Differential Incorporation and Removal of Antiviral Deoxynucleotides by Human DNA Polymerase γ*

Mitochondrial toxicity can result from antiviral nucleotide analog therapy used to control human immunodeficiency virus type 1 infection. We evaluated the ability of such analogs to inhibit DNA synthesis by the human mitochondrial DNA polymerase (pol γ) by comparing the insertion and exonucleolytic removal of six antiviral nucleotide analogs. Apparent steady-stateK m and kcat values for insertion of 2′,3′-dideoxy-TTP (ddTTP), 3′-azido-TTP (AZT-TP), 2′,3′-dideoxy-CTP (ddCTP), 2′,3′-didehydro-TTP (D4T-TP), (-)-2′,3′-dideoxy-3′-thiacytidine (3TC-TP), and carbocyclic 2′,3′-didehydro-ddGTP (CBV-TP) indicated incorporation of all six analogs, albeit with varying efficiencies. Dideoxynucleotides and D4T-TP were utilized by pol γ in vitro as efficiently as natural deoxynucleotides, whereas AZT-TP, 3TC-TP, and CBV-TP were only moderate inhibitors of DNA chain elongation. Inefficient excision of dideoxynucleotides, D4T, AZT, and CBV from DNA predicts persistencein vivo following successful incorporation. In contrast, removal of 3′-terminal 3TC residues was 50% as efficient as natural 3′ termini. Finally, we observed inhibition of exonuclease activity by concentrations of AZT-monophosphate known to occur in cells. Thus, although their greatest inhibitory effects are through incorporation and chain termination, persistence of these analogs in DNA and inhibition of exonucleolytic proofreading may also contribute to mitochondrial toxicity.

More than 36 million people are infected by the human immunodeficiency virus worldwide, where 5.3 million new infections occurred during 2000 (1). Although antiviral therapy effectively extends the life of individuals, the death toll continues to rise; 3 million people, the highest number since the epidemic began, died from AIDS in 2000 (1). Nucleoside analogs utilized in antiviral therapy are readily incorporated into DNA by the HIV-1 1 reverse transcriptase. Although viral replication is effectively inhibited by DNA chain terminators, cellular side effects also result. The continuous antiviral therapy required to keep the HIV infection under control has increased the chance for severe antiviral analog induced toxicity. Current antiviral nucleoside analog therapy against HIV clearly results in compromised mitochondrial function due to inhibition of the mitochondrial DNA polymerase (2,3).
AZT was the first analog to be approved for anti-HIV therapy in 1985. In 1990, Dalakas et al. (4) were the first to report mitochondrial myopathies in HIV-infected individuals undergoing AZT treatment. Control studies thereafter demonstrated that these induced myopathies, most notably visualized histologically as ragged red fibers, were indeed caused by AZT treatment and were not a consequence of the HIV infection (5). This study revealed reduced amounts of mitochondrial DNA in AZT-treated skeletal muscle (5). Further clinical evidence has demonstrated that mitochondrial myopathy slowly and cumulatively develops during AZT treatment (6).
The second class of antiviral nucleoside analogs approved for HIV therapy are the dideoxynucleoside analogs ddI and ddC. These chain terminators also cause toxic side effects by inhibiting mitochondrial function. The use of 2Ј-3Ј-dideoxycytidine (ddC) causes a reversible peripheral neuropathy in many patients (7). Treatment of human Molt-4 cells with ddC results in delayed cytotoxicity with a concomitant loss of mitochondrial DNA (8), indicating the cellular target is likely mitochondrial DNA replication. Treatment of human CEM cells with ddC, D4T, and ddI results in a significant decrease of mtDNA and ultrastructural changes of the mitochondria (9). Both AZT and ddC treatment result in depletion of mitochondrial DNA, and both drugs have been shown to cause an increase in mtDNA deletions (10).
Mitochondrial DNA is replicated by an assembly of proteins and enzymes including DNA polymerase ␥, single-stranded DNAbinding protein, DNA helicase, multiple transcription factors, and a number of accessory proteins (11). In vitro analysis from several laboratories has demonstrated that among the cellular replicative DNA polymerases, the mitochondrial DNA polymerase ␥ is the enzyme most sensitive to the antiviral nucleotide analogs currently approved to control HIV-1 infection (12)(13)(14)(15)(16)(17)(18)(19)(20)(21).
