Single-stranded DNA Scanning and Deamination by APOBEC3G Cytidine Deaminase at Single Molecule Resolution*♦

Background: Apo3G, an ssDNA-dependent C deaminase, inactivates HIV-1 in T cells by C to T hypermutation. Results: smFRET is used to detect Apo3G scanning and C-deamination on ssDNA. Conclusion: Apo3G scans ssDNA randomly and bidirectionally, favoring nonrandom 3′ to 5′ deamination. Significance: This smFRET study describes a broadly applicable approach to visualize motion and catalysis in real time by an enzyme that scans ssDNA. APOBEC3G (Apo3G) is a single-stranded (ss)DNA cytosine deaminase that eliminates HIV-1 infectivity by converting C → U in numerous small target motifs on the minus viral cDNA. Apo3G deaminates linear ssDNA in vitro with pronounced spatial asymmetry favoring the 3′ → 5′ direction. A similar polarity observed in vivo is believed responsible for initiating localized C → T mutational gradients that inactivate the virus. When compared with double-stranded (ds)DNA scanning enzymes, e.g. DNA glycosylases that excise rare aberrant bases, there is a paucity of mechanistic studies on ssDNA scanning enzymes. Here, we investigate ssDNA scanning and motif-targeting mechanisms for Apo3G using single molecule Förster resonance energy transfer. We address the specific issue of deamination asymmetry within the general context of ssDNA scanning mechanisms and show that Apo3G scanning trajectories, ssDNA contraction, and deamination efficiencies depend on motif sequence, location, and ionic strength. Notably, we observe the presence of bidirectional quasi-localized scanning of Apo3G occurring proximal to a 5′ hot motif, a motif-dependent DNA contraction greatest for 5′ hot > 3′ hot > 5′ cold motifs, and diminished mobility at low salt. We discuss the single molecule Förster resonance energy transfer data in terms of a model in which deamination polarity occurs as a consequence of Apo3G binding to ssDNA in two orientations, one that is catalytically favorable, with the other disfavorable.

Incidental deamination of dC 3 dU occurs frequently, especially on single-stranded (ss)DNA (1), which is a potential source of spontaneous C 3 T mutations. The elimination of U⅐G mismatches by base excision repair ensures that deamination-initiated mutations are minimized in genomic DNA (1,2). However, dC deaminations also occur enzymatically, as a regulated function of the immune system (3,4). For the HIV-1 host restriction factor, APOBEC3G (Apo3G), 5 this entails deaminating C residues on newly reverse transcribed HIV-1 cDNA (minus strand DNA), most often in 5Ј-CCC motifs, mainly at 3Ј-C (underlined), although occasionally at the middle C in the motif (5,6). The virus can be neutralized by the catalytic action of Apo3G in at least two ways. The presence of U might induce degradation of the HIV-1 minus strand by concerted action of uracil DNA glycosylase and apurinic/apyrimidinic endonuclease (2,7,8), or G 3 A mutations that occur following synthesis of the plus strand may destroy viral infectivity (5,6,9). In the event that Apo3G does not induce sufficient mutagenesis to result in HIV-1 inactivation, it may contribute to HIV-1 evolution and development of drug-resistant quasi-species (10). It is important to establish the biochemical mechanisms of how Apo3G induces multiple mutations in the HIV-1 genome per se, and also to support ongoing efforts to develop Apo3G-based HIV therapies (11).
Our objective is to examine how Apo3G locates and then deaminates C motifs on an ssDNA substrate. Apo3G contains two deaminase domains. The N-terminal domain (CD1) is not catalytically active, but it is required for HIV-1 virion encapsidation and can bind DNA and RNA (12,13). The C-terminal domain (CD2) contains the active site (14). Previous "bulk" biochemical studies showed that Apo3G catalyzes processive deaminations on linear ssDNA prior to acting on another substrate molecule while displaying a pronounced catalytic polarity (15,16). Trinucleotide motifs are deaminated with increased efficiency when located nearer to the 5Ј-end of the DNA, and the terminal 33 nt at the 3Ј-end of an ssDNA substrate are barely deaminated, creating a deamination "dead zone" (16). The current model for the Apo3G scanning mechanism that has resulted from bulk biochemical studies has yet been unable to answer several questions regarding the scanning mechanism. These questions are addressed here using single molecule FRET (smFRET). From a biological perspective, it has been suggested that Apo3G deamination polarity offers a likely explanation for the presence of localized mutational gradients in the HIV-1 genome that appear to be important in viral inactivation (17).
