DNA mutagenic activity and capacity for HIV-1 restriction of the cytidine deaminase APOBEC3G depend on whether DNA or RNA binds to tyrosine 315

APOBEC3G (A3G) belongs to the AID/APOBEC protein family of cytidine deaminases (CDA) that bind to nucleic acids. A3G mutates the HIV genome by deamination of dC to dU, leading to accumulation of virus-inactivating mutations. Binding to cellular RNAs inhibits A3G binding to substrate single-stranded (ss) DNA and CDA activity. Bulk RNA and substrate ssDNA bind to the same three A3G tryptic peptides (amino acids 181–194, 314–320, and 345–374) that form parts of a continuously exposed protein surface extending from the catalytic domain in the C terminus of A3G to its N terminus. We show here that the A3G tyrosines 181 and 315 directly cross-linked ssDNA. Binding experiments showed that a Y315A mutation alone significantly reduced A3G binding to both ssDNA and RNA, whereas Y181A and Y182A mutations only moderately affected A3G nucleic acid binding. Consistent with these findings, the Y315A mutant exhibited little to no deaminase activity in an Escherichia coli DNA mutator reporter, whereas Y181A and Y182A mutants retained ∼50% of wild-type A3G activity. The Y315A mutant also showed a markedly reduced ability to assemble into viral particles and had reduced antiviral activity. In uninfected cells, the impaired RNA-binding capacity of Y315A was evident by a shift of A3G from high-molecular-mass ribonucleoprotein complexes to low-molecular-mass complexes. We conclude that Tyr-315 is essential for coordinating ssDNA interaction with or entry to the deaminase domain and hypothesize that RNA bound to Tyr-315 may be sufficient to competitively inhibit ssDNA deaminase-dependent antiviral activity.

oligodeoxynucleotides have been obtained for CD2 of A3G and A3B revealing potential ssDNA coordination residues that lie within and proximal to the catalytic groove (42,43). In recent studies (10), cross-linking of full-length and catalytically active A3G to deoxyoligonucleotides and mass spectrometry (MS) of tryptic peptides showed A3G interactions with ssDNA within peptides at amino acids (aa) 181-194, 314 -320, and 345-374. In three-dimensional reconstruction, these peptides modeled proximally to the catalytic center of CD2 and along a continuously exposed surface on the back of A3G (relative to the catalytic face) toward CD1 in the N terminus (10,44).
In this study, we used site-directed mutagenesis and functional end-point analysis to show that tyrosines 181 and 315 were responsible for cross-linking to nucleic acids. A novel finding in our studies is that Tyr-315 in the C terminus of A3G is essential for RNA binding. Y315A mutants had a comparable fold as native A3G but did not form high-molecular-mass RNP and had an impaired ability to assembly with HIV virions. Although other studies have presumed RNA only binds to the N terminus of A3G (a single RNA-binding domain model), the data reported here reveal a significant role of Tyr-315 in determining RNA-dependent oligomerization of A3G in RNP critical to viral particle assembly of A3G and cellular cytoplasmic RNP formation. The distribution of RNA-binding peptides within the N-and C-terminal domains of A3G suggests that a dual domain RNA-binding model may more accurately account for the diversity of functional outcomes due to different RNA sequences bound to A3G. We hypothesize that Tyr-315 plays an essential role for both RNA and ssDNA binding whose occupancy may be key to gating entry of ssDNA to the active site and accounts for the observed RNA competitive inhibition of ssDNA binding and deaminase activity (10,36).

