The Local Dinucleotide Preference of APOBEC3G Can Be Altered from 5′-CC to 5′-TC by a Single Amino Acid Substitution

Abstract

APOBEC3A and APOBEC3G are DNA cytosine deaminases with biological functions in foreign DNA and retrovirus restriction, respectively. APOBEC3A has an intrinsic preference for cytosine preceded by thymine (5′-TC) in single-stranded DNA substrates, whereas APOBEC3G prefers the target cytosine to be preceded by another cytosine (5′-CC). To determine the amino acids responsible for these strong dinucleotide preferences, we analyzed a series of chimeras in which putative DNA binding loop regions of APOBEC3G were replaced with the corresponding regions from APOBEC3A. Loop 3 replacement enhanced APOBEC3G catalytic activity but did not alter its intrinsic 5′-CC dinucleotide substrate preference. Loop 7 replacement caused APOBEC3G to become APOBEC3A-like and strongly prefer 5′-TC substrates. Simultaneous loop 3/7 replacement resulted in a hyperactive APOBEC3G variant that also preferred 5′-TC dinucleotides. Single amino acid exchanges revealed D317 as a critical determinant of dinucleotide substrate specificity. Multi-copy explicitly solvated all-atom molecular dynamics simulations suggested a model in which D317 acts as a helix-capping residue by constraining the mobility of loop 7, forming a novel binding pocket that favorably accommodates cytosine. All catalytically active APOBEC3G variants, regardless of dinucleotide preference, retained HIV-1 restriction activity. These data support a model in which the loop 7 region governs the selection of local dinucleotide substrates for deamination but is unlikely to be part of the higher level targeting mechanisms that direct these enzymes to biological substrates such as HIV-1 cDNA.

Introduction

Human cells have the capacity to express up to nine enzymes with DNA cytosine to uracil (C-to-U) deaminase activity. Activation-induced deaminase (AID) deaminates immunoglobulin gene variable and switch region DNA cytosines to initiate somatic hypermutation and class switch recombination, respectively^1,2. Apolipoprotein B mRNA editing protein catalytic subunit 1 (APOBEC1) edits RNA cytosines, but it is also a potent DNA deaminase that may have biological roles in suppressing retroelement replication^3–9. Similarly, numerous studies combine to support a model in which the seven APOBEC3 (A3) enzymes have overlapping functions as innate immune effectors that prevent the replication of a broad number of endogenous and exogenous DNA-based substrates^10–14. For instance, many retroviruses including HIV-1 are susceptible to restriction by one or more of the A3s and, under some circumstances, single- and double-stranded DNA viruses may also be targeted^6,15–33.

Hundreds of studies over the past decade have shed light on the mechanism of HIV-1 restriction by APOBEC3G (A3G)^10–13. Current models posit that cytoplasmic A3G binds RNA and thereby associates with the viral protein Gag to preferentially access the cores of assembling viral particles. Then, in a Trojan horse type mechanism, encapsidated A3G inhibits virus replication in a susceptible target cell by physically interfering with the progression of reverse transcription and by deaminating nascent viral cDNA cytosines to uracils. As reverse transcription proceeds to completion, viral cDNA uracils template the insertion of genomic-strand adenines. This series of events explains the well-established phenomenon of G-to-A hypermutation. A3G is the only family member with a strong intrinsic preference for 5′-CC dinucleotides, and this property accounts for a significant fraction of observed HIV-1 hypermutations in vivo (i.e., 5′-GG-to-AG mutations)^15,34,35. APOBEC3D (A3D), APOBEC3F (A3F), and APOBEC3H (A3H) strongly prefer 5′-TC substrates, and these enzymes combine to explain the remaining fraction of hypermutations (i.e., 5′-GA-to-AA mutations^36,37).

Additional studies have revealed that APOBEC3A (A3A) and several other family members (but not A3G) have the capacity to trigger the clearance of naked foreign DNA from cells by a C-to-U deamination mechanism^38–41. In vivo, foreign DNA may come in a variety of forms, such as nuclear DNA from apoptotic or necrotic cells, mitochondrial DNA from organelle recycling, or microbial DNA from bacteria, viruses, and/or fungi. However, in cell culture, foreign DNA restriction can be modelled by transfecting plasmid DNA produced in E. coli into human cells and quantifying subsequent A3-dependent hypermutation and degradation^38,41. A3A is the most effective foreign DNA restriction enzyme by this assay and, like several enzymes described above (but not A3G), it strongly prefers 5′-TC single-stranded DNA substrates^23,38,41,42. A biological role in foreign DNA restriction is concordant with the fact that A3A is expressed exclusively in myeloid lineage cells such as macrophages, it is the most interferon-inducible of all of the human A3 family members, and it can accommodate both cytosine and 5-methyl-cytosine single-stranded DNA substrates^{23,38,41,43–45}. Thus, A3A appears specialized for biological function in foreign DNA clearance, although a role in virus restriction in myeloid cell types is also plausible^46,47.

