APOBEC ENZYMES AS TARGETS FOR VIRUS AND CANCER THERAPY

Abstract

Human DNA cytosine-to-uracil deaminases catalyze mutations in both pathogen and cellular genomes. APOBEC3D, APOBEC3F, APOBEC3G, and APOBEC3H restrict human immunodeficiency virus-1 (HIV-1) infection in cells deficient in the viral infectivity factor (Vif), and have the potential to catalyze sub-lethal levels of mutation in viral genomes in Vif proficient cells. At least two APOBEC3 enzymes, and in particular APOBEC3B, are sources of somatic mutagenesis in cancer cells that drive tumor evolution and may manifest clinically as recurrence, metastasis, and/or therapy resistance. Consequently, APOBEC3 enzymes are tantalizing targets for developing chemical probes and therapeutic molecules to harness mutational processes in human disease. This review highlights recent efforts to chemically manipulate APOBEC3 activities.

eTOC Summary

APOBECs are single-stranded-DNA cytosine-to-uracil deaminases that perform essential roles in innate immunity by restricting foreign DNA; however, their aberrant activities can drive mutagenesis of virus and cancer genomes. Here, Olson et al. review chemical approaches to harness APOBEC mutagenesis as a new strategy to control genome evolution in human disease.

Introduction

Mutation contributes to genomic variation, and may result in beneficial, neutral or harmful consequences for an organism. Well-established sources of genomic mutation are DNA replication errors, such as misincorporation of a nucleotide by a polymerase, and environmental mutagens, such as certain chemicals and ionizing radiation. The APOBEC3 (A3) subfamily of single-stranded DNA (ssDNA) cytosine-to-uracil deaminases has emerged recently as an innate, enzymatic source of mutation in humans with significant roles in disease.

A3 enzymes belong to the larger apolipoprotein B mRNA-editing enzyme catalytic polypeptide-like (APOBEC) family. Humans have 11 polynucleotide cytosine-to-uracil (C-to-U) editing enzymes including activation induced deaminase (AID), APOBEC1 (A1), APOBEC2 (A2), seven A3s (A3A/B/C/D/F/G/H), and APOBEC4 (A4). This review will focus exclusively on the chemical biology of the human A3 enzymes.

APOBEC Structure and Mechanism

A3 catalytic activity is dependent on zinc (Zn)-mediated hydrolysis of the 4-NH₂ group on cytosines in ssDNA. The A3s contain a consensus Zn-binding motif of histidine(His)-X-glutamic acid(Glu)-X_23–28-proline(Pro)-cysteine(Cys)-X_2–4-cysteine(Cys), where X represents any amino acid and the His and Cys residues coordinate Zn (Conticello et al., 2005; Jarmuz et al., 2002; LaRue et al., 2009; Wedekind et al., 2003). The active site also contains a water molecule, yielding a tetrahedrally coordinated Zn.

A3 catalytic domains can be divided into Z1, Z2, and Z3 phylogenetic groups (Figure 1A), which are defined by conserved amino acid differences (Conticello, 2008; LaRue et al., 2009). Moreover, A3s may be expressed as either single domain or double domain enzymes (LaRue et al., 2009). A3A, A3C, and A3H are characterized as single domain A3s, although each has a different Z-domain type (i.e., A3A-Z1, A3C-Z2, A3H-Z3). A3B, A3D, A3F, and A3G contain two structural domains, where A3B and A3G are structurally defined as having a Z2-Z1 organization and A3D and A3F have a Z2-Z2 organization.

Figure 1. Structure, Organization, and Enzymatic Activity of Human A3 Enzymes.

(A) The seven human A3 family members are distinguished by their number of structural domains (represented as one or two arrows), the phylogenetic grouping of each domain (Z1, green; Z2, orange; Z3, blue), and sub-cellular localization. Rendered X-ray structures (PyMOL) depict the catalytic domains of A3B (pdb 5CQD), A3F (pdb 5HX5) and A3G (pdb 3V4K), and the full-length structures of A3A (pdb 5SWW) and A3C (pdb 3VM8) (Li et al., 2012; Shaban et al., 2016; Shi et al., 2017; Shi et al., 2015). Structurally, each A3 domain has six alpha helices (red) and five beta strands (yellow). The flexible and variable loops are depicted in green. The active site coordinates a single zinc ion (grey sphere). (B) The proposed mechanism of A3-mediated ssDNA C-to-U deamination. (C) A3A-ssDNA X-ray co-crystal structure (pdb 5SWW) that shows the DNA substrate binds in a U-shaped confirmation with the −1 based flipped out of the active site and the target DNA cytosine interacting with the catalytic zinc (Shi et al., 2017). The −1 base forms specificity-conferring H-bonding contacts with residues of A3A.

The proposed mechanism of deamination is based on previous structural studies of bacterial (Betts et al., 1994; Johansson et al., 2002) and yeast (Ireton et al., 2003; Ko et al., 2003; Xie et al., 2004) cytidine deaminases, and relies on the activation of the active site water molecule through deprotonation by a conserved, catalytic Glu that acts as a general acid/base in the deamination mechanism (Figure 1B). The resulting Zn-stabilized hydroxide ion (⁻OH) attacks the 4-position of the DNA cytosine nucleobase yielding an unstable tetrahedral intermediate. A DNA uracil nucleobase is achieved through formation of the C4-O olefin and the release of NH₃.

