Lack of Mutational Hot Spots during Decitabine-Mediated HIV-1 Mutagenesis

Abstract

Decitabine has previously been shown to induce lethal mutagenesis of human immunodeficiency virus type 1 (HIV-1). However, the factors that determine the susceptibilities of individual sequence positions in HIV-1 to decitabine have not yet been defined. To investigate this, we performed Illumina high-throughput sequencing of multiple amplicons prepared from proviral DNA that was recovered from decitabine-treated cells infected with HIV-1. We found that decitabine induced an ≈4.1-fold increase in the total mutation frequency of HIV-1, primarily due to a striking ≈155-fold increase in the G-to-C transversion frequency. Intriguingly, decitabine also led to an ≈29-fold increase in the C-to-G transversion frequency. G-to-C frequencies varied substantially (up to ≈80-fold) depending upon sequence position, but surprisingly, mutational hot spots (defined as upper outliers within the mutation frequency distribution) were not observed. We further found that every single guanine position examined was significantly susceptible to the mutagenic effects of decitabine. Taken together, these observations demonstrate for the first time that decitabine-mediated HIV-1 mutagenesis is promiscuous and occurs in the absence of a clear bias for mutational hot spots. These data imply that decitabine-mediated G-to-C mutagenesis is a highly effective antiviral mechanism for extinguishing HIV-1 infectivity.

INTRODUCTION

RNA viruses often mutate at high rates, ranging from 10⁻⁴ to 10⁻⁶ mutations/bp/cycle (1), enabling rapid adaption to constantly changing environments. However, rapid adaptability comes at a price—relatively small increases in the mutation rate can drive the viral population into error catastrophe, during which genetic information cannot be stably maintained and the viral population collapses (2–4). As a result, RNA viruses have been hypothesized to be susceptible to compounds that would elevate the viral mutation rate, a therapeutic strategy referred to as lethal mutagenesis (5). Lethal mutagenesis has now been demonstrated in cell culture with a wide variety of RNA viruses and mutagenic agents (6, 7). While experimental results in cell culture have been promising, the clinical translation of lethal mutagenesis has been a slow process due largely to a lack of strong lead compounds. Notably, ribavirin may inhibit hepatitis C virus in vivo at least in part through a lethal mutagenesis mechanism, but this topic remains controversial (8–11). The nucleoside analog KP-1212 has been shown to inhibit human immunodeficiency virus type 1 (HIV-1) replication by lethal mutagenesis in cell culture (12). However, in phase II clinical trials, KP-1461 (a prodrug of KP-1212) did not reduce viral loads and demonstrated only slight evidence of mutagenic effects despite sufficient bioavailability (13, 14). The nucleobase analog favipiravir (T-705) exerts broad-spectrum antiviral activity in cell culture and is currently being investigated in phase III clinical trials for influenza in the United States and in phase II clinical trials for Ebola virus in West Africa (15, 16). T-705 has been shown to elicit antiviral mutagenesis against influenza virus in cell culture and against norovirus in a mouse model of infection (17, 18). However, T-705 has also been shown to directly inhibit viral RNA polymerases (19–21); therefore, it remains unclear whether T-705 acts in vivo primarily as a lethal mutagen or as a direct inhibitor of viral replication.

Most lethal mutagens investigated thus far have been nucleobase or nucleoside analogs that base-pair promiscuously due to properties such as conformational flexibility, tautomerization, structural rearrangement, or ionization (7). For HIV-1 lethal mutagenesis, several active compounds have been identified, including decitabine (5-aza-2′-deoxycytidine), 5-azacytidine (5-AZC; the ribonucleoside equivalent of decitabine), 5,6-dihydro-5-aza-2′-deoxycytidine (KP-1212), and 5-hydroxy-2′-deoxycytidine (5, 12, 22, 23). Notably, HIV-1 lethal mutagenesis based on mutagenic nucleosides is similar to APOBEC3-mediated hypermutation that occurs naturally to restrict viral replication in the absence of Vif (24, 25). Thus, targeting the Vif-APOBEC3 interaction (or other necessary interactions for APOBEC3 degradation, such as between Vif and CBF-β) represents an additional strategy to achieve lethal mutagenesis, and this avenue is also actively being explored (26–29).

Decitabine is a particularly promising candidate for the lethal mutagenesis of HIV-1 because it exhibits potent antiviral activity (50% effective concentration [EC₅₀] of ≈200 nM) in the absence of significant cytotoxicity (22). Furthermore, decitabine synergizes with low, noncytotoxic concentrations of several antimetabolites, including gemcitabine and resveratrol (22, 30). The combination of decitabine and gemcitabine was also shown to exhibit potent antiretroviral activity in a murine leukemia virus (MLV)-based mouse model for AIDS (31). Although decitabine exhibits poor bioavailability when administered orally, a prodrug form called decitabine divalerate was recently synthesized that demonstrates improved stability and permeability while retaining anti-HIV-1 activity (32). Previous studies found that decitabine increased the mutant frequency of HIV-1 by inducing predominantly G-to-C transversions, which rarely occurred in the absence of the drug (22). However, the factors that determine the susceptibilities of individual sequence positions in HIV-1 to decitabine have not yet been analyzed in great depth. To examine this, we performed Illumina high-throughput sequencing of five amplicons prepared from proviral DNA that was recovered from untreated or decitabine-treated cells infected with HIV-1. We found that decitabine induced an ≈4.1-fold increase in the total mutation frequency of HIV-1, largely due to an ≈155-fold increase in the frequency of G-to-C transversions. Intriguingly, decitabine also resulted in an ≈29-fold increase in the frequency of C-to-G transversions. G-to-C frequencies varied significantly (up to ≈80-fold) between different sequence positions, but unexpectedly, mutational hot spots (defined as upper outliers within the distribution of mutation frequencies) were not observed. We also found that every single guanine position examined (134/134) and nearly every cytosine position examined (163/169) were significantly susceptible to the mutagenic effects of decitabine. Taken together, these observations show, for the first time, that decitabine-mediated lethal mutagenesis of HIV-1 is promiscuous and occurs without a clear bias for mutational hot spots. These data imply that decitabine-mediated mutagenesis is a highly effective strategy for extinguishing the infectivity of HIV-1 and support further preclinical evaluation of decitabine in HIV-1 infection, including studies in animal models.

