Oligomerization transforms human APOBEC3G from an efficient enzyme to a slowly dissociating nucleic acid binding protein

Abstract

The human APOBEC3 proteins are a family of DNA-editing enzymes that play an important role in the innate immune response and have broad activity against retroviruses and retrotransposons. APOBEC3G is a member of this family that inhibits HIV-1 replication in the absence of the viral infectivity factor Vif. Inhibition of HIV replication occurs by both deamination of viral single-stranded DNA and a deamination-independent mechanism. Efficient deamination requires rapid binding to and dissociation from ssDNA. However, a relatively slow dissociation rate is required for the proposed deaminase-independent roadblock mechanism in which APOBEC3G binds the viral template strand and blocks reverse transcriptase-catalyzed DNA elongation. Here we show that APOBEC3G initially binds ssDNA with rapid on-off rates and subsequently converts to a slowly dissociating mode. In contrast, an oligomerization-deficient APOBEC3G mutant did not exhibit a slow off rate. We propose that catalytically active monomers or dimers slowly oligomerize on the viral genome and inhibit reverse transcription.

TEXT

APOBEC3 proteins are DNA-editing enzymes that are part of the innate human immune response to viral pathogens, including retroviruses and retrotransposons^1–4. Of all the A3 proteins, A3G⁵ is the most potent inhibitor of HIV-1 replication^1,4,6, reducing viral infectivity by several orders of magnitude in the absence of the HIV-1 viral infectivity factor Vif^7–9. In fact, the function of Vif is to specifically counteract the antiviral activity of A3G^1,9.

Although A3G is the most studied of all the APOBEC proteins, the molecular mechanism for A3G-mediated HIV-1 restriction is still not fully understood. A3G is a deoxcytidine deaminase, which converts deoxycytidine bases in single-stranded DNA (ssDNA) to deoxyuridine^5,10–14. A3G deamination of minus-strand viral DNA formed during reverse transcription results in G to A hypermutation in the plus-strand^11,12,15, which effectively impairs viral replication. However, there are several lines of evidence that suggest that a deaminase-independent mechanism is also involved^1,6,20. First, A3G catalytic mutants retain antiviral activity^16–19. Second, A3G inhibits hepatitis B virus replication without G to A hypermutation²¹. Third, A3A inhibits LINE-1 and Alu retrotransposition^22–28 and parvovirus replication^24,29 independent of deaminase activity. Further evidence for a non-editing mechanism is based on the reduction of minus-strand viral DNA levels in HIV-1 particles during endogenous reverse transcription³⁰, inhibition of reverse transcriptase (RT)-catalyzed viral DNA elongation in vitro by catalytic A3G mutants^31,32, inhibition of strand transfer reactions in vitro and in cell-based assays^32–34, and A3G-induced inhibition of reverse transcription in viruses from human CD4⁺ T cells³⁵. A roadblock model, in which A3G molecules bind the template strand at one or a few locations and physically block viral DNA synthesis, has therefore been proposed as a molecular mechanism for deaminase-independent inhibition³².

Since only 7 (± 4) A3G molecules are incorporated into each vif-deficient virion³⁶, RT inhibition by an A3G roadblock requires a slow A3G off-rate from single-stranded nucleic acids. In contrast, these few A3G molecules must have fast on-off rates to deaminate up to 1000 sites in several minutes¹⁴ using a rapid search mechanism on viral ssDNA^37,38. To resolve this apparent paradox, we hypothesize that A3G exhibits fast binding kinetics as a monomer or dimer in order to function as an efficient enzyme, and slow kinetics upon oligomerization in order to block RT from elongating viral DNA. To test this idea, we used optical tweezers to monitor A3G binding kinetics on a single DNA molecule.

RESULTS

Single molecule measurements of A3G binding to ssDNA

For these studies, a single double-stranded λ-DNA molecule was tethered to two polystyrene beads, with one bead held in an optical trap and another on a micropipette tip. As the fixed bead is gradually moved away from the optically-trapped bead, the force on the DNA molecule is measured at each extension, yielding a force-extension curve (Fig. 1, solid black line). The solution surrounding the single DNA molecule can be exchanged in order to measure the effects of DNA binding ligands on the properties of DNA.

Figure 1.

