the DNA glyco-sylases, for which enzyme inhibition by the abasic DNA product often limits the value of multiple-turnover measurements (26).

Discussion
A wide variety of alkylated nucleobases have been shown to be substrates of AlkB (2), but a quantitative understanding of its substrate specificity has been impeded by differences in how studies have been performed. For example, different studies of AlkB have used variable reaction conditions, DNA substrates of various lengths and sequences, and an array of assays. Not surprisingly, this diversity of substrates, reaction conditions, and Figure 6. Steady-state competition experiments for AlkB-catalyzed oxidative dealkylation. A, schematic for a direct competition experiment. The relative k cat /K m values for two competing substrates (S1 and S2) present in the same AlkB reaction mixture can be determined from the initial rates (V) and the initial substrate concentrations. B, a relative k cat /K m value of 9.3 Ϯ 0.9 was determined for 1mA⅐T-25mer with respect to the reference substrate ⑀A⅐T-19mer. C, a relative k cat /K m value of 1.3 Ϯ 0.1 was determined for ⑀A⅐T-25mer with respect to ⑀A⅐T-19mer, indicating a slight length preference for the 25mer. Reactions contained 5 nM AlkB and 200 nM of each dsDNA. The average Ϯ S.D. of two reactions is shown.

DNA repair kinetics of AlkB
experimental approaches has given rise to highly variable results that are difficult to interpret, as illustrated by the disparate kinetic parameters that have been reported for ⑀A and 1mA substrates of AlkB and assembled in Table S4.
To analyze AlkB's substrate specificity quantitatively, we systematically compared its repair of ⑀A and 1mA. The two alkylated bases were embedded in oligonucleotides of identical length and sequence in both ssDNA and dsDNA and assessed as substrates for AlkB in the same reaction conditions, using the same DNA glycosylase-coupled assay. Several limitations associated with previous studies of AlkB were overcome by using rapid reaction methods and a defined order of addition to investigate single-turnover reactions, in conjunction with competition assays to compare representative multiple-turnover reactions.

1mA in ssDNA is a preferred substrate of AlkB
E. coli alkB mutants are defective in processing methylated ssDNA (9), and 1mA is widespread in ssDNA that has been treated with alkylating agents (10). Consistent with these observations, our single-turnover repair comparisons revealed a substantial preference of AlkB for 1mA in ssDNA. AlkB's substrate specificity constant k max /K1 ⁄ 2 is 18-fold larger for 1mA in 25-nucleotide ssDNA than for 1mA in the same dsDNA sequence (Table 1; 1mA-25mer versus 1mA⅐T-25mer). In contrast, k max / K1 ⁄ 2 values for ss ⑀A-25mer and ds ⑀A⅐T-25mer are within 2-fold of each other, indicating that DNA strand context matters little for repair of ⑀A by AlkB. This difference in the importance of strand context for the 1mA and ⑀A substrates may be due to differences in their base pairing with T (Fig. 7). Although methylation of N1 of A prevents Watson-Crick base pairing between 1mA and T, they instead form a Hoogsteen base pair (27,28). Thus, hydrogen-bonding interactions between 1mA and T must be broken before AlkB can flip 1mA into its active site in dsDNA, but not in ssDNA, which may factor into AlkB's preference for 1mA in ssDNA. There are no hydrogen-bonding interactions between ⑀A and T in dsDNA, so the energetic barrier for flipping ⑀A into AlkB's active site is expected to be less dependent on strand context; this may account for the similar k max /K1 ⁄ 2 values for repair of ss ⑀A-25mer and ds ⑀A⅐T-25mer by AlkB. The k max /K1 ⁄ 2 ratios in Table 1 also reveal that although AlkB discriminates between 1mA and ⑀A in ssDNA by a factor of 160, it has only a modest preference (6.0 Ϯ 3.0) for 1mA over ⑀A when both lesions are present in dsDNA. Importantly, multiple-turnover competition reactions confirmed the modest preference (7.2 Ϯ 0.9) for 1mA⅐T-25mer over ⑀A⅐T-25mer.
Although the specificity of AlkB for each of the four substrates differs as described above, its maximal rate constants for their repair are all very similar (k max values between 0.75 and 1.7 s Ϫ1 ; Table 1). This high reactivity may come into play when AlkB is induced in E. coli as part of the adaptive response to alkylation damage (1): high AlkB concentrations promote single-turnover repair of lesions like ⑀A, despite the weak specificity of AlkB for them. Consistent with this idea, AlkB was shown to play a major role in repair of a plasmid containing ⑀A damage that had been introduced into E. coli cells, with increased repair detected in cells in which the adaptive response was induced (11).