MATERIALS AND METHODS

Production of viral stocks.

The production and titering of HIV-1 viral stocks for single-cycle infections and Illumina sequencing were previously described (33). Briefly, we used the envelope-deficient HIV-1 vector, pNL4-3 HIG (23), which contains a cassette inserted into nef that encodes heat-stable antigen (HSA), an internal ribosomal entry site (IRES), and enhanced green fluorescent protein (EGFP). Viral stocks were produced by cotransfecting 10 μg of pNL4-3 HIG and 5 μg of pNL4-3 Env per 10-cm plate of 293T cells. For each viral stock, five 10-cm plates were transfected, and the resulting supernatants were pooled and concentrated (≈10-fold) using 100,000 molecular weight cutoff (MWCO) filtration columns (Vivaproducts, Littleton, MA). Viral stocks were then treated with 10 U/ml of DNase I (New England BioLabs, Ipswich, MA) for 2 h at 37°C to degrade residual plasmid DNA from transfections. Lastly, the titers of the viral stocks were determined in U373-MAGI-X4 cells based on EGFP expression at low infectivities as described previously (33).

Preparation of genomic DNA from infected cells.

In order to prepare samples for Illumina sequencing, 1 × 10⁶ U373-MAGI-X4 cells were infected at a multiplicity of infection (MOI) of 1.0 as described previously (33). Cells were pretreated (or not pretreated) with 2 μM decitabine divalerate (an ≈EC₆₅ concentration) for 2 h before infection. Uninfected cells and cells infected with heat-inactivated viruses (i.e., virus stocks that were incubated at 65°C for 1 h) were included as negative controls. The medium was replaced 24 h postinfection, and the cells were harvested for genomic DNA extraction at 72 h postinfection. Notably, only a single round of replication can occur within this assay due to the lack of envelope expression in target cells. All infections were performed in triplicate, using independently produced viral stocks. Infections in the presence or absence of decitabine were performed using the same viral stock for each replicate. Genomic DNA was isolated from all collected cells using the High Pure PCR template preparation kit (Roche, Basel, Switzerland) following the manufacturer's instructions and eluted in 150 μl buffer. Genomic DNA was analyzed by quantitative PCR (qPCR) in order to quantify the amount of plasmid carryover from transfections and was found to be ≈0.2% of the proviral copy number, a level that would not significantly impact measured mutation frequencies.

Proviral DNA amplification and Illumina sequencing.

PCR was performed as described previously (33) to generate five small (150 to 170 bp) amplicons (Gag, Vif, HSA, EGFP-1, EGFP-2) for each sample, representing a mixture of viral and marker gene targets. All forward primers contained barcodes for later demultiplexing of samples. Primer and barcode sequences for the decitabine samples are listed in Table S1 in the supplemental material, while others have been previously reported (33). PCR was performed using Phusion Hot Start II high-fidelity DNA polymerase (Fisher Scientific, Pittsburgh, PA). PCRs were performed with 8 μl of genomic DNA per 50-μl reaction and cycling conditions of 98°C for 30 s, 30 cycles of 98°C for 10 s/56°C for 30 s/72°C for 15 s, and a final extension of 72°C for 10 min. All PCRs were performed in triplicate, pooled after amplification, and gel purified. Plasmid control amplifications from pNL4-3 HIG were performed in parallel under closely matched conditions to measure background error due to PCR and Illumina sequencing. For each sample, all amplicons were pooled in an equimolar fashion to normalize coverage between amplicons. Next, sequencing libraries were constructed from each sample (9 samples in total: 3 each for HIV-1, HIV-1 + decitabine divalerate, and HIV-1 plasmid control) using the TruSeq Nano DNA sample preparation kit and following the manufacturer's instructions. Libraries were quantified using the Qubit dsDNA BR assay kit, size-confirmed with Agilent DNA 1000 chips (Agilent Technologies, Santa Clara, CA), and pooled in an equimolar fashion to normalize library coverage. Before sequencing, a PhiX library was added in at 25% of total mass to improve sequence diversity. The pooled libraries were then subjected to 2 × 150 paired-end sequencing on the Illumina MiSeq.

Illumina sequencing data processing.