As a thymidylate analog, AZT-TP is a competitive inhibitor for dTTP with pol ␥ (16). Partially purified human DNA pol ␥ is strongly inhibited by dideoxynucleotide triphosphates and D4T-TP, whereas AZT-TP, 3TC-TP, and CBV-TP inhibit pol ␥ to a lesser but significant degree (13). Purified recombinant yeast pol ␥ can readily incorporate dideoxynucleotides and didehydroCTP, but this enzyme is less efficient in the incorporation of AZT (18). 3TC-TP has also been shown to be a substrate for human DNA pol ␥ as well as for HIV-RT (22). These results clearly show that pol ␥ is a primary cellular target for analog-induced mitochondrial toxicity. This acquired mitochondrial toxicity may be caused by 1) direct inhibition of DNA pol ␥ without incorporation, 2) chain termination by incorpo-ration of these analogs into mitochondrial DNA by DNA pol ␥, 3) alteration of the fidelity of DNA synthesis by pol ␥, 4) the persistence of these analogs in mtDNA due to inefficient excision, or 5) any combination thereof. To understand the mechanism of this acquired mitochondrial toxicity, a thorough understanding of the interaction of nucleotides and analogs with pol ␥ is needed. Such information may help in the design of nucleotide analogs that selectively inhibit the HIV reverse transcriptase without inducing mitochondrial dysfunction.
We and others have cloned and overproduced the catalytic subunit of human DNA polymerase ␥ in insect cells via a recombinant baculovirus (23)(24)(25). In this report, we have determined the insertion efficiency of the currently approved anti-HIV analogs into DNA by purified recombinant human DNA polymerase ␥, and we have investigated the efficiency of removing these analogs from DNA by the intrinsic 3Ј-5Ј exonuclease activity of pol ␥.
Incorporation of Antiviral Nucleotide Analogs into DNA by pol ␥-The kinetics of antiviral nucleotide analog insertion into DNA by Exo Ϫ pol ␥ was measured using the gel-based oligonucleotide extension assay (26,27) as modified for incorporation of antiviral nucleoside analogs (17). The primer-template sets used for each type of analog follow. The positions for analog insertion are in bold type. The gel-purified 18-mer was labeled at the 5Ј end with [␥-32 P]ATP and annealed in a 1:1.4 ratio to the respective templates in 10 mM Tris-HCl, pH 7.5, by heating to 90°C for 5 min followed by slow cooling to room temperature. One pmol of primer:template ( 32 P-end-labeled primer) was incubated with 1.4 ng of Exo Ϫ pol ␥ (10 fmol of enzyme) in a 10-l reaction containing 25 mM Hepes-OH, pH 7.5, 2 mM 2-mercaptoethanol, 0.1 mM EDTA, 5 mM MgCl 2 , 50 g/ml acetylated bovine serum albumin, and variable dNTP concentrations (3 nM to 1.0 mM). Reactions were incubated at 37°C for 10 min and stopped on ice by the addition of 10 l of formamide loading dye. The products were separated on a 15% polyacrylamide-urea gel and visualized as described above.
Inhibition of Single Nucleotide Incorporation by pol ␥ with Antiviral Nucleotide Analogs-The inhibition of a single nucleotide incorporation was performed with the primer-template sets described above in the same buffer conditions but with 100 nM [␣-32 P]dCTP, [␣-32 P]dTTP, or [␣-32 P]dGTP. Antiviral nucleotide analog triphosphate was added to these reactions as indicated. The products were separated on a 15% polyacrylamide-urea gel and visualized as described above. Ten fmol of 32 P-end-labeled 38-mer was added to all reactions and used to normalize the products from gel loading error.