Using smFRET, we visualize Apo3G movement and catalysis on linear ssDNA in real time. We investigate how motif location and identity, i.e. "hot" versus "cold" motifs, influence Apo3G motion, C deamination efficiency, and DNA contraction. The smFRET data provide new insights into a model in which Apo3G is presumed to scan ssDNA in a symmetric, bidirectional manner, yet causes spatially polar deaminations by binding asymmetrically to ssDNA in two orientations, one that is catalytically active and another that is essentially inactive (18), with pseudo-localized scanning in the vicinity of a 5Ј hot motif.

EXPERIMENTAL PROCEDURES
Apo3G Labeling with Cy5 Fluorescent Dye-To remove surface-exposed cysteine residues (confirmed by bioinformatic structural modeling of Apo3G), we used a QuikChange sitedirected mutagenesis protocol to mutate cysteine 139, 243, and 308 to leucine, alanine, and leucine, respectively. Only one surface-exposed residue (Cys-356) is available for labeling with Cy5-maleimide. A baculovirus-expressed mutated GST-Apo3G protein variant was prepared as described previously (15,16). The Cy5 maleimide mono-reactive dye kit (GE Healthcare) protocol was used to achieve 10 -20% labeling efficiency, to prevent nonspecific labeling of protein side chains. Free label was removed by gel filtration on Bio-Gel P-6 (Bio-Rad). Protein concentration was determined by Bradford assay and BSA protein standardization. Labeling efficiency was determined by UV (280 and 552 nm). Mass spectrometry analysis of Cy5-Apo3G confirmed that the predicted surface-exposed Cys-356 was labeled (Harvard University Mass Spectrometry and Proteomics Core). We have further verified by photobleaching that Apo3G has a single Cy5-label. Cy5-Apo3G retained activity and deamination properties of native Apo3G. We have also verified, based on previous atomic force microscopy data (16,18), that Apo3G is predominantly monomer under smFRET conditions. Pfu DNA Polymerase Expression, Purification, and Labeling-Two rounds of site-directed mutagenesis (QuikChange sitedirected mutagenesis kit, Stratagene) of Pfu DNA polymerase expression vector pET-30a-PFU (Dr. Stephen Bell, Cambridge, UK) were performed (D125A, I48C) to make an exo-Pfu suitable for labeling. The resulting vector (pPFUe-I48C) was transformed into BL21 Rosetta(DE3) pLysS (Novagen). Cells were grown at 37°C until an A 500 of ϳ0.5 was reached. Isopropyl-1thio-␤-D-galactopyranoside was added (1 mM final concentration), and cells were grown at room temperature (ϳ25°C) overnight and then centrifuged at 4°C (4,000 rpm for 15 min) and resuspended in 25 mM HEPES, pH 7.8, 500 mM NaCl, 5 mM imidazole, 1 mM phenylmethanesulfonyl fluoride, protease inhibitor mixture, 50 mg/ml lysozyme. After sonication on ice (10 ϫ 15-s pulses), 20 units of DNase I and 10 g/ml RNase A were added, and cells were incubated at 37°C for 30 min and then boiled for 15 min. Samples were centrifuged twice (17,000 ϫ g for 15 min, 4°C), the supernatant was loaded onto a 5-ml nickel column (nickel-nitrilotriacetic acid His-Bind resin, Novagen) and washed four times with 25 mM HEPES, pH 7.8, 500 mM NaCl, 5 mM imidazole. Protein was eluted with 25 mM HEPES, pH 7.8, 500 mM NaCl, 200 mM imidazole, and fractions were analyzed by SDS-PAGE and Coomassie Blue staining. Protein concentration of fractions (ϳ90 kDa) was determined by near-UV absorbance (A 280 ). Pfu D125A/I48C was labeled using the Cy5 maleimide mono-reactive dye kit (GE Healthcare) according to provided protocols.
smFRET-Quartz microscope slides were passivated with methoxy-PEG (M r ϭ 5,000; Laysan Bio Inc.) and 1% biotin-PEG (M r ϭ 3,400; Laysan Bio Inc.) to minimize nonspecific binding (19,20), and ssDNAs were surface-immobilized by using a common anchor DNA (supplemental Table S1) with 5Ј-Cy3 and 3Ј-biotin (see Fig. 2A). ssDNAs (2 M) were annealed to the anchor DNA (1 M) in standard buffer (10 l, 50 mM MOPS, pH 7.4, 5 mM MgCl 2 , 60 mM NaCl, 2% (v/v) 2-mercaptoethanol, and 2 mM Trolox) by heating to 90°C for 45 s and cooled to room temperature over 15 min. Solution was serially diluted for surface immobilization (25-50 pM) of partial double-stranded DNA in standard buffer. Excess DNA was washed out, Cy5labeled Apo3G (1 nM) was introduced in a standard buffer, immediately followed by data acquisition on a home-built prism-based total internal reflection single molecule fluorescence microscope at 1-s or 30-ms time resolution (21). Movies were recorded for ϳ10 min in five different areas of the reaction channel.