Illumina sequencing data were subjected to multiple read-level and base-level quality filtering steps as described previously (33). Samples were first demultiplexed based on perfectly matching barcodes. Read pairs with poor quality, improper mapping, or alignment to target amplicon sequences were excluded from analysis. After all filtering steps, we obtained ≈413,000 to 494,000 read pairs per sample (see reference 33; see also Table S2 in the supplemental material). We then used a custom algorithm based on the Genome Analysis Toolkit (GATK) walker framework (34) to identify mutations and calculate mutation frequencies for total mutations as well as every possible mutational subclass. Mutations and wild-type bases were required to be identified on forward and reverse reads, as read pairs were mostly overlapping. Mutations and wild-type bases were also required to have a Q-score of ≥30 on the two reads. Furthermore, background error hot spots were identified using the plasmid controls and masked prior to mutational analysis as described previously (33). Most (≈83%) of the mutations excluded by this process were G-to-T or C-to-A transversions, and only one of the excluded mutations was a G-to-C or C-to-G transversion. Mutation frequencies (defined as mutations per base pair [m/bp]) were calculated by dividing the number of mutations passing filters by all reference bases (mutations + wild-type bases) passing filters, combined across amplicons and separated by amplicon. We also determined the relative percentage of each mutational type (mutation spectra), numbers of mutants, dinucleotide contexts of mutants, and the mutation frequencies at individual sequence positions within the target amplicons.

Biostatistical analysis of Illumina DNA sequencing data.

To test for variables that may influence mutation frequencies, generalized linear mixed effects models were applied to the data that came from our Illumina data processing pipeline. The raw counts for each type of mutation (e.g., transitions) were modeled as Poisson random variables with an offset given by the total number of reference bases. The type of sample (i.e., HIV-1 without drug, HIV-1 + decitabine, and HIV-1 plasmid control), the type of amplicon, and their interactions were treated as fixed effects while the replicate was treated as a random effect. The logarithmic link was used as is standard for Poisson outcomes, and penalized quasi-likelihood was used to estimate the model parameters (35). These computations were conducted using R v3.1.0 and the MASS package. Mutation frequencies at individual sequence positions were compared using the Mantel-Haenszel test. All figures and tables were created in GraphPad Prism v6.0 (GraphPad Software, Inc., La Jolla, CA) or Microsoft Office for Mac 2011 v14.3.8 (Microsoft Corp., Redmond, WA).

Sequencing data accession number.

All Illumina sequencing data supporting the results of the manuscript have been deposited into the NCBI Sequence Read Archive (SRA) under accession code SRP059680.

RESULTS

Illumina sequencing of HIV-1 proviral DNA produced in the presence or absence of decitabine.

In order to investigate the impact of decitabine on HIV-1 mutagenesis, we prepared samples for Illumina amplicon sequencing as described previously (see Materials and Methods and reference 33). Briefly, we produced viral stocks by cotransfecting an envelope-deficient HIV-1 vector with an HIV-1 X4-tropic Env expression plasmid and then infected 1 million U373-MAGI-X4 cells at an MOI of 1.0. In this assay, only a single cycle of viral replication can occur, as neither producer cells nor target cells can be reinfected. U373-MAGI-X4 cells were pretreated (or not pretreated) with 2 μM (an ≈EC₆₅ concentration) of decitabine divalerate, a prodrug form of decitabine, for 2 h before infection. The medium was replaced 24 h postinfection, and cells were collected 72 h postinfection. Genomic DNA was purified from infected cells and used to prepare five amplicons by PCR, including viral (Gag and Vif) and marker (HSA, EGFP-1, and EGFP-2) gene targets. The EGFP-1 and EGFP-2 amplicons represent nonoverlapping portions of the egfp gene. Plasmid amplifications were included to measure the level of background error due to PCR and Illumina sequencing. Amplicons from the same sample were pooled and used to prepare sequencing libraries, which were then subjected to 2 × 150 paired-end sequencing on the Illumina MiSeq. In total, nine samples were analyzed, including three experimental replicates each of HIV-1, HIV-1 exposed to decitabine, and HIV-1 plasmid controls. For the decitabine samples, Illumina sequencing resulted in ≈425,000 read pairs/sample, representing ≈125,000 mutations and ≈45 million reference bases (wild-type bases + mutations) per sample after all bioinformatics processing steps (see Table S2 in the supplemental material). The Illumina sequencing data for the controls (HIV-1 without drug and HIV-1 plasmid control) were previously described elsewhere but were reanalyzed for the purposes of this study (33). Using this data, we determined mutation frequencies (expressed as mutations per base pair [m/bp]) for all samples, for total mutations and every possible mutational subclass (e.g., substitutions, transitions, transversions, etc.). The data were analyzed combined across all five amplicons and separated by amplicon.

Decitabine induces primarily G-to-C transversion mutations during HIV-1 replication.

We found that decitabine increased the total mutation frequency of HIV-1 from ≈6.9 × 10⁻⁴ m/bp to ≈2.9 × 10⁻³ m/bp (Fig. 1A), a difference of ≈4.1-fold (P < 0.001). Decitabine increased the transversion frequency of HIV-1 by ≈23-fold (P < 0.001), while transition frequencies were not significantly affected by the presence of the compound (P = 0.23). Upon dividing out the eight possible types of transversions, we found that decitabine increased the frequency of G-to-C transversions from 1.1 × 10⁻⁵ m/bp to 1.7 × 10⁻³ m/bp, a striking difference of ≈155-fold (P < 0.001). As a result, the HIV-1 mutational spectrum shifted significantly in the presence of the drug, with an increase in the relative percentage of G-to-C transversions from 2% to 57% (Fig. 1B). Intriguingly, we also observed an ≈29-fold increase (P = 0.003) in the frequency of C-to-G transversions in the presence of decitabine (Fig. 1A), which was not observed in a previous report analyzing the activity of decitabine against HIV-1 (22) (see Discussion). The relative percentage of C-to-G transversions increased from 2% in the absence of decitabine to 10% in the presence of decitabine (Fig. 1B). Decitabine-mediated G-to-C transversions have been previously hypothesized to result from incorporation into the minus strand viral DNA, followed by structural rearrangement and mispairing during plus strand synthesis (22), while C-to-G transversions may be caused by plus strand incorporation of decitabine or minus strand incorporation of decitabine decomposition products (see Discussion). The preponderance of G-to-C transversions over C-to-G transversions is consistent with the hypothesis that most of the decitabine incorporated into the plus strand viral DNA is removed by host repair processes (see Discussion).