A3G tyrosines 181 and 315 are cross-linked to ssDNA in A3G/BrdU-ssDNA samples
We employed protein/nucleic acid photocross-linking coupled with MS and applied a comparative approach for MS analysis of tryptic peptides as candidate ssDNA-binding sites (44). Three A3G peptides were determined previously as being involved in ssDNA binding: aa 181-194, 314 -320, and 345-374 (10). In this study, we took advantage of bromodeoxyuridine (BrdU)-modified ssDNA cross-linked to full-length and native A3G (supplemental Fig. S1) to identify amino acid residues that were cross-linked to nucleotides in ssDNA.
Previously, studies of protein and peptides cross-linked to 5-bromouracil (BU) established that its photoreactivity may be dependent on cross-linking parameters and the peptide microenvironment (45)(46)(47). At low-energy excitation, crosslinking may yield the reactive BU triplet state (n,* state), created by initial single electron transfer from an oxidizable group in an amino acid residue (e.g. cross-linking to the aromatic ring of Trp or Tyr, followed by ion radical coupling and elimination of bromine (HBr) to yield the cross-link). Alternatively, excitation with a higher energy singlet state, a ,* state, may yield a uracil radical and a bromine atom in competition with intersys-tem cross-linking to the triplet state. Subsequent reaction of the uracil-bromine atom doubled radical pair may yield cross-linking to a nearby amino acid. These radicals were proposed to have additional reactivities. In the second scenario, excitation of the BU chromophore with UV may result in C-Br bond homolysis in an upper singlet state and single electron transfer from a neighboring group in the triplet state. Even though BrdU (HBr) was often lost after protein cross-linking to nucleic acids (48), it also was possible to observe BrdU modifications in cross-linked peptides. In addition, because bromine was at position C5 in BrdU, it may be stabilized as was suggested for APOBEC3A in respect of edited nucleotide that is engaged in T-shaped -stacking interaction with Tyr-130, Thr-31, and His-70 (42). Importantly, because of the naturally occurring bromine stable isotopes ( 79 Br and 81 Br, typically present at an ϳ1:1 ratio), tryptic peptides cross-linked to BrdU-containing ssDNA may be distinguished in MS mass-to-change (m/z) spectra as double peaks of about similar intensity with a 1.998 atomic unit difference for monoisotopic peptides or 0.999 m/z difference for ϩ2 charge peptides, etc. Although the presence of stable 13 C, 15 N, and 2 H isotopes may complicate MS analysis, using BrdU-modified ssDNA may facilitate the identification of cross-linked peptides in MS samples and individual crosslinked residues within the MS/MS tryptic peptide fragmentation data.
A 25-nt ssDNA with a single centrally located BrdU adjacent to the dC for editing was cross-linked to A3G, and the crosslinked product in the reaction was resolved (gel shifted) from uncross-linked protein by SDS-PAGE. Protein subjected to UV irradiation and migrating with the molecular weight of monomeric A3G served as control for maximum MS peptide identification over the primary sequence of A3G. Extensive nuclease treatment of these samples prior to MS analysis ensured that they would only have one nucleotide modification at the crosslinked position. Gel-shifted and control samples were analyzed using Q Exactive (Thermo Fisher Scientific) mass spectrometer, and the MS data were initially searched for differences in peptides identified in the ssDNA cross-linked in A3G sample due to altered m/z compared with those identified in control A3G lacking ssDNA cross-linking. Peptides corresponding to these differences were analyzed for 79 Br and 81 Br BrdU-modified double peak peptides with Compass Software (Isotope-Pattern, Bruker Daltonics) and the chemistry of modification set as corresponding to BrdU, C 9 H 11 N 2 O 8 P 1 Br. Two sets of peaks, with m/z values in the range of 1111-1113 (charge ϩ4) and 663-665 (charge ϩ3), were identified that showed MS1 peaks matching the predicted m/z values of the A3G-BrdUmodified peptides, aa 181-213, sequence YYILLHIMLGEILR-HSMDPPTFTFNFNNEPWVR (one missed trypsin cleavage) (Fig. 1A) and aa 314 -326, sequence IYDDQGRCQEGLR (one missed trypsin cleavage) (Fig. 1B), suggesting that these peptides contain cross-linked 79 Br/ 81 Br-modified amino acid residues (see details under "Experimental procedures" and in Fig. 1,  A and B, legend). MS/MS fragmentation data of these peaks were in agreement with corresponding peptide sequences (Fig.  2, A and B). To identify individual amino acid residues crosslinked to BrdU within A3G peptides determined by Compass IsotopePattern Mascot, searches were performed with param-eters set to identify BrdU modifications of residues commonly known to cross-link to nucleic acids: cysteine, tyrosine, and phenylalanine and also tryptophan, histidine, methionine, ser-ine, arginine, and lysine (49). Tyr-181 in tryptic peptide aa 181-194 and Tyr-315 in tryptic peptide aa 314 -320 were identified as BrdU-modified residues in cross-linked samples (MS/MS   79 Br and 81 Br, including 13 C (mostly) 2 H, and 15 N, the MS1 pattern of each shown A3G peptide analyzed by Xcalibur and Compass is represented by several peaks (typically 2-4, depending on peptide change, with decreasing relative intensity). As evident from Compass IsotopePattern analysis (panel c), we observed eight peptide peaks in an area of 1111-1113 m/z that at the charge 4ϩ and peak distance ϳ0.5 m/z (not counting common 13 C-double peaks) may constitute two identical peptides that differ only by ϳ2.0 m/z that corresponds to bromine-isotope difference of 79 79 Br and 81 Br, including 13 C (mostly), 2 H, and 15 N, the MS1 pattern of each shown A3G peptide analyzed by Xcalibur and Compass is represented by several peaks (typically 2-4, depending on peptide change, with decreasing relative intensity). As evident from Compass IsotopePattern analysis (panel c), we observed six peptide peaks in an area of 663-665 m/z that at the charge 3ϩ and peak distance ϳ0.67 m/z (not counting common 13 C-double peaks) may constitute two identical peptides that differ only by ϳ2.0 m/z that correspond to BrdU-isotope difference of 79 Br and 81 Br in BrdU cross-linker. spectra of these peptides with indicated positions of modified tyrosines are shown in Fig. 2, A and B). For example, the m/z value of b1 1ϩ ion corresponding to Tyr-181 in peptide aa 181-194 was found to be 548.01, which is much higher than the expected value of m/z 163.06 for tyrosine alone, and the difference between the observed and expected (tyrosine alone) m/z value was equal to the m/z of BrdU. Similarly, an observed m/z value for the b2 2ϩ ion, Ile-314, Tyr-315 in peptide aa 314 -320, was 331.11 and not 138.57 as if Tyr-315 was not cross-linked to BrdU (note that this peptide has 2ϩ charge). The supplemental Tables 1 and 2 list all observed ions to complement MS/MS fragmentation data presented in Fig.   2. Both tyrosines at positions 181 and 315 were confirmed to be modified by cross-linked BrdU using BioTools software to manually examine the fragmentation patterns of the corresponding peptide spectra observed during liquid chromatography separation and fragmentation (MS/MS data).

A3G tyrosines 181, 182, and 315 are involved in ssDNA binding
To obtain definitive proof that Tyr-181 and Tyr-315 in A3G actually are involved in ssDNA binding, point mutants were created that replaced Tyr-181, Tyr-182, and Tyr-315 to neutral alanine residues. Tyr-182 was included in our studies even though it was not identified by MS as being cross-linked to BrdU because it was adjacent to Tyr-181, and we wanted to evaluate its potential to bind ssDNA when Tyr-181 was mutated. The corresponding mutant constructs were expressed as full-length proteins in Hi-Five insect cells using the Baculovirus protein expression system and purified by nickel-nitrilotriacetic acid affinity resin with a final concentration of 1-3 mg/ml soluble protein (supplemental Fig. S1).
Thermal fluorimetry (thermal shift assays) determines the temperature-dependent ordered unfolding of proteins as detected by SYPRO Orange dye binding to exposed hydrophobic regions and was used here to assess mutant protein folding relative to wild-type (WT) A3G. Purified WT A3G and all three single tyrosine replacement proteins showed near identical dissociation curves as indicated by detection of comparable amounts of SYPRO Orange dye binding across a temperature gradient with only minor difference in melting temperature (supplemental Fig. S2). Such minor differences in melting temperature (not exceeding 2-3°C) typically are associated with pipetting variations and not with altered protein stability (50).
WT and mutant A3G protein nucleic acid binding was quantified by fluorescence anisotropy (FA) over a range of A3G input (Fig. 3A). The binding assays showed that the tyrosine mutants had diminished affinity for substrate ssDNA (Fig. 3A). Y181A mutant had a k d ϭ 4.18 Ϯ 0.79 (n ϭ 3) for binding to 25 nt of ssDNA that was about 2-fold lower than the k d value observed for WT A3G, 2.18 Ϯ 0.17 (n ϭ 3). Y182A mutant protein had a k d value for ssDNA binding that was similar to Y181A (data not shown). In marked contrast, the Y315A mutant showed substantially lower affinity to ssDNA, k d ϭ 45.3 Ϯ 9.65 (n ϭ 3). The data suggested that Tyr-315 was critical for ssDNA binding in vitro and Tyr-181/Tyr-182 that are more distal from CD2 had less significant roles in coordinating ssDNA binding.