Hidden Markov Model and TDP Analysis-Long binding (from 25 s to 10 min), single molecule FRET time trajectories were analyzed using a hidden Markov model, as described (22). Transitions were divided into 10 virtual macrocanonical states separated by 0.1 FRET units, which do not correspond to specific conformational states, but rather to an ensemble of states with the protein located at a certain distance from the 5Ј-end. FRET trajectories were compiled into transition density plots (TDPs) showing the number of transitions observed between a given initial and final FRET value. All initial binding events originate from zero FRET and are located on the y axis.
Exponential Excursion Length Analysis-The excursion distance is calculated using the initial binding FRET value (ϳ0.2) and Förster's equation E FRET ϭ 1/[1ϩ(R/R 0 ) 6 ], where R is the observed distance and R 0 is the Förster distance corresponding to 50% energy transfer (55 Å for Cy3, Cy5, assuming 2 ϭ 2/3). Final FRET values in the TDP upper diagonal represent the location of the protein after each transition, which is converted into a distance using Förster's equation. The TDP provides the fraction of transitions at this distance. The fraction of transitions is plotted as a function of the distance to the initial binding site to estimate the distance scanned by the protein, which decays exponentially, indicating random excursions. Because we are calculating differences in distances, the 2 ϭ 2/3 approximation only minimally affects our results.
Post-synchronization Analysis-To synchronize the long binding events with respect to their initial bindings, we analyzed the long binding FRET trajectories from 10 s before the initial binding (threshold 0.15 FRET) until 10 s after the dissociation (0 FRET). Using a MATLAB script kindly provided by Jody Puglisy (Stanford University), we binned the trajectories (0.05 FRET bins, 3-s time bins) to determine how many trajectories are located at a given FRET and time value and then generate the histograms.
smFRET Deamination Assay-DNA substrates were surfaceimmobilized as described above using deamination buffer (50 mM MOPS, pH 7.4, 125 mM NaCl). Cy5-labeled Pfu exo Ϫ (50 nM) and unlabeled Apo3G (1 nM) were introduced on to the surface-immobilized substrate DNA immediately followed by data acquisition with 1-s time resolution (21) for the first 10 min. For each substrate tested, we analyzed 100 -200 single molecule trajectories (N), and for each substrate, we found the number of DNA molecules (n) showing any specific Cy5-labeled Pfu exo Ϫ binding (t on Ն 1 s), and the bound percentage was calculated as n/N ϫ 100%. A small background (ϳ10%) of nonspecific binding (measured in the absence of Apo3G) was subtracted for each substrate.
smFRET DNA Conformational Dynamics Experiments-To study the Apo3G-induced ssDNA conformational dynamics, we immobilized ssDNAs with 3Ј-Cy5-to the 5Ј-Cy3-labeled anchor DNA as described above. Data acquisition from five different areas of the reaction channel was done in the absence and presence of Apo3G (1 nM) with 1-s time resolution (21) for 10 min each. The smFRET histograms were obtained by time-binning Ͻ100 time trajectories and fit to a Gaussian distribution, y ϭ A exp [Ϫ(FRET Ϫ FRET 0 ) 2 /2 2 ] using Igor (WaveMetrics, Lake Oswego, OR).