FIG 1.

Decitabine primarily induces G-to-C transversion mutations during HIV-1 replication. (A) Mutation frequency analysis. Using Illumina sequencing data, mutation frequencies were calculated by dividing the number of mutations by the number of reference bases (mutations + wild-type bases) and are expressed as mutations per base pair (m/bp). Mutation frequencies were determined for HIV-1 in the presence or absence of decitabine (2 μM, an ≈EC₆₅ value) as well as for a plasmid control to determine the level of background errors. Fold increases in mutation frequencies in the presence of decitabine are indicated. (B) Mutation spectra analysis. Mutation spectra were determined by dividing the frequency of each type of mutation by the total mutation frequency, with the results expressed as a percentage of total mutations. (A, B) Data represent the mean of three experimental replicates, with error bars indicating the standard deviation. Asterisks denote statistically significant differences between HIV-1 and HIV-1 + decitabine (**, P < 0.01; ***, P < 0.001; N.S., not significant). The actual numbers of read pairs, mutations, and reference bases are listed in Table S2 in the supplemental material and in a previous report (33).

Most decitabine-induced G-to-C and C-to-G mutant read pairs have relatively low mutational loads.

Decitabine may lead to high levels of G-to-C and C-to-G transversions by inducing high numbers of mutations in relatively few read pairs or by inducing low numbers of mutations in many read pairs. To address this, we determined the number of G-to-C or C-to-G mutations per mutant read pair (Fig. 2). We found that 12% of all read pairs (≈149,000/1,272,000) contained at least one G-to-C transversion in the presence of decitabine (Fig. 2A), whereas 3% (≈33,000/1,272,000) contained at least one C-to-G transversion (Fig. 2B). As expected, the total numbers of G-to-C and C-to-G mutants were much higher in the presence of decitabine than those in its absence. Most of the decitabine-induced mutant read pairs contained a single G-to-C or C-to-G transversion, with single mutants representing ≈71% of all G-to-C mutants and ≈83% of all C-to-G mutants. However, decitabine treatment also led to a significant number of read pairs with multiple G-to-C or C-to-G transversion mutations. While most of these multiple-mutated sequences contained two or three mutations, we identified rare sequences that were excessively mutated, with as many as 13 G-to-C or 9 C-to-G transversions in a single read pair. Decitabine-induced mutants with multiple G-to-C or C-to-G transversions accounted for ≈53% of all G-to-C transversions and ≈33% of all C-to-G transversions, indicating their important contribution to decitabine-mediated HIV-1 mutagenesis. Notably, given the small size of Illumina read pairs (≈120 bp after processing), the full-length proviral genomes (≈11.2 kb including marker genes) likely contained much higher levels of G-to-C and C-to-G transversions in the presence of decitabine.

FIG 2.

Most decitabine-induced G-to-C and C-to-G mutant read pairs have relatively low mutational loads. For all G-to-C (A) or C-to-G (B) mutants, the numbers of G-to-C or C-to-G mutations per read pair were determined in order to evaluate mutational loads. The means, medians, and ranges are also indicated. Note that due to the short length of the Illumina read pairs (≈120 bp after processing), the total numbers of G-to-C and C-to-G mutations in the full-length viral genome may be much higher. The figure represents compiled data across three experimental replicates.

All amplicons are susceptible to decitabine-mediated HIV-1 mutagenesis.

In order to further examine the distribution of decitabine-induced mutations, we analyzed the mutational data separately for each of the five amplicons (Gag, Vif, HSA, EGFP-1, and EGFP-2). We hypothesized that differences in amplicon primary sequences, nucleic acid folding, or positioning within the viral genome may potentially lead to various susceptibilities to the drug. We found that all five amplicons were highly sensitive to decitabine-induced G-to-C transversions (P < 0.001), with increases ranging from 77-fold (Gag) to 248-fold (EGFP-2) (Fig. 3A). The EGFP-1 and EGFP-2 amplicons were most sensitive to G-to-C transversions, in terms of absolute G-to-C frequencies and in terms of fold induction. The absolute G-to-C frequencies were 2- to 3-fold higher for these amplicons than for the Gag, Vif, and HSA amplicons (P < 0.001). Likewise, all five amplicons were significantly susceptible to decitabine-mediated C-to-G transversions (P < 0.001) with increases ranging from 15-fold (EGFP-2) to 45-fold (Gag) (Fig. 3A). The absolute frequencies of C-to-G transversions varied between amplicons up to ≈2.1-fold (Gag versus EGFP-2). Small differences in the susceptibilities of the five amplicons to G-to-C or C-to-G transversions may be due to the varying nucleotide content of the amplicon sequences. We found that guanine content varied from 20% to 31% while cytosine content varied from 15% to 43% between amplicons. To account for these differences, we divided G-to-C and C-to-G transversion frequencies by the guanine or cytosine content of each amplicon, respectively, and normalized the resulting data to the amplicon with the highest transversion frequency (EGFP-1 for G-to-C and Vif for C-to-G) (Fig. 3B). After normalization, the general trends remained unchanged: EGFP-1 and EGFP-2 demonstrated ≈2.0- to 2.5-fold higher G-to-C frequencies than Gag, Vif, and HSA while, conversely, EGFP-1 and EGFP-2 demonstrated the lowest C-to-G frequencies.