The force-dependent difference in length between DNA and a saturated A3G-DNA complex allows us to measure A3G binding. Typical extension (solid black) and return (dashed black) of a single DNA molecule. At 61.0 ± 0.5 pN, the molecule undergoes a force-induced melting transition from dsDNA (green, Eq. S1) to ssDNA (blue). A3G-saturated ssDNA (200 nM A3G, t > 500 s, data points fit to Eq. S4, solid purple) is longer than dsDNA (Δx_b below the melting transition) and shorter than ssDNA (Δx_a above the melting transition). A3G-saturated ssDNA is significantly shorter than ssDNA only (blue, Eq. S2), which suggests that A3G may wrap ssDNA upon binding.

In the absence of binding ligands, force-induced melting occurs at a constant force of 61.0 ± 0.5 pN, generating ssDNA, either by peeling from the ends or forming melting bubbles³⁹. Although a stretched form of dsDNA called S-DNA may form upon overstretching at high salt, it is well-established that overstretching is force-induced melting in the presence of ssDNA binding proteins and at the ionic strength used in this work^39,40. At the end of the overstretching transition, the bead movement reverses direction and DNA tension is gradually released (Fig. 1, dashed black line). The minimal hysteresis, or mismatch between DNA extension and release, indicates that the ssDNA generated by force reanneals immediately into dsDNA during the return. At any given point along the transition, the molecule is a well-characterized combination of dsDNA and ssDNA (Supplementary Eq. S3)⁴⁰. Pausing at fixed extension during the melting transition allows precise control of the fraction of ssDNA substrate available for protein binding.

The force-extension curve probes the length of the captured DNA molecule at a given force. At forces above 7 pN, ssDNA is longer than dsDNA. We exploit this force-dependent difference in length to measure the ssDNA binding properties of A3G. A3G-saturated ssDNA, obtained at high protein concentration by first overstretching DNA and allowing A3G to fully bind and stabilize the DNA in its single-stranded form, is longer than dsDNA and shorter than ssDNA (Fig. 1). Therefore, A3G bound to ssDNA increases the molecule in length by Δx_b below the melting transition, and decreases the length by Δx_a above the melting transition.

In the presence of 50 nM A3G, the extension curve follows the DNA-only curve before the melting transition, reflecting no measurable binding to dsDNA (Fig. 2, panel a, solid line). A3G only binds after force-induced melting generates ssDNA. Based on the observed hysteresis (Fig. 2, panel a, dashed line), most of the protein does not dissociate upon DNA release and prevents the two strands from fully reannealing. A3G-bound ssDNA is longer than dsDNA (Fig. 1), so the change in length at a given force (Fig. 2, panel a, Δx_t) describes the total fraction of A3G bound to ssDNA (f_total, see Supplementary Fig. 1, panel a online).

Figure 2.

Single molecule method to measure fast and slow fractions of A3G binding. (A) Without protein (black), a single DNA molecule reanneals immediately upon release, exhibiting minimal hysteresis, or mismatch between stretch (solid) and release (dashed) curves. In the presence of 50 nM A3G, the stretch curve (solid green) follows the dsDNA-only curve, indicating negligible A3G-dsDNA binding. A3G binds the exposed ssDNA, and prohibits the DNA strands from reannealing, resulting in hysteresis (dashed green). For a given force (40 pN shown), there is a corresponding change in DNA length Δx_t between A3G-free dsDNA (left arrow, drawing 1) and partially A3G-bound ssDNA (right arrow, drawing 2). This force-dependent length change measures A3G-ssDNA binding (Supplementary Fig. 1). (B) The second stretch (solid blue) lies between the first stretch and release curves, distinguishing the fraction of A3G that remains bound (f_slow) from the fraction that dissociated (f_fast) before the second stretch. The A3G that dissociates rapidly allows the strands to reanneal immediately into dsDNA (drawing 3), resulting in length decrease Δx_f. (C) Pausing at fixed DNA extension after incubating ssDNA with 50 nM A3G results in additional binding (drawing 4), indicated by the corresponding length increase Δx_i measured during DNA release. (D) A3G binding increases with total exposure time to ssDNA (dashed lines).