RESULTS
Processive Apo3G-catalyzed C Deamination on ssDNA-Apo3G was incubated with phage M13mp2 circular DNA containing a series of in-frame 5Ј-aaaCCCaaa hot motifs embedded in lacZ␣ reporter sequence located within a single-stranded gapped region of M13 dsDNA (Fig. 1A). Following transfection of the DNA in Ung Ϫ Escherichia coli, Apo3G-catalyzed deaminations are identified as C 3 T mutations in DNA isolated and sequenced from individual mutant phage clones (23). M13 mutant phage (white plaques) comprise Ͻ2% of the total plaques, mutant and wild type (blue plaques), so that virtually all of the individual DNA clones were deaminated by at most one Apo3G molecule, in accord with Poisson statistics (24,25). The mutations occur as singletons and in clusters containing 2-5 consecutively deaminated motifs, with deaminations observed predominantly at C in 5Ј-aaaCCCaaa (Fig. 1B). The data suggest that a single Apo3G molecule scans ssDNA pro-FIGURE 1. Apo3G-initiated mutation patterns. A, graphic demonstrating the structure of the gapped M13 vector containing the series of aaaCCCaaa hot motifs in-frame with the lacZ␣ reporter sequence. The hot motifs were inserted into the EcoRI site of the lacZ␣ as three cassettes of 12 repeats separated by short spacers. Reactions were carried out under enzyme single-hit conditions (24,25) by reacting M13 gap DNA (ϳ1 mM) with Apo3G (110 pM) for 5 s at 37°C in the presence of MgCl 2 (5 mM) in a reaction volume of 90 l followed by transfection into host cells and plating on X-gal-containing medium. Deamination of any target C results in a stop codon during transcription and a white plaque following transfection. B, the target area is represented schematically with CCC motifs represented by a black dot and Apo3G-catalyzed deamination events of the 3Ј-C represented by a red T. The presence of mutational singletons interspersed with clusters of consecutively mutated motifs, as shown in representative clones, is consistent with a model wherein Apo3G moves bidirectionally on the DNA, predominantly by sliding, which gives rise to the consecutive deaminations, whereas hopping and intersegmental transfers could also reposition the enzyme on the ssDNA.
cessively, deaminating C 3 U haphazardly, reminiscent of the processive stochastic deamination patterns observed for activation-induced deoxycytidine deaminase (23), which is also an Apobec family protein (3,4). We have purified native Apo3G to apparent homogeneity and have previously obtained a highresolution x-ray structure for the catalytic CD2 domain (14). We now examine the scanning behavior of native full-length Apo3G at single molecule resolution using smFRET.
Apo3G ssDNA Scanning Observed by smFRET-We investigated the scanning behavior of Apo3G on a surface-immobilized ssDNA labeled at the 5Ј-end with Cy3, containing an ata-CCCaaa hot motif (15) located near the 5Ј-end (pdT 5Ј hot) and used a Cy5-labeled Apo3G to scan the DNA ( Fig. 2A, supplemental Table S1). Based on previous atomic force microscopy data (16,18), we have verified that Apo3G is predominantly a monomer under the smFRET conditions. The smFRET trajectories (Fig. 2B) show Apo3G binding and scanning motion as changes in apparent FRET efficiency, binding as an abrupt increase from zero, and dissociation as a sharp decrease to zero. Motions toward 5Ј and 3Ј directions are observed as increases and decreases in FRET, respectively. Representative traces, including anticorrelated FRET trajectories, are presented as supplemental material (supplemental Fig. S1).
The FRET trajectories reveal two populations, consistent with bulk experiments (16). About half show short Apo3G binding, Ͻ25 s (FRET ϳ0.2), with apparent binding and dissociation pseudo-first order rate constants k on ϭ 0.01 Ϯ 0.01 s Ϫ1 and k off ϭ 0.23 Ϯ 0.04 s Ϫ1 , respectively (supplemental Fig. S2). Long binding trajectories (Ն25 s) exhibit more complex dynamics indicating rapid scanning (Fig. 2B, Scanning trajectories). The FRET efficiencies oscillate rapidly between ϳ0.  Table S2). Because specific FRET states could not be identified, a hidden Markov Model (HMM) (22) was used to determine FRET densities corresponding to the locations of Apo3G relative to the 5Ј-end. Thus, each "FRET state" corresponds to an ensemble of configurations characterized by the distance between the protein and the 5Ј-end. The HMM is used to generate a TDP showing the number of transitions observed between initial and final FRET states (Fig. 2B, TDP). All initial binding events stem from zero FRET on the y axis, with most occurring at ϳ0.2 FRET, indicating that Apo3G binds preferentially away from the tethered 5Ј-end, confirmed by switching the tethered DNA to the 3Ј-end, in which case favored binding still occurs near the free end (i.e. 5Ј-end) (supplemental Fig. S4). Movements toward the 5Ј-and 3Ј-ends appear as peaks above and below the diagonal, respectively. The transitions are symmetric in both directions (Fig. 2B, TDP), indicating that Apo3G scans ssDNA without directional preference. Analysis of the dwell times between FRET states shows that scanning transitions occur with similar rate constants ϳ1 s Ϫ1 , measured at 33 frames/s (supplemental Fig. S5; supplemental Table S3).