FIG 3.

All amplicons are susceptible to decitabine-mediated HIV-1 mutagenesis. (A) Analysis of mutation frequencies separated by amplicon. G-to-C and C-to-G transversion frequencies were determined in all five amplicons subjected to Illumina sequencing. The EGFP-1 and EGFP-2 amplicons represent nonoverlapping portions of the egfp gene. Fold increases in mutation frequencies in the presence of decitabine are indicated. (B) Analysis of amplicon mutation frequencies normalized for variations in amplicon nucleoside content. To account for varying nucleoside content across amplicons, G-to-C and C-to-G transversion frequencies were divided by the proportion of guanines or cytosines, respectively, within each amplicon. These values were then normalized to the amplicon with the highest G-to-C or C-to-G frequency (EGFP-1 and Vif, respectively). (A, B) Data represent the mean of three experimental replicates, with error bars indicating the standard deviation. Asterisks denote statistically significant differences between HIV-1 and HIV-1 + decitabine (***, P < 0.001).

Decitabine-induced mutations in HIV-1 are not caused by DNA methyltransferase-mediated hydrolysis.

Decitabine-induced mutations in genomic DNA have been previously found by one group to be heavily biased toward CpG dinucleotides, suggesting that DNA methyltransferases (DNMTs) were involved in their mechanism of formation (36). DNMTs were proposed to attack the C6 position of decitabine, leading to ring opening and the formation of a premutagenic lesion capable of base-pairing with cytosine. However, decitabine is also known to be highly unstable in aqueous solution, undergoing rapid spontaneous hydrolysis and deformylation into a variety of ring-opened derivatives (37–39). To determine the mechanism of hydrolysis most relevant to the antiviral activity of decitabine, we analyzed the dinucleotide context of all G-to-C and C-to-G transversions that occurred in the presence of the drug. We found that only 19% of G-to-C transversions occurred within CpG dinucleotides (as opposed to TpG, ApG, and GpG dinucleotides [sites of mutations are underlined]) (Fig. 4). In particular, after taking into account the various frequencies of the dinucleotides within the amplicon sequences, there was no significant bias of G-to-C transversions toward any particular dinucleotide (see Fig. S1 in the supplemental material). Likewise, only 11% of C-to-G transversions occurred within the CpG dinucleotide context (as opposed to CpA, CpC, and CpT dinucleotides [sites of mutations are underlined]) (Fig. 4). Even after taking into account the lower frequency of CpG dinucleotides within the amplicon sequences, only 18% of C-to-G transversions occurred within the CpG context, although a possible bias toward CpT dinucleotides was observed (see Fig. S1 in the supplemental material). Taken together, these data support the hypothesis that most decitabine-mediated transversions in HIV-1 are due to spontaneous hydrolysis, deformylation, and mispairing rather than DNMT-mediated hydrolysis.

FIG 4.

Decitabine-induced mutations in HIV-1 are not caused by DNA methyltransferase-mediated hydrolysis. Decitabine-induced mutations in eukaryotic genomic DNA can either be biased toward CpG dinucleotides, implicating formation via DNA methyltransferases (36), or not be biased toward CpG dinucleotides, implicating spontaneous structural rearrangement (43). In order to investigate the mechanism most important for decitabine-mediated HIV-1 mutagenesis, the dinucleotide contexts of all G-to-C and C-to-G transversions observed in the presence of decitabine were determined. Note that G-to-C transversions likely arise from decitabine incorporation in place of deoxycytidine in the minus strand viral DNA, such that a 5′-CpG-3′ context on the minus strand corresponds to a 5′-CpG-3′ context on the plus strand (sites of mutation are underlined). The data represent the mean of three experimental replicates.

Lack of decitabine-induced G-to-C transversion hot spots.

In order to investigate whether certain sequence positions were more susceptible to decitabine-mediated mutagenesis than others, the G-to-C frequency was determined at every individual guanine position (134 in total), and the C-to-G frequency was determined at every individual cytosine position (169 in total). G-to-C frequencies were highly variable, ranging from ≈2.0 × 10⁻⁴ m/bp to ≈1.7 × 10⁻² m/bp (a median of 7.1 × 10⁻³ m/bp), a maximal difference of ≈83-fold (Fig. 5A). However, we failed to identify G-to-C mutational hot spots, which we defined as upper outliers within the mutation frequency distribution using the 1.5 × interquartile range (IQR) rule. For C-to-G transversions, we also identified a wide range of frequencies in the presence of decitabine, varying from 0 to 3.5 × 10⁻³ m/bp (median of 9.0 × 10⁻⁴ m/bp) (Fig. 5B). Using the 1.5 × IQR rule, we identified four decitabine hot spots for C-to-G transversions, all located within the Gag and Vif amplicons. We next examined the locations and coding effects of the most frequently mutated guanines and cytosines in order to determine whether they occurred in specific amplicons and to gauge the effects of the mutations on gene function. We found that the 20 most frequent decitabine-induced G-to-C transversions all occurred within the EGFP-1 and EGFP-2 amplicons (Table 1), which is consistent with the higher G-to-C frequencies observed earlier in these amplicons (Fig. 3). Most (18/20) of these G-to-C transversions would lead to nonsynonymous missense mutations, and notably G-to-C mutations cannot introduce premature stop codons. However, decitabine induced a wide array of semiconservative and nonconservative amino acid changes that may potentially disrupt protein function. In contrast to G-to-C transversions, the most common C-to-G transversions were located in the Gag and Vif amplicons (Table 2), which is in agreement with the higher C-to-G frequencies observed in these amplicons (Fig. 3). Although C-to-G mutations may potentially introduce stop codons, the most frequent C-to-G transversions did not result in stop codons but instead introduced missense mutations. Overall, these findings demonstrate that (i) decitabine-induced G-to-C transversion mutation frequencies vary widely among sequence positions but without clear mutational hot spots and that (ii) decitabine triggers HIV-1 lethal mutagenesis mainly through missense mutations.