The seven human A3 proteins can be divided into three phylogenetic subgroups based on conserved amino acids present in each enzyme’s catalytically active zinc-coordinating domain^48,49. A3A, A3B, and A3G have Z1-type active site domains, A3C, A3D, and A3F have Z2-type active site domains, and A3H is the lone enzyme with a Z3-type active site domain. AID and APOBEC1 comprise separate phylogenetic groups. Chimeras and mutants have been used to gain insights into the amino acids that govern the intrinsic dinucleotide specificity of several of these enzymes. Original studies by Langlois and coworkers created A3F mutants with altered dinucleotide deamination preferences and thereby implicated a number of amino acids including loop 7 residues⁵⁰. Kohli and coworkers grafted A3F and A3G loop 7 residues into AID and reported a corresponding transfer of dinucleotide deamination preferences⁵¹. Carpenter and coworkers and Wang and coworkers also implicated loop 7 by grafting sequences from A3s into AID^52,53. Additionally, Kohli and coworkers (in a subsequent paper) and Wang and coworkers showed that the resulting AID chimeras were functional for somatic hypermutation and class switch recombination, with correspondingly altered mutational patterns (i.e., 5′-purine-C preference of wildtype AID had become 5′-CC or 5′-TC)^53,54. These local substrate targeting data are concordant with a crystal structure of the distantly related deaminase TadA bound to its tRNA substrate, in which the analogous loop region interacts with the nucleotide immediately 5′ of the target adenosine^55,56.

Here, we take advantage of the fact that A3A and the active domain of A3G are closely related, belonging to the same phylogenetic Z1 subgroup, and yet still elicit distinct local dinucleotide preferences, preferring 5′-TC and 5′-CC, respectively. This differential substrate specificity was used as a phenotype to determine the amino acids responsible and, further, to ask whether local substrate preferences may be linked to biological activity. Guided by high-resolution structures of the A3G catalytic domain and a predicted structure of A3A, we constructed a series of A3G chimeras in which putative DNA binding loops were replaced with the corresponding regions of A3A. Biochemical activity assays and HIV-1 restriction experiments combined to show that loops 1 and 3 influence enzymatic activity and loop 7 alone governs 5′-nucleotide selection relative to the cytosine targeted for DNA deamination. Replacing a single amino acid in A3G, D317, with the corresponding residue in A3A, Y132, was sufficient to endow 5′-TC DNA deamination activity. In addition, regardless of local dinucleotide preference, all of the enzymatically active A3G mutants showed potent HIV-1 restriction activity indicating that the precise amino acid composition of these loop regions is dispensable for engaging physiological targets.

Results

Structure-guided construction of A3G chimeras with loops from A3A

To better understand the differential dinucleotide substrate specificities of A3G and A3A, ClustalW⁵⁷ was used to align the amino acid sequences of the catalytic domain of A3G (residues 197–384) and wildtype A3A (Fig. 1a and Supplementary Fig. S1). As noted previously^40,58,59, the greatest divergence between these two Z1-type DNA deaminases is concentrated in the putative DNA binding loop regions, here named loops 1, 3, and 7 to correspond with regions located between conserved α-helices and β-strands (Fig. 1a and Supplementary Fig. S1). The Loop 7 region of A3G has been strongly implicated in dinucleotide substrate selectivity (see Introduction), but the precise single amino acid determinants have yet to be determined and reconciled with structural information.

Fig. 1. Alignment and schematics of A3A, A3G, and A3G loop mutants.

(a) A schematic of A3A (red) and A3G C-terminal domain (ctd; blue) showing loop 1, loop 3, and loop 7 amino acids. A3G structural elements are labelled below for reference.

(b) Schematics of A3G and derivatives made by replacing loop regions with the structurally corresponding residues from A3A.

(c) Ribbon schematic of an NMR structure of A3A⁶¹ showing loop 1 in green, loop 3 in blue, and loop 7 in red. The active site zinc is depicted as a purple sphere.

(d) Ribbon schematic of a crystal structure of A3G-197-380⁶⁰ showing loop 1 in green, loop 3 in blue, and loop 7 in red. The active site zinc is depicted as a purple sphere.

Based on these alignments, we constructed a complete series of chimeras, in which loops 1, 3, and 7 of the A3G catalytic domain were replaced with the corresponding loop regions from A3A (loop sequences in Fig. 1a; construct schematics in Fig. 1b). DNA sequencing and immunoblotting were used to confirm the integrity of all constructs (below). Loops 1, 3, and 7 are located adjacent to the active site of A3G⁶⁰ and A3A⁶¹ (colored green, blue, and red, respectively; Fig. 1c and d).

