Deoxyguanosine-Linked Bifunctional Inhibitor of SAMHD1 dNTPase Activity and Nucleic Acid Binding

Abstract

Sterile alpha motif histidine-aspartate domain protein 1 (SAMHD1) is a deoxynucleotide triphosphohydrolase that exists in monomeric, dimeric, and tetrameric forms. It is activated by GTP binding to an A1 allosteric site on each monomer subunit, which induces dimerization, a prerequisite for dNTP-induced tetramerization. SAMHD1 is a validated drug target stemming from its inactivation of many anticancer nucleoside drugs leading to drug resistance. The enzyme also possesses a single-strand nucleic acid binding function that promotes RNA and DNA homeostasis by several mechanisms. To discover small molecule inhibitors of SAMHD1, we screened a custom ∼69 000-compound library for dNTPase inhibitors. Surprisingly, this effort yielded no viable hits and indicated that exceptional barriers for discovery of small molecule inhibitors existed. We then took a rational fragment-based inhibitor design approach using a deoxyguanosine (dG) A1 site targeting fragment. A targeted chemical library was synthesized by coupling a 5′-phosphoryl propylamine dG fragment (dGpC₃NH₂) to 376 carboxylic acids (RCOOH). Direct screening of the products (dGpC₃NHCO-R) yielded nine initial hits, one of which (R = 3-(3′-bromo-[1,1′-biphenyl]), 5a) was investigated extensively. Amide 5a is a competitive inhibitor against GTP binding to the A1 site and induces inactive dimers that are deficient in tetramerization. Surprisingly, 5a also prevented ssDNA and ssRNA binding, demonstrating that the dNTPase and nucleic acid binding functions of SAMHD1 can be disrupted by a single small molecule. A structure of the SAMHD1-5a complex indicates that the biphenyl fragment impedes a conformational change in the C-terminal lobe that is required for tetramerization.

Homotetrameric SAMHD1 is a deoxynucleotide triphosphohydrolase enzyme (dNTPase) that is highly expressed in nondividing cells of the myeloid lineage, and its activity is allosterically controlled by binding of GTP and dNTPs to eight activator sites on the tetramer (four GTP/dGTP-specific A1 sites and four dNTP-specific A2 sites¹⁻³). Genetic studies have shown that inherited mutations in SAMHD1 cause the severe neurodegenerative disorder Aicardi-Goutières syndrome (AGS) that is characterized by chronic inflammation and genomic instability.⁴ Extensive genetic, cell biology, biochemical, and clinical evidence has linked SAMHD1 deficiency to interferon-inducible chronic inflammation, Chilbains syndrome, and systemic lupus erythematosus (SLE).⁵⁻⁹ Further, its increased expression is associated with clinical resistance to nucleoside-based anticancer chemotherapies and poor survival.¹⁰ Thus, SAMHD1 is an attractive anticancer drug target, but there are no bioactive SAMHD1 inhibitors yet reported. In general, small-molecule probes targeting SAMHD1 could act by different mechanisms such as competitive binding to catalytic and activator sites or interaction at subunit interfaces and could perturb the diverse functions and oligomeric states of SAMHD1. Given the importance of the A1 site to activity, we focused on a high-throughput inhibitor design strategy that targeted the A1 site with a deoxyguanosine molecular fragment linked to other functional groups, allowing high-throughput screening of the tethered library for high affinity compounds.

Inhibitors of SAMHD1 could impact different aspects of cancer therapy. Mutated SAMHD1 has been strongly linked to the development of chronic lymphocytic leukemia (CLL)^11,12 and other solid tumors,¹³ and overexpression of SAMHD1 has been established as the major resistance mechanism to cytosine arabinoside (AraC) chemotherapy for the treatment of acute myeloid leukemia (AML).¹⁰ In mouse models of retroviral AML transplantation, as well as in retrospective analyses of adult patients with AML, the response to AraC was negatively correlated with SAMHD1 expression,^14,15 indicating that a drug targeting SAMHD1 would enhance therapeutic outcomes. Recently, a similar role for SAMHD1 in resistance to the AML drug decitabine triphosphate was also reported.¹⁶ With AraC, as well as other nucleoside-based drugs, SAMHD1 hydrolyzes the active triphosphate form, leading to drug inactivation (Figure 1).¹⁷

Figure 1.