Unlike the bacterial and yeast cytidine deaminases, the A3 enzymes cannot turnover single cytidine nucleosides/nucleotides. For deamination, A3 enzymes require a minimum of a 5-mer DNA oligomer, where the targeted DNA cytosine is preceded by three nucleotides on its 5′ end and one nucleotide on its 3′ end (Harjes et al., 2013; Nabel et al., 2013). Importantly, because the A3s require substrate oligomers, designed transition state analogues zebularine and tetrahydrouridine, which are known inhibitors of cytidine deaminase (Cohen and Wolfenden, 1971; Wolfenden and Kati, 1991), fail to inhibit the A3s when tested against the purified enzymes (unpublished results from our labs). Accordingly, tetrahydrouridine also fails to inhibit APOBEC1 (Petersen-Mahrt and Neuberger, 2003). However, these studies should be repeated in a cell-based assay for A3 inhibition as the nucleoside analogues may be converted to their triphosphate forms and incorporated into DNA in a cellular context. Cytosine methyltransferases, which require a double-stranded DNA substrate of a specific sequence, have been successfully inhibited by 5′-azacytidine and zebularine through this strategy (Friedman, 1981; Marquez et al., 2005).

The A3s can also be differentiated by strong individual preferences for distinct dinucleotide motifs. Specifically, A3G is unique in targeting DNA cytosines that are immediately preceded (5′) by another DNA cytosine. The other A3s (A3A/B/C/D/F/H) prefer to deaminate DNA cytosines preceded by a 5′-DNA thymine (Carpenter et al., 2010; Kohli et al., 2010; Rathore et al., 2013; Wang et al., 2010). In addition to a strong preference for the preceding nucleobase, the minus-two and plus-one bases also affect A3 deamination efficiency, though specific trends are less concrete (Holtz et al., 2013; Nabel et al., 2013; Rausch et al., 2009; Yu et al., 2004). These observations are supported by recently solved A3A-ssDNA and A3Bctd-ssDNA co-crystal structures that reveal a U-shaped DNA substrate confirmation with the −1, specificity-conferring nucleotide flipped out of the active site and the target DNA cytosine deep within the zinc-coordinating active site pocket (Figure 1C) (Kouno et al., 2017; Shi et al., 2017). Importantly, the −1 base makes direct hydrogen bonding contacts with the protein, accounting for the unique dinucleotide sequence preferences of the A3s. In short, structural differences between A3 family members should impart sufficient chemical features for developing specific enzyme inhibitors, which will be a primary focus in subsequent sections.

HIV-Restrictive APOBECs

The seven A3s constitute vital defense enzymes of the innate immune system that respond to infection from exogenous DNA-based viruses and endogenous transposons by catalyzing cytosine-to-uracil (C-to-U) deamination of foreign ssDNA. While HIV-1 is the prototypical retrovirus associated with A3 activity, the A3s also restrict other viruses, such as hepatitis B, human T-cell leukemia virus type 1, and human papilloma virus [reviewed by (Harris and Dudley, 2015)]. C-to-U deamination affects genomic mutation as the resulting uracil lesions template adenines during subsequent DNA synthesis, which in the case of human immunodeficiency-1 (HIV-1) cDNA, results in genomic strand guanine-to-adenine (G-to-A) hypermutation (Harris et al., 2012; Malim and Bieniasz, 2012; Malim and Emerman, 2008). For example, A3D/F/G/H potently inhibit HIV-1 replication in the absence of the viral countermeasure virion infectivity factor (Vif). Uncontested, the A3 enzymes can deaminate up to 10% of viral cDNA cytosines in a single round of replication, which is lethal to the virus (reviewed by (Harris and Dudley, 2015; Refsland and Harris, 2013). A3 inhibition of virus replication has also been shown to occur through deamination-independent mechanisms (Bishop et al., 2008; Iwatani et al., 2007; Newman et al., 2005). Specifically, A3s can directly bind viral genomic RNA and sterically block the progression of reverse transcriptase (Bishop et al., 2006; Bishop et al., 2008; Iwatani et al., 2007; Iwatani et al., 2006; Mbisa et al., 2007; Newman et al., 2005). Deamination dependant restriction, however, is likely the dominant mechanism for HIV-1 restriction by A3G (Browne et al., 2009; Miyagi et al., 2007; Schumacher et al., 2008).

The accepted model of A3-catalyzed HIV-1 restriction posits that A3D/F/G/H, which are cytosolic, incorporate into budding HIV-1 particles through a binding event between their N-terminal domain and HIV-1 RNA (Figure 2). This binding event enables A3 transport with the virus until fusion with a target cell (Alce and Popik, 2004; Bogerd and Cullen, 2008; Burnett and Spearman, 2007; Khan et al., 2005; Luo et al., 2004; Refsland and Harris, 2013; Schafer et al., 2004; Svarovskaia et al., 2004). Upon HIV-1 reverse transcriptase-mediated cDNA synthesis, A3D/F/G/H deaminate viral cDNA cytosines to uracils during reverse transcription (Harris et al., 2003; Lecossier et al., 2003; Mangeat et al., 2003; Yu et al., 2004; Zhang et al., 2003). The resulting uracilated cDNA is either degraded or immortalized as G-to-A hypermutations in the viral genome (Harris et al., 2003; Lecossier et al., 2003; Mangeat et al., 2003; Mbisa et al., 2007; Yang et al., 2007; Zhang et al., 2003).

Figure 2. Model for A3-Mediated HIV-1 Restriction.

In an infected cell, a sublethal number of A3s incorporate into budding viral particles and hitch-hike to virus naïve cells. HIV-1 modulates A3 expression through Vif, which forms an E3 ubiquitin ligase complex that polyubiquitinates A3s and triggers their degradation at the 26S proteasome. The A3s restrict viral replication through both deamination-dependant mutagenesis and deamination-independent RT inhibition mechanisms. Adapted from (Harris and Dudley, 2015).

Vif is required by HIV-1 to evade this A3 immune defense network (Fisher et al., 1987; Gabuzda et al., 1992; Kan et al., 1986; Lee et al., 1986; Sodroski et al., 1986; Strebel et al., 1987; von Schwedler et al., 1993). To maintain the integrity of the viral genome, Vif nucleates the formation of an E3 ubiquitin ligase complex by heterodimerizing with the transcription co-factor CBF-β, then recruiting CUL5/NEDD8, ELOB, ELOC, and RBX2. This ubiquitin ligase complex directly binds the A3s, flagging the enzymes for degradation at the 26S proteasome (Conticello et al., 2003; Guo et al., 2014; Jager et al., 2012; Marin et al., 2003; Mehle et al., 2004; Sheehy et al., 2003; Yu et al., 2003).

