Interactions of APOBEC3s with DNA and RNA

Abstract

APOBEC3 enzymes are key enzymes in our innate immune system regulating antiviral response in HIV and unfortunately adding diversity in cancer as they deaminate cytosine. Seven unique single and double domain APOBEC3s provide them with unique activity and specificity profiles for this deamination. Recent crystal and NMR structures of APOBEC3 complexes are unraveling the variety of epitopes involved in binding nucleic acids, including at the catalytic site, elsewhere on the catalytic domain and in the inactive N-terminal domain. The interplay between these diverse interactions is critical to uncovering the mechanisms by which APOBEC3s recognize and process their substrates.

Introduction:

APOBEC3s (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like 3) is a family of human cytidine deaminases that catalyze the deamination of cytosine to uracil in the single stranded DNA(ssDNA) or ssRNA[1–5]. The APOBEC3 (A3) family consists of seven members; three of these enzymes (A3A, A3C, A3H) have a single zinc binding domain while the other four (A3B, A3D, A3F, A3G) have two domains. The two-domain A3s have a catalytically active C terminal domain (CTD) and a pseudo-catalytic N terminal domain (NTD) that binds to nucleic acids but does not have deaminase activity. Although NTDs have no deamination activity, they appear to be key for regulating the catalytic activity through increasing ssDNA binding affinity and promoting oligomerization[6]. These seven unique A3’s provide unique activity and specificity profiles for deamination.

A3s were first discovered due to their antiviral restriction of HIV[7–10]. HIV-1 counteracts A3 proteins by the accessory proteins Vif, which targets A3s for proteasomal degradation [8,11–16]. In addition to activity against retroviruses (including HIV-1), A3s are involved in the restriction of endogenous retrotransposons, especially LINE-1 elements. A3s also restrict DNA viruses including nuclear replicating ssDNA viruses such as adeno-associated virus[17] and dsDNA viruses such as hepatitis B virus, herpes viruses and HPV[18–21]. However, A3 activity can be a double-edged sword. In addition to inducing mutations on single-stranded viral genomes, A3s can cause mutations in host genomes when localization and/or activity of A3s is mis-regulated. Recent studies found that A3A, A3B and A3H could cause heterogeneities of human genome in various types of human cancer, including breast, bladder, head and neck, cervical, and lung cancer[22–30]. Moreover, when coupled with CRIPSR/Cas9, A3s are being leveraged as novel base editors to treat genetic diseases[31,32]. Therefore, insights of structures of A3 proteins and their complexes with substrates may contribute toward developing more effective antivirals and cancer therapeutics as well as base editing systems for gene therapy.

The structure of the catalytic domain of A3G (A3G-CTD) was the first domain to be solved by NMR and X-ray crystallography in 2008[33,34]. Since then more than fifty structures of A3 domains have been deposited in Protein Data Bank (PDB) (Supplementary Table 1) as of July 2020. However, these structures have been challenging to acquire due to poor solubility and tendency of these enzymes to aggregate. In the last several years however, complexes of A3 domains with DNA/RNA and full length A3 structures have finally been successfully determined.

In this review we will summarize the structural features and differences of individual A3 domains. These structural features provide the determinants of specificity for the deamination target sequences. We will further highlight the unique interactions RNA makes with A3H and finally discuss the most recently determined full-length double domain A3G structures. Thus this review will gives an overview of the structural basis for how these unique A3’s interact with DNA and RNA.

Structural features of individual A3 domains:

The consensus A3 domain structure contains six α-helices and five β-strands forming one β-sheet (Fig. 1a–c), and one Zn²⁺ is chelated by the histidine and cysteines in the deaminase motif which is essential for catalysis[1]. Each domain possesses consensus (H/C)-(A/V)-E-X_24–30-P-C-X₂-C cytidine deaminase motif although the N-terminal domain of the double-domain A3s are catalytically inactive[35]. The catalytic site centered by Zn²⁺ forms a positively charged pocket (Fig. 1a). Although the overall fold is conserved, subtle sequence differences among A3s have resulted in variations in loops length, structure, and flexibility as well as variations in surface charge, active site interactions, and oligomeric tendency (Figs. 1d and e). These variations underlie the functional characteristics of each A3 protein.