A3G Tyr-315 is critical for the DNA mutator phenotype
We reasoned that because alanine substitutions of A3G Tyr-181, Tyr-182, and Tyr-315 reduced substrate ssDNA binding to A3G, the corresponding A3G mutants might also have reduced deaminase activity. WT and mutant A3Gs were expressed in an Escherichia coli DNA mutator reporter system that quantifies growth of bacterial colonies on plates under rifampicin selection due to mutations in the reporter ␤-subunit of RNA polymerase gene rpoB that confer rifampicin resistance (51)(52)(53). Plasmids containing WT and mutant A3G were transformed into reporter strain, and transformants expressing similar amounts of full-length WT or mutant A3G (supplemental Fig.  S3) were plated onto rifampicin selection media.
WT A3G produced on average 25 Rif R colonies per 10 8 cells (Fig. 4). In agreement with the slight reduction in ssDNA binding seen with both single Y181A and Y182A proteins, these mutants had ϳ2-fold fewer rifampicin-resistant colonies relative to WT A3G. Bacterial transformants expressing A3G Y181A/Y182A double mutation did not further decrease the number of rifampicin-resistant colonies. The data suggested that although MS analysis identified Tyr-181 as predominantly involved in ssDNA binding, we cannot rule out that Tyr-181 or  Mutational activity of A3G mutants due to the expression of cytidine deaminases in E. coli was determined by using a bacterial reverse rifampicin resistance reporter assay (␤-subunit of RNA polymerase) (51,53). Mutational frequency for each WT or mutant A3G strain was calculated as the number of rifampicin-resistant colonies per 10 8 of viable cells after making correction for the number of colonies on control plates and shown as scatter diagrams with the median value drawn through (red lines). Each experiment was repeated four times. Bacterial strains tested were vector (vector alone); wild type (wild-type A3G); Y181A/Y182A (A3G with double Y181A and Y182A mutations); Y181A (A3G with a single Y181A mutation); Y182A (A3G with a single Y182A mutation); Y315A (A3G with a single Y315A mutation); Y181A/Y182A/Y315A (A3G with a triple Y181A, Y182A, and Y315A mutations).
Tyr-182 could support ssDNA binding and deaminase activity. The activity of single-site mutants versus double-site mutants suggested that simultaneous occupancy of Tyr-181 and Tyr-182 by ssDNA was not required.
Few or no rifampicin-resistant colonies were induced by the A3G Y315A mutant that was similar to the number of Rif R colonies seen with empty vector control. The data showed that A3G Y315A mutant lost Ͼ90% of WT A3G DNA mutagenic activity. Taken together, the DNA mutator assay results suggested that binding of ssDNA to Tyr-315 was essential for A3G deamination of dC to dU in ssDNA substrates, consistent with the highly conserved position of this residue at an entry point for ssDNA relative to catalytic residues in zinc-dependent deaminase domain (1,42,54).

Tyr-315 mutation reduces the formation of A3G-containing high-molecular-mass RNPs in mammalian cells
We recently reported that RNA as well as ssDNA binds to A3G tryptic peptides in the CD2 domain (10). Quantitative FA data showed that WT A3G had a slightly higher affinity for the 25-nt RNA than for ssDNA (k d ϭ 1.39 Ϯ 0.14 and k d ϭ 2.18 Ϯ 0.17, (n ϭ 3) respectively) (Fig. 3B). Our FA experiments also showed that the Y181A mutant had about 2.5-fold reduced affinity for RNA binding than WT A3G (similar to the difference in ssDNA binding). In contrast to prior studies that used non-quantitative RT-PCR to assess the role of Tyr-315 in RNA binding (15), the Y315A mutant showed a 10-fold reduction in k d values for RNA as compared with WT A3G (k d ϭ 13.85 Ϯ 1.75 (n ϭ 3), see Fig. 3B). It is noteworthy that the Y315A mutation reduced binding to ssDNA to a greater extent than to RNA (ϳ20-fold versus ϳ10-fold, respectively). We suggest that this may not be solely due to the relative higher affinity of A3G to RNA, but rather it could have been due to the greater number of RNA-binding residues distributed in the N-and C-terminal domains of A3G compared with the three peptides found to bind to ssDNA.
Velocity sedimentation of A3G WT, Y181A, Y182A, and Y315A mutant proteins from the cell extracts of transfected HEK293T cells was evaluated to confirm the effect of these mutations on RNA binding in cells. Cellular A3G typically is assembled in high-molecular-mass RNP complexes that are formed through RNA-bridged protein-protein interactions in cytoplasm processing P-bodies (20 -25). A3G ribonucleoproteins are of megadalton molecular mass and therefore sediment to the high-density fractions in glycerol gradients during velocity sedimentation analysis (34,44,55). The majority of WT A3G was recovered in high-density fractions and at the bottom of the glycerol gradient (Fig. 5, left panels). Y181A and Y182A mutants sedimented as more evenly distributed complexes and were recovered throughout the gradient. Therefore, relative to WT A3G these two mutants had reduced capacity for RNA-dependent oligomerization into high-molecular-mass complexes. RNase A digestion of cell extracts prior to fractionation markedly reduced the sedimentation of Y181A and Y182A mutants to that comparable with WT A3G following treatment with RNase A (Fig. 5, right panels). Similar sedimentation and RNase A digestion sensitivities were observed for the double Y181A/ Y182A mutant (data not shown).
In contrast, Y315A mutant protein showed a significantly altered profile of sedimentation as compared with WT A3G or Y181A and Y182A (Fig. 5, bottom left panels). Y315A sedimented in fractions typically seen only for WT A3G after RNase A digestion. RNase A digestion of Y315A cell extract resulted in relatively minor changes in the sedimentation of the mutant protein. These findings suggest that most if not all of the HEK293T cellular RNA binding to A3G and A3G oligomerization was dependent on the interaction of RNAs with Tyr-315. Moreover, the sedimentation of Y315A without or with RNase digestion to low-molecular-mass regions of the gradient comparable with WT A3G following RNase digestions supported the conclusion from thermal fluorimetry that Y315A was not aggregated. Given the essential role of Tyr-315 in ssDNA deamination, bulk cellular RNA binding to Tyr-315 may account for the inhibition of A3G antiviral activity reported in the literature when A3G becomes sequestered in P-bodies. Although previously it was believed that residues in A3G N-terminal CD1 were primarily responsible for RNA binding, A3G oligomerization, virion incorporation, and sequestration as RNP in P-bodies (21)(22)(23), mutations of residues within the N terminus predicted to bind RNA did not all affect RNA binding nor were they able to uniformly affect A3G binding to all RNA sequences (7, 15, 33, 56 -58). Our data suggested a novel and unanticipated role for Tyr-315 in RNA binding and support a two RNA binding domain model for A3G (59).