Dinucleotide deamination preferences of A3G loop chimeras in cell extracts

Fluorescence-based biochemical assays provide robust measures of DNA deaminase activity^41,62,63 (schematic in Fig. 2a). The activity of A3G and its loop replacement derivatives was quantified by expressing each construct in 293 cells, preparing soluble lysates, and incubating each lysate with 5′-CC or 5′-TC containing single-stranded DNA substrates (primary deamination target underlined hereon; Fig. 2b). DNA cytosine deamination followed by uracil excision and hydrolytic cleavage results in the release of the fluorescent 6-FAM group from the TAMRA quench. Previous studies have shown that A3G has an intrinsic preference for 5′-CC, whereas A3A prefers 5′-TC^{15,23,34,35,38,41,42}. A3G catalytic activity was severely compromised when the analogous region from A3A was used to replace loop 1, resembling the phenotype of the catalytic mutant E259Q. In contrast, replacement of A3G loop 3 with the homologous A3A loop caused a consistent and significant increase in 5′-CC deaminase activity. Concordant with prior literature (see Introduction), placing A3A loop 7 into A3G changed the enzyme’s substrate preference to 5′-TC, albeit with lower overall activity. Since loop 3 from A3A made A3G more active and loop 7 altered its substrate specificity, we assessed the activity of the combined loop mutant. Interestingly, this double loop mutant had both new properties, showing enhanced 5′-TC deamination activity and undetectable 5′-CC editing activity (Fig. 2b). Immunoblots showed that all of A3G loop chimeras expressed similarly and that observed activity and substrate differences are likely due to the intrinsic properties of the proteins themselves (Fig. 2b).

Fig. 2. DNA deaminase activity of A3G loop constructs in 293 cell extracts.

(a) Schematic of the fluorescence-based ssDNA cytosine deamination assay. A3G deaminates C-to-U, UDG excises the U, NaOH breaks the phosphodiester backbone, and the 5′ fluorophore 6-FAM releases from the 3′ quench TAMRA. The resulting fluorescent read-out provides a direct measure of DNA deaminase activity because UDG and NaOH are provided in excess.

(b) A histogram reporting the relative activity of A3G and loop replacement derivatives on 5′-CC or 5′-TC single-stranded DNA substrates (blue and red bars, respectively). Immunoblots of A3G (anti-V5) and tubulin (anti-TUB) in the corresponding 293 lysates are shown below. Data are mean +/− standard deviation of three independent experiments.

HIV-1 restriction by A3G loop mutants

Multiple studies have shown that DNA deamination activity is required for full levels of HIV-1 restriction by human A3G^16,64–66. Single-cycle virus infectivity assays were therefore done in the presence of wildtype A3G or each loop mutant to ask whether altering the local deamination preference had an impact on the restriction activity of this enzyme. Immunoblotting showed that each of the V5-tagged constructs was similarly expressed in 293 cells and packaged into viral particles (Fig. 3a). Low ratios of A3G to proviral DNA were used to minimize potential deaminase-independent effects⁶⁷. For instance, transfection of 50 and 200 ng of wildtype A3G caused intermediate (67%) and strong (91%) levels of restriction, respectively, whereas equivalent amounts of the E259Q catalytic mutant had little effect. Interestingly, all of the loop mutants showed significant restriction activity ranging from modest (loop 1 and loop 1/7 constructs) to intermediate (loop 1/3) to near wildtype levels (loop 3, loop 7, loop 3/7, and loop 1/3/7 constructs) (Fig. 3a). These restriction capacities correlated with deaminase activity, as determined above using cell extracts and single-stranded DNA substrates. However, it is notable that these restriction assays appeared to provide more sensitive read-outs than cell extract activity assays because even the loop 1 construct was partially active and its activity could be boosted by combining it with loop 3.

Fig. 3. HIV restriction and hypermutation capabilities of A3G loop replacement constructs.

(a) A histogram reporting the average infectivity of HIV-GFP produced in the presence of 0, 50 or 200 ng of transfected plasmid expressing empty vector (EV), wildtype A3G (WT), catalytic mutant A3G (E259Q), or the indicated A3G loop replacements (n=3; mean and SD shown for each condition). Representative immunoblots are shown below for A3G (anti-V5) and tubulin (anti-TUB) expression in cell lysates and A3G (anti-V5) and Gag (anti-p24) in viral particles. Data are mean +/− standard deviation of three independent experiments.

(b) G-to-A mutation profiles of HIV proviruses from the infectivity experiment shown in ‘A’. Each tick represents a single G-to-A mutation, and colors depict dinucleotide context.