Figure 1: Recognition of the deamination target sequences.

a) Apo-form structure of the A3G-CTD (PDB ID: 2KEM). Electrostatic surface distribution showing positive (+5 kT/e, blue) and negative (−5 kT/e, red) regions. Helices and strands are labeled, and sidechains of His257, Glu259, Cys288 and Cys291 are shown as stick model. Zn²⁺ ion is shown as a purple sphere. b) Co-crystal structure of A3A with ssDNA (PDB ID: 5KEG). Three nucleotides of the deamination target ssDNA bound as U-shaped conformations. c) Co-crystal structure of a A3G catalytic domain variant (CTD2*) with ssDNA (PDB ID: 6BUX). ssDNA have more open and extended shape providing extra interaction with protein. In both A3A and A3G-CTD2*, the target cytosine (C⁰) is flipped out, and tightly packed under the Zn²⁺ ion by stacking aromatic rings with the Zn²⁺-chelating residue His70 and His257 in A3A and A3G, respectively. Proteins are represented as yellow cartoons; C, N, and O atoms are colored yellow, blue, and red, respectively, for amino acid residues of the protein. DNAs are represented as cartoons; nucleotides are colored blue and phosphate backbone are colored orange. d) and e) Summary of interactions between protein and ssDNA for A3A and A3G-CTD2*, respectively. f) Sequence alignment and structural representation of active site loops in APOBEC3 family. The sequence alignments of active site loops in A3 family are mapped on the ssDNA bound A3A crystal structure (PDB: 5KEG). The sequence similarity is colored according to figure legend.

Particularly, sequence differences in active site loops (loop 1, loop 3, loop 5 and loop 7) that surround the active site pocket (Fig. 1f) of catalytically active domains mainly contribute to the differential substrate specificity, binding affinity and deamination activity for ssDNA, as well as the distinct physiological functions in A3s[36]. For instance, comparing the structures of Z1 A3 domains (A3A, A3B-CTD and A3G-CTD), A3A has the shortest loop-1 appeared to keep the catalytic site open for substrate access, which may relate to A3A’s strong deaminase activity[37]. A3A and A3G-CTD share a unique feature in β-strand 2 as it is divided to two sections connected with a short loop [33,37,38]. Early structures of A3G-CTD suggested importance of loop-1 and loop-7 residues, including W211, R213, R215 and H216[33,39] and D316 and D317[34], respectively, for ssDNA-binding and deamination. A3B has been related to mutations of human genome in some cancer cells[27,28], yet A3B’s catalytic activity is low in vitro, which may be due to the closed conformation of its catalytic site seen in both solution NMR and crystal structures[40–42]. Therefore, systematic analysis of these active site loops will help reveal the structural mechanism of these functionally overlapping but distinct A3 proteins.

Structural determinants selecting deamination target sequences in ssDNA: A3A, B and G.

Although A3 domains share similar structure and overall fold, they have different catalytic activities, substrate preferences (Supplementary Table 2), and sequences specificities. A3 proteins deaminates 5’-TC and 5’-CC hot spots in single-stranded DNA (ssDNA) as well as HIV-1 proviral DNA[43–46]. Time-resolved NMR deamination assay indicated that A3G-CTD clearly preferred the 3’ cytidine as deamination target compared with the middle cytidine within a 5’-CCC motif[47]. Recent years, co-crystal structures of A3 proteins and substrate ssDNAs have emerged. The A3A-ssDNA complexes were published by both the Aihara and the Matsuo/Schiffer labs in 2017[48,49]. The Aihara lab co-crystallized inactive (E72A mutation) and C-terminal 4-residues truncation version of human A3A (Residues 1–195) with 15-nt ssDNA containing a 5’-TC deamination target (5’-AAAAAAATCGGGAAA)[48]. This crystal structure has 4 monomeric complexes in the asymmetric unit, and each shows clear electron density for either 5 nucleotides(5’-A⁻²T⁻¹C⁰G⁺¹G⁺²) or 6 nucleotides (5’-A⁻²T⁻¹C⁰G⁺¹G⁺²G⁺³) centered on the target cytosine (C⁰). The Schiffer and Matsuo labs also used an inactive A3A variant (E72A/C171A) for co-crystallization with 15-nt ssDNA (5’-TTTTTTTCTTTTTTT). There was a single A3A–ssDNA complex in the asymmetric unit of the 2.2 Å resolution structure where the target sequence 5’-T⁻¹C⁰ was well ordered. In both structures, A3A-bound ssDNA adopts a U-shaped conformation anchored by the target cytosine C⁰ and T⁻¹, with up and down-stream ssDNA bent away from the active site (Fig. 1b, d). The C⁰ and T⁻¹ are accommodated in a deep groove formed by Loops-1, −3, −5 and −7 of A3A. The bound DNA adopts an irregular conformation, flipping out the nucleobase of the target cytidine (C⁰), to encircle the side chain of H29, which serves as a gate keeper to the active site (Fig. 1b, d).