Tyr-315 is required for robust A3G antiviral activity
A3G binding to HIV RNA (11,13,16,21,60) and host cell 7SL RNA (11,12,14,16,30,61,62) are thought to be essential for assembly of A3G with nascent viral particles (in addition to interactions with HIV Gag protein (60,63,64)). We evaluated the importance of tyrosines 181, 182, and 315 for A3G incorporation with viral particles using pseudotyped HIV-1 virus particles produced in HEK293T cells. These studies were conducted with pseudotyped HIV lacking Vif expression to prevent Vif-mediated A3G degradation thus preserving the cellular abundance of A3G necessary for assembly with virions. The expression of WT and mutant A3G proteins in transiently transfected HEK293T cells was quantified by determining the ratio of the whole-band densitometry for A3G-specific Western blotting signals in cell extracts relative to densitometry for control GAPDH-specific western blotting signals present in each lane (supplemental Fig. S4, top panel, ratio values are shown below each lane/sample). Based on such assessment, A3G protein expression in transfected cells varied no more than 2.5-fold between WT and mutants proteins. Viral particles assembled and released from each transfected cell culture were analyzed for A3G content relative to HIV p24 capsid protein by calculating the ratios of the specific western blotting signal densitometry for these proteins in the same lane (supplemental Fig.  S4, bottom panel, ratio values are shown below each lane). Both Y181A and Y182A mutant proteins were detected in total cell extracts and within viral particles in similar relative abundance compared with WT A3G. This suggested that the RNA-binding activity of A3G Y181A and Y182A mutants was sufficient for their assembly with virions and that RNA binding contributed by these residues is not essential for A3G incorporation within virions. In contrast, recovery of the Y315A mutant protein with viral particles was reduced as compared with its abundance in cell extracts (ratios 0.5 and 2.0, respectively). These results suggest that although Tyr-315 was required for RNA binding, the Y315A mutation markedly reduced but did not completely abolish A3G assembly with virions. In fact, the literature suggests that A3G association with the viral particle also depends on its interaction with the nuclear capsid portion of Gag (18,63,65). Antiviral activity of A3G requires its assembly within the viral core in association with HIV genomic RNA placing A3G proximal to reverse transcription complexes and immediate access to ssDNA for CDA activity (20,66). It is anticipated that A3G mutations that reduce its packaging within virions will preserve the infectivity of those virions. To evaluate the effects of tyrosine replacement mutants on A3G antiviral activity, p24 normalized viral particles were used to infect HeLa-based TZM-bl reporter cells that express luciferase-reporter from a TAT-transactivated HIV LTR promoter.
As anticipated, WT A3G assembled with viral particles markedly suppressed their infectivity (Fig. 6). The infectivity profiles of viral particles containing Y181A and Y182A mutants were similar to those observed for virions containing WT A3G. This finding was consistent with recovery of these A3G mutants with virions and their deaminase activity in the bacterial DNA mutator assay. Virions containing A3G Y315A mutant showed substantial infectivity (greater than 60% of control virions infectivity that lacked A3G). Although Y315A mutant had significantly diminished ssDNA binding and CDA activity, low levels of Y315A were incorporated with virions. Therefore, we cannot rule out some antiviral activity through a deaminaseindependent mechanism (67-71), or residual deaminase activity may result from the Y315A proteins assembled with virions.