FIG 5.

Lack of decitabine-induced G-to-C transversion mutation hot spots. In order to determine whether certain sequence positions in HIV-1 were more susceptible to decitabine than others, the G-to-C frequency at each individual guanine position (134 in total [A]) and the C-to-G frequency at each individual cytosine position (169 in total [B]) were determined. The box plots indicate the observed minima, maxima, and medians. We also investigated whether decitabine induced mutational hot spots, which we defined as upper outliers within the mutation frequency distributions using the 1.5 × interquartile range rule. Decitabine was not found to induce any G-to-C mutational hot spots, although several C-to-G mutational hot spots (denoted by ×) were observed. The data represent the mean of three experimental replicates.

TABLE 1.

Locations and effects of common decitabine-induced G-to-C transversions in HIV-1^a

Rank	Positionb	Amplicon	Frequencyc	Mutation typed	Amino acid change
1	324	EGFP-1	1.65E−02	N	K108N
2	417	EGFP-1	1.56E−02	N	E139D
3	383	EGFP-1	1.49E−02	N	G128A
4	608	EGFP-2	1.45E−02	N	S203T
5	427	EGFP-1	1.43E−02	N	E143Q
6	416	EGFP-1	1.41E−02	N	G139A
7	429	EGFP-1	1.40E−02	N	E143D
8	388	EGFP-1	1.34E−02	N	D130H
9	415	EGFP-1	1.33E−02	N	G139R
10	368	EGFP-1	1.32E−02	N	R123P
11	606	EGFP-2	1.29E−02	S	NAe
12	555	EGFP-2	1.29E−02	N	Q185H
13	552	EGFP-2	1.25E−02	N	Q184H
14	352	EGFP-1	1.22E−02	N	D118H
15	350	EGFP-1	1.21E−02	N	G117A
16	626	EGFP-2	1.19E−02	N	S209T
17	534	EGFP-2	1.18E−02	N	Q178H
18	624	EGFP-2	1.17E−02	S	NA
19	592	EGFP-2	1.16E−02	N	D198H
20	615	EGFP-2	1.15E−02	N	Q205H

^aThe data indicate the positions, frequencies, and coding effects of the 20 most common G-to-C transversions observed in HIV-1 exposed to decitabine, although these were not technically defined as mutational hot spots by our statistical criterion.

^bAll numbers refer to the position relative to the start of the egfp gene.

^cThe G-to-C transversion frequency (mutations/bp) was determined at every individual guanine position, though only the top 20 are shown here. The data represent the average of three experimental replicates.

^dN, nonsynonymous mutation; S, synonymous mutation.

^eNA, not applicable.

TABLE 2.

Locations and effects of common decitabine-induced C-to-G transversions in HIV-1^a

Rank	Positionb	Amplicon	Frequencyc	Mutation typed	Amino acid change
1*	438	Gag	3.50E−03	S	NAe
2*	446	Gag	3.30E−03	N	P149R
3*	392	Vif	3.00E−03	N	P131R
4*	431	Vif	3.00E−03	N	S144C
5	452	Gag	2.70E−03	N	T151S
6	379	Vif	2.60E−03	N	R127G
7	412	Gag	2.60E−03	N	L138V
8	388	Gag	2.40E−03	N	Q130E
9	356	Vif	2.30E−03	N	A119G
10	369	Vif	2.30E−03	S	NA
11	437	Gag	2.20E−03	N	A146G
12	398	Gag	2.20E−03	N	P133R
13	414	Gag	2.20E−03	S	NA
14	415	Vif	2.10E−03	N	H139D
15	118	HSA	2.00E−03	N	P40A
16	368	Vif	2.00E−03	N	T123S
17	325	Vif	2.00E−03	N	L109V
18	445	Gag	2.00E−03	N	P149A
19	160	HSA	1.90E−03	N	L54V
20	461	Gag	1.90E−03	N	A154G

^aThe data indicate the positions, frequencies, and coding effects of the 20 most common C-to-G transversions observed in HIV-1 exposed to decitabine, although only the top four of these (those marked by asterisks) were mutational hot spots, defined as upper outliers within the mutation frequency distribution using the 1.5 × interquartile range rule.

^bAll numbers refer to the position relative to the start of the indicated gene (gag, vif, or hsa).