Exonucleolytic Removal of Antiviral Analogs from the 3Ј-Termini of DNA-To produce single-stranded substrates for exonuclease assay with various antiviral analogs at the 3Ј terminus, the 18-mer primer was 32 P-end-labeled on the 5Ј termini with T4 polynucleotide kinase and then extended at the 3Ј terminus with the different analogs using terminal deoxynucleotidyltransferase (TdT) in one-Phor-all buffer (Amersham Pharmacia Biotech). The products of the reaction were desalted and purified on a Sephadex G-25 column followed by centrifugation in a Microcon-3 microconcentrator. Purified HIV-RT was used to label the 3Ј termini of the 18-mer in the 18/36-mer duplex with analog by incubating with 1 mM of the indicated analog triphosphate for 1 h at 37°C. The reactions were heat-inactivated and duplex DNA purified on a Sephadex G-25 spin column and washed in a Microcon-3 microconcentrator with six volumes of distilled H 2 O. To examine pol ␥ exonuclease activity with these analog-containing primer-templates, 0.2 pmol of 32 P-end-labeled oligonucleotide containing the designated analog at the 3Ј end was incubated with 70 -840 fmol of wild type pol ␥, as indicated. The exonuclease activity was carried out in 33 mM Hepes-OH, pH 7.5, 13 mM KCl, 1.3 mM DTT, and 3.3 mM MgCl 2 at 37°C for 30 min. The products were separated by denaturing PAGE and visualized as described above.
Inhibition of Exonuclease Activity of pol ␥ by AZT Mono-and Diphosphate-To determine the inhibition of pol ␥ exonuclease activity by AZT mono-and di-phosphate, 21 fmol of wild type pol ␥ was incubated with 0.5 pmol of 32 P-labeled 18-mer primer in a reaction containing 25 mM Hepes-OH, pH 7.5, 50 g/ml acetylated bovine serum albumin, 2 mM 2-mercaptoethanol, 0.01 mM EDTA, and 10 mM MgCl 2 at 37°C for 15 min in the presence of nucleotide mono-or diphosphate, as indicated. The reaction was stopped by heat denaturation and separated on a gel as described above.

RESULTS
We sought to identify the mechanisms by which AZT-TP, ddCTP, 3TC-TP, D4T-TP, and carbovir-TP inhibit the human DNA polymerase ␥. All of these analogs lack the 3Ј hydroxyl group and consequently act as chain terminators once incorporated into DNA. For comparison and reference the structures of these analogs are shown in Fig. 1. We used the purified recombinant human DNA polymerase ␥ overproduced in baculovirusinfected insect cells. This recombinant DNA polymerase ␥ and an exonuclease-deficient catalytic subunit has been characterized previously in our laboratory and shown to possess polymerase properties identical to the native catalytic subunit of DNA polymerase ␥ (24). To simplify the analysis of analog incorporation into DNA without the complication of proofreading, we generated a histidine-tagged exonuclease-deficient DNA polymerase. The specific polymerase activity of both the wild type and exonuclease-deficient histidine-tagged pol ␥ was 32 units/ng in the poly(rA)/oligo(dT) assay (data not shown). We first addressed the inhibition and incorporation of these antiviral nucleotide analogs into DNA, and then we tested the efficiency of excising analogs from DNA by the 3Ј-5Ј exonuclease activity.
Inhibition of Human DNA pol ␥ by Antiviral Nucleotide Analogs-As a first approximation, the IC 50 values for inhibiting DNA synthesis were determined with two different assays. We designed these assays to specifically measure the ability of analogs to inhibit the incorporation of the cognitive nucleotide. First, the incorporation of a single normal ␣-32 P-labeled dNMP into an 18/36-mer primer-template was assayed in the presence of increasing concentrations of competing antiviral nucleotide analog. Inhibition was monitored by gel electrophoresis and quantified to determined the IC 50 concentrations. Graphical results for all five analogs are shown in Fig. 2A. These results demonstrated that both dideoxycytidine and D4T-TP had strong inhibition profiles, whereas AZT-TP, 3TC-TP, and CBV-TP showed modest inhibition. The IC 50 for ddNTP and D4T-TP was 8 and 20 M, respectively, while 3TC-TP and CBV-TP had an IC 50 of 80 M. The analog AZT-TP had an IC 50 of 130 M.
The severity of inhibition is better demonstrated in the second assay, which measures multiple incorporation events. Inhibition by the thymidine analogs was determined here in our standard poly(rA)⅐oligo(dT) assay using 25 M dTTP. On this substrate pol ␥ has a K m for ddTTP of 4.5 M (28). Results are shown in Fig. 2B and demonstrate that ddTTP and D4T-TP are potent inhibitors in vitro while AZT-TP required higher levels to inhibit pol ␥. Addition of AZT-TP resulted in an IC 50 of ϳ25 M, which was also the concentration of normal dTTP in this assay. Dideoxythymidine triphosphate and D4T-TP showed IC 50 at ϳ15 and 150 nM, respectively, more than 2 orders of magnitude lower than AZT-TP. Given this relative ranking as inhibitors, we wanted to determine the mode of inhibition for each analog. We specifically wanted to determine whether chain termination was the primary mechanism of inhibition or whether inhibition of polymerase activity could occur prior to incorporation of the analog into DNA. Additionally, once incorporated into DNA how efficiently could the analogs be removed by the 3Ј-5Ј exonuclease activity of pol ␥?