An estimate of scanning distance in either direction is obtained using the number of transitions in the TDP and Förster's equation. The number of transitions decays exponentially from the initial binding site, consistent with random excursions (Fig. 2B, Scanning lengths) with a half-distance l ϭ 12 Å (ϳ9 nt) (26). To eliminate data blurring from asynchronous binding, long binding trajectories were post-synchronized by alignment at the initial binding event (27). The post-synchronization histogram (PSH) shows an initial FRET increase to ϳ0.2 (Fig. 2B,  PSH), consistent with the initial binding observed in the TDP. The FRET ratio increases to ϳ0.4 followed by oscillations between 0.2 and 0.7. Therefore, Apo3G moves in a bidirectional random manner yet hovers in the vicinity of the 5Ј hot motif, as shown by trajectories that remain synchronized up to 150 s (Fig.  2B, PSH).
Moving the hot motif closer to the 3Ј-end shows that scanning is strongly influenced by motif location, where Apo3G goes to high FRET values (Ͼ0.4) less often for the 3Ј hot motif (Fig. 2C). The number of transitions from the initial binding site decays exponentially with a half-distance l ϭ 7 Å (ϳ5-nt excursions) (Fig. 2C, Scanning lengths), about half when compared with the 5Ј hot motif. The post-synchronization histogram for the 3Ј hot trajectories confirms initial binding at ϳ0.2 FRET followed by oscillations between 0.2 and 0.6 FRET, but with more transitions in the 0.2-0.4 range and fewer in the 0.4 -0.6 range (Fig. 2C, PSH). The trajectories remain synchronized up to 50 s (Fig. 2C, PSH) when compared with ϳ150 s for the 5Ј hot motif.
The local sequence surrounding the target CCC influences scanning as shown by replacing the 5Ј hot (ataCCCaaa) motif with a 5Ј cold (tttCCCttt) motif (15) (Fig. 2D). The initial binding nearer the 3Ј-end (ϳ0.2 FRET), excess transitions toward the 5Ј-end (FRETϾ0.4), and excursion half-distance 11 Å (Fig.  2D, Scanning lengths) are similar to the 5Ј hot motif. However, the trajectories remain synchronized for only about half as long, 75 s when compared with 150 s for the 5Ј hot motif (Fig. 2D,  PSH), suggesting that Apo3G hovers in the vicinity of the 5Ј hot motif (ϳ0.4 FRET), about twice as long when compared with the 5Ј cold motif, because for the cold motif, much more of the scanning occurs at lower FRET states (0.2-0.3) away from the target motif (Fig. 2D, PSH). Scanning toward the 5Ј-end (FRETϾ0.4) remains when the hot 5Ј-CCC is replaced by the 5Ј-CCU deaminated product, and the scanning excursion (9 Å) is intermediate (Fig. 2E, Scanning lengths). However, the synchronized trajectories, reflecting hovering, near the 5Ј-CCU are 75 s (Fig. 2E, PSH), which is half as long as the 5Ј hot motif, but similar to the 5Ј cold motif. In the absence of a deamination target (poly(dT), Fig. 2F), the scanning is similar to the 3Ј hot DNA (pdT 3Ј hot) with respect to traces, TDP, and displacement half-distance from the initial binding site.
Notably, however, the histogram for poly(dT) (pdT) shows that there is no longer a discernible spatial localization of Apo3G on the DNA (Fig. 2F, PSH). In other words, Apo3G hovering is absent when there is no deamination motif. The long binders exhibit similar average residence times for each of the constructs (supplemental Table S2). In summary, the data show that Apo3G scans ssDNA bidirectionally over the entire molecule, favoring movement in the vicinity of a 5Ј hot motif.
Apo3G Contracts ssDNA in a Deamination Motif-dependent Manner-With fluorescent probes located at 5Ј-and 3Ј-ends, we can use smFRET to detect DNA contraction by measuring end-to-end distances in the presence of unlabeled Apo3G (Fig.  3A). In the absence of Apo3G, the histogram for pdT 5Ј hot has a narrow distribution centered at 0 (Fig. 3B), indicating that the DNA is in an extended conformation with its ends separated by Ͼ90 Å. In the presence of Apo3G, the FRET distribution shifts to 0.5 and broadens (0.2-0.8), indicative of contraction (Fig.  3C). The broad distribution indicates the presence of rapid con-formational dynamics exceeding our time resolution (1 s) (28). Pronounced differences are observed for the pdT 3Ј hot, where the FRET distribution is centered at 0.4 and the width narrows (0.2-0.6) (Fig. 3D). A further shift to 0.3 and narrowing (0.1-0.4) occurs with pdT 5Ј cold (Fig. 3E), 5Ј-CCU product (Fig. 3F), and pdT (Fig. 3G). These data show that Apo3G contracts ssDNA in a motif-dependent manner. Contraction is most pronounced during long excursions from the initial binding site in the vicinity of a 5Ј hot motif (Fig. 2).