Despite the apparent, unrelenting counter defense that HIV-1 Vif wages against A3D/F/G/H enzymes, analyses of the mutational spectrum of HIV-1 clinical isolates yields a more complex relationship between the A3s and Vif. An enrichment of adenines compared to human genomic DNA are typically found in samples from HIV-1 positive patients despite the fact that Vif can often wholly neutralize A3-catalyzed lethal mutagenesis (Janini et al., 2001). This observation implies that the A3s are continuously providing a mechanism for HIV-1 to mutagenize its genome and achieve maximal fitness. As such, we hypothesize that HIV-1 is “addicted” to the A3 enzymes to accomplish a high mutation rate and viral fitness (Harris, 2008). Consequently, chemical perturbation of A3D/F/G/H activity, including both agonism and antagonism, may shift the favorable balance of HIV-1 mutagenesis to levels that affect viral fitness, and perhaps, render HIV-1 more susceptible to antiretroviral therapies (Figure 3).

Figure 3. Therapeutic Strategies to Target the A3-Vif Interface.

(A) In a clinical infection, A3-catalyzed restriction is inhibited by Vif-mediated ubiquitylation and proteasomal degradation. Despite this inhibition, a sublethal level of A3 deamination is observed. Thus, the activities of Vif and the A3s are balanced to achieve optimum viral fitness. (B) Therapy by hypermutation: This therapeutic strategy seeks to inhibit Vif and/or block the Vif-A3 interface thereby reinstating the restrictive capabilities of the A3s (Haché et al., 2006). Upon Vif inhibition, the relevant A3s can lethally mutate the viral genome. (C) Therapy by hypomutation: Inhibition of the A3s may deprive the virus of a needed mutation source. Inhibiting viral fitness may enable immunological or antiretroviral HIV-1 clearance (Harris, 2008).

APOBECs and the APOBEC-Vif Interface as Therapeutic Targets

Mounting evidence suggests that promutagenic A3 enzymes contribute appreciable amounts of mutation that are detrimental to human health in the context of HIV-1 and cancer. As a result, A3s are emerging drug targets. In the case of HIV-1 therapy two possible strategies are envisioned: (1) direct inhibition of Vif to restore the protective capacity of A3D/F/G/H (via “therapy by hypermutation” and/or deamination independent mechanisms) and (2) inhibition of A3D/F/G/H deaminase activity to starve HIV-1 of essential genetic variation (“therapy by hypomutation”) (Harris, 2008).

Agonism of the de-uracilating enzyme uracil DNA glycosylase (UNG/UDG) could serve as a strategy to enhance the anti-HIV-1 activity of A3-catalyzed cytosine deamination. Evidence for this strategy has been demonstrated through the use of raltitrexed, a thymidylate synthase (TS) inhibitor that increases the cellular dUTP:dTTP ratio, and in turn, the uracil content of viral cDNA. (Weil et al., 2013). UNG excises DNA uracils, causing a reduction in proviral integration and viral protein expression. Accordingly, small molecule UNG agonists could promote degradation of A3-edited HIV-1 cDNA to restrict viral replication. While this may be a viable strategy, detailed discussion of this approach is outside this scope of this review. In the subsequent sections, we will focus on the current status of inhibitor development for the Vif-A3 interface and the A3G enzyme.

Vif Inhibition: Therapy by Hypermutation

The ability of A3D/F/G/H to lethally mutagenize HIV-1 in the absence of Vif indicates that this natural defense network can be reinstated if the A3s are protected from degradation (Haché et al., 2006; Harris, 2008). As such, multiple laboratories have investigated Vif inhibition (Figure 4). Vif is an attractive therapeutic target because it has no known mammalian homologues and possesses sole responsibility for A3 degradation.

Figure 4.

Chemical Structures of Small Molecules that Function through “Therapy by Hypermutation”.

The earliest efforts to inhibit Vif employed lentiviral delivery of small interfering RNAs (siRNAs) for vif gene silencing (Barnor et al., 2005; Barnor et al., 2004). When used synergistically with another anti-HIV-1 siRNA, >80% inhibition of HIV-1 replication in Jurkat cells was observed. These initial studies offered proof-of-concept that vif siRNA could be employed for gene therapy; however, siRNAs for the viral capsid, integrase, protease, and tat/rev open-reading frames are closer to clinical translation (Centlivre et al., 2013; Spanevello et al., 2016).

Multiple laboratories have undertaken programs to discover small molecule Vif inhibitors. The first, RN-18, was identified through cell-based high through screening (HTS) of 30,000 small molecules for the sustained fluorescence of A3G-YFP in the presence of Vif (Nathans et al., 2008). Of 25 hits, benzamides RN-18 and RN-19 exhibited Vif antagonism and HIV-1 inhibitory activity in non-permissive cell lines with IC₅₀ potencies of 6 μM and 25 μM in H9 cells, respectively. RN-18 and RN-19 were also found to increase cellular levels of A3G and promote A3G encapsidation into budding viral particles. Subsequent work explored the structure-activity relationship (SAR) of RN18, which yielded the identification of tolerated isosteric and water-solubilizing modifications (Ali et al., 2012; Mohammed et al., 2012). These analogues, however, only offered modest potency improvements over RN-18. A more recent SAR study of RN-18 reported a 100-fold potency improvement in antiviral activity in non-permissive H9 cells (Zhou et al., 2017). Incorporation of a water-solubilizing glycine prodrug, to yield 13a, improved the EC₅₀ for inhibition of viral replication to 0.25 μM.