Incorporation of Antiviral Nucleotide Analogs into DNA by DNA Polymerase ␥-To determine relative efficiencies with which these analogs could be incorporated into DNA, we performed primer extension reactions and analyzed the products by gel electrophoresis. We used the exonuclease-deficient DNA polymerase ␥ in this assay to avoid degradation of the primer by the proofreading function and to simplify interpretation of results. This strategy became imperative due to the relatively high amount of enzyme and the longer incubation times required to detect incorporation with some of these analogs. Fig.  3 depicts the incorporation of dTMP, ddTMP, AZT-MP, and D4T-MP into DNA. Rate was determined as the fraction of primer extended by one nucleotide per unit time, and Michaelis-Menten kinetic constants were determined by plotting the rate as a function of nucleotide analog concentration (26). Human pol ␥ displayed high affinity (low apparent K m ) for normal nucleotides (Tables I-III), which is in agreement with other kinetic studies of pol ␥ (13)(14)(15)18). All the analogs could be incorporated into our DNA substrate, but different concentrations of the analog were required. The dideoxynucleotide analogs ddCTP and ddTTP were the easiest to incorporate and had k cat values similar to their normal nucleotide counterpart (Tables I-II). The apparent K m was 2-5-fold higher than the normal nucleotide. The effect of these analogs on competitive incorporation can be assessed by taking the ratio of the kinetic constants, (f in ). This value is equivalent theoretically to measurements made using competing substrates. Thus, ddTTP would get incorporated one in four incorporation events if the concentration of TTP and ddTTP were equal (Table I). The apparent K m for D4T-TP incorporation was similar to dideoxynucleotides but had a slightly decreased k cat (Table I).
These results predict D4T to be as inhibitory as the dideoxynucleotides. The f in values for these analogs determined with pol ␥ followed the general trend of inhibition observed in Fig. 2. Pol ␥ exerted most of its discrimination through K m effects, whereas the k cat was only modestly reduced for most of these analogs. Incorporation of AZT-TP, 3TC-TP, and CBV-TP required much higher concentrations than ddNTP or D4T-TP. However, apparent K m values were still in the micromolar range (Tables I-III) dideoxynucleotides and D4T. The dCTP analog, 3TC-TP, had the lowest k cat as well as a high apparent K m (Table II).
Exonucleolytic Removal of Antiviral Analogs from DNA by DNA Polymerase ␥-The inhibitory effect of a chain terminator is limited by its ability to persist in DNA once incorporated. The persistence of all of these analogs in DNA has largely been ignored in the literature. This concept is critical to the understanding the toxicity because, even though many of these analogs are only moderate inhibitors in vitro and are not readily incorporated, their resistance to exonucleolytic removal increases their ability to thwart DNA replication and presumably cause cytotoxicity. Since DNA polymerase ␥ has an intrinsic 3Ј-5Ј exonuclease function, we investigated whether human DNA polymerase ␥ could remove these antiviral nucleotide analogs from DNA termini. Single-stranded DNA substrates bearing the analog at the 3Ј terminus were constructed with TdT. Since TdT did not effectively add AZT-MP to the ends of DNA we used HIV-1 reverse transcriptase to insert AZT-MP onto the 3Ј end of the 18/36-mer substrate. The exonuclease activity by pol ␥ on this dsDNA substrate was compared with the degradation of the normal 18/36-mer substrate, as well as a 19/36-mer dsDNA bearing either D4T-MP or ddTMP. At an equal molar ratio of wild type pol ␥ and either ssDNA or dsDNA, we observed efficient exonucleolytic removal of normal nucleotides, but very little detectable removal of the analogs with the exception of 3TC. At this stoichiometric level of polymerase and substrate, 10 -20% of the 3TC was removed from the 3Ј termini as compared with the control. Pol ␥ did not remove detectable amounts of the other analogs at these enzyme concentrations (data not shown). However, when enzyme concentrations exceeded substrate concentrations, we detected removal of the analogs from the 3Ј terminus (Fig. 4A). A 1.3fold molar excess of enzyme was only able to remove 10% or the terminally incorporated analogs for single or double-stranded substrate. The exception was 3TC, where Ͼ50% of the analog was removed in 30 min (Fig. 4B). The remaining terminally incorporated analogs required a 3-4-fold molar excess of pol ␥ to remove Ͼ50% in 30 min.