Bifunctional inhibitors of SAMHD1 could increase the efficacy of antimetabolite chemotherapy by two distinct mechanisms. Cytarabine (araC) is phosphorylated by kinases into its active form araCTP that is incorporated into DNA by a replicative polymerase, where it precipitates replication fork (RF) arrest. SAMHD1 dNTPase activity effectively hydrolyzes araCTP, preventing its toxic effects at replication forks. In addition, the dNTPase-independent DNA repair activity of SAMHD1 promotes fork progression and diminishes the release of immunostimulatory ssDNA from stalled forks that would otherwise activate the cGAS/STING pathway for type I interferon signaling (see text).

Given its described function as a dNTPase, it has been mysterious why loss-of-function mutations in SAMHD1 are associated with AGS, SLE, oncogenesis, and chronic DNA damage in cells.^5,11,13,18 Insight into this question was provided by two studies that revealed secondary DNA damage repair (DDR) functions of SAMHD1, that if inhibited, could lead to additional antitumor effects.¹⁹ One repair function of SAMHD1 is to promote homologous recombination by its association with double strand break (DSB) foci,^20,21 which helps explain the association of SAMHD1 with oncogenesis and DNA damage. Another role for SAMHD1 is in coordinating the restart of stalled replication forks (RFs). In this capacity, a distinct phosphorylated form of SAMHD1 (pSAMHD1) recruits MRE11 exonuclease to stalled forks and promotes degradation of nascent DNA, thereby preventing cytosolic ssDNA accumulation and activation of the cyclic GMP-AMP synthase (cGAS)/stimulator of interferon genes (STING) pathway (Figure 1).²² This function of SAMHD1 provides an explanation for the chronic interferon stimulation observed in SAMHD1 deficient AGS and SLE patients. These DDR functions of SAMHD1 are likely related to its known ssDNA binding activity.²³ It is now clear that SAMHD1 not only is a major regulator of genome stability through the maintenance of dNTP pools²⁴ but also plays a direct role in DDR and suppression of innate immune signaling through the STING pathway.²⁵

Based on the above biological activities, we view SAMHD1 as a bifunctional target for increasing the efficacy of fork stalling agents such as nucleoside drugs. In this bifunctional model, inhibition of SAMHD1 dNTPase activity would first prevent nucleoside drug resistance and increase the levels of stalled forks through increased drug incorporation (Figure 1). An inhibitor that targets its fork-restart function would potentially sensitize tumors to detection by the immune system in a highly targeted way that is not possible with current STING agonists.^26,27 Ideally, a single molecule that targets both dNTPase and the nucleic acid functions of SAMHD1 would be most effective. There is reason to believe that this approach would have tumor specificity because most solid tumor cells are characterized by aneuploidy and oncogene-induced replication stress, frequently leading to tumor-cell-intrinsic DNA replication fork stalling.²⁸⁻³⁰ Here, we report on (i) a large high-throughput screening effort that highlights the challenges of targeting SAMHD1 using such approaches and (ii) a rational fragment-based design approach that specifically targets the A1 activator site. The fragment approach yielded an inhibitor that prevents activation by physiological concentrations of GTP and dNTPs and simultaneously inhibits nucleic acid binding. We discuss the inherent challenges in targeting an enzyme that exhibits profound hysteresis in its activation mechanism and how further extensions of the linked fragment approach could be advantageous.

Results and Discussion

Barriers to High-Throughput Screening Success

We initially attempted to identify inhibitors against SAMHD1 from a curated 69 000 library of small molecules using our previously validated malachite green (MG) colorimetric assay for dNTPase activity.³¹ The MG assay indirectly detects the triphosphate product of SAMHD1 catalyzed hydrolysis of dNTPs through the inclusion of E. coli pyrophosphatase (PPase), which converts PPP_i into 3P_i, producing the colorimetric signal. For simplicity, the screening reaction uses 200 μM dGTP as a substrate because this is the only dNTP that can activate its own hydrolysis in the absence of GTP, and the reaction conditions are designed to be zero-order in [PPase].³¹ Library compounds were screened at a concentration of 10 μM. For screening, the reaction mixture of dGTP and inhibitors were combined prior to the addition of SAMHD1. Thus, at the reaction initiation, all ligands (dGTP, inhibitors) compete for their respective binding sites on the apoenzyme. This single concentration of the A1 site activator/substrate dGTP saturates the A1 site but leaves the A2 and active sites partially saturated.¹ Thus, this design is more likely to detect competitive inhibitors that target the A2 and active sites.