An independent HTS of 8634 small molecules using a similar assay identified IMB-26 and IMB-35 as Vif inhibitors (Cen et al., 2010). At 2 μM, IMB-26/35 reduced HIV-1 infectivity by 97% in H9 cells. Interestingly, IMB-26/35, like RN-18, stabilized A3G expression and packaging in the presence of Vif. Surface plasmon resonance (SPR) experiments indicated that IMB-26 and IMB-35 directly bind A3G, affording the hypothesis that IMB-26/35-A3G binding prevents Vif from targeting A3G for degradation. Due to the structural similarity between IMB-26/35 and RN-18, RN-18 may inhibit Vif through a similar mechanism; however, this hypothesis has not been probed experimentally.

HTS of 20000 compounds (Enamine) using a cell-based A3G-GFP stability assay (reported previously (Conticello et al., 2003; Harris et al., 2003) identified another unique class of Vif-inhibitors (Matsui et al., 2014). Primary hits were confirmed by secondary screening that measured the stability of A3G-Luc in the presence of Vif in 293T cells. Hits that preserved chemiluminescence, in addition to fluorescence, yielded compounds for further study. MM-1 and MM-2 were discovered to inhibit HIV-1 replication, increase A3G expression, and promote A3G packaging into Vif-proficient viral particles. Immunoprecipitation studies were performed to determine if MM-1 and MM-2 inhibited A3G-Vif binding. Interestingly, co-precipitation was unaffected by the inhibitors, suggesting that MM-1 and MM2 does not modulate A3G-Vif binding. Immunoblotting with an anti-ubiquitin antibody demonstrated that MM-1 and MM-2 do not inhibit Vif-mediated ubiquitylation of A3G. MM-1 and MM-2 also do not inhibit proteasomal activity. As a result, the target of the MM series remains elusive.

Interestingly, an independent laboratory also employed the A3G-GFP reporter assay to screen the >20000 compound Enamine library (Pan et al., 2015). Of the 372 initial hits, 133 were found to inhibit HIV-1 replication in non-permissive H9 cells. A benzimidazole hit, ZBMA-1, was followed-up due to 50% inhibition at 5 μM. Extensive SAR studies identified compound 14 as the most potent Vif inhibitor discovered to date, with an IC₅₀ value of 3.5 nM for HIV-1 replication in H9 cells. Like the other Vif inhibitors, compound 14 protects cellular levels of A3G by dampening Vif-mediated ubiquitylation and degradation of A3G. There is notable structural similarity between MM-1 and compound 14 (Figure 4). Accordingly, one could envision similar mechanisms of inhibition. In SPR experiments with compound 14, however, a modest reduction (17%) in Vif-A3G binding was observed. Although this observation is conservative, the results suggest that compound 14 and its analogues could bind at the Vif-A3G interface; co-immunoprecipitation experiments with compound 14 revealed no effect on Vif binding to ELOB, ELOC and CUL5.

A time-resolved fluorescence resonance energy transfer (FRET) assay was developed to screen for inhibition of the A3G-Vif binding interaction (Pery et al., 2015b). Using this assay, 307,520 small molecules from the NIH Molecular Libraries Small Molecule Repository (MLSMR) were screened, and one compound, N.41, was identified as an 8.4 μM inhibitor of HIV-1 replication from 3,650 unique chemical hits after extensive secondary screening.

Mechanistically, N.41 was found to increase cellular levels of A3G and promote incorporation of A3G into virions. Early SAR studies have improved upon the potency of N.41 two-fold. Redoxal, a small molecule previously studied for the treatment of rheumatoid arthritis, Candida albicans, West Nile virus, and apoptosis, was also identified as an inhibitor of HIV-1 replication through this screen (Pery et al., 2015a). Redoxal is a known inhibitor of de novo pyrimidine synthesis, and in this study, supplemented uridine or orotate reduced A3G-dependant restriction of HIV-1, overcoming the ability of Redoxal to stabilize A3G protein levels and viral packaging. Importantly, the authors elucidated a key relationship between pyrimidine biosynthesis and viral replication, which appears A3 dependent.

Employing a quenched FRET assay, an analogue of the FDA-approved camptothecin, O2-16, was found to exhibit strong anti-HIV activity, through Vif inhibition (Bennett et al., 2016). Molecular modelling predicted that O2-16 directly binds Vif near its PPLP motif.

Chemical refinement of these small molecule inhibitors has been limited by a lack of Vif structural data; particularly, no Vif-inhibitor or Vif-A3 co-crystal structures have been solved to date. In 2013, Zhou and co-workers explored binding modes of the RN-18 series by computer modeling using a previously reported Vif-homology model (Lv et al., 2007; Zhou et al., 2013). Blind and focused docking, binding free energy calculations, and molecular dynamics simulations predicted a common binding conformation at the C-terminal interface of Vif-ELOC-CUL5 for RN-18 and analogues (Zhou et al., 2013). These computational studies further supported the hypothesis that RN-18 and IMB-26/35 have unique mechanisms of action, despite having chemical similarities. The authors subsequently suggested a number of structural modifications to RN-18 to improve binding affinity for Vif. A similar in silico study was completed with an in-house designed Vif homology model. The homology model was employed to evaluate the ability of RN-18 to block the Vif-A3G protein-protein interface (Sinha et al., 2015). Ultimately, 18 analogues predicted to bind Vif with better affinity than RN-18 were proposed.

As a general approach, it is critically important to consider the development of resistance when designing Vif inhibitors, as Vif is a virally-encoded protein that undergoes rapid evolution. The clinical implementation of Vif inhibitors would likely require combination therapies, as is now standard practice in antiretroviral therapy, to raise the barrier for the development of drug-inactivating Vif variants.