The Aihara group also determined a co-crystal structure of a 7-nt ssDNA (5′-TTTTCAT) and a non-catalytic mutant (E255A) of A3B-CTD in which loop-1 was swapped with A3A loop-1 (A3Bctd-QMΔloop3-A3Aloop1) (PDB ID: 5TD5)[48]. The A3Bctd-QMΔloop3-A3Aloop1-ssDNA complex has a single nucleoprotein complex in the asymmetric unit and clear electron density for 4 nucleotides (5′-T⁻²T⁻¹C⁰A⁺¹). The overall DNA conformation and active site interactions look identical to that observed in aforementioned DNA-bound A3A structure[48], which was likely because this A3Bctd-QMΔloop3-A3Aloop1 protein contained a histidine from A3A (H29 in A3A) that provided ssDNA critical interactions to form the U-shape (Fig 2a, b). The Schiffer lab uncovered the underlying specificity of A3B by combining molecular modeling, dynamics simulations and deamination assays to discover that R211 in A3B-CTD plays the critical gate keeper role in DNA binding and deamination target specificity[50] (Fig 2c, d). The loop-1 amino acid sequence of A3B-CTD (203-NNDPLVLRRRQT-214) is significantly different from that of A3A (23-NNGIGRHKT-31) (Fig 2e). This study filled the gap for how ssDNA binds to the catalytic site of wild-type A3B-CTD, but more can provide a strategy to uncover substrate binding in other wild-type A3 domains.

Figure 2: Structural comparison of A3A-ssDNA and A3B-ssDNA.

The structural comparison of A3A-ssDNA (PDB: 5KEG) (a) and A3B-ssDNA model (c). Proteins are shown as cartoon representation (iceblue for A3A while green for A3B). ssDNAs are shown as orange sticks. Gatekeeper residues are shown as salmon sticks with critical intermolecular interactions shown as gray dashes. The intermolecular interactions of gatekeeper residue for coordinating DNA binding (b) H29 in A3A and(d) R211 in A3B. The different loop-1 residues in A3Bctd-QMΔloop3-A3Aloop1 structure (PDB: 5TD5; yellow sticks, cyan cartoon) and wild type A3B-ssDNA model are shown in (e).

Since A3G-CTD has relatively low affinity to ssDNA, to solve a co-crystal structure of A3G-CTD and ssDNA complex[51] the Matsuo lab generated a soluble, highly active variant of A3G-CTD with higher DNA binding affinity, namely CTD2. Previously in vitro experiments had found that A3G-CTD deaminated 5-nt 5’-TCCCA as fast as longer ssDNA substrates and deamination speed depended on pH with the fastest at pH 5.5, and identified H216 as the determinant of this pH dependency[52]. The CTD2-ssDNA complex structure supported these previous finding as all 5 nucleotides in the 5’-T⁻³C⁻²C⁻¹C⁰A⁺¹ target sequence had interaction with protein, and H216 had a π-π stacking interaction with A⁺¹ which can be enforced by deprotonation of the histidine imidazole ring at lower pH (Fig. 1c, e). The role of H216 in the target sequence binding is similar to the corresponding H29 in A3A, and indeed A3A also deaminate faster at lower pH[53]. As Holden and co-workers suggested in their crystal structure of A3G-CTD[34], D316 and D317 recognized C⁻²C⁻¹ through specific hydrogen bonds with nucleobases. W211 is unique to A3G-CTD, and it allows additional interaction with T⁻³ through π-π stacking interaction (Fig 1c, e). The Xiong lab took a different approach to generate co-crystals of A3G-CTD and ssDNA as they used a ssDNA binding protein, namely Pot1, to fuse with a soluble variant of A3G-CTD, and captured ssDNA through the Pot1-binding DNA sequence[54]. However, the Pot1-A3G-CTD fusion protein could not bind the deamination target sequence, instead an adenine of ssDNA from a neighboring asymmetric unit was interacting with W211, which authors suggested as an interaction during the search of target sequences. Recently, Cao lab has published NMR structures of A3G-CTD in complex with deamination product ssDNAs[55], presenting various nucleotide-protein interactions which may be used for non-catalytic binding. Although the A3-ssDNA complex structures have revealed specific interactions between protein and deamination target sequences during catalytic reaction, DNA-interactions during the search of the target sequence on a ssDNA are still elusive. Single molecule experiment such as optical tweezer[56] and atomic force microscopy[57] may be combined with structural information to show how A3 proteins “hops and slides” on ssDNA[58].