Single-turnover repair measurements reveal fast oxidative dealkylation by AlkB
Our single-turnover repair measurements show that AlkB is a much faster oxidative dealkylase than has been appreciated. Previously reported or estimated k cat values for repair of 1mA and ⑀A in ssDNA by AlkB differ over wide ranges, from 0.007 to 0.2 s Ϫ1 for 1mA and from 0.001 to 0.03 s Ϫ1 for ⑀A (Table S4). We used rapid reaction methods to measure the k max value of 0.90 s Ϫ1 for repair of 1mA in ssDNA, which exceeds reported k cat values by 5-100-fold (Table 1 and Table S4). Likewise, the k max value of 0.75 s Ϫ1 for repair of ⑀A in ssDNA is 30 -800-fold larger than reported k cat values (Table 1 and Table S4).
It is likely that loss of activity of AlkB over time and AlkB inhibition due to the order of addition of reaction components may have contributed to both the variability and the apparent underestimation of previous maximal rate measurements. We observed a rapid loss of AlkB activity during multiple-turnover repair that prevented us from measuring steady-state kinetic parameters for repair of 1mA and ⑀A (Fig. 5 and Fig. S3). Inactivating modifications of AlkB may be responsible for the observed slowdown in repair rate, because autocatalyzed oxidative modifications are well-known for Fe(II)/2OG-dependent dioxygenases (3)(4)(5) and oxidized AlkB variants have been found (24,25). Variable kinetic parameters such as those that have been reported for ⑀A and 1mA substrates of AlkB (Table  S4) are typical for enzymes that lose activity during reactions, because rarely shown experimental details such as the number and spacing of reaction time points and the timescales of reactions can have large effects on measurements. For example, if only a single reaction time point is collected, then enzyme inactivation cannot be detected. Another experimental concern is the order of addition of reaction components used in previous AlkB studies. We observed that multiple-turnover reactions,

DNA repair kinetics of AlkB
like single-turnover reactions, are inhibited by OA1, the simultaneous addition of DNA substrate, iron, and 2OG to AlkB (Fig.  3 and Fig. S2). Insufficient experimental detail is provided to know the order of addition that was used in most of the studies in Table S4, but the inhibitory OA1 was described in at least one of them.

Comparison of repair of ⑀A in dsDNA by DRR and BER enzymes
Whereas 1mA in ssDNA is a preferred substrate, AlkB also repairs ⑀A in dsDNA ( Table 1). The DNA glycosylase AlkA from E. coli initiates repair of ⑀A in dsDNA by excising it (12,13), allowing the repair strategies of the two enzymes for this shared substrate to be compared. AlkB and AlkA have almost identical overall repair efficiencies for ⑀A in dsDNA, with k max /K1 ⁄ 2 values of 2.0 ϫ 10 5 M Ϫ1 s Ϫ1 for dealkylation by AlkB (Table 1) and 1.6 ϫ 10 5 M Ϫ1 s Ϫ1 for base excision by AlkA (29). However, to achieve this repair efficiency, AlkB couples high reactivity (k max ϭ 1.7 s Ϫ1 ) with weak recognition of ⑀A⅐T-DNA (K1 ⁄ 2 ϭ 8400 nM), whereas AlkA has an almost 400-fold lower k max value (0.0045 s Ϫ1 ) but a much higher substrate affinity (K1 ⁄ 2 ϭ 29 nM).
Both AlkA and AlkB are induced in E. coli as part of the adaptive response to alkylation damage, but AlkB appears to be the most important for in vivo repair of ⑀A damage under standard growth conditions (11). Given the chemical differences in the AlkA-and AlkB-initiated pathways, it may be that each pathway contributes under different conditions. Cytoplasmic pH and availability of oxygen are two critical factors governing the relative proficiency of DRR and BER. Although the catalytic efficiency of AlkA matches that of AlkB under optimal glycosylase conditions at pH 6.1 (29), above this pH value, AlkA shows a log-linear decrease in glycosylase activity (30). AlkB requires oxygen as a substrate, and therefore DRR is only available under aerobic growth conditions. It is interesting to speculate that the coexistence of two separate DNA repair pathways, AlkA-initiated BER and AlkB-initiated DRR, provides broader coverage for repair of alkylated DNA over a wider range of physiological conditions.