A second stretch of the same molecule does not retrace the release curve, revealing that some fraction of the protein has dissociated during the 30 s incubation at zero force between stretch-release cycles (Fig. 2, panel b). As soon as any A3G dissociates at forces below the force-induced melting transition, the two strands reanneal into dsDNA, which is shorter than A3G-bound ssDNA (Fig. 2, panel b, Δx_f). Therefore the second stretch reflects the fraction of A3G that remains ssDNA-bound (f_slow), which allows the fraction that dissociates quickly (f_fast) to be quantified as f_fast = f_total − f_slow. The DNA was held at zero force between the first release and the second stretch for 30 s, but longer wait times up to 120 minutes do not lead to further measurable dissociation (data not shown for clarity).

A3G was exposed to ssDNA for 50 s in this experiment. However, A3G oligomerization observed in bulk experiments occurs on much longer timescales³⁷. To measure slow binding, A3G was incubated for 250 s with ssDNA generated by force-induced melting (Fig. 2, panel c). The DNA release curve obtained after this incubation exhibits a length increase relative to the initial release curve (Fig. 2, panel c, Δx_i) because additional A3G binds ssDNA during incubation. This effect increases at longer incubation times (Fig. 2, panel d), approaching the A3G-saturated ssDNA curve (shown in Fig. 1).⁴¹ Fits to the DNA release curves at increasing incubation time yield f_total(t), while the subsequent stretch (data not shown for clarity) quantifies f_slow(t), and f_fast(t) is the difference between the two. These measurements for 50 nM A3G are presented in Fig. 3, panel a.

Figure 3.

Quantifying A3G binding reveals association and dissociation rates for fast and slow binding modes. (a) Total binding at 50 nM A3G (f_total, red) separated into a fast fraction (f_fast, blue) and slow fraction (f_slow, green), as a function of ssDNA-A3G incubation time. Fits to the binding model (solid lines, Supplementary Eqs. S9–S11) yield observed rates k_fast and k_slow. (b) Slow fraction bound as a function of time for five A3G concentrations. Solid lines are fits to Supplementary Eq. S10. Error bars (panels a, b) are standard error (N≥3) for 50–200 nM A3G and propagated error for 10–20 nM A3G. (c) Fast rates (k_fast, blue data points) obtained from fits to the binding model (shown in panel a for 50 nM A3G). The linear fit (solid blue line, Supplementary Eq. S13) yields k₁ and k₋₁. k₁c (purple data points) and k₋₁ (red data points) were also calculated from the binding model. Linear fits (solid lines, Supplementary Eqs. S15 and S17) yield consistent values of k₁ and k₋₁. (d) Slow rates (k_slow, green data points) from fits to the binding model (panel b). Fits to Supplementary Eq. S21 (solid green line, see Supplementary Fig. 2, panel b) yield k₂ and k₋₂. Separate calculations of k₂ (purple) and k₋₂ (red) from the binding model (Supplementary Eqs. S24 and S25) are also shown.

Quantitative binding model

The slow binding component increases at the expense of the fast component, suggesting that the A3G-ssDNA reaction may be modeled as a two-step process:

$$A3G+\mathit{ssDNA}\underset{k_{-1}}{\overset{k_{1}c}{\rightleftharpoons }}{(A3G/\mathit{ssDNA})}_{\mathit{fast}}\underset{k_{-2}}{\overset{k_2}{\rightleftharpoons }}{(A3G/\mathit{ssDNA})}_{\mathit{slow}},$$ 1

in which an initial bimolecular process leads to a fast complex that converts to a slow, more stable complex in the second unimolecular step. Binding rates were obtained from fits to this model (Supplementary Eqs. S9–S11) at five A3G concentrations (Fig. 3, panel b). (A3G precipitates at high concentrations¹⁸, so experiments were calibrated using force-extension curves at known protein concentrations.) As expected, the observed fast rate k_fast and the on rate k₁c are both linear with A3G concentration (Fig. 3, panel c). The bimolecular rate constant k₁ = 1.5 (± 0.1) × 10⁵ M⁻¹s⁻¹ and off rate k₋₁ = 1.2 (± 0.1) × 10⁻² s⁻¹ are consistent with single molecule FRET⁴² and fluorescence spectroscopy³⁷ measurements, considering differences in solution conditions. The observed slow rate k_slow saturates at high A3G concentration (Fig. 3, panel d), and both the on and off rates for the second, unimolecular step are concentration-independent (k₂ = 6.7 (± 0.6) × 10⁻³ s⁻¹ and k₋₂ = 2.8 (± 0.5) × 10⁻⁵ s⁻¹). Elementary reaction rates were obtained from the data in several different ways, and agreement of the resulting values (see Supplementary Table 2 online) supports the binding model.