Distinguishing between Apo3G Scanning and ssDNA Contraction-The FRET fluctuations shown in Fig. 2, with the FRET donor situated at the 5Ј-tethered end of the ssDNA, are interpreted as Apo3G scanning the ssDNA. However, a possible alternative explanation for these fluctuations could be that Apo3G binds at a fixed position on the DNA and causes DNA conformational changes, e.g. by contracting the DNA, which would also result in distance changes between the FRET donor on the 5Ј-ssDNA end and acceptor on Apo3G, which binds preferentially near the untethered 3Ј-end (Fig. 2).
To establish that Apo3G does not remain bound at a fixed position, but instead moves along the ssDNA, we relocated the FRET donor to the 3Ј-end of the ssDNA substrate (Fig. 4A). If Apo3G were to bind nearer the 3Ј-tail and contract the ssDNA without scanning, then the FRET ratio should remain approximately constant. However, the observed smFRET trajectories and TDP analysis for the pdT 5Ј hot ssDNA (Fig. 4B) show rapid, bidirectional FRET changes (in the range of FRET 0 -1), independently indicating that Apo3G moves along the entire ssDNA.
Another important distinction between Apo3G ssDNA scanning and contraction was obtained by repeating smFRET scanning measurements using poly(dA) with a 5Ј hot motif (Fig.  5, pdA 5Ј hot). Because of significantly greater base stacking, pdA is considerably stiffer than pdT and therefore should decrease contraction while maintaining scanning. Indeed, when unlabeled Apo3G binds pdA 5Ј hot that is labeled on each end (Fig. 5A), ssDNA contraction is significantly reduced (FRET ϳ0.2). Fig. 5B (Scanning trajectories) shows representative FRET time trajectories of Cy5-labeled Apo3G moving along the pdA 5Ј hot ssDNA. Similar FRET fluctuations are observed as on pdT 5Ј hot (Fig. 2B), indicating that Apo3G also scans the stiffer ssDNA (Fig. 5B). The resulting TDP shows symmetric FRET oscillations between ϳ0.2 and ϳ0.8, indicating that Apo3G also scans poly(dA) 5Ј hot randomly and bidirectionally (Fig. 5B).
Once again, to confirm that Apo3G scans the DNA and does not gain access to different DNA regions by contracting the DNA while remaining bound at a fixed position, the Cy3 donor label was placed at the 3Ј-end of ssDNA. Here again, the FRET trajectories and TDP analysis show that Apo3G randomly and bidirectionally scans the entire ssDNA (FRET 0 -1; Fig. 5C). These data confirm that the FRET fluctuations observed in the time trajectories reflect the scanning behavior of the enzyme on the DNA for the less flexible pdA 5Ј hot (Fig. 5B) as it did for the more flexible pdT 5Ј hot (Fig. 2B).
Reduced Apo3G Mobility on ssDNA at Low Salt-The catalytically inactive N-terminal CD1 domain has a predicted large net positive charge (ϩ11), in contrast to the catalytically active CD2 domain (Ϫ4.5), and is likely to govern the mobility of Apo3G on ssDNA, which should depend on metal ion concentration. When the experiments for 5Ј hot and poly(dT) DNA were repeated in "low" salt conditions (30 mM NaCl, 0 mM MgCl 2 ), when compared with "high" salt (60 mM NaCl, 5 mM MgCl 2 ), there was a marked reduction in scanning speed, accompanied by lengthy pauses in the FRET trajectories (Fig.  6). Although bidirectional motion still occurs over the entire ssDNA at low salt, there are far more transitions confined to low FRET values near the initial binding site (Fig. 6, A and B,  Table S2; the intensity bar represents the frequency scale of FRET transitions (blue, low frequency; yellow, high frequency). Scanning length analysis displays the distribution of the distances (x axis) Apo3G travels from the initial binding site and the frequency of those transitions relative to all transitions. The percentage of intensity (y axis) gives the relative probability of each step Apo3G traveled, which decays exponentially with distance, fit to single exponential curves to calculate scanning lengths (l) Ϯ error from the fitting. The PSH is compiled using the FRET transitions (y axis) over a given time (x axis); the intensity bar represents the frequency scale of these particular FRET states (beige, low frequency; red, high frequency). Apo3G tends to have more frequent and longer transitions in the vicinity of a hot motif, showing that Apo3G tends to scan more often in the vicinity of a 5Ј hot motif, referred to under "Results" as quasi-localized motion.  MAY 4, 2012 • VOLUME 287 • NUMBER 19 TDP), and the average excursion distance for the pdT 5Ј hot is reduced to 8 Å when compared with 12 Å at high salt (Figs. 2B and 6A, Scanning lengths). The scanning transition rate constants are reduced 10 -100-fold at low salt (supplemental Table  S4), and the localized motions of Apo3G near the 5Ј hot motif at high salt (Fig. 2B, PSH) are absent at low salt (Fig. 6A, PSH). The same restricted motion is observed for the pdT substrate, with a 6 Å excursion distance (Fig. 6B, Scanning lengths). Low salt also diminishes DNA contracting, as shown in histograms having narrower distributions centered at lower FRET values for the pdT 5Ј hot (Fig. 6C). These nonphysiological salt concentrations were chosen to illustrate changes in Apo3G scanning mobility. The large increase in the mobility of Apo3G on ssDNA at higher salt is likely caused by a partial shielding of the electrostatic interactions between the strongly positively charged CD1 domain and the negatively charged DNA phosphate backbone, possibly augmented by an increased DNA base stacking (29) that could further facilitate Apo3G movement by providing a more ordered ssDNA backbone.