Interactions with RNA:

A3H, a single domain A3 and capable of restricting HIV-1 infection, has a unique structural feature as it forms dimer through binding a double-stranded RNA (dsRNA) without protein-protein interaction[30,59,60] (Fig. 3a). In the crystal structure of chimpanzee A3H (cpzA3H), bound 9-nt dsRNA was well ordered, and critical interactions between RNA and cmpA3H were determined, which included Y23 and W115 through π–π stacking interactions with nucleobases, and seven R/K residues in loop-1 fitting in the major groove of the dsRNA (Fig. 3b) forming non-specific interactions with the phosphate backbone[60]. Subsequently, Chen lab reported a crystal structure of a monomeric form of human A3H haplotype II variant that contained nine amino acid substitutions, among them W115A was the key substitution to be monomeric in solution[61]. The interactions of A3H with RNA involving residues remote from the active site indicates the more extensive interplay of the enzymes surface with recognizing nucleic acid sequences.

Figure 3: Crystal structure of the cpzA3H dimer with an RNA duplex.

A3H molecules are represented as yellow and green cartoons; C, N, and O atoms are colored yellow/green, blue, and red, respectively, for amino acid residues of the protein.; RNAs are represented as cartoon as well with color orange for phosphate back bone and light blue for ribose and bases. Zn²⁺ is shown as a purple sphere. a) Two molecules of cpzA3H (yellow and green) bound to dimeric dsRNA (PDB ID: 5Z98). b) Enlarge view of cpzA3H-dsRNA complex showing important residues involved in protein-RNA interactions. Two aromatic residues, Y23 and W115, provide critical driving forces through π–π stacking interactions with the aromatic rings of nucleobases. The arginine cluster fits in the dsRNA major groove. R16, R20, R21 and R26 in both chains lock the phosphate backbone. Surface of the dsRNA is shown in red.

Full-length double domain A3 structures:

The role of the inactive N-terminal domain of A3B, A3D, A3F and A3G in the full-length enzyme has been largely elusive. Since the N-terminal domain of human A3G (A3G-NTD) is profoundly insoluble, structure determination had been difficult. The Matsuo lab generated a soluble (and HIV-1 Vif binding) variant of A3G-NTD by substituting residues with corresponding ones in relatively soluble A3 domains, and solved solution NMR structure, mapped Vif-binding surfaces formed by loop-1, 3 and 7[62]. The Chen lab solved the crystal structure of a rhesus macaque A3G-NTD (rmA3G-NTD) variant with loop-8 replaced with four residues from human A3G-CTD to improve solubility[63]. The rmA3G-NTD variant formed a stable dimer in both solution and crystal, seems to be stabilized by RNA-binding at the dimerization interface[63].

In 2020, the Chen lab determined structures of two soluble variants of full-length rhesus macaque A3G (rmA3G) for the first time, and these structures suggested mechanisms for both dimerization and possible RNA-binding surfaces of rmA3G[64] (Fig. 4a). Both NTD and CTD were folded in canonical A3 domain structures and connected through a short linker (R194-D198), although relative orientation between the NTD and the CTD were different by ~29° in these two rmA3G variant structures suggesting flexibility in relative orientation of two domains. These rmA3G variant formed different dimers in crystals through NTD-NTD (PDB ID: 6P3X) or NTD-CTD (PDB ID: 6P40) interactions[64]. The dimerization interface of 6P3X involves loop-1 (R24), loop-7 (Y124, Y125, F126, W127), helix-6 (N176, N177, K180, H181) from both subunits (Fig. 4a). Interestingly R24, K180 and H181 form a continuous positively charged patch at the dimerization interface (Fig. 4a). As previously suggested[65–67], W127 appeared on surface and accessible for direct interactions with RNA. The positively charged patch at the dimerization interface was important for RNA binging and stable dimerization of rmA3G. Amino acid sequence similarity and in vitro RNA-binding assays suggested the similar RNA binding/dimerization mechanism for human A3G although the importance of the positively charge patch and the dimerization for the restriction of HIV-1 infection was still elusive[64].