Oligomerization is responsible for slow binding

To determine whether slow binding is due to A3G oligomerization, we expressed and purified the F126A/W127A A3G mutant (A3G FW), which is severely defective in oligomerization⁴¹. When this mutant was incubated with ssDNA for 1050 s, the release curve exhibited minimal hysteresis (Fig. 4, panel a), and all the bound protein dissociated prior to the subsequent stretch (Fig. 4, panel b). A direct comparison of the hysteresis observed for both wild type and mutant A3G is shown in Fig. 4, panel c. The lack of a slow ssDNA bound fraction observed for the mutant, along with the striking difference between the hysteresis observed for the two proteins, shows that the oligomerization-defective mutant does not exhibit slow ssDNA binding kinetics. Thus, we conclude that the slow kinetics observed for wild type A3G is due to oligomerization.

Figure 4.

Oligomerization-defective mutant F126A/W127A (FW) demonstrates that the slow kinetics observed for wild type A3G is due to oligomerization. (a) In the absence of protein (black), a single DNA molecule reanneals immediately upon release, exhibiting minimal hysteresis between extension (solid) and release (dashed). In the presence of 50 nM F126A/W127A A3G (orange), the stretch curve (solid) follows the dsDNA-only curve, indicating no measurable A3G FW binding to dsDNA (drawing 1). Pausing at fixed DNA extension after the melting transition to incubate the ssDNA with the protein results in ssDNA binding (drawing 2), indicated by the corresponding increase in length Δx_t measured during DNA release at a given force (shown for 40 pN). (b) The subsequent stretch (dark blue) follows the initial stretch curve (solid orange), indicating that all the mutant A3G bound during incubation dissociates rapidly (drawing 3), resulting in a decrease in length Δx_f. (c) Wild type A3G (drawing 4) exhibits a greater change in length Δx_o relative to the FW mutant (drawing 2) at 1050 s incubation due to oligomerization on ssDNA.

DISCUSSION

Here we use a single molecule method that allows us to quantify two distinct modes of A3G binding to ssDNA and characterize the conversion of a fast state into a slow state. These results suggest a binding mechanism in which monomers or dimers initially bind ssDNA and rapidly reach equilibrium (1/k_fast = 24 ± 1 s at 200 nM), before slowly converting to oligomers (1/k_slow = 206 ± 20 s) (Fig. 5, panel a). Previous bulk solution experiments have established that A3G oligomerizes in the presence of single-stranded nucleic acids^41,43, which inhibits efficient deaminase activity⁴¹. We also demonstrate that an A3G mutant (F126A/W127A) that is severely compromised in oligomerization, but retains deaminase activity, does not exhibit slow binding kinetics. Furthermore, a recent study shows that a similar oligomerization-defective mutant (W127A) inhibits deamination-independent viral restriction⁴⁴. These observations support the hypothesis that A3G oligomerization is responsible for deaminase-independent inhibition of viral replication.

Figure 5.

Models for A3G oligomerization (A) in vitro and (B) in virio. (a) Initially monomers or dimers bind ssDNA with on rate k₁c and off rate k₋₁. These forward and backward rates are on similar timescales (1/k₁c = 33 ± 1 s at 200 nM A3G, and 1/k₋₁ = 85 ± 5 s), so fast binding reaches equilibrium before the monomers or dimers convert to oligomers (1/k₂ = 149 ± 13 s) on ssDNA. Oligomer dissociation is significantly slower (1/k₋₂ = 10 ± 2 h) in vitro. (b) Inside the virus, A3G oligomerizes on the RNA genome, blocking minus-strand DNA synthesis by RT (panel 1). Once the oligomer dissociates, the monomers or dimers released bind ssDNA within a second, allowing rapid enzymatic activity (panel 2) until A3G oligomerizes on the ssDNA template in 150 s and blocks plus strand synthesis (panel 3).