(c) A histogram summarizing the fraction of G-to-A mutations observed within each dinucleotide context (data from ‘b’).

(d) A histogram summarizing the observed G-to-A mutation frequency of the indicated constructs (data from ‘b’).

Next, we asked whether the local dinucleotide deamination preferences delineated using cell extracts would translate into similarly altered HIV-1 proviral DNA hypermutation patterns. This was done by PCR amplifying, cloning, and sequencing a 470 bp region of the GFP cassette from the integrated viruses that resulted from each single-round infectivity experiment. As expected from prior studies (see Introduction), the vast majority of viral plus strand mutations caused by A3G were 5′-GG-to-AG (87%) and a minority were 5′-GA-to-AA (12%) (Fig. 3b and c). Loop 1, loop 3, and loop 1/3 constructs also yielded over 50% 5′-GG-to-AG mutations. However, in stark contrast, loop 7, loop 1/7, loop 3/7, and loop 1/3/7 constructs all inflicted over 70% 5′-GA-to-AA mutations (Fig. 3b and c). The average G-to-A mutation loads were also interesting with a minimum of 6 mutations per kilobase (loop 1 construct) and a maximum of 46 mutations per kilobase (loop 3/7 construct), in comparison to wildtype A3G at 12 mutations per kilobase (Fig. 3d). Overall, these data confirm and extend prior studies by unambiguously demonstrating that loop 7 alone is responsible for selecting the 5′-base relative to the target cytosine. Our studies additionally show that DNA deaminase activity is influenced by loops 1 and 3 with negative and positive effects, respectively.

Activity of A3G loop 7 single amino acid substitution mutants

To further delineate the intrinsic determinants of dinucleotide selection, the six amino acids that differ between A3G and A3A in loop 7 were replaced individually. The resulting A3G constructs were expressed in 293 cells and lysates were prepared and used in a triplicate series of fluorescence-based activity assays as described above. Changing A3G D317 into the corresponding residue of A3A (Y132) proved particularly informative, as this substitution was the only one to cause A3G to adopt a more A3A-like 5′-TC preference (Fig. 4a). In comparison, four of the single amino acid substitution derivatives retained the intrinsic 5′-CC preference of wildtype A3G (Q318D, R320L, C321Y, Q322K) and one had no activity above background (G319P). Similar results were obtained by replacing individual loop 7 residues in the context of the hyperactive A3G loop 3 construct (Supplementary Fig. S2).

Fig. 4. Activities of A3G loop 7 single amino acid substitution mutants.

(a) A histogram reporting the relative activity of A3G and the indicated amino acid substitution mutants on 5′-CC or 5′-TC single-stranded DNA substrates (blue and red bars, respectively). Immunoblots of A3G (anti-V5) and tubulin (anti-TUB) in the corresponding 293 lysates are shown below. Data are mean +/- standard deviation of three independent experiments.

(b) A histogram reporting the average infectivity of HIV-GFP produced in the presence of 0, 50 or 200 ng of transfected plasmid expressing empty vector (EV), wildtype A3G (WT), catalytic mutant A3G (E259Q), A3G loop 7, or the indicated loop 7 amino acid substitution mutants (n=3; mean and SD shown for each condition). Representative immunoblots are shown below for A3G (anti-V5) and tubulin (anti-TUB) expression in cell lysates and A3G (anti-V5) and Gag (anti-p24) in viral particles. WT and D317Y data are the same as those in Fig. 4 and shown again here for comparison. Data are mean +/− standard deviation of three independent experiments.

(c) G-to-A mutation profiles of HIV proviruses from the infectivity experiment shown in ‘B’. Each tick represents a single G-to-A mutation, and colors depict dinucleotide context.

(d) A histogram summarizing the fraction of G-to-A mutations observed within each dinucleotide context (data from ‘c’).

(e) A histogram summarizing the observed G-to-A mutation frequency of the indicated constructs (data from ‘c’).

The importance of D317 for 5′-CC targeting was further tested through a series of HIV restriction and hypermutation experiments. As above for the whole loop constructs, all of the single amino acid substitution mutants were expressed and packaged similarly into viral particles (Fig. 4b). Moreover, apart from the inactive G319P construct, all of the proteins inhibited HIV-GFP infectivity similar to wildtype A3G (Fig. 4b). Viral DNA that had been produced in the presence of each of these constructs showed additional evidence for altered dinucleotide targeting preferences (Fig. 4c). Again, wildtype A3G and four of the mutant derivatives elicited less than 18% 5′-TC-to-TT mutations, which are reflected on the viral genomic strand as 5′-GA-to-AA mutations. In contrast, the D317Y construct inflicted over twice as many of this type of mutation (43%) (Fig. 4d). The lower activity of this single amino acid substitution mutant and incompletely transformed intrinsic dinucleotide preference in comparison to the entire loop 7 construct indicate that both the identity and the topology of other loop 7 amino acid residues are also contributing factors (compare dinucleotide hypermutation patterns in Figs. 3 and 4).