A second strategy to reinstate A3-catalyzed lethal mutagenesis focuses on interrupting essential Vif interaction surfaces, such as those that bind CUL5, ELOC, and CBF-β to form the E3 ubiquitin ligase complex. Being host proteins, these targets are inherently more stable – undergoing a drastically reduced rate of evolution. There are four protein-protein interactions (PPI) that are required for degradation of the A3s: A3-Vif (Conticello et al., 2003; Marin et al., 2003), CBF-β-Vif (Jager et al., 2012; Zhang et al., 2011), ELOC-Vif (Yu et al., 2003), and CUL5-Vif (Yu et al., 2003). Mutagenesis studies indicate that disrupting any one of these interfaces prevents complex assembly and A3 degradation. As proof of concept, genetic knockdown of CBF-β diminishes Vif expression and function (Jager et al., 2012).

Achievement of the first Vif/CBF-β/CUL5/RBX2/ELOB/ELOC X-ray structure has offered an enormous amount of information on the binding interfaces in this complex (Figure 5A) (Guo et al., 2014). As such, drug discovery efforts targeting these interfaces should benefit immensely. Notwithstanding, structural information regarding the A3-Vif interface remains a considerable gap in the field, and inhibition of this interface may be difficult due to the discontinuous surface through which Vif engages the A3s. While previous work has identified putative interacting structural motifs (Albin and Harris, 2010; Kitamura et al., 2012), an X-ray or cryo-EM structure has yet to be achieved. Moreover, recent evidence suggests that A3G may engage Vif through an interface unique to the other A3s [recently reviewed by (Harris and Anderson, 2016); recent work by (Guo et al., 2014; Letko et al., 2015; Nakashima et al., 2015; Refsland et al., 2014; Richards et al., 2015)]. As a result, a PPI inhibitor that disrupts one A3-Vif interface may not inhibit the others.

Figure 5. Candidate Inhibitors of Vif Interaction Surfaces.

(A) X-ray structure of the Vif/CBF-β/CUL5/ELOB/ELOC complex (pdb 4N9F) (Guo et al., 2014). Color Scheme: Vif (magenta), CBF-β (yellow), CUL5 NTD (green), ELOC (orange), ELOB (cyan), Zn (grey). (B) Chemical structures of VEC-5, a putative Vif-ELOC PPI inhibitor, SN-1 and -2 and Baculiferin L and M, Vif-A3 PPI inhibitors.

However, prior to knowledge of the Vif/CBF-β/CUL5/RBX2/ELOB/ELOC structure, Zuo and coworkers used a Vif-ELOB/ELOC homology model to virtually screen 1.2 million small molecules from the Available Chemicals Directory, identifying a benzoylindolizine, designated VEC-5, as a Vif-ELOC PPI inhibitor (Lv et al., 2007; Zuo et al., 2012) (Figure 5B). VEC-5 inhibited HIV-1 replication only in non-permissive (A3G-positive) cells, stabilized cellular levels of A3G, augmented A3G packaging into viral particles, and reduced HIV-1 infectivity with an IC₅₀ value of 24 μM (Zuo et al., 2012). Co-immunoprecipitation assays and label-free biolayer interferometry (BLI) demonstrated that VEC-5 treatment abolished the binding interaction of Vif to ELOC/CUL5, while having no affect on the Vif-A3G interface Subsequent efforts to improve upon the potency of VEC-5, however, have yielded only modest improvements (IC₅₀ of 24 μM to 11 μM) (Huang et al., 2013a; Huang et al., 2013b). One important caveat to the discovery of VEC-5 is that the work was performed prior to the knowledge that CBF-β is required for stable formation of the Vif/CBF-β/CUL5/RBX2/ELOB/ELOC complex.

Although drug resistance is less likely to occur in host enzymes, targeting ELOC, CUL5, and/or CBF-β may also have liabilities. Specifically, small molecules that target the E3 ubiquitin ligase complex may interfere with essential cellular processes, yielding off-target effects or toxicities. Inhibitors that function via these mechanisms must be thoroughly vetted to characterize their safety. For example, CBF-β is required for healthy blood and bone development [reviewed by (Imperato et al., 2015; Lin and Hankenson, 2011)].

A final therapy by hypermutation strategy seeks to identify molecules that directly agonize A3G activity (Cen et al., 2010; Ejima et al., 2011; Smith et al., 2009). Inspired by the finding that N,N,N′,N′-tetrakis(2-pyridylmethyl)ethane-1,2-diamine (TPEN) leaches Zn from Vif, preventing A3G-Vif binding, known Zn chelators SN-1 and SN-2 were tested for disruption of Vif-A3 binding (Figure 5B) (Ejima et al., 2011). The authors found that SN-2 increased A3G expression levels at a concentration of 5 μM, though had no affect on Vif expression or function (Ejima et al., 2011).

Another class of compounds purported to bind at the Vif-A3 interface are the baculiferins (Figure 5B), DOPA-derived pyrrole alkaloids from the Chinese marine sponge Lotrochota baculifera (Fan et al., 2010). After isolation, baculiferins C, EH, and K-N were found to inhibit HIV-1 replication in MT4 and MAGI cells using the p24 antigen detection assay (IC₅₀’s = 1.4 – 8.4 μg/mL; MT4 and 0.1 – 4.4 μg/mL; MAGI). To illuminate possible mechanisms of inhibition, binding affinities of the baculiferins for Vif and A3G were determined using SPR. Baculiferins L and M exhibited strong binding affinities for both Vif and A3G, suggesting that these compounds act at the Vif-A3G interface (Fan et al., 2010). These compounds have yet to be improved upon.

Although much excitement exists for the hypermutation strategy in HIV-1 therapeutic design, small molecules that restore the pro-mutagenic capacity of A3s still remain at relatively early stages of development. Most notably, there is an absence of data regarding the characterization of specific Vif-inhibitor binding interactions, especially with regards to small molecule-Vif X-ray structures. Additionally, none of the discussed studies investigated the selection of viral resistance mutants by the small molecule inhibitors. Resistance remains the gold standard for validating that a small molecule specifically targets a viral protein. Failure to identify resistance mutants within the intended target implies that the small molecule functions via a non-specific mechanism of action. It is also important to note that although this section is presented as ‘therapy by hypermutation’, Vif inhibition could also confer deamination-independent viral restriction, such as A3-mediated direct inhibition of RT.