^cThe C-to-G transversion frequency (mutations/bp) was determined at every individual cytosine position, though only the top 20 are shown here. The data represent the average of three experimental replicates.

^dN, nonsynonymous mutation; S, synonymous mutation.

^eNA, not applicable.

Virtually all guanine and cytosine positions are susceptible to decitabine-mediated mutagenesis.

Intriguingly, we observed that the minimal G-to-C and C-to-G frequencies in the presence of decitabine were similar to the highest G-to-C and C-to-G frequencies in the absence of drug (Fig. 5), suggesting that most guanines and cytosines are susceptible to decitabine-mediated mutagenesis. In order to more thoroughly examine this possibility, we determined the fold difference (decitabine/no drug) in G-to-C or C-to-G transversion frequency at every individual guanine or cytosine position, respectively. Fold differences were not calculated for a few positions (1 of 134 for G-to-C and 6 of 169 for C-to-G) because they were not mutated at all in the absence of decitabine. We found that G-to-C transversion frequencies increased by ≈15- to ≈1,010-fold (median of ≈178-fold) in the presence of decitabine (Fig. 6). Likewise, C-to-G transversion frequencies increased by ≈3- to ≈152-fold (median of ≈27-fold) in the presence of decitabine. In order to determine whether these increases were statistically significant, we compared G-to-C and C-to-G frequencies in the presence or absence of decitabine at every possible sequence position using the Mantel-Haenszel test (see Materials and Methods). We found that 100% of the guanine positions (134/134) were significantly susceptible to decitabine-mediated G-to-C transversions (P < 0.001). Similarly, 96% of the cytosine positions (163/169) were significantly susceptible to decitabine-induced C-to-G transversions (P < 0.001). Overall, these data suggest that virtually all guanines and cytosines within the HIV-1 genome are susceptible to the mutagenic effects of decitabine, although full genome sequencing will be required to address this in more detail (see Discussion).

FIG 6.

Virtually all guanine and cytosine positions are susceptible to decitabine-mediated mutagenesis. In order to further investigate the susceptibilities of individual nucleotides in HIV-1 to decitabine, fold differences in mutation frequencies (decitabine/no drug) were calculated at every possible sequence position. Fold differences were not calculated for a small number of positions (1 of 134 for G-to-C and 6 of 169 for C-to-G), as they were not mutated in the absence of decitabine. Mutation frequencies at individual sequence positions were also compared statistically using the Mantel-Haenszel test (see Results).

DISCUSSION

In this study, we have shown that decitabine is strongly mutagenic to HIV-1, inducing ≈4.1-, 23-, and 155-fold increases in total, transversion, and G-to-C mutation frequencies, respectively (Fig. 1). We also observed an ≈29-fold increase in C-to-G transversions in the presence of decitabine, which was not found in a previous report on the mutagenic and antiviral effects of decitabine (22). These discrepancies are likely due to differences in the sequencing methodologies applied. Previously, cells harboring mutant proviruses (HSA+, EGFP−) were isolated, and the mutated marker gene (egfp) was amplified and analyzed by Sanger sequencing. In total, 82 mutations were identified, 31 (i.e., 38%) of which were G-to-C transversions. Here, we found that the frequency of C-to-G transversions in the presence of decitabine was ≈3.0 × 10⁻⁴ m/bp, while the G-to-C transversion frequency was ≈1.7 × 10⁻³ m/bp, a 5.5-fold difference (Fig. 1). However, we also found that G-to-C mutations were overrepresented and C-to-G mutations were underrepresented in the EGFP-1 and EGFP-2 amplicons compared to the other amplicons we examined (Fig. 3). In these amplicons, the G-to-C transversion frequency was ≈12-fold higher than the C-to-G transversion frequency. Thus, on the basis of our results, we would expect only two to three C-to-G transversions to have been identified in the earlier study involving Sanger sequencing, a level that would likely be too low to result in a statistically significant increase in C-to-G transversions. Additionally, it is possible that decitabine and decitabine divalerate (the prodrug form used here) may exhibit small differences in mutagenicity due to the improved permeability and stability of decitabine divalerate (32).

The observation that decitabine induces primarily G-to-C transversions in HIV-1 is consistent with a previously proposed model (22) for decitabine, as follows. (i) Decitabine is incorporated in place of deoxycytidine during minus strand synthesis by the HIV-1 reverse transcriptase (RT). (ii) Decitabine undergoes structural rearrangement (hydrolysis and deformylation), resulting in a premutagenic lesion. (iii) The ring-opened derivative of decitabine mispairs with deoxycytidine during plus strand synthesis, leading to the fixation of G-to-C transversions. In contrast, while decitabine is presumably also incorporated during plus strand synthesis by the HIV-1 RT, the ring-opened derivatives of decitabine can be correctly removed and replaced by the host repair process after the import of viral DNA into the nucleus, preventing fixation of C-to-G transversions. However, we demonstrated here that decitabine increased the C-to-G transversion frequency of HIV-1 by ≈29-fold, clearly demonstrating that this process is more complex. While the mechanism responsible for decitabine-induced C-to-G transversions is presently unknown, one possible explanation is that they result from the incorporation of decitabine during plus strand synthesis, followed by structural rearrangement and inefficient repair by host processes. In the absence of repair by host processes, ring-opened derivatives of decitabine may lead to the fixation of C-to-G transversions upon replication of the eukaryotic genomic DNA during cell division or during PCR. In some cases, decitabine may be incorporated but not immediately undergo structural rearrangement, potentially evading host repair processes. However, the in vitro half-life of decitabine is quite short (≈11 h) under physiological conditions (39), and cells were not collected for genomic DNA extraction until 72 h postinfection. Nonetheless, the stability of decitabine may significantly vary between double-stranded DNA, single-stranded DNA, and unincorporated forms. In the future, it would be of great interest to better define the host repair processes and enzymes that act on decitabine and its derivatives, as such processes may greatly influence the mutagenicity of decitabine against HIV-1 proviral DNA and eukaryotic genomic DNA. Notably, 2′-deoxyriboguanylurea (dRGU), the primary decomposition product of decitabine, was recently shown to be mutagenic in human cells and found to base pair with deoxycytidine and deoxyguanosine in Taq-based PCRs (40). Thus, decitabine-mediated transversions (G-to-C and C-to-G) in HIV-1 may also result from direct incorporation of decomposition products by RT during minus strand synthesis. For example, C-to-G transversions may result from the incorporation of dRGU in place of deoxyguanosine during minus strand synthesis followed by base pairing with deoxyguanosine during plus strand synthesis. However, the ability of dRGU to directly induce the mutagenesis of HIV-1 has not yet been examined in cell culture or in assays with purified RT.