Figure 4: Crystal structures of a full-length rhesus macaque A3G variant and full-length human A3G variant.

a) Electrostatic surface distribution of the dimeric structure of rmA3G E/Q variant (PDB ID: 6P3X). Loop-1, loop-3 and helix-6 of the N-terminal domain (NTD) of both subunits form a positively charged dimerization interface where RNA likely binds. Orange dotted line shows a possible RNA-binding pathway. Sidechains of R24 and W127 of both subunits are presented as stick. b) Electrostatic surface distribution of the monomeric structural model of wild-type A3G generated based on the structure of sA3G* (PDB ID: 6WMA). 5’-dCdC binds near the active site interacting with residues including W211, R215, H216, T218 and D316. The dinucleotide is represented as sticks; atoms in nucleotides are colored blue, navy blue, red, and orange for C, N, O, and P, respectively. A positively charged channel is formed by NTD loop-1 and CTD loop-3, and R24 is located at the center of this channel. Orange dotted line indicates a possible ssDNA-binding pathway.

Most recently, the Matsuo/Schiffer labs determined a co-crystal structure of a soluble human A3G variant and deoxy-cytidine dinucleotide (PDB ID# 6WMA) (Fig. 4b)[68]. Two deoxy-cytidines, namely 5ʹ-C1C2 hereafter, were bound near the catalytic site, but distant from Zn²⁺, not positioned for deamination. The Watson-Crick face of C1 interacts with sA3G by three hydrogen bonds with V212 (loop-1) and D316 (loop-7), and the C1 ribose forms a hydrogen bond with W211. Additionally, the C1 has a π–π stacking interaction with W211. The Watson-Crick face of C2 also interacts with sA3G through three hydrogen bonds with R215, H216 and T218 through an ordered water molecule. Potentially the position and interactions of 5ʹ-C1C2 present a snapshot of the search of the deamination target sequence as A3G is moving along a ssDNA. The N-terminal domain of A3G has been suggested to provide additional affinity for ssDNA, and the sA3G:5ʹ-C1C2 complex structure indicated a positively charged channel between two domains which was directly connecting to the bound dinucleotide (Fig. 4b). R24 located in the middle of this channel appeared to be important for ssDNA binding as the substitution of R24 to alanine decreased both affinity for ssDNA and deamination efficiency in vitro. To reveal detail interactions between the NTD and DNA as well as the relative orientation of the NTD and the CTD upon the binding to a ssDNA, additional structures of full-length A3G bound to longer ssDNA must be determined.

Conclusions:

The A3 family of enzymes interactions with DNA and RNA are a complex combination of catalytic specificity and potentially nonspecific recognition whose role in specificity is still being characterized. Active site loops (loop 1, 3, 5 and 7), which have direct contacts with ssDNA, have shown the most conformational changes in complex compared to apo structures. The dynamics of these loops might be the key for defining the substrate specificities and functional variation among A3s. In addition, these structures revealed differences in the conformation of bound ssDNA (U-shape in A3A and chimeric A3B; linear in A3G). Hence, the differences in the secondary structure of substrate DNA may provide fundamental insights into the mechanisms by which A3s recognize their specific substrates. Interactions with nucleic acids remote from the active site, both within the catalytic domain as observed in A3H or within the non-catalytic domain NTD domain of A3G and likely the other double domain A3B, A3D and A3F, likely fine tune the specific interactions of the enzymes with their substrates within the cellular compartment within which they exist.

There are still many unanswered questions into the specificity and mechanisms of actions by which A3’s function and the role of the non-catalytic domain. Most likely single molecule experiments such as atomic force microscopy, fluorescence spectroscopy and optical tweezers combined with mutational analyses will provide invaluable insights, which are missing from ensemble studies or static structures of A3 enzymes. In addition, the elusive structures of the A3-HIV-Vif-E3 ligase complexes are beginning to be accessible through both single molecule Cryo electron microscopy and tomography. These technologies potentially even permitting the visualization of snapshots of A3s deaminating HIV-1 ssDNA in HIV-1 capsid or within a cellular compartment in near future. Through such characterization A3’s will likely be leveraged as therapeutic targets, either for inhibition as in cancer or in exquisitely specific base editors when coupled with Cas9.

Highlights.

• Specificity of APOBEC3 proteins is determined by 4 active site loops

• ssDNA adopts different conformations dependent on the APOBEC3 enzyme

• Regions remote from the active site, including the N-terminal domain, contribute to substrate DNA and RNA binding sites.