Single Molecule FRET Studies of Apo3G Scanning ssDNA
Observing Apo3G C Deamination Polarity with smFRET-Apo3G exhibits a marked deamination polarity favoring the 5Ј direction on linear ssDNA measured in bulk solution (15,16) (Fig. 7A). Both native and Cy5-labled forms of Apo3G favor deamination of the 5Ј-CCC motif located toward the 5Ј-end of the ssDNA substrate by about 2-fold when compared with 5Ј-CCC situated nearer the 3Ј-end. Both enzymes also show correlated double deaminations, indicative of enzyme processivity (15,16), as also illustrated by the clustered deamination patterns observed in Fig. 1B. In contrast, the C-terminal catalytically active CD2 domain is neither catalytically polar nor processive in the absence of the catalytically inactive CD1 domain (18) (Fig. 7A).

DISCUSSION
Apobec family proteins are classified in two distinct groups, having either two deaminase domains, such as Apo3G, or , representative smFRET scanning trajectory shows the FRET ratios (y axis, black) with the HMM fit (red) over a given time (x axis). C, the TDP was prepared by HMM analysis of the long binding FRET transitions (n ϭ 66); the intensity bar relates to the frequency of FRET transitions at specific states (blue, low frequency; yellow, high frequency). Scanning trajectories and TDP show that Apo3G scans the entire DNA and does not remain bound to the 3Ј-end. FIGURE 5. Apo3G scans stiffer ssDNA with reduced contraction. Apo3Ginduced ssDNA contraction for pdA 5Ј hot DNA was monitored using a setup similar to Fig. 3A. A, single molecule FRET histograms in the presence (panel 1) and absence (panel 2) of Apo3G reveal less protein-induced DNA contraction as a small change in the center of the FRET distribution and width. B, representative long binding smFRET time trajectory and TDP for pdA 5Ј hot DNA scanning (n ϭ 68) show similar scanning to the pdT 5Ј hot (setup as in Fig. 2A). C, representative long binding smFRET time trajectory and TDP for scanning 3Ј-Cy3-labeled (reverse Fl-labeled) pdA 5Ј hot DNA (n ϭ 50) show that Apo3G scans this stiffer DNA and does not remain bound to the 3Ј-end. one deaminase domain, e.g. activation-induced deoxycytidine deaminase (3,4). When present as a monomer, Apo3G is composed of a catalytically active C-terminal domain, CD2, with a negative charge (Ϫ4.5), and a catalytically inactive CD1 domain, with a predicted large positive charge (ϩ11), needed for DNA and RNA binding (12,13,18). A model to explain 3Ј 3 5Ј deamination polarity suggests that Apo3G binds ssDNA in either an active or an inactive orientation, with equal probability. The active orientation occurs when CD2 faces the 5Ј-ssDNA end, whereas the inactive orientation has CD2 facing the 3Ј-end (18) (Fig. 8).