An experimental foundation for future studies of AlkB and its human homologs
The rapid reaction methods and DNA glycosylase-coupled assay that we developed and used to systematically compare 1mA and ⑀A substrates can be extended to additional substrates to assess AlkB's substrate specificity more broadly. These approaches can also be applied to AlkB homologs, including the human proteins ALKBH2 and ALKBH3, which catalyze oxidative dealkylation of damaged DNA (2,31). Such studies will help to resolve discrepancies in reported kinetic parameters for ALKBH2 and ALKBH3, like those compiled for AlkB in Table  S4, that currently obscure their substrate specificities. The differences in substrate preference that are uncovered for different AlkB homologs and for their site-specific variants will help to identify the molecular basis for discrimination between substrates, allowing rules for substrate selection by oxidative dealkylases to be defined and used to predict their activity toward novel forms of alkylation damage. Our results show that singleturnover repair, with its short timescales and low DNA concentrations, is well-suited for evaluating substrate specificity for this family of repair enzymes, which may be prone to inactivation during multiple turnover repair like their prototypical member AlkB.

Purification of AlkB
A pET19-derived expression construct for E. coli AlkB (15) was transformed into BL21(DE3) E. coli, and protein was expressed and purified as described previously (20). Purified protein was concentrated to 650 M as determined by the absorbance at 280 nm using the predicted extinction coefficient of 32,430 M Ϫ1 cm Ϫ1 . AlkB was purified without iron and stored in 1 mM EDTA to ensure protein stability.

Dialysis and ICP-MS of AlkB
Dialysis and ICP-MS were performed to determine the concentration of purified recombinant AlkB that is competent for binding zinc, for comparison with the concentration that was determined by absorbance at 280 nm. The dialysis buffer contained 25 mM HEPES (pH 7.5), 150 mM NaCl, and 0.1 mM TCEP, and buffer in the reservoir (250 ml) was supplemented with 0.25 M ZnSO 4 . AlkB was prepared at 25 M in dialysis buffer with or without 250 M ZnSO 4 . Dialysis was performed at 4°C in a 10-well microdialysis chamber (Spectra/Por) with each well containing 50 l of analyte solution. Three wells contained AlkB with ZnSO 4 , three wells contained AlkB without supplemental metal, two wells contained dialysis buffer with 250 M ZnSO 4 , and two wells contained dialysis buffer only. The free Zn 2ϩ concentration was estimated to be 0.43 M, 60-fold less than the concentration of AlkB in the wells but a 14-fold molar excess over total protein in the system. Dialysis proceeded for 2 days to reach equilibrium. Protein concentration was remeasured by absorbance at 280 nm as described above. Samples were then diluted in 0.5% nitric acid, and the zinc concentration was determined against a standard curve using a Nexion 2000 ICP mass spectrometer (PerkinElmer Life Sciences).

Synthesis and purification of oligonucleotides
DNA substrates were synthesized by Integrated DNA Technologies or the Keck facility at Yale and were purified using denaturing PAGE as described previously (32). Sequences are provided in Table S2. ssDNA concentrations were determined from the absorbance at 260 nm using calculated extinction coefficients. For oligonucleotides containing ⑀A, the extinction coefficient was calculated for the same sequence with an A in place of ⑀A and corrected by subtracting 9400 M Ϫ1 cm Ϫ1 to account for the weaker absorbance of ⑀A (33). For oligonucleotides containing 1mA, the extinction coefficient was calculated for the same sequence with A in place of 1mA. Duplexes were annealed with 1.5-fold unlabeled complement.