Insertion of other aromatic amino acids at A3G position 317 also skews deamination toward 5′-TC containing single-stranded DNA substrates

We next used HIV restriction and hypermutation experiments to test whether other aromatic amino acid substitutions at position 317 would also favor 5′-TC deamination. All of the constructs were expressed and packaged similarly and all inhibited the replication of Vif-deficient HIV-1 (Fig. 5a). Interestingly, each aromatic amino acid substitution construct showed a strong tendency for 5′-GA-to-AA hypermutation (Fig. 5b). The D317W construct in particular yielded a higher fraction of 5′-GA-to-AA hypermutations in comparison to the original D317Y construct, 60% versus 40%, respectively, suggesting that the larger aromatic amino acid may be even more favourable for 5′-TC deamination (Fig. 5c). This tendency also correlated with a slightly increased number of G-to-A mutations per kilobase (Fig. 5d). In terms of natural amino acid variation, it is important to note that human A3A and A3B have a tyrosine and other 5′-TC preferring deaminases, A3C, A3D, and A3F, have a phenylalanine at this position.

Fig. 5. Activities of A3G derivatives with aromatic amino acids at position 317.

(a) A histogram reporting the average infectivity of HIV-GFP produced in the presence of 0, 50 or 200 ng of transfected plasmid expressing empty vector (EV), wildtype A3G (WT), or the indicated amino acid substitution mutant (n=3; mean and SD shown for each condition). Representative immunoblots are shown below for A3G (anti-V5) and tubulin (anti-TUB) expression in cell lysates and A3G (anti-V5) and Gag (anti-p24) in viral particles. Data are mean +/− standard deviation of three independent experiments.

(b) G-to-A mutation profiles of HIV proviruses from the infectivity experiment shown in ‘A’. Each tick represents a single G-to-A mutation, and colors depict dinucleotide context.

(c) A histogram summarizing the fraction of G-to-A mutations observed within each dinucleotide context (data from ‘b’).

(d) A histogram summarizing the observed G-to-A mutation frequency of the indicated constructs (data from ‘b’).

Structural model for local dinucleotide selection

To investigate the effect of D317Y on A3G structure, we performed a series of all-atom explicit-solvent MD simulations. The overall structure of wildtype A3G was well maintained throughout the entire duration of each MD simulation (Supplementary Fig. S3). Interestingly, in two separate MD experiments, the D317 side chain formed a persistent hydrogen bond with the R320 backbone (Fig. 6). This interaction creates an α-helix cap, known to stabilize α-helices in other proteins^68–70. In contrast, this helical cap was unable to form in MD simulations of A3G D317Y. Instead, a local refolding event occurred in which the loop 7 region adopts a conformation that partly occludes the active site (Fig. 6b). Additionally, full or partial formation of a new 2-fold α-helix occurred in the loop 1 region during two separate MD simulations of A3G D317Y.

Fig. 6. Structural model for 5′-CC dinucleotide selection by A3G.

(a) Ribbon schematics of the wildtype A3G catalytic domain structure obtained from MD simulations of the apo system. The initial structure (gray) and the final snapshot (orange) of a 150 ns MD simulation are shown superimposed. Zinc is shown as a green sphere. D317 is depicted in stick representation, with carbon atoms in cyan and oxygen atoms in red.

(b) Ribbon schematics of A3G D317Y catalytic domain structure obtained in MD simulations of the apo system. The initial structure (gray) and the final snapshot (purple) of a 150 ns MD simulation are shown superimposed. Zinc is shown as a green sphere. Y317 is depicted in stick representation, with carbon atoms in cyan and oxygen atoms in red.

(c) A close-up view of the persistent helix-capping interaction between D317 and the backbone of R320 observed in MD simulations of the wildtype A3G in its apo form.

(d) Binding pose of a 5′-CC dinucleotide in the active site of wildtype A3G, based upon the final snapshot of a 400 ns MD simulation of 5′-CC-bound A3G. Nitrogen, oxygen and phosphorus atoms are depicted as sticks in blue, red and violet colors, respectively. Carbon atoms of the dinucleotide and the amino acids are shown in orange and cyan, respectively. Only the polar hydrogen atoms are depicted in light gray for clarity.