Discussion
The mechanism by which RNA inhibits A3G ssDNA deaminase activity has long been of interest and thought to solely involve RNA-bridged oligomerization of A3G through its N terminus. In this report, site-directed mutagenesis of candidate nucleic acid-binding residues was informed by MS of tryptic peptides from native and full-length A3G cross-linked to ssDNA and by binding assays of the mutant proteins to ssDNA and RNA. We demonstrate that Tyr-315 is essential for ssDNA binding and consequently deaminase activity on ssDNA. RNA binding to the C-terminal CD of A3G, and directly to Tyr-315, is shown for the first time. The data shift the paradigm in the field from a single RNA-binding domain model to a dual RNAbinding domain model (59) and provide a rational explanation for why RNA is a competitive inhibitor of A3G ssDNA binding and deaminase activity on ssDNA (2,10,36). MS and FA revealed Tyr-181 and Tyr-182 as novel N-terminal ssDNAbinding residues. In this regard, the predicted coordination of ssDNA resulting from binding to Tyr-181/Tyr-182 provided a rational explanation for why full-length A3G is required for maximal deaminase and ssDNA-binding activity even though the N-terminal CD is catalytically inactive.
We began this study by determining the amino acids in A3G bound to ssDNA. MS data suggested that BrdU containing ssDNA substrates cross-linked to A3G at tyrosines Tyr-181 and Tyr-315. Replacement of each of these tyrosines for alanines did not affect the general protein folding but showed that Tyr-315 was essential for ssDNA binding and deaminase activity. Photo-cross-linking can only occur when the residues in A3G protein have close proximity with nucleotides approaching 0 Å (49). Therefore, we do not rule out that residues other than Tyr-181, Tyr-182, and Tyr-315 also have roles in coordinating nucleic acid binding or in the mechanism for dC to dU deamination.
Crystal structure of the catalytic CD2 domain of A3G and structures of CD2 co-crystallized with mononucleotide or CCC-trinucleotide predicted several different possibilities for substrate ssDNA binding depending on how DNA bends into Figure 6. Infectivity assays of A3G WT and tyrosine mutants showed that Tyr-315 mutant lost antiviral function. Infectivity of TZM-bl cells is measured by the RLU from the expression of a stably integrated HIV LTR-driven luciferase gene. TZM-bl reporter cells were infected with 500 ng of p24-normalized pseudotyped virus particles containing WT A3G in presence of Vif and WT and mutant A3G without Vif. SteadyGlo Reagent (Promega) was used to detect the level of luciferase expression and RLU were quantified. The relative percent change in infectivity of WTA3G and mutants without Vif was compared with infectivity with WT A3G in the presence of Vif, which was set as the maximum infectivity (100%); n ϭ 3 with error bars representing standard deviations. the corresponding protein groove: "brim" model (31), "kinked" model (38,40), and "straight" model (39). Interestingly, all these models predicted an involvement of peptide sequence in loop 7, RIYDDGQR, aa 313-320, particularly residues Arg-313, Tyr-315, Asp-316, and Asp-317 in ssDNA binding. Loop 7 of A3 proteins has been proposed to be a nucleotide specificity box (72,73). A3F structural studies predicted that residues Tyr-307/308, Phe-309, and Trp-310 in loop 7 may form favorable stacking interactions with the deaminase hot spot (5Ј-TTCA-3Ј) in ssDNA substrates. These aromatic residues are not conserved across the A3 family and are replaced by YYFQ and YDDQ in A3C and A3G, respectively (74). Mutating the YDDQ box in A3G to the aromatic-rich YYFW of A3F resulted in a change of substrate preference from 5Ј-CC-3Ј to 5Ј-TC-3Ј or 5Ј-GC-3Ј, suggesting that mutating these four residues are sufficient in determining the nearest neighbor nucleotide context for deamination (73). Similarly, a D317Y mutation in loop 7 of the C terminus of A3G (Tyr-132 is the homologous position in A3A) was sufficient to alter the local dinucleotide preference for deamination from 5Ј-CC to 5Ј-TC (75). The other A3G residues within CD2 predicted to coordinate binding of ssDNA are His-216 (in the kinked model) and His-367 (in the straight model) along with Asn-244, Trp-285, Arg-374, and Arg-376 (38 -40).
Arg-374/376 residues located on helix 6 in the C terminus of A3G with amino acid sequence LDEHSQDLSGRLR have been predicted to interact with ssDNA (38,40). We consistently observed A3G C-terminal peptide aa 345-374 as "missing" in our MS analysis of BrdU ssDNA cross-linked samples as compared with uncross-linked A3G, suggesting that it was chemically modified after cross-linking to ssDNA (and RNA) (10). This peptide is 30 residues long (molecular mass without modification is 3514 Da), so high confidence identification of the modified residue has been problematic due to MS technical limitations. Similarly, Arg-376-containing tryptic peptide is only a dipeptide, LR, and too small for accurate MS detection; therefore, a different approach is required for identification of cross-linked residue in A3G.
While this paper was being reviewed, APOBEC3A (contains only one CD domain) and an A3A loop 1-substituted A3B cocrystal structure with a ssDNA oligodeoxynucleotide were reported (42). The structures suggested that the deaminase hot spot nearest neighbor preference in ssDNA, 5Ј-TC, was accommodated by bending the ssDNA substrate such that the 5Ј-T is rotated out of the pocket through hydrogen binding with flexible loop domains, and the deaminated C is positioned deep with the catalytic cleft through interactions with several residues key to zinc coordination and catalysis. It is of interest to note that Tyr-315 is conserved among A3 proteins and was shown positioned in the co-crystal structure with 5Ј-TTTT-CAT-3Ј deoxyoligonucleotide where it was predicted to have rotational flexibility and serve in gating ssDNA access to the catalytic site, creating an opened or closed catalytic cleft conformation (42).
Although we found no evidence for BrdU cross-linking to catalytic residues in CD2 known to be involved in zinc ion coordination, these interactions may not have been stable enough for cross-linking or not positioned appropriately for cross-link-ing to the BrdU nucleotide. It is important to keep in mind that A3G models based on crystal structures of half of the protein may be conditionally accurate and that Tyr-315 becomes an essential high affinity binding site for ssDNA only in the context of full-length A3G, which retains the capability of forming holoenzyme oligomers that are required for robust deaminase activity (34,76). We cannot rule out that there may have been local misfolding due to tyrosine substitution that affected RNA or DNA binding even though we demonstrated expression of soluble full-length A3G mutant proteins with appropriate sedimentation characteristics (without and with RNA) and with near WT A3G thermo fluorescent dye-binding profiles.
MS also identified candidate RNA-binding sites. Earlier computational modeling of APOBEC structures for exposed residues that may bind to nucleic acids (15,38,40,77) predominantly focused on mutagenesis of the N terminus of A3G. Tyr-315 was evaluated in these studies but found not to be important for RNA binding but important for deaminase activity and viral restriction (15). Non-conservative mutation of Tyr-315 had only a moderate decrease in binding to RNAs co-assembled with A3G in virions such as hY1 RNA and no significant change in hY4 and 7SL RNA binding (15). These analyses were based on fixed cycle number RT-PCR detection of RNAs. We propose these studies may have overestimated how much RNA was bound to Tyr-315 mutants as the template-saturating conditions of RT-PCR are not quantitative. We cannot rule out that RNA primary sequence, RNA size, and RNA secondary structure differences may have contributed to the discrepancy in experimental findings.
Tyr-181 (Tyr-182) in the N terminus of A3G also crosslinked to ssDNA and had a moderate contribution to ssDNA binding. Although these mutations did not impair packaging in viral particles, each resulted in an ϳ50% loss of deaminase activity. Earlier A3G site-directed mutagenesis studies suggested that mutations within the CD1 domain moderately reduced CDA activity of the full-length A3G (29) and affected A3G oligomerization required for processive deamination (15,78). In fact the C-terminal catalytic domain had little or no deaminase activity when expressed without the N terminus (29,79). The importance of the N terminus for deaminase activity was underscored by the high viral load and decreased CD4 ϩ T-cell phenotype associated with the naturally occurring mutation H186R in A3G (80). Although A3G-H186R retained the ability to incorporate into HIV-1 virions and retained catalytic activity (as did the Y181A mutant), it had a decreased efficiency in mutagenesis due to reduced scanning along ssDNA (78). A3G CD1 W94A/W127A double substitutions altered ssDNA substrate specificity from 5Ј-CC to 5Ј-TC (56). W127A substitution also reduced the ability of A3G to scan ssDNA for new deamination sites presumably because of the loss of protein homodimerization (81). Considering these findings Tyr-181/ 182 may be part of a domain that serves as an ssDNA coordination site for bending ssDNA bound to the catalytic site (42) or facilitates ssDNA-dependent oligomerization of A3G as dimeric ssDNA recognition complexes (34) or higher order holoenzyme complexes (33,34,82). Residues in the CD2 ssDNA entry site may coordinate with Tyr-315 in the context of full-length A3G to stabilize ssDNA at the catalytic site and thereby enhance the catalytic efficiency of A3G.
Our data on A3G mutant binding to nucleic acids point to physical evidence for how RNA competitive inhibition of ssDNA binding could be mediated by RNA occupancy and displacement of ssDNA from Tyr-315 within the C terminus of A3G. Previously, we demonstrated through kinetic competition binding assays that RNA directly inhibited A3G CDA activity (36). FA, EMSA, and near equilibrium binding studies showed that A3G assembly and disassembly on ssDNA was an ordered process involving A3G dimers and multimers but that RNA stochastically dissociated A3G dimers and higher-order oligomers from ssDNA (10). Given that Y315A mutation significantly affected binding of both ssDNA and RNA, we suggest that the mechanism of inhibition of ssDNA binding and CDA inactivation by RNA can be explained by direct competitive binding to Tyr-315.
Y315A mutation also had a profound inhibitory effect on A3G assembly with virions in pseudo-type virus-infected cells and the formation of high-molecular-mass RNP in uninfected cells. Y181A and Y182A only moderately reduced the aggregate sedimentation of RNP but were not able to prevent A3G assembly with virions. Interestingly, Pan et al. (83) confirmed the existence of two separate RNA-binding sites, with different affinity to A3G by using atomic force microscopy.
MS analysis of A3G cross-linked to RNA suggested that in addition to peptides in the C terminus, multiple peptides in the N terminus of A3G cross-linked to RNA (10). Mutations of Trp-94 and Trp-127 in A3G significantly reduced binding to several RNAs tested, namely HIV, 7SL, hY1, and hY3 RNAs as determined by A3G immunoprecipitation and RT-PCR (11,17,56,84). S28E, Y124A, and F126L mutants showed alterations in binding to these RNAs (84). A3G W94A/W127A double replacement mutant protein remained localized to P-bodies and exerted DNA mutator activity in bacterial reporter strain but showed no to low HIV restriction despite packaging with virions (56). Molecular modeling suggested that residues Trp-94 and Tyr-124 -Trp-127 were involved in protein-protein intermolecular interaction in an A3G homodimer formation during nucleic acid binding (27,84). A critical distinction of Y315A mutant phenotype contrasting with the W94A/W124A double mutant is the requirement of Tyr-315 for A3G cellular and antiviral functions. Sedimentation in glycerol density gradient (Fig. 5) and thermal shift assays (supplemental Fig. S2) suggested that the phenotype of Y315A was not due to aggregation or misfolding. Further study of the region containing Tyr-181, Tyr-182, and Tyr-315 may provide understanding of how A3G coordinates ssDNA substrates for process deamination and how it is regulated as a host defense factor and potentially targeted therapeutically.
In this study, we have used an AU-rich RNA probe to identify candidate general RNA-binding sites on A3G. An important open question that can now be addressed in future studies will be whether there are residues in either the N or C terminus of A3G that selectively bind to different RNA sequences or secondary structures and how this determines A3G cellular and antiviral functions.