Out of the 69 000 compounds, only 124 inhibited SAMHD1 by at least 40%, constituting a 0.2% hit rate (Figure 2A). Of these hits, most were excluded based on their relatively weak inhibitory potency, lack of activity in secondary assays, or unfavorable properties. Seventy-two hits were brought forward for testing in secondary screens and, in some cases, dose–response studies (see Supporting Information). An overwhelming majority of these were found to be weak SAMHD1 inhibitors, contained chemical structures associated with promiscuous inhibition (PAINS),^32,33 had chemical properties that were unfavorable for use in biological applications,³⁴ or were highly complex and unsuited for further chemical optimization. Ultimately, none of the screening hits were deemed promising for further development. To document these results, we have included a summary spreadsheet in the Supporting Information that details the library compounds, the screening results, and the follow-up studies.

Figure 2.

Large scale, high-throughput inhibitor screen and characterization of guanine analogs for A1 site targeting. (A) High-throughput screening of over 69 000 compounds for inhibition of SAMHD1 dNTPase activity was performed using a previously described MG colorimetric activity assay. (B) The effects of guanine (G) analogs on the oligomeric state of SAMHD1. Each G analog (500 μM) was added to a solution containing 0.6 μM SAMHD1. Reactions were then cross-linked using glutaraldehyde, and the monomer (M), dimer (D), and tetramer (T) forms of SAMHD1 were resolved by gel electrophoresis. (C) Fluorescence anisotropy competitive binding assay using the A1 site probe 5′-FAM-pdGpdGpdG-3′. (D) Competition displacement measurements using the indicated G analogs. The K_d values are reported in Table 1. (E) G analogs inhibit SAMHD1 dNTPase activity as determined using the MG colorimetric assay. The G analogs were preincubated with SAMHD1 prior to the addition of 50 μM GTP and 50 μM of the substrate dATP. All error bars represent an SD of at least two independent replicate measurements.

Fragment-Based Inhibitor Design Using dG

Given the negative outcome from high-throughput screening, we shifted to a rational design approach that took advantage of the guanine specificity of the A1 site. Each monomer subunit of SAMHD1 contains one A1 site that specifically binds GTP (or dGTP). GTP binding to the A1 site induces dimerization and creates a partial A2 site that is competent to bind dTTP, dATP, dCTP, or dGTP. However, the A2 site is not fully formed until tetramerization because a majority of the enzyme interactions with the A2 dNTP arise from a third monomer. The guanine specificity of the A1 site and its central role in promoting oligomerization led to the hypothesis that simple guanine analogs might be employed to target this site and be used as a binding handle to attach other chemical domains (for increased affinity and specificity) through high-throughput tethering with flexible linkers.

To identify the best guanine (G) analog for targeting the A1 site, we first used our established glutaraldehyde protein cross-linking assay to assess how changes in the guanine nucleotide structure impacted the ability of SAMHD1 to oligomerize into dimers (D) or tetramers (T).^1,35 In this survey, we investigated guanosine (rG), GMP, GDP, GTP, deoxyguanosine (dG), dGMP, dGDP, acyclovir (an acyclic guanine antiviral nucleoside), 8-oxo-dG, 8-Br-dG, and 7-Me-GTP (all at 500 μM concentrations) and dGTPαS (100 μM). A representative subset of these analogs is shown in Figure 2B, and the remaining cross-linking results are shown in Supplementary Figure S1.