The smFRET data for the ssDNA substrate speak directly to the asymmetric deamination model. The C deamination efficiency is about 2.3-fold higher for the 5Ј hot motif when compared with the 3Ј hot motif (Fig. 7D). The TDPs are symmetrical (Fig. 2), showing that Apo3G scans along ssDNA with equal probability toward 5Ј and 3Ј directions. Therefore, the deamination polarity cannot be caused by motion favoring the 5Ј direction. Instead, the likely source of the 3Ј 3 5Ј polarity is that Apo3G is active mainly when bound with CD2 facing the 5Ј-end of the ssDNA (see e.g. Fig. 8). The presence of a dead zone, a 33-nt region at the 3Ј-end of the DNA in which deaminations are barely detectable (16) (Fig. 7D), would occur whenever catalytically inactive CD1 binds at the 3Ј-end, thereby precluding access by CD2. Binding in the opposite direction, with CD1 facing the 5Ј-end, would allow unrestricted access of CD2 to the 3Ј-end (dead zone), but in a catalytically inactive orientation (Fig. 8). It is the 3Ј-C of the 5Ј-CCC-3Ј motif that is almost always deaminated, whereas the 5Ј-C is never deaminated (6,16). Based on an x-ray structural analysis of CD2 (14), it was proposed that the 3Ј-C is properly positioned nearby the catalytic zinc hydroxylation site only when CD2 is facing the 3Ј-C of the 5Ј-CCC motif (see the legend for Fig. 8), in agreement with the asymmetrical deamination model derived from the biochemical data (18).
The catalytic asymmetry is likely a consequence of the double domain structure of Apo3G. In the absence of CD1, deaminations catalyzed by CD2 have no dead zone (18) and are nonpolar when acting on linear ssDNA (18) (Fig. 7A). Activation-induced deoxycytidine deaminase, which has just a single catalytically active domain, also has no dead zone and catalyzes deaminations with equal efficiencies in 5Ј and 3Ј directions (16) (supplemental Fig. S7). Notably, it is the use of a linear ssDNA construct that imposes the type of end constraint that facilitates identification of dual catalytic orientations, active and inactive. As predicted, there was no polarity observed on circular ssDNA (16). However, annealing a complementary DNA to a circle restored the polarity (16), which may explain the presence of localized regions showing 3Ј 3 5Ј deamination polarity that are observed on the HIV-1 cDNA during reverse transcription in vivo (17), where cDNA synthesis and RNA template degradation occur concurrently, leaving ssDNA for Apo3G to act on proximal to multiple RNA/DNA hybrid regions.
There is a paucity of dynamic data for enzymes that scan ssDNA. The smFRET data provide an initial picture describing ssDNA scanning by Apo3G. The magnitude of the temporal FIGURE 6. Low salt restricts Apo3G scanning. A and B, representative low salt long binding smFRET time trajectories, TDP, scanning length analysis, and PST, for 5Ј hot DNA and poly(dT) DNA (n ϭ 69 and 67, respectively), as indicated. Experimental set up is as in Fig. 2A. Decreasing the salt concentration increases the number and duration of pauses between transitions and reduces the amount of high FRET transitions (TDP), the average scanning length (Ϯ error from the fitting), and the pseudo-localized motion, especially for 5Ј hot DNA, indicating that low salt slows scanning down. C, single molecule FRET histograms reveal less ssDNA contraction in low salt, but still in a motif-dependent manner.
variations in the FRET signal accompanying Apo3G scanning trajectories reveals that the entire 72-nt ssDNA region is traversed in a random, bidirectional manner (Fig. 2). While moving on average equally in either direction, Apo3G nevertheless spends excess time moving in the vicinity of the deamination motifs (Fig. 2, TDP, bright spots, ϳFRET 0.4). This quasi-localized motion is most pronounced proximal to the 5Ј hot motif and least pronounced nearby the 3Ј hot motif. Hovering is absent for poly(dT) lacking a deamination motif. While scanning, Apo3G induces motif-dependent contraction of the ssDNA, which is greatest with the 5Ј hot motif (Fig. 3). The sequence-dependent contraction may serve to reposition the enzyme and traverse sizable distances in sequence space, perhaps via hopping or intersegmental transfer (32)(33)(34). However, the catalytic events per se are likely to happen during sliding because we observe clusters of deaminations occurring in adjacent motifs (Fig. 1B). By reversing the Cy3 donor label from the tethered 5Ј-ssDNA end to the free 3Ј-end, we continue to see random, bidirectional, traversal of the entire ssDNA substrate (Figs. 4, B and C, and 5C). These data are further evidence that the FRET signal changes cannot be attributed to confor-mational changes caused by contracting the DNA by an immobile Apo3G. Clearly, Apo3G scans the ssDNA.
From a biological perspective, Apo3G deamination polarity has been suggested as a likely explanation for the presence of localized mutational gradients in the HIV-1 genome that appear to have an important functional role in viral inactivation (17). The demonstration that random scanning of ssDNA generates nonrandom catalysis supports a model in which Apo3G binds in an asymmetric catalytically active orientation (18).