AZT-monophosphate Inhibits pol ␥ Exonuclease Function-Through uptake and phosphorylation, AZT-MP is known to accumulate in millimolar concentrations in cells (29,30). Since normal deoxynucleoside monophosphates inhibit the 3Ј-5Јexonucleolytic activity by product inhibition, we tested the ability of the mono-and diphosphate forms of AZT to inhibit the exonuclease activity of pol ␥. Inhibition of exonucleolytic digestion of the 18-mer by increasing concentrations of TMP, AZT-MP, TDP, or AZT-DP was monitored by gel electrophoresis. The fraction of 18-mer was plotted for the indicated concentration of each analog (Fig. 5). The TMP inhibited the human pol ␥ exonuclease activity at similar concentrations to what has been observed for the inhibition of Drosophila pol ␥ by AMP (31). AZT-MP and TMP inhibited pol ␥ exonuclease at similar concentrations, indicating an IC 50 of 2.5 mM. AZT-DP and TDP caused some inhibition of exonuclease but only at high concentrations. AZT-TP and TTP did not inhibit the exonuclease activity in this assay (data not shown). Thus, the inhibition of exonuclease activity was specific to either dTMP or AZT-monophosphate and occurred at concentrations similar to those observed in vivo (29,30). DISCUSSION Antiviral nucleoside analogs have been implicated to cause mitochondrial toxicity in patients being treated for the HIV-1 viral infection. The cause of this toxicity is the inhibition or perturbation of mitochondrial DNA synthesis. We sought to determine the mode of inhibition of DNA polymerase ␥, the only known polymerase in animal cell mitochondria. We found that all of the currently approved antiviral nucleoside analogs that we tested were incorporated into DNA by pol ␥, and all inhibited DNA synthesis by pol ␥ in vitro to varying degrees. The dideoxynucleoside triphosphates and D4T-TP were incorporated most readily into DNA by pol ␥ and also exerted the strongest inhibition.
Pol ␥ was able to incorporate the analogs in the following order of efficiency: ddNTP Ͼ D4T-TP Ͼ CBV-TP Ͼ 3TC-TP Ͼ AZT-TP. Our data indicate that 3TC-TP was one of analogs least likely to be incorporated and yet was one of those most efficiently removed. Taken together, this may explain the low  0.003, 0.01, 0.03, 0.1, and 0.3 M dTTP; 0, 0.01, 0.03, 0.1, 0.3, 1, and 3 M ddTTP or D4T-TP; or 0, 10, 30, 100, 300, 438, and 657 M AZT-TP in lanes 0 -6, respectively, at 37°C for 10 min. The products were separated on 15% polyacrylamide-urea gel and quantitated on a Molecular Dynamics PhosphorImager as described under "Experimental Procedures." a f in ϭ (k cat /K m (TTP))/(k cat /K m (analog)). a f in ϭ (k cat /K m (dCTP))/(k cat /K m (analog)). mitochondrial toxicity induced by 3TC in vivo. The low incorporation of 3TC may also be attributed to its existence as the (Ϫ)-isomeric form. Our data also suggest that AZT-TP is the least likely to be incorporated into DNA by pol ␥ but once incorporated, it was not efficiently removed from DNA. The inefficiency of pol ␥ to remove AZT from DNA may help to explain the AZT-induced mtDNA depletion observed in vivo.