APOBEC Inhibition: Therapy by Hypomutation

Despite the ability of Vif to effectively counteract A3-catalyzed lethal mutagenesis, strong evidence supports the hypothesis that the A3s are being exploited by HIV-1 to achieve high rates of evolution and overall improvements in fitness (Haché et al., 2006; Harris, 2008; Kim et al., 2014; Kim et al., 2010). In clinical infections, sublethal levels of A3-catalyzed mutation persist (Kim et al., 2010; Mulder et al., 2008). Thus, genetic variation attributable to the A3s likely contributes to the high mutation rate of HIV-1, its ability to evade immune clearance mechanisms, and its rapid evolution of antiretroviral resistance. Therefore, current antiretroviral therapies may benefit from removing the A3s as enablers of HIV-1 evolution. This strategy, termed “therapy by hypomutation” (Harris, 2008), may be the key to offset the unprecedented balance between HIV-1 mutagenesis and pathogenesis, as a small molecule that limits viral evolution, in combination with host immune clearance mechanisms and/or the antiretroviral drugs, may offer a strategy for curing infections.

Our laboratories have pioneered the discovery and development of small molecule A3 inhibitors. To initiate these studies, a high-throughput biochemical screening assay was developed (Figure 6A). Recombinant A3 is incubated with a ssDNA oligomer containing a 5′-6-FAM fluorophore and a 3′-TAMRA quenching molecule (Li et al., 2012). In the absence of a small molecule inhibitor, A3 enzymes deaminate the target DNA cytosine to a uracil, which is then excised by uracil DNA glycosylase (UNG/UDG). Addition of aqueous NaOH or heat cleaves the phosphodiester backbone at the abasic site, releasing the 6-FAM fluorophore from the TAMRA quench yielding quantifiable fluorescence that is directly proportional to enzyme activity. Utilizing this HTS assay, a library of 1280 pharmacologically active compounds (LOPAC, Sigma) was screened to discover the first small molecule inhibitor of A3G, MN30. Of 34 HTS hits against A3G, multiple compounds containing a catechol substructure were observed and MN30 was characterized as a representative member of this class (Figure 6B). Through site directed mutagenesis and mass spectrometry experiments, MN30 was determined to covalently engage Cys321 of A3G upon auto-oxidation from the catechol to the ortho-quinone. Electrophoretic mobility shift assays (EMSAs) demonstrated that although MN30 binds near the A3G active site, A3G/ssDNA binding is not affected. An energy-minimized model of MN30 bound to C321 suggested that MN30’s steric bulk forces Y315 to flip and negatively contact W285. Y315 and W285 are conserved and essential for enzyme function. Thus, despite the ability of A3Gntd to bind a ssDNA substrate, the conformational change induced by MN30 likely prevents A3Gctd from adopting a productive conformation for binding single-stranded DNA and performing catalysis (Li et al., 2012). Importantly, this work identified that a cysteine residue located within 8 Å of the Zn-containing active site is a unique structural feature of A3G. This finding led to the discovery of additional classes of electrophilic small molecules that specifically inhibit A3G through Cys321 versus other A3 family members lacking a comparable residue (Olson et al., 2015; Olson et al., 2013).

Figure 6. HTS for Small Molecules that Function through “Therapy by Hypermutation”.

(A) Schematic of a fluorescence-based C-to-U deamination assay for high-throughput screening. Uninhibited A3-catalyzed deamination results in a high fluorescence readout, while potent A3 inhibition reads as background fluorescence. (B) Chemical structures of published A3G inhibitors.

Subsequent studies by our groups reported that small molecules based on MN256.0102 and MN132.0262 inhibit A3G with low micromolar (3.9–21 μM) potencies (Figure 6B) (Olson et al., 2015; Olson et al., 2013). These inhibitors were also identified using the aforementioned fluorescence-based assay in screens of over 325,000 compounds of the NIH Molecular Libraries Probe Production Centers Network (MLPCN)/MLSMR collection and 21,126 compounds from University of Minnesota collections, including compounds from the Sigma LOPAC (LO1280), Tocriscreen, Prestwick Chemical, NIH Clinical Collection and MicroSource Discovery commercial libraries. When screening the NIH MLSMR collection, 21% (96/469) of the identified A3G-specific HTS hits included the 4-amino-1,2,4-triazole-3-thiol pharmacophore. MN256.0102, the parent molecule of this inhibitor class, is hypothesized to primarily inhibit A3G by covalently binding C321 through a disulfide bond. MN256.0102, like MN30, fails to fully inhibit A3G C321A, a catalytically active A3G mutant, highlighting the importance of the nucleophilic C321 residue (Li et al., 2012). The mechanism of action for the inhibitor class headlined by MN132.0262 has yet to be elucidated, though this compound may also inhibit in a covalent manner given the presence of its electrophilic Michael acceptor. Ultimately, these early identified inhibitors function as useful tools to study A3G inhibition in vitro, but their electrophilic functionalities are too cross-reactive for studies in cells [despite most passing PAINS filters (Baell and Holloway, 2010)]. On-going work in our laboratory seeks to develop tuneable covalent A3G inhibitors with the ultimate goal of delivering chemical probes that are active in cell culture and animal models.

With knowledge that A3s bind directly to Vif, 46 Vif-derived peptides spanning the full-length of the protein were investigated for A3G inhibition using a gel-based in vitro deamination assay. From this study, Vif_25–39 and Vif_105–119 were identified as sub-micromolar inhibitors of A3G activity, 0.6 μM and 0.1 μM, respectively (Britan-Rosich et al., 2011). Subsequent work by our laboratories with de novo, in-house synthesized peptides, however, found Vif_25–39 did not to inhibit A3G deaminase activity and Vif_107–115 only inhibited A3G non-specifically (Richards et al., 2017). These results demonstrated that Vif-derived peptides are unlikely to serve as leads for A3 inhibitor development. This conclusion is further supported by recent findings that Vif interacts with A3 enzymes through multiple, discontinuous surfaces (Guo et al., 2014; Letko et al., 2015; Nakashima et al., 2015; Refsland et al., 2014; Richards et al., 2015).