Construction of A3G mutants
The N-His 6 -HA-A3G construct containing the wild-type full-length human A3G cDNA with N-terminal His 6 and HA epitope tag was previously cloned in our laboratory into pIRES vector (85). It should be noted that the His 6 tag did not affect A3G function because tagged protein binding affinity to nucleic acids and its CDA activity in bacterial test system was not affected. For protein production, the N-terminal His 6 -A3G DNA fragment was subcloned into EcoRI/XhoI restriction sites of pFastBac1 vector (Thermo Fisher Scientific), and the resultant construct was verified by DNA sequencing. His 6 -A3G mutant constructs containing Y181A, Y182A, and Y315A replacements in vector pUC57 were obtained from GeneScript, Inc. (Piscataway, NJ), and then subcloned into EcoRI/XhoI restriction sites of pFastBac1 and separately in the same sites of pIRES vectors. In addition, A3G WT and mutant sequences were subcloned into EcoRI/XhoI sites of pTrc99A vector under IPTG-inducible promoter to use in DNA mutator assays.

A3G protein expression and purification
To produce A3G WT and mutant proteins, the corresponding His 6 -A3G constructs in pFastBac1 were transformed into E. coli DH10Bac helper strain to produce the corresponding bacmids that then transfected into Sf9 insect cells with X-tremeGENE HP DNA transfection reagent (Roche Applied Science). All other procedures with insect cells were performed as described in the Bac-to-Bac manual (Thermo Fisher Scientific), including obtaining the initial viral stock and transducing the High-Five insect cells with recombinant viruses for protein production. A3G protein purification was carried out as described previously (10), frozen in liquid nitrogen, and stored at Ϫ80°C until needed.

Thermo fluorometry (thermal shift assays)
A3G WT and mutant protein thermal shift assays were performed using Stratagene Mx3000P real-time quantitative PCR system instrument (Hauptman-Woodward Medical Research Institute, Buffalo, NY). Reactions were carried out in Micro-Amp Optical 96-well reaction plates in 30-l reactions with 1ϫ PBS buffer containing 5 M protein (final concentration) and 2 l of 1:500 diluted SYPRO Orange dye (Sigma). The assays were performed by heating the plates at ϳ2°C/min from 25 to 95°C and reading the fluorescence of the sample. Protein samples were quantified in triplicate. Protein dissociation curves were plotted from the data using GraphPad Prism 6 software.

A3G assembly with nucleic acids and EMSA
Complexes containing A3G bound to nucleic acids were prepared by incubating varying amounts of A3G and varying molar ratios of AlexaFluor647-labeled oligonucleotides in deaminase buffer (1ϫ DB, 40 mM Tris, pH 7.2, 50 mM NaCl, 10 mM MgCl 2 , 1 mM DTT, 0.1% Triton X-100, 2% glycerol) for 20 min at 37°C. Complexes assembled in each reaction condition were resolved on a 5% native gel (36) and visualized using a Typhoon TM 9410 Trio Imager (GE Healthcare) by excitation at 633 nm wavelength and measuring fluorescence at 670 nm.