To further investigate and structurally rationalize the unique preference of A3G for 5′-CC substrates, we combined information from A3G catalytic domain crystal structure (3IR2), which lacks an active site ligand, and from a mouse cytidine deaminase crystal structure (2FR6), with the nucleobase positioned in the active site (Materials and Methods). The MD simulations starting from the modelled cytidine-bound A3G structure suggested that A3G has two distinct cytidine binding sites: one in a pocket near the catalytic glutamate (E259) and another in an adjacent pocket near C321 and other loop 7 residues (Supplementary Fig. S4). These preferred binding sites are consistent with the target cytosine needing to be positioned within the active site for deamination and the upstream cytosine needing to be accommodated nearby.

Combining the two preferred binding poses of cytidine, a 5′-CC dinucleotide was modelled into an MD-generated A3G catalytic domain structure. MD simulations of the 5′-CC-bound A3G catalytic domain structure showed that this dinucleotide adopted a binding pose that persisted for the entire 400 ns simulation time in two of the three experiments (Fig. 6d). In the third experiment, the same pose was maintained for about 300 ns and, after that period, the cytidine at position 0 came out of the pocket and sampled other conformations while the cytidine at position −1 remained bound. In contrast, no persistent nucleotide binding pose was observed in three separate MD simulations of A3G with 5′-TC.

Discussion

Originally suggested by E. coli mutagenesis experiments and subsequently demonstrated in several biochemical and biological assay systems, AID, APOBEC1, and various APOBEC3 proteins elicit distinct local DNA cytosine deamination preferences^3,71 (Introduction). A loop adjacent to the active site (loop 7 in most family members) has been strongly implicated in local selection of the nucleobase immediately 5′ of the target cytosine with, for instance, A3G preferring a 5′ cytosine and A3A a 5′ thymine^{15,23,34,35,38,41,42}. Here, a series of loop replacements was used to demonstrate that loop 7 alone is responsible for governing the intrinsic preference of A3G for 5′-CC dinucleotides. A full-length A3G variant with loop 7 from A3A loses 5′-CC deamination activity and gains a strong preference for 5′-TC substrates. Deaminase activity is enhanced and local preferences are maintained by simultaneous replacement of A3G loop 3 with the homologous region of A3A. A series of single amino acid exchanges in loop 7 led to the identification of A3G D317 as a major determinant of 5′ cytosine recognition. Replacement of D317 with tyrosine (naturally occurring in A3A and A3B loop 7), phenylalanine (naturally occurring in A3C, A3D, and A3F), or tryptophan causes A3G to adopt a strong preference for single-stranded DNA substrates with 5′-TC. Extensive multi-copy molecular dynamics simulations suggest a model in which 5′-CC dinucleotide-containing substrates are preferred by A3G because D317 acts as an α-helix cap, constraining the mobility of loop 7 and revealing a novel binding pocket near Cys321 that favorably accommodates cytosine, but not thymine or larger purine nucleobases. Stabilization of helices by helix dipole capping is well-established^68–70. MD simulations further indicate that the loss of the helix-capping residue in the A3G D317Y mutant may induce local refolding events in loops 1 and 7, which may ultimately combine to reshape the −1 nucleobase binding pocket. Thus, the single amino acid substitution D317Y appears to cause A3G to lose its unique 5′-CC binding site, modifying its loop regions to adopt a DNA binding pose similar to other APOBEC3 enzymes that prefer 5′-TC dinucleotide-containing substrates.

The full replacement of loop 7 of A3G with the homologous region of A3A caused full conversion to a 5′-TC preference, whereas the single amino acid substitution D317Y caused a strong but partial conversion (compare Figs. 3b and 4c). These observations suggest that at least one other residue in loop 7 and/or the overall topology of loop 7 also contributes to 5′ base selection, substrate binding, and ultimately target cytosine deamination. Multiple substrate and product bound crystal structures may be required to appreciate the full set of molecular contacts required for single-stranded DNA binding and target cytosine deamination.

Overall, the results with the fluorescence-based DNA C-to-U deamination assays correlated very well with the HIV-1 restriction and hypermutation data. However, some minor differences can be reconciled by the fact that the fluorescence-based assay has a relatively low signal to noise making weaker activities hard to differentiate. In contrast, the HIV restriction and hypermutation assays are more sensitive, but they must also be interpreted carefully because similar losses of infectivity can be obtained with different levels of hypermutation (e.g., wildtype versus loop 3/7 construct in Fig. 3). Part of this may be due to the fact that it theoretically takes only 1 base substitution mutation to inactivate GFP (the infectivity assay read-out), whereas sequencing provides a more direct measure of enzymatic activity.