While early efforts have identified the first small molecule A3G inhibitors, considerable work is required to advance this concept. In particular, effective “therapy by hypomutation” would likely require the simultaneous inhibition of A3D/F/G/H; however, only A3G contains a Cys residue appropriately positioned within 8 Å of the active site. Thus, non-covalent A3 inhibitors also need to be pursued. In this vein, the field needs additional assays amenable to A3 ligand discovery to be developed in addition to the aforementioned fluorescence-based deamination assay.

Cancer-Associated APOBECs

Cancer progression, recurrence, and therapy resistance require continual acquisition of new somatic mutations. Multiple A3 enzymes have now been implicated in cancer mutagenesis, although to varying degrees [in addition to AID, reviewed by (Robbiani and Nussenzweig, 2013)]. Next-generation sequencing (NGS) has revealed an APOBEC mutation signature in >50% of human cancer types, with variable impact within each tumor ranging from <5% to >90% of all base substitution mutations (Alexandrov et al., 2013; Burns et al., 2013b; Roberts et al., 2013). Additionally, the A3s can catalyze clustered hypermutation, or kataegis, in cancer cell genomes; however, A3-catalyzed mutation clusters tend to be only ~10% or less of the total mutation load (Chan and Gordenin, 2015; Nik-Zainal et al., 2012). As a result, the A3s can catalyze modest changes in the cancer genome to promote tumor evolution; one consequence being the development of therapeutic resistance. Our recent studies on the role of A3B-catalyzed mutation in breast cancer provide an illustration of this concept.

Analysis of breast cancer cell lines and primary tissue samples demonstrated that A3B mRNA is up-regulated in 65% of patient-derived primary breast cancer specimens and 90% of breast cancer cell lines (Burns et al., 2013a). Moreover, tumors that express A3B exhibited twice as many overall mutations compared to A3B-low expressing breast tumors, including mutations in the TP53 tumor suppressor gene.

Genetic knockdown of A3B caused reduced RNA genomic uracil content, decreased overall mutational loads, and reduced frequencies of C-to-T transitions. Consequently, this model of chronic endogenous mutation may explain the heterogeneity, rapid evolution and therapeutic resistance characteristics of breast cancer and potentially of many other tumor types (Figure 7A). Higher levels of A3B expression correlate with decreased survival rates in estrogen receptor-positive breast cancer patients (ER⁺) (Cescon et al., 2015; Sieuwerts et al., 2014). High levels of A3B also correlate with a reduced efficacy of tamoxifen (TAM) therapy in patients with recurrent ER+ breast cancer (Law et al., 2016). Strikingly, this clinical correlation has also been corroborated in a xenograft model for ER⁺ breast cancer (Figure 7B). In brief, MCF-7L cells have an endogenous level of A3B analogous to that found in human breast cancer specimens. As depicted in Figure 7B, TAM therapy initially yields favorable therapy outcomes in this model until resistance almost invariably occurs. However, MCL-7L cells with A3B-low levels, created with a short hairpin-directed against A3B, resulted in substantially more durable responses to tamoxifen therapy (Law et al., 2016). Consequently, chemical modulation of A3B is an exciting new strategy to harness this newly identified endogenous source of cancer mutation that when inhibited may dramatically improve therapy outcomes. A3A and A3H have also been implicated in mutagenizing cancer genomes, albeit to a lesser extent [recently reviewed by (Hutchinson, 2016; Knisbacher et al., 2016; Mertz et al., 2017; Salter et al., 2016)].

Figure 7. Model Depicting the Impact of A3B-Catalyzed Base Substitution Mutation in Cancer.

(A) Mutation catalyzed by A3B can accelerate tumor cell growth, metastasis, and the development of therapeutic resistance. A3B preferentially deaminates DNA cytosines in a 5′-TCA context. The resulting uracil templates the insertion of adenine during complementary strand synthesis and uracil base excision repair will convert the U-A base pair to T-A (a C-to-T transition mutation). (B) Schematic representation of tumor volume during an A3B knockdown study in an ER⁺ breast cancer xenograft model (Law et al., 2016). At 50 days, mice were injected with tumor cells expressing a shRNA control or a shRNA to knockdown endogenous A3B. At 125 days, TAM treatment was initiated to suppress growth of similarly sized tumors. However, by 300 days, most of the control (A3B expressing) tumors had become resistant to TAM therapy, whereas the growth of most of the A3B depleted tumors was still suppressed. This study demonstrates that A3B contributes to the development of tamoxifen resistance.

APOBEC Inhibition: An Emerging Strategy for Cancer Therapy

As discussed above, it is clear that misregulated A3s contribute to somatic mutation in multiple human cancers. Thus, inhibiting these enzymes may prevent pro-cancerous mutations from occurring, such as those involved in tumor recurrence, metastasis, and drug resistance. Importantly, A3B, which is hypothesized to be a principal driver of A3-catalyzed somatic mutation, is a non-essential enzyme in humans, as confirmed by the existence of a full gene deletion allele in the general population (Kidd et al., 2007). Due to the relatively recent discovery of APOBEC mutagenesis in cancer, efforts to identify inhibitors for these purposes are still in the early stages of development. The release of the first A3A-ssDNA and A3Bctd-ssDNA co-crystal structures (Figure 1C), however, will aid in on-going small molecule inhibitor design, as specific binding interactions between the A3s and relevant ssDNA substrates have been elucidated (Kouno et al., 2017; Shi et al., 2017). With these recent publications, structure-based design of A3 inhibitors can be pursued for the first time.