Equilibrium fluorescence anisotropy
Binding of A3G WT and mutant proteins were quantitatively measured by using equilibrium fluorescence anisotropy. Alexa-Fluor647-labeled ssDNA or RNA was added to each reaction at a fixed 2 nM concentration and incubated with purified 0 -40 nM A3G proteins at 37°C for 20 min in 1ϫ DB buffer. Fluorescence anisotropy measurements were carried out in Fluoromax-4 fluorometer (Horiba Scientific, Edison, NJ), excitation 647 nm, and emission 670 nm with 5-nm band passes, each sample was done in triplicate with the mean values calculated. The change in anisotropy was calculated by subtracting the average anisotropy of the free 5Ј-labeled ssDNA or RNA from the average anisotropy of the A3G/nucleic acid assembly reactions and plotted against A3G concentration. The K d value was calculated through non-linear regression by fitting the data to the following equation, where A3G/NA is the fraction of A3G bound to nucleic acid (NA) and B max is the enzyme's capacity to bind substrate: A3G/NA ϭ B max ([A3G])/(K d ϩ [A3G]).

UV-induced cross-linking of A3G to ssDNA and sample preparations for MS
Photo-cross-linking of proteins to nucleic acids is frequently used to map interacting regions/sites (49). The amino acids commonly identified in photo-adducts are cysteine, methionine, serine, arginine, lysine, histidine, phenylalanine, tryptophan, and tyrosine. Pyrimidines, especially uridine, are typically 10 times more effective for cross-linking to amino acids as compared with purines. Photo-cross-linking in combination with high-resolution MS provides an opportunity for direct identification of interacting sites (86). However, often because of chemical heterogeneity in the composition of the cross-linked entity, it has been challenging to identify the amino acid-nucleosides within cross-linked peptides.
A3G was assembled with 25-nt ssDNA containing BrdU at a 4:1 molar ratio (protein/ssDNA) in 1ϫ PBS at 37°C for 20 min, cooled on ice, and irradiated for 20 min with 302-nm wavelength UV light. The A3G cross-linked products were separated onto 12% SDS-PAGE and visualized after staining with Coomassie SimpleBlue SafeStain (Thermo Fisher Scientific) as proteins bands with slower mobility in the gel as compared with monomeric A3G. A3G cross-linked to RNA showed a marked reduction in mobility causing the bands to be positioned near the top of the resolving gel. Monomeric A3G and A3G crosslinked protein bands were cut out from SDS-polyacrylamide gels, washed three times in 25 mM Tris, pH 7.5, 2 mM EDTA, and processed as described previously (44). Briefly, ssDNAcross-linked samples were digested with 10 g of DNase I (Sigma) and 300 units of micrococcal nuclease (Thermo Fisher Scientific) for 4 -6 h at 37°C in presence of Mg 2ϩ and Ca 2ϩ . The gel slices were washed three times with 25 mM ammonium bicarbonate (AmBc), dehydrated with solution containing 50% acetonitrile and 25 mM AmBc, and dried out in a vacuum-speed centrifuge. The gel slices were rehydrated with 25 mM AmBc, and then two more rounds of the dehydration-rehydration procedure were performed to remove any residual amounts of nucleases or digested nucleic acids. Samples were then treated with 10 mM DTT at 55°C for 1 h and iodoacetamide at room temperature in the dark for 45 min. The samples then were washed with 25 mM AmBc, dehydrated with 50% acetonitrile and 25 mM AmBc solution, vacuum-dried, and digested overnight with MS-grade trypsin (G-Biosciences, 0.5 g/l) in 25 mM AmBc at 1:20 ratio overnight. The extracted peptides were then purified with C18 resin (Thermo Fisher Scientific) as suggested by the manufacturer and concentrated to 10 l in a vacuum-speed centrifuge.

Analysis of A3G tryptic peptides by mass spectrometry
MS analysis of tryptic peptides was performed using either an LTQ Orbitrap XL or Q Exactive Plus Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific). Samples were adjusted to 0.1% formic acid in water and vortexed before autosampler loaded onto a home-pulled and home-packed C 18 analytical column. Columns were pulled to a tip of ϳ10 m with a Sutter laser puller and packed using a pressure bomb to 10 cm with C 18 AQ 5 m of 200 Å media (Michrom), ending with an internal column diameter of 75 m. Columns were equilibrated to initial run conditions prior to loading the sample on the column. Solvent A was 0.1% formic acid in water, and Solvent B was 0.1% formic acid in acetonitrile. Peptides were eluted with the following chromatographic profile: 0% B for 2 min, ramping to 40% B over 23 min then to 70% B over 1 min, remaining at 70% B for 3 min, and finally returning to initial run conditions. Data were collected as RAW files and converted to MGF files using Bioworks Browser. Initially, when analyzing peptide spectra we took an advantage of the fact that bromine in BrdUcontaining ssDNA is present in two naturally occurring stable isotopes ( 79 Br and 81 Br, usually at ϳ1:1 ratio); therefore, the cross-linked peptides should produce double MS1 peaks with similar intensity and with about 1.998 atomic units difference for ϩ1 m/z charge peptides, about 0.999 unit for ϩ2 m/z charge peptides, etc. It should be noted that because of the presence of other stable isotopes besides 79 Br and 81 Br in the samples, including 13  To identify BrdU-modified residues in A3G tryptic peptides, we used MASCOT software (Matrix Science). Search parameters included the following: trypsin as an enzyme; three missed cleavages; 0.05 Da for MS and MS/MS data; 1 for #13C, 2ϩ; ϩ3; ϩ4 for charge state; decoy search and acceptance criteria of a minimum one peptide greater than identity score; minimum score of 20; and false discovery rate less than 5% (87). To identify individual amino acid residues cross-linked to BrdU within A3G peptides determined by Compass IsotopePattern software, Mascot searches were performed for BrdU modifications at cysteine, tyrosine, phenylalanine, tryptophan, histidine, methionine, serine, arginine, or lysine that are all known as being involved in cross-linking to nucleic acids (49). Also, to facilitate identification of such peptides, we created a custom database that included only A3G protein. To verify the A3Gidentified cross-linked peptides, we used BioTools software (Bruker Daltonics) to manually analyze MS/MS fragmentation patterns and to determine individual cross-linked residues within these peptides by matching the observed peak values with anticipated modified residues.