Decitabine is primarily used clinically for the management of myelodysplastic syndromes as an inhibitor of DNA methyltransferases (DNMTs), by which decitabine leads to the reactivation of aberrantly silenced genes (41). DNMTs can attack decitabine at the C6 position, resulting in the formation of trapped covalent complexes and depletion of functional DNMTs in the cell. In a previous study, decitabine was found to induce C-to-G and G-to-C transversions in eukaryotic genomic DNA mostly within CpG dinucleotides, suggesting that the mutations resulted from DNMT attack followed by hydrolysis and deformylation (36). However, other studies found that decitabine did not increase the mutation frequency of eukaryotic genomic DNA (42) or that decitabine was mutagenic through spontaneous rather than DNMT-mediated hydrolysis (43). We predicted that most G-to-C transversions would not be caused by DNMT attack, as the preponderance of G-to-C over C-to-G transversions suggests that they mostly result from decitabine incorporation and rearrangement during reverse transcription, which takes place in the cytoplasm. Nonetheless, DNMTs may still be involved in the formation of C-to-G transversions and of some G-to-C transversions. In this study, we found that there was no bias of G-to-C or C-to-G transversions toward CpG dinucleotides even after correcting for the various frequencies of dinucleotides in our amplicon sequences (Fig. 4; see also Fig. S1 in the supplemental material). These data support the hypothesis that most decitabine-induced mutations result from spontaneous hydrolysis and deformylation.

Upon examining the distribution of decitabine-mediated mutations, we observed significant variability between amplicons and between individual sequence positions (Fig. 3 and 5). The positions most susceptible to G-to-C transversions were located in the EGFP-1 and EGFP-2 amplicons, whereas the positions most susceptible to C-to-G transversions were located in the Gag and Vif amplicons (Tables 1 and 2). Notably, many previous studies on HIV-1 mutagenesis have reported wide variability in the distribution of mutations across the sequences of mutational targets, even in the absence of a mutagen (44–46). Unfortunately, the factors that create favorable positions for RT-mediated substitutions are still very poorly understood, although homopolymeric runs are known to trigger most insertions and deletions. We hypothesize that certain sequence positions were more susceptible to decitabine than others because (i) they occurred within primary sequence contexts that favored decitabine incorporation, (ii) they occurred within nucleic acid secondary structures that favored decitabine incorporation, or (iii) they occurred within regions of the HIV-1 genome more susceptible than others to decitabine-mediated mutagenesis. More specifically, the positioning of the amplicon within the viral genome impacts the amount of time the minus strand viral DNA remains single-stranded during reverse transcription, which has previously been proposed to drive differences in APOBEC3-mediated hypermutation across the genome (47, 48). Decitabine that is incorporated into regions of the genome that remain single-stranded longer than others may more efficiently undergo structural rearrangement prior to plus strand synthesis. The differences in G-to-C frequencies that we observed between amplicons are in general agreement with this hypothesis (Fig. 3). However, full genome sequencing will be required to determine the extent to which these factors can explain the variability of decitabine-mediated mutagenesis across amplicons and sequence positions.

In order to more thoroughly investigate the susceptibility of individual sequence positions to decitabine, we determined fold changes in mutation frequencies at individual positions and also evaluated the results statistically. Strikingly, we found that every guanine position examined (134/134) and nearly every cytosine position examined (163/169) were significantly susceptible to the mutagenic effects of decitabine, with increases in G-to-C and C-to-G transversion frequencies ranging from ≈15- to 1,000-fold and ≈3- to 150-fold, respectively (Fig. 6). These results were obtained by sequencing five different amplicons representing multiple genes, nucleotide contents, and positions within the genome. Thus, these findings support the hypothesis that the vast majority of guanines and cytosines within the HIV-1 genome are susceptible to decitabine-mediated mutagenesis, although full genome sequencing will be required to examine this possibility in more detail. Overall, these findings demonstrate for the first time that decitabine-mediated mutagenesis of HIV-1 is widely promiscuous and occurs in the clear absence of mutational hot spots. These data imply that decitabine-mediated G-to-C mutagenesis is a highly effective mechanism for extinguishing HIV-1 infectivity and encourage further evaluation of decitabine in animal models for the induction of HIV-1 lethal mutagenesis.