In light of these quantitative results in vitro, we propose a model for A3G’s effects on HIV-1 replication. A3G oligomerizes on viral RNA when it is packaged inside the virion^43,45, stalling RT during minus-strand synthesis. RT pauses until the oligomer dissociates, or switches to the other RNA template strand, circumventing the A3G roadblock and leading to partial inhibition of reverse transcription³⁵. In contrast to our in vitro studies, the maximum size of the A3G oligomer may be limited by the small number of A3G molecules packaged in virions³⁶. These limitations on oligomer size may lead to higher off rates in virio. Once the A3G oligomer dissociates during minus-strand synthesis, the monomers or dimers that are released have low affinity for the newly formed RNA-DNA duplex^14,18. As RNase H activity exposes the minus strand, all A3G molecules bind the ssDNA template within a second, as indicated by 1/k₁ = 0.7 ± 0.1 s when fast binding rates are extrapolated to the estimated 13 ± 8 μM A3G concentration in the virion³². A3G remains bound in a rapid sliding mode^37,42 for 1/k₋₁ = 85 ± 5 s, which allows high deamination rates^37,38 until oligomerization forms a roadblock to plus-strand synthesis after 1/k₂ = 149 ± 13 s (Fig. 5, panel b).

The in virio model that we propose based on our measurements is consistent with the available data on A3G function. Recent cell-based experiments demonstrate that A3G uniformly blocks minus-strand synthesis in the absence of preferred RT termination sites along the viral template³⁵. However, this does not necessarily indicate that A3G binds throughout the length of the genomic RNA. Oligomerization is nucleic acid sequence-independent and therefore we expect the roadblock to form at a small number of random sites along the viral genome, as observed in virio. Although A3G may bind RT in an RNA-independent interaction⁴⁶, a recent report concludes that there is no direct interaction between A3G and RT³¹. In any case, this type of binding is equally probable for any of the 100 RT heterodimers present in the virion⁴⁷ and the probability that one of 7 ± 4 A3G molecules binds and inhibits the catalytically active RT molecule is therefore less than 10%. Thus, a possible A3G-RT interaction appears unlikely to be primarily responsible for A3G-induced inhibition of reverse transcription.

The data we present resolve seemingly contradictory mechanisms for A3G inhibition of viral replication by demonstrating that A3G can function as both a fast deaminase and a slow nucleic acid binding protein. Regulation of enzymatic activity via protein oligomerization may be a general property of other APOBEC family members that inhibit replication of retroviruses and retrotransposons independent of deaminase activity. The single molecule method described here does not require labeling of DNA or protein and is optimal for measuring the biologically important process of slow protein oligomerization on ssDNA.

METHODS

A3G preparation and purification

Recombinant WT A3G and A3G FW were expressed in a baculovirus expression system using an N-terminal glutathione S-transferase (GST) tag and purified. The GST tag was then removed using a Novagen Enterokinase Cleavage Capture Kit. The deaminase activity of purified wild type A3G was verified using a gel-based uracil DNA glycosylase assay after removal of the GST tag. A similar method was used to verify the enzymatic activity of purified GST-tagged A3G FW. Additional protein preparation details are supplied in the Supplementary Methods.

Single molecule experiments

Biotin-labeled bacteriophage λ DNA was captured between two streptavidin-coated polystyrene beads, one on a fixed micropipette tip and the other held in an optical trap, as described in Supplementary Methods. The DNA was extended by moving the micropipette tip at a rate of 100 nm/s, and the resulting force on the bead in the trap was recorded in order to obtain the force-extension curve for DNA alone. The buffer surrounding the DNA molecule was then exchanged for a solution of fixed protein concentration. Force-extension curves in the presence and absence of protein were fit to the Worm-Like Chain model for dsDNA (Supplementary Eq. S1) and the Freely-Jointed Chain model (Supplementary Eq. S2) for ssDNA. The A3G-saturated ssDNA force-extension curve was fit to Supplementary Eq. S4. Measurements of the fraction of ssDNA bound by protein were then determined by fitting force-extension curves to the linear combination of dsDNA extension and A3G-bound ssDNA extension, as shown in Supplementary Eqs S6 and S7.

Two-step binding model

The kinetics of protein-DNA binding was fit to a two-step binding model, in which it was assumed that an initial fast bimolecular binding event was followed by a slow unimolecular binding event, as described by Supplementary Eqs. S9–S11, yielding measured fast and slow binding rates k_fast and k_slow. By then fitting the concentration-dependent measurements of k_fast and k_slow to the reaction shown in Eq. 1, we determine the elementary reaction rates for each step k₁, k₋₁, k₂, and k₋₂, as well as the elementary equilibrium constants for each step K₁ and K₂, as described in Supplementary Methods.