The largest determinant in pol ␥ discrimination against these analogs, especially AZT, was through apparent K m effects with only modest changes in k cat . Although the exact level of AZT-TP in mitochondria is unclear, it appears that the concentration required for incorporation can be obtained in vivo (29,30). Exonucleolytic Removal of Analogs from DNA-Human DNA polymerase contains an intrinsic 3Ј-5Ј exonuclease active site in the 140-kDa polypeptide (23,24). DNA polymerase ␥ is a family A polymerase, which is best typified by the Escherichia coli pol I (32). Mutagenesis studies and the three-dimensional structure of the E. coli pol I exonuclease active site show that the 3Ј-OH group of the terminal nucleotide plays a key role in the exonucleolytic mechanism by forming a hydrogen bond with a glutamic acid side chain (Glu-200 in human pol ␥) facilitating this residue to makes an ionic bond to the catalytic Mg 2ϩ and a hydroxide anion (33)(34)(35). Stabilized by the metal cation, the activated hydroxide anion attacks the phosphodiester bond, inverting the configuration of the phosphate to release of the terminal nucleoside monophosphate. We sought to determine if the absence of the hydroxyl group in these chain terminators would affect their excision. We found that the pol ␥ exonuclease was inefficient at removing these chain terminators at molar equivalents and removal required a 3-4-fold molar excess of polymerase over 3Ј termini. Excision of ddNMP and other analogs from the 3Ј terminus by yeast and porcine DNA pol ␥ has also been shown to be inefficient when enzyme is the limiting component in the reaction (18,36). The exception was the excision of 3TC, which was only 2-fold less efficient than a normal nucleotide with our highly active pol ␥. Gray and colleagues (22) have also demonstrated that 3TC-MP can be removed from the 3Јterminus of DNA, but only after 2 h of incubation. 3TC is in the (Ϫ)-enantiomer form; therefore, the possibility exists for the ribose oxygen to maintain hydrogen bonding with the active site glutamic side chain through a water molecule, accounting for its efficient excision. Due to the role of the 3Ј OH in the exonucleolytic catalysis, the inefficient excision of chain terminators by polymerases with intrinsic exonucleases may be a general phenomena. Indeed, the nuclear DNA polymerases ␦ and ⑀ have also been shown to remove AZT-MP inefficiently from the 3Ј termini (19). Clearly, the 3Ј-azido group in AZT cannot substitute for the 3Ј-OH group in H-bonding Glu-200. The monophosphate inhibition experiment shown in Fig. 5 suggests, at least with AZT, that these analogs bind with similar affinity in the exonuclease active site. Thus, the 3Ј-azido group may interfere sterically with the ability of FIG. 4. Exonucleolytic removal of antiviral analogs from the 3 termini of DNA. A, gel image of the pol ␥ exonuclease reactions that test for the removal of antiviral nucleotide from the 3Ј end. Radiolabeled 19-mer containing the 3Јterminal analog as indicated in the figure was incubated with wild type pol ␥ for 30 min. Substrate was prepared using TdT and exonuclease reactions performed as described under "Experimental Procedures." The different substrates (175-220 fmol) were incubated with 0, 70, 140, and 280 fmol of wild type pol ␥ for the 18-mer control substrate (lanes 0 -3 for 18-mer, respectively), or 0, 280, 560, and 840 fmol of wild type pol ␥ for the 3Ј-terminal analog substrates (lanes 0 -3 under the indicated analog substrate, respectively). B, graph of the relative efficiency of exonucleolytic removal of antiviral analogs from the 3Ј terminus of DNA with 280 fmol of enzyme. Open bars represent the exonuclease activity on the single-stranded oligonucleotide substrates, and filled bars represent the exonuclease activity with the double-stranded primer-template substrates. Activities on the singlestranded and double-stranded primertemplates were normalized to wild type pol ␥ activity with the control 18-mer or 18/36-mer primer-template. 100% activity for wild type with either double-stranded or single-stranded substrates represents 152 fmol of 3Ј terminus degraded with 280 fmol of wild type enzyme in 30 min.
Glu-200 to hydrogen bond with the attacking OH anion, and provide added resistance to exonucleolytic cleavage.