As a final note, one could also envision a ‘therapy by hypermutation” strategy for cancer. Although more challenging than “therapy by hypomutation” (A3B inhibition), small molecule agonism of A3-catalyzed mutation could promote cancer cell killing by increasing the overall mutational load in the cell to toxic levels (therapy by hypermutation).

Considerations for APOBEC Inhibitor Design

Achieving specific A3 inhibitors is the ultimate goal from a chemical probe perspective, although pan-A3 inhibitors may be valuable for certain therapeutic applications. In the case of developing A3 inhibitors to affect HIV-1 mutation, a pan-A3D/F/G/H inhibitor would be particularly desirable to eliminate any possible source of pro-evolutionary A3 mutation for the virus (as opposed to individually targeting A3D, A3F, A3G, or A3H). Conversely, a selective A3B inhibitor would be ideal for constraining A3-catalyzed mutation in cancer since the majority of tumor mutations involve A3B (and A3B is non-essential). This scenario would also leave intact the majority of A3 enzymes to restrict pathogenic DNA as part of the body’s innate immune response. Avoiding the inhibition of AID is of particular importance given its essential role in antibody diversification (Di Noia and Neuberger, 2007). Screening for AID inhibition is an important anti-target to be monitored during A3 inhibitor discovery programs.

The development of specific A3 inhibitors that avoid AID inhibition is a realistic outcome given the unique phylogenetic subgroups of the A3s and their specificities for deaminating discrete dinucleotide sequences. Recall that A3 catalytic domains can be divided into Z1, Z2, and Z3 phylogenetic groups (Figure 1), which are defined by differences in the amino acid sequences of the Zn-coordinating active site (Conticello, 2008; LaRue et al., 2009). Additionally, A3 preferences for deaminating specific dinucleotide sequences is conferred by direct A3-substrate binding interactions between residues of the active site and the +2, +1, and −1 bases (Holtz et al., 2013; Nabel et al., 2013; Rathore et al., 2013; Rausch et al., 2009; Yu et al., 2004). The molecular characterization of how A3s engage their ssDNA substrates resulting from the recent achievement of A3A-ssDNA and A3Bctd-ssDNA X-ray co-crystal structures will undoubtedly catalyze structure-based inhibitor discovery programs. Notably, AID uniquely prefers a 5′-purine (A/G) prior to the deamination target cytosine, which is distinct from the 5′-pyrimidine (T/C) preferences of all A3s (i.e., 5′-CC or 5′-TC). Taken together, the unique active site characteristics of A3s and AID should be exploitable to achieve chemical probes, and ultimately, therapeutic molecules with appropriate specificity profiles.

Concerns are sometimes raised as to whether the inhibition of multiple A3s would leave patients susceptible to infection by viruses and/or retrotransposition. While we acknowledge the delicate balance between these immunological defenses and erroneous mutation of the host genome, a therapeutic window is likely to exist for the inhibition of detrimental, A3-catalyzed endogenous mutation, during which benefits will greatly outweigh the potential detriments of inhibiting only one component of the body’s expansive innate immune system. The next-generation of A3 inhibitors that are developed will be invaluable chemical probes for cellular and animal studies that will address these important questions. Further refinement to achieve specific A3 modulators will yield even better inhibitors with less probability of off-target side effects.

APOBEC Modulators: Past Challenges and Future Promises

The field of A3 chemical probe/therapeutic inhibitor discovery is still relatively young. For example, the discovery of A3B as an endogenous source of cancer-promoting mutation was first reported in 2013 (Burns et al., 2013a), which spurred our own interest in developing A3B inhibitors. As research groups become more aware of the “APOBEC problem” in cancer, the desire for cell culture- and animal model-active A3 chemical probes will continue to grow. Additionally, it is likely that A3B inhibitors (and to a lesser extent A3H and A3A inhibitors), when used in concert with existing cancer therapies, will slow or possibly prevent the development of drug resistance, metastasis, and therapy failure across multiple human cancers. The development of these first-in-class clinical candidates and their implementation into first-in-human clinical trials will be a fascinating proof-of-concept test for validating this newly identified, endogenous source of cancer mutation.

Until recently (references below), no X-ray co-crystal structures for A3s engaged to their ssDNA substrates or for the Vif/CBF-β/CUL5/RBX2/ELOB/ELOC complex were available. This lack of structural information forced the sole use of HTS for A3 and Vif inhibitor discovery. In the case of A3 inhibitor discovery, we encountered issues (typically) associated HTS efforts, such as false positives, PAINS molecules, and inhibitors that fail to translate activity to a cellular context (Li et al., 2012; Olson et al., 2015; Olson et al., 2013). In the case of Vif inhibition, most of the reported studies were performed prior to the knowledge that CBF-β is essential for Vif stability. However, given the new structural information available for A3-ssDNA binding (Kouno et al., 2017; Shi et al., 2017) and the Vif/CBF-β/CUL5/RBX2/ELOB/ELOC complex (Guo et al., 2014), the field is now much better equipped for the rational design of A3- and Vif-modulating compounds for anti-mutation applications.

Finally, as progress is made towards the development of clinically viable A3 and Vif modulators, the field requires the development of robust animal models for evaluating their safety and efficacy. To date, only a few examples of animal models engineered to express different human A3s have been reported (Law et al., 2016; Nakaya et al., 2016). The continued development of animal models to enable the evaluation of candidate A3 and Vif inhibitors is essential for their clinical realization.

Conclusions

The discovery that human A3 enzymes function as sources of mutation detrimental to human health offers new and exciting opportunities for therapeutic intervention. Although research in this field is still at early stages, current efforts have discovered leads for Vif and A3G inhibition. Combined with the recent advances in Vif and A3 structural information, the stage is now set for significant advances in A3 chemical probe and therapeutic molecule development that may one day yield next-generation HIV-1 and cancer therapies.