As expected, GTP induced complete dimerization and the nonhydrolyzable substrate analog dGTPαS induced tetramerization due to its ability to bind the A1, A2, and active sites (Figure 2B).^1,36 However, the gamma phosphate groups of GTP and dGTP are not required for dimerization and tetramerization, respectively, because GDP can induce dimerization and dGDP induced tetramerization (although much less efficiently than dGTP). In addition, removal of the gamma and beta phosphates to give GMP and dGMP did not impact the ability to induce dimers, nor did the removal of all three phosphates (dG). In contrast, rG does not induce SAMHD1 dimerization. Indeed, inhibition by the acyclic nucleoside shows that an intact sugar ring is not required to inhibit dimerization (Figure 2B).

After exploring the effects of guanine analogs on oligomerization, we then employed a fluorescence anisotropy competitive binding assay to determine the relative binding affinities of selected guanine analogs to the A1 site. This assay utilizes a 5′-FAM-labeled deoxyguanosine oligonucleotide probe (5′-FAM-pdGpdGpdG-3′) that binds to the A1 site (K_D = 1.4 μM; Figure S2). Ligands that target the A1 site displace 5′-FAM-pdGpdGpdG-3′ and result in a decrease in fluorescence anisotropy of the probe (Figure 2C).³⁷ As expected, and serving as a positive control for A1 site binding, we found that GTP displaced the probe efficiently with a K_d = 61 ± 11 μM (Figure 2D; Table 1). Surprisingly, this quantitative binding assay revealed that dGMP and dG bind with comparable affinities (Figure 2D), leading to the conclusion that the charged 5′-phosphate group is not required for binding or dimerization (Table 1). In contrast, base substitutions such as 7-methylguanosine, 8-oxo-deoxyguanosine, and 8-bromo-deoxyguanosine substantially decreased binding affinity (Table 1).

Table 1. Inhibitory and Binding Properties of A1 Site Ligands.

A1 site ligand	dNTPase inhibition (%, [I] = 500 μM)a	IC50 (μM)b	binding Kd (μM)c	D | Tr (%)d
none				30 | 0
rG	6.0 ± 8.5	n.d.	500 ± 185	64 | 0
GMP	7.0 ± 0.7	n.d.	110 ± 41	87 | 0
GDP	e	e	210 ± 32	88 | 0
GTP	e	e	61 ± 11	64 | 30
7-Me-GTP	n.t.f	n.t.	>1000	n.t.
dG	65 ± 2.4	130 ± 89	79 ± 20	80 | 0
dGMP	75 ± 1.6	110 ± 47	140 ± 43	91 | 0
dGDP	e	e	130 ± 28	39 | 56
acyclovir	68 ± 31	n.d.	160 ± 27	78 | 0
8-oxo-dG	n.t.	n.t.	>1000	37 | 0
8-bromo-dG	n.t.	n.t.	>1000	62 | 0
dAMP	10 ± 0.02	n.t.	>1000	32 | 0
3′-dGpdA-3′	82 ± 2.6	102 ± 37	41 ± 4.7	90 | 0
3′-dApdA-3′	7.0 ± 0.9	n.t.	>1000	33 | 0
5′-dGpdA-5′	50 ± 8.0	n.t.	70 ± 13	93 | 0
cGAMP	6.5 ± 3.6	>1000	>1000	40 | 0
4	n.t.	140 ± 110	360 ± 150	46 | 0
1023	n.t.	220 ± 150	>500	17 | 0
5a	n.t.	8.6 ± 4.1	5.2 ± 1.2	100 | 0g

^aConditions were 1 mM concentration of each ligand and 50 μM GTP and 50 μM dATP.

^bIC₅₀ values were obtained from 11-point dose response curves using the MG colorimetric assay with 50 μM GTP and 50 μM dATP.

^cCompetition binding measurements were performed using the 5′ FAM-pdGpdGpdG-3′ probe (see text).

^dEach ligand (500 μM) was incubated with 600 nM SAMHD1 before performing glutaraldehyde cross-linking. D = dimer, T = tetramer.

^eThis ligand is an activator of SAMHD1 dNTPase activity.

^fn.t., not tested.

^g200 μM 5a was used in this experiment. All reported errors are SD from at least two replicate experiments.