A3G mutational activity (DNA mutator assay)
The DNA mutator assay measures the induction of rifampicin resistance in E. coli that is caused by the accumulation of mutations in the rpoB gene encoding the ␤-subunit of RNA polymerase (51,53). Briefly, E. coli strain BW310 Hfr  relA1 spoT1 thi-1 ung-1 deficient for uracil DNA glycosylase was transformed with pTrc99A vector alone or plasmids encoding WT or mutant A3G. Parental strain KL16, which expresses normal uracil-DNA glycosylase and can repair deaminase-induced dC-to-dU mutations, was used as a control. Single colonies of each transformation in triplicates were grown overnight in the presence of carbenicillin and 1 mM IPTG, and the aliquots of 10-fold dilutions were plated on carbenicillinand rifampicin-containing plates. Mutation frequency was calculated as the number of rifampicin-resistant colonies per 10 8 of viable cells after making corrections for the number of colonies on control plates. Each experiment was repeated four times independently.

Glycerol density gradient fractionation
Glycerol density gradient sedimentation was used to determine the association of A3G with RNA as RNP complexes (34,44). WT and mutant A3G constructs subcloned in pIRES vector (85) were transfected into HEK293T human cells using X-tremeGENE 9 DNA transfection reagent (Roche Applied Science) with puromycin added the next day to 10-cm dishes for selection. Cells were collected at 60 -64 h post-transfection time; cell extracts were prepared in buffer 1ϫ DB, 0.5% Triton X-100 with protease inhibitors (Roche Applied Science), and then cell extracts from each dish were split into two halves with one-half treated with RNase A and T1 for 30 min to ensure maximum digestion of RNA. Cell extracts (500 g) were centrifuged at 200,000 ϫ g for 10 h at 5°C through 11-ml linear 10 -50% glycerol gradients, and 500-l fractions were collected from the top of each gradient. The absorbance at 280 and 260 nm was determined for each fraction. The aliquots of each third fraction were separated on 12% SDS-polyacrylamide gels, Western blotted, and probed with anti-A3G (C-terminal) rabbit polyclonal antibody (AIDS Reagent Program (National Institutes of Health), catalog no. 10201) to determine the sedimentation distribution of A3G WT and mutants in the gradient (three separate cell extracts independently). The sedimentation of ovalbumin (42 kDa, 3.5S), BSA (67 kDa, 4.2S), aldolase (160 kDa, 7.4S), catalase (250 kDa, 11.3S), and thyroglobulin (660 kDa, 19S) and in vitro assembled spliceosomes (60S) were sedimented as gradient calibration controls. The densitometric quantification of WT A3G from Western blotting signals for each gradient fraction without and with RNase digestion served to calibrate the gradient for the maximum and minimum RNAdependent A3G sedimentation.

HIV-1 infectivity assay
WT and each of the A3G mutants (N-terminally HA-tagged) were subcloned into pIRES-P vector (85) under CMV promoter and transfected into HEK293T cells with FuGENE HD transfection reagent (Promega) in 6-well dishes. Cells were transfected with a 5-10-fold range of plasmid concentrations to ensure that we determined the amount of plasmid DNAs to produce the same level of protein expression. It should be noted that HEK293T cells do not naturally express A3G. 48 h posttransfection, the cells were lysed with Reporter Lysis buffer (Promega), and whole-cell extracts together with H9 human T-cell extract as control were probed by Western blotting with anti-A3G (C-terminal) rabbit polyclonal antibody (AIDS Reagent Program (National Institutes of Health), catalog no. 10201). The Western blotting signals were quantified by scanning densitometry to identify the amount of transfecting plasmids that enabled A3G and mutants expression equal to that expressed naturally in H9 cells.
The antiviral activity of A3G mutants was measured by a single-round infectivity assay with pseudotyped HIV-1 produced in HEK293T cells. The HIV proviral vector pDHIV3-GFP encodes for all HIV genes except nef (replaced with enhanced GFP) and env. The p⌬Vif-DHIV3-GFP proviral vector in addition contains an early stop codon within the Vif gene. Use of this ⌬Vif virus was necessary so that A3G could be assembled with virions and not subjected to Vif-dependent degradation. HEK293T cells were co-transfected with the p⌬Vif-DHIV3-GFP, plasmid encoding VSV-G coat protein (pVSV-G) that provides the env gene function in trans position and WT A3G or A3G mutant constructs. Media on the plates were replaced 4 h after transfection, and then media were harvested 20 h later for viral particle isolation. The pseudotyped virus was normalized by p24 ELISA (PerkinElmer Life Sciences), and 500 pg of protein in p24 equivalents were used to infect the reporter TZM-bl cells in 96-well plates in triplicate. In this reporter cell line infectivity is proportional to the expression of a stable firefly LTR-luciferase gene that is driven by the HIV LTR promoter (88). 48 h after infection SteadyGlo Reagent (Promega) was added to each well for 30 min. Luminescence units (RLU) were quantified, and the relative percent change in infectivity due to WT A3G or RNA-binding mutants in cells without Vif was determined from the range established by the infectivity of WT A3G with Vif containing virus set as the maximum infectivity.
A3G WT and mutant protein expression levels in cell extracts were evaluated by Western blottings with HA antibody and normalized by the level of GAPDH protein expression. Viral particles (50 ng of protein in p24 equivalents) were isolated by passing the media through 0.45-micron filters, clearing by 1000 ϫ g centrifugation for 10 min, and then pelleting through a 20% sucrose cushion by ultracentrifugation (Beckman SW 41 rotor, 25,000 rpm, 2 h at 4°C). The content of A3G in each viral preparation was determined by densitometric scans of Western blottings probed with monoclonal anti-p24 (AIDS Reagent Program (National Institutes of Health), catalog no. 3537) as loading control and anti-HA antibody to detect either WT or mutant A3G proteins.

Statistical analysis
The statistical analyses for nucleic acid binding to A3G and functional end point data were done using GraphPad Prism 6 (GraphPad, La Jolla, CA) programs.