It is noteworthy that at least two distinct classes of small molecules block A3G enzymatic activity through covalent attachment to Cys321, which is near Asp317 in loop 7^63,72. Select compounds defined by catechol or 4-amino-1,2,4-triazole-3-thiol moieties have the capacity to covalently bind Cys321 and block substrate single-stranded DNA cytosine from entering the active site. It is therefore possible that a deeper appreciation of the molecular ligands required for 5′ base selection may help inform the design of deaminase inhibitor compounds. Future applications for DNA deaminase inhibitors could include infectious disease, autoimmune disease, and possibly even cancer therapies. For instance, small molecules that block the activity of the 5′-TC genomic DNA deaminase A3B⁷³, which has a catalytic domain 92% similar to A3A, could have utility as adjuvants to existing chemotherapeutics by dampening the rate of tumor evolution and thereby decreasing the probability of a tumor becoming drug resistant.

Materials and Methods

Plasmid constructs

Wildtype A3A and A3G cDNA sequences used in this study match the GenBank reference sequences, NM_145699 and NM_021822, respectively. APOBEC3 expression constructs were expressed in pcDNA3.1 vectors with C-terminal V5 or MycHis tags. To prevent expression of the highly mutagenic human A3A protein in E. coli, an intron was inserted into A3A coding region between exons 2 and 3, as described³⁶. Loop graft variants were created by overlap extension PCR, and single amino acid substitution variants were made by QuikChange site-directed mutagenesis (Stratagene). Restriction digests and DNA sequencing were used to verify the integrity of all constructs. Primer sequences are available upon request.

DNA C-to-U activity assays

Semi-confluent 293T cells in 6 well plates were transiently transfected with 2 μg of the indicated constructs using TransIT transfection reagent (Mirus Bio). Two days post-transfection, cytoplasmic extracts were made by lysing the cells in GST lysis buffer (25 mM HEPES pH 7.4, 150 mM NaCl, 0.5% Triton X-100, 1 mM EDTA, 1mM MgCl₂,1 mM ZnCl₂,10% Glycerol, 50μg/ml RNase A, protease inhibitor). Clarified cell lysates were used for fluorescence-based single-stranded DNA cytosine deaminase activity assays using Förster resonance energy transfer (FRET) from a fluorescein (FAM) label to a carboxytetramethylrhodamine (TAMRA) fluorophore^62,63. Briefly, 20 μl of each cell lysate was incubated for 2 hrs at 37°C with 10 pmol of ssDNA oligonucleotide substrate 5′-FAM-AAACCCTAATAGATAATGTGA-TAMRA or 5′-FAM-ATAATCAAATAGATAAT-TAMRA (Biosearch Technologies, Inc.), 0.02 units of uracil DNA glycosylase (UDG; NEB), and 15 μl of 50 mM Tris-Cl (pH7.4),10 mM EDTA in Nunc 384-well black plates. Deamination of cytosine results in uracil, which is excised by UDG. The resulting abasic site is subjected to hydrolytic cleavage by adding 0.1 M NaOH, followed by mixing and incubating at 37°C for another 30 min. 3 μl of 4N HCl and 37 μl of 2M Tris-Cl (pH7.9) was then added for neutralization. Upon cleavage, the FAM and TAMRA labels are physically separated resulting in loss of FRET and increase in FAM signal. Activity was therefore quantified using a fluorescence plate reader with excitation at 490 nm and emission at 520 nm (Synergy Mx monochromator-based Multi-Mode Reader (BioTek Instruments Inc.) or LJL Analyst AD (LJL BioSystems Inc.).

Single cycle HIV-1 infectivity assays

293T cells were cultured in Dulbecco modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum and 0.5% penicillin-streptomycin. At 50% confluency, cells were transfected (TransIT; Mirus Bio) with 100 ng of pMDG(vesicular stomatitis virus G protein [VSV-G]-ENV), 300 ng of pCS-CG, and 300 ng of pΔNRF (gag-pol-rev-tat)¹⁹ along with gradient of either V5 tagged APOBEC3 (0, 50 and 200 ng) expression construct or with an appropriate empty vector using TransIT (Mirus Bio). After 48 hours, virus-containing supernatants were harvested and purified by 0.45 μm PVDF filters to remove any remaining producer cells. A portion of the clarified viral supernatants was used to infect fresh 293T target cells and immunoblotted for the incorporated A3. Transfection efficiency of the producer cells was measured by flow cytometry for GFP (FACSCanto II Ruo; BD Biosciences). Producer cell lysates were immunoblotted to measure protein expression levels. 72 hours post-infection, target cells were harvested and infectivity (GFP) was measured by flow cytometry. Data were analyzed using FlowJo flow cytometry analysis software, version 8.7.1. Quantification was done by first gating the live cell population, followed by gating on the GFP-positive cells. All experiments were performed in triplicate with means and standard deviations shown.