DNA glycosylase-coupled assay for direct repair by AlkB
Single-turnover reactions contained 50 nM to 6.4 M AlkB and a 10 -50 nM concentration of a fluorescein-labeled ss or ds 25mer DNA substrate (Table S2), and typical multiple turnover reactions contained 5 nM AlkB and a 1.2-1.4 M concentration of the same DNA substrates. Reactions were carried out at 37°C DNA repair kinetics of AlkB in reaction mixtures containing 25 mM HEPES (pH 7.5), 100 mM NaCl, 0.1 mg/ml BSA, 1 mM TCEP, 2 mM sodium L-ascorbate, 1 mM 2OG, and 40 M ammonium iron (II) sulfate.
Kinetic parameters reported in the paper are from reactions that were initiated by adding DNA substrate to AlkB reactions after iron and 2OG had already been added (OA2; Fig. 3A). However, for comparison, some reactions were carried out using an alternate order of addition in which DNA substrate, iron, and 2OG were all added to AlkB reactions at the same time (OA1; Fig. 3A), and the results of this comparison for singleturnover reactions are shown in Table S3, Fig. 3, and Fig. S2.
For the fastest reactions, dealkylation is complete within seconds, requiring a rapid and efficient quench to stop the reaction. EDTA alone did not immediately stop reactions of AlkB, but a combination of EDTA and concentrated sodium hydroxide effectively stopped even the fastest reactions (Fig. S4). Therefore, repair reaction aliquots were typically quenched with an equal volume of a solution of 12 mM EDTA in 200 mM sodium hydroxide, and these quenched aliquots were subsequently neutralized to pH ϳ7.2 with 0.5 volumes of 400 mM HEPES free acid. B. subtilis AAG was added to the neutralized samples to give a Ն2-fold excess of glycosylase over the DNA present in the neutralized samples, and an excess of the complementary DNA strand was also added at this time to samples that contained ssDNA substrates. After overnight incubation at room temperature to allow bAAG to convert any unrepaired DNA into abasic DNA, sodium hydroxide was added to a final concentration of 225 mM, and samples were heated at 70°C for 12 min to convert abasic sites into single-strand breaks. This glycosylase treatment converts unrepaired 25mer DNA into 12mer DNA, but 25mers that have been repaired by AlkB remain intact and appear as slower-migrating product bands on gels (22). Hydrolyzed samples were mixed with formamide/ EDTA loading buffer and analyzed by denaturing PAGE on 15 or 20% (w/v) gels containing 6.6 M urea. Gels were scanned using a Typhoon imager (GE Healthcare), and emission was measured with a 525BP40 filter following excitation of the fluorescein label at 488 nm. Fluorescence intensities of gel bands were quantified using ImageQuant TL (GE Healthcare) and corrected for the amount of background signal. The data were converted to fraction product (P) and then fit by the single exponential shown in Equation 1. Mock DNA repair reactions were performed for each DNA substrate in the absence of AlkB as a positive control for the DNA glycosylase activity of bAAG and to reveal the maximum amount of repairable DNA in each initial sample (end pt), which was 95% for the ⑀A-25mercontaining substrates and 70 -80% for the more synthetically challenging 1mA-25mer-containing substrates (Fig. S1) (19,34).
Fraction P ϭ end pt͑1 Ϫ exp͑Ϫk obs t͒ ϩ zero pt (Eq. 1) The concentration dependence of the single turnover rate constant (k obs ) was fit by a hyperbolic dependence (Equation 2), k obs ϭ ͑k max ͓E͔͒/͑K 1/ 2 ϩ ͓E͔͒ (Eq. 2) where k max is the maximal single turnover rate constant, E is AlkB, and K1 ⁄ 2 is the concentration of AlkB at which k obs is 50% of the k max value.
When reactions were too fast to be followed using time points taken by hand, rapid mixing experiments were performed in a KinTek RFQ-3 quench-flow apparatus. For OA2, one sample loop contained AlkB, iron, and 2OG, and the other contained DNA substrate (Fig. 3A), all at twice the final reaction concentration. Both samples were in reaction buffer, and the drive syringes contained the same solution. Loaded reactants were allowed to equilibrate for 90 s to reach 37°C. Reactions were initiated by mixing of 45 l of each sample and were quenched at the desired times by mixing with 90 l of 12 mM EDTA in 200 mM hydroxide. Quenched aliquots were then neutralized and analyzed using the DNA glycosylase-coupled assay as described above.
To ensure that enzyme activity was retained during the time required to set up and then carry out typical reactions, we performed preincubation controls for AlkB. For all single-and multiple-turnover AlkB studies, the preparation time was less than 30 min, and reactions proceeded for a maximum of 30 min, so a 90-min preincubation time was tested. We compared otherwise identical multiple-turnover reactions of 1mA⅐T-25mer using AlkB that had not been preincubated, AlkB that had been preincubated at 37°C for 90 min in reaction buffer and salt only, and AlkB that had been preincubated at 37°C for 90 min in all reaction components except DNA. The reaction progress curves were indistinguishable (Fig. 5C), indicating that AlkB is stable for at least 90 min in our reaction conditions before DNA substrate is added.

Competition assays to determine relative k cat /K m values for AlkB
The relative specificity of AlkB for dsDNA substrates containing ⑀A and 1mA was determined using competition assays, with the substrate ⑀A⅐T-19mer used as a reference (Table S2). As above, reactions were carried out using OA2 at 37°C in reaction mixtures containing 25 mM HEPES (pH 7.5), 100 mM NaCl, 0.1 mg/ml BSA, 1 mM TCEP, 2 mM sodium L-ascorbate, 1 mM 2OG, and 40 M ammonium iron (II) sulfate. Reactions initially contained 5 nM AlkB, 200 nM ⑀A⅐T-19mer, and 200 nM either 1mA⅐T-25mer or ⑀A⅐T-25mer, and dealkylation activity was followed up to a maximum of 10% repair of either substrate. Samples were quenched and analyzed using the DNA glycosylase-coupled assay described above. Initial velocities are proportional to relative k cat /K m values as described in Equation 3 (35).

Data availability
All data are contained within the article and accompanying supporting information.