Besides the antiviral nucleoside triphosphates, the precursor forms of these analogs have also been implicated in altering cellular DNA replication (37,38). Furman et al. measured the intracellular concentrations of AZT and the various phosphorylated forms. They found that AZT accumulates at high level in the monophosphate form (Ͼ1 mM), whereas the concentration of the triphosphate was found to be only 2 M (29, 30). We found that AZT-MP inhibited the exonuclease function as efficiently as its normal counterpart, dTMP. This was surprising, since AZT was the least favored substrate to be removed from DNA, and suggests that free AZT-MP or terminally incorporated AZT binds readily in the exonuclease active site preventing efficient catalysis. Thus, intracellular levels of AZT-MP are expected to inhibit the exonuclease activity of mitochondrial DNA polymerase and possibly lower its fidelity. Although AZT-MP has been shown to inhibit the exonuclease function of DNA polymerase ␦ (38) and SV40 replication in vitro (37), it was not shown to increase the mutation frequency in the SV40 replication-fidelity assay (37). However, this assay was not specific for pol ␥ and did not score mutations in mitochondrial DNA. Inhibition of pol ␥ proofreading by monophosphates or inactivation of exonuclease activity results in as much as a 20-fold decrease in replication fidelity in vitro (39,40). 2 In vivo, the exonuclease-deficient yeast and mouse DNA polymerase ␥ transgenes have conferred a mutator phenotype in mitochondrial DNA (41,42). Thus, inhibition of pol ␥ exonuclease function by AZT-MP is likely to result in mutations within the mitochondrial DNA. Mutations in mitochondrial DNA cause a wide range of mitochondrial diseases due to the resulting de-fects in oxidative phosphorylation (43,44). When oxidative phosphorylation is disrupted electrons can leak into the mitochondria matrix, react with oxygen, forming reactive oxygen species (45). We find it intriguing that both an increase in reactive oxygen species and oxidative DNA damage have been noted in patients treated with AZT (46,47). AZT incorporation and chain termination may also produce reactive oxygen species as a result of mtDNA depletion and loss of oxidativephosphorylation function. Interestingly, AZT has been shown to be mutagenic in animal and cellular models (48 -54). Future studies of the fidelity of pol ␥ in the presence of nucleoside monophosphate analogs may offer insight into the mechanism of AZT induced mitochondrial toxicity seen in treated patients.
Relevance to Clinical Symptoms and Observed Mitochondrial Toxicity-Acquired mitochondrial toxicity in patients taking antiviral nucleoside analogs is generally accepted to occur as a consequence of incorporation into mitochondrial DNA and/or inhibition of mitochondrial DNA replication (55). This toxicity requires five steps to present a clinical phenotype: uptake of analog into the cell, conversion to the triphosphate form, transport into the mitochondria, incorporation into mitochondrial DNA, and persistence in mitochondrial DNA. The transport of these analogs into the mitochondria may occur before or after phosphorylation, but one study suggests that intramitochondrially phosphorylated analogs are preferentially incorporated into DNA (56). In contrast to this finding, another study suggests that ddC is phosphorylated in the cytoplasm and transported into mitochondria prior to exerting its inhibitory effect on mtDNA synthesis (57). Our present study focused only on the latter steps leading to toxicity: the incorporation, inhibition, and persistence in DNA. If all of the analogs we studied showed equivalent cellular adsorption, phosphorylation and transport, then we could draw a direct correlation with cellular toxicity. However, the intramitochondrial concentration and phosphorylation state of these analogs is unclear. One of these analogs, 3TC, may not be transported efficiently into the mitochondria and may even block transport of other dCTP analogs (58 -61). For carbovir, no mitochondrial toxicity has been observed in tissue culture cells (62), and it remains to be seen whether it will cause any mitochondrial toxicity in patients. Our results suggest that the observed AZT toxicity in vivo may be the combined effect of moderately efficient incorporation and very inefficient removal, resulting in persistence in mtDNA. Additionally, a high in vivo concentration of AZT-MP may inhibit pol ␥ proofreading, thereby causing an increase in mtDNA mutations. Our data suggest that, unlike AZT, the cytotoxicity from dideoxynucleosides and D4T is primarily due to incorporation and persistence in mtDNA. FIG. 5. Inhibition of the pol ␥ exonuclease activity by AZT monophosphate. Activity was graphed relative to no added mono-or diphosphate. The exonuclease activity of pol ␥ was determined by incubation of 0.5 pmol of 32 P-end-labeled 18-mer with 21 fmol of pol ␥ and increasing amount of nucleoside mono-or diphosphate. 100% activity represented 0.4 pmol of 3Ј end removed in 15 min with no added monoor diphosphate. Open and filled squares depict the inhibition with TMP and TDP, respectively; open and filled circles depict the inhibition with AZT-MP and AZT-DP, respectively. The concentrations of nucleoside mono-or diphosphate are indicated in the x axis.