Immunoblotting

Transfected 293T cells were washed in phosphate-buffered saline (PBS) and lysed directly in 2.5 X Laemmli Sample Buffer (25mM Tris, pH 6.8, 8% glycerol, 0.8% SDS, 2% 2-mercaptoethanol, 0.02% bromophenol blue). The mixture was boiled at 95°C for 30 minutes. Virus-like particles were isolated from culture supernatants by purification through 0.45 μm PVDF filters (Millipore) followed by centrifugation (13,000 rpm for 2 hours) through a 20% sucrose cushion and lysis directly in 2.5X Laemmli sample buffer. Samples were separated by SDS-PAGE on 12.5% resolving gels (Bio-Rad Criterion) at 150 V for 90 min. Proteins were transferred to PVDF membranes (Bio-Rad) at 90V for 2 hours. Membranes were blocked in 4% milk in phosphate-buffered saline (PBS), 0.1% Tween-20 prior to overnight incubation with mouse anti-V5 (Invitrogen) to detect A3G-V5 (Invitrogen) or mouse anti-α-Tubulin (Covance), p24/capsid (NIH ARRRP 3537 courtesy of B. Chesebro and K. Wehrly). Anti-mouse horseradish peroxidase (HRP)-conjugated secondary antibodies (Bio-Rad) were detected using Hyglo HRP detection reagents (Denville Scientific). Blots were incubated in stripping buffer (0.2M glycine pH 2.2, 1.0% SDS, 1.0% Tween-20) before re-probing.

Proviral DNA sequence analysis

Genomic DNA was prepared from the HIV-1 infected 293T cell cultures using the DNeasy kit (Qiagen). A 712-bp amplicon from the GFP gene of integrated proviruses was amplified by Taq polymerase (Roche) for 20 cycles using degenerate primers 5′-ACTACAACARCCACAACRTCTATATC and 5′-AGGGTGTAAYAAGYGGGTGTTYT. Nested PCRs was done using a degenerate set of primers 5′-CTTCAARATCCRCCACAAC and 5′-GGAAGTAGYYTTGTGTGTGGTAGAT to amplify 514-bp amplicons. PCR products were separated on agarose gels and visualized by ethidium bromide staining. The amplicons were purified and cloned into pJET1.2 using the CloneJET PCR cloning kit (Fermentas). The amplicons were sequenced at the BioMedical Genomics Center at the University of Minnesota. Sequences were analyzed using Sequencher, version 4.6 (Gene Codes Corp.). Duplicate sequences potentially due to PCR treated were only analyzed once.

Molecular Dynamics Simulations

A3G catalytic domain molecular dynamics (MD) simulations were based on crystal structure 3IR2⁶⁰. Amino acid substitutions introduced for protein solubility and crystallization were reverted in silico and the intermolecular zinc atom was removed to obtain a virtual wildtype structure. For cytidine-bound A3G simulations, the binding pose of cytidine was modelled using mouse cytidine deaminase as a template (2FR6)⁷⁴ after which the distorted transition-state-like structure of cytidine was replaced with a geometry-optimized cytidine structure. The crystallographic water molecules were retained in each system solvated in a TIP3P water box with 10 Å buffer⁷⁵. The residues coordinating zinc were modelled using the cationic dummy atom method⁷⁶. Residues Glu254 and Glu259 that are in close proximity to the zinc site were protonated to comply with the suggestions of the cationic dummy atom model. The protonation states of other titratable residues were determined using the Whatif Web Interface⁷⁷. MolProbity web server was used to check and correct histidine, asparagine and glutamine side chain orientations⁷⁸. Systems were simulated at 200 mM salt concentration. Each solvated A3G structure consisted of about 31,000 atoms. Amber FF99SB force field was used to build the systems⁷⁹.

Each system was first relaxed using 36,000 steps of minimization and four consecutive restrained 250 ps MD simulations. Subsequently, two copies of 150 ns MD simulations were performed at 1 atm and 310 K using NAMD2.7 suite⁸⁰. This was done for each apo form of wildtype A3G, the apo form of A3G-D317Y, and the cytidine-bound form of wildtype A3G.

The two distinct binding modes of cytidine in our cytdine-bound MD simulations were combined in order to model 5′-CC and 5′-TC dinucleotides in the virtual wildtype A3G active site. For each of the dinucleotide-bound systems, three copies of 400 ns MD simulations were performed at 1 atm and 310 K using Amber12 suite (http://ambermd.org)⁸¹ on the GPU-based Keck II Center. RMSD plots indicate stability of the presented MD simulations (Supplementary Fig. S3).

Highlights.

• APOBEC3G is a DNA cytosine deaminase with an intrinsic 5′CC preference

• APOBEC3A is a DNA cytosine deaminase with an intrinsic 5′TC preference

• The intrinsic DNA deamination preference of both enzymes maps to loop 7

• APOBEC3G loop 7 D317 is required for 5′CC dinucleotide selection

• MD simulations suggest D317 stabilizes loop 7 and creates a pocket for binding 5′CC