We also explored binding and dNTPase inhibition by guanine-containing dinucleotides with a 5′,5′ internucleotide linkage (3′-dGpdA-3′), a 3′,3′ linkage (5′-dGpdA-5′), and combined 5′,3′ and 5′,2′ linkages (i.e., 2′,3′ cGAMP; Table 1). The dinucleotides with 3′,3′ and 5′,5′ linkages induced dimerization, but cGAMP did not (Figure S1). With respect to binding and inhibition, only the dinucleotides 3′-dGpdA-3′ and 5′-dGpdA-5′ bound or inhibited SAMHD1 at a concentration less than 500 μM (Figure 2D,E). Because 3′-dApdA-3′ did not induce dimerization, nor did it inhibit the enzyme at concentrations as high as 1 mM, we concluded that a guanine nucleotide was an essential element for activity of the dinucleotides (Table 1, Figure S1). In conclusion, these studies revealed that (i) dG is the minimal binding unit that can promote dimerization and (ii) the 5′-position of dG is a potentially useful site for appending modifications.

Synthesis and Screening of dG-Linked Library

With the above structure–activity results for G analogs, we designed a rapid synthesis strategy to couple a library of carboxylate compounds to the 5′ position of dG through a flexible linker (Supplementary Chart 1). To facilitate synthesis of the dG synthon, the commercially available protected deoxyguanosine 5′ phosphoramidite (1) was coupled to 2,2,2-trifluoro-N-(3-hydroxypropyl)acetamide in the presence of 5-ethylthio-1H-tetrazole to yield product 2 (Scheme 1). The acid labile 3′ dimethoxytrityl group of 2 was then removed, and the resulting product 3 was purified by column chromatography. The amine and phosphate protecting groups of 3 were removed to yield 4.

Scheme 1. Synthesis of Deoxyguanosine Phosphoryl Propylamide Library.

(a) 5-Ethylthio-1H-tetrazole, 2,2,2-trifluoro-N-(3-hydroxypropyl)acetamide, ACN, rt, 4 h; (b) t-BuOOH, rt, 30 min (over 2 steps); (c) 2% TFA, triisopropylsilane, DCM, rt, 15 min; (d) NH₄OH, 4 °C, 4 h (50% yield overall from 1); (e) R-COOH, HBTU, HOBt, DIPEA, DMF, rt, 6 h (5a, 31% yield).

The screening library was generated by coupling the primary amine of 4 to a library of 376 carboxylates (RCOOH) to give linked compounds with the general structure dGpC₃NHCO-R. The coupling efficiencies were evaluated for three representative library members, which fell in the range 80 to 100% (average 94%). These results provide reasonable confidence that most of the library reactions occurred with similar efficiencies. These yields were deemed sufficiently high to allow direct screening of the crude synthetic reactions for activity against SAMHD1. Control experiments established that none of the reagents or intermediates in the synthesis inhibited SAMHD1 at the concentrations used in the screening (Figure S3). The dGpC₃NHCO-R library was screened for dNTPase inhibition at two concentrations, 25 μM and 80 μM using the MG colorimetric assay (Figure 3A). The fluorescence anisotropy competition assay was also used as a high-throughput secondary screen to assess binding to the A1 site (Figure 3B). Based on strong inhibitory and binding effects in the two assays, three compounds were explored further (5a, 5b, 5c). These compounds consisted of structurally similar biphenyl (5a), biphenyl ether (5b), or benzophenone (5c) substituents (Scheme 1). Complete 10 point binding isotherms for 5a, 5b, and 5c were performed using the fluorescence anisotropy competition assay, which provided relative K_d values of 10 μM (5a), 13 μM (5b), and 29 μM (5c; Figure 3C). (Since the crude reaction mixtures were used for these binding studies, the absolute binding affinities of 5a, 5b, and 5c are not given by these measurements, and the relative values are based on an estimated synthetic coupling efficiency of 90% for each compound.) Given its favorable properties, we focused on the biphenyl analog 5a for scaled-up studies using the purified compound.

Figure 3.

Screening of the dGpC3NHCO-R library. (A) SAMHD1 activity remaining in the presence of 25 μM of each library compound. The enzyme was preincubated with each compound before the addition of 50 μM GTP A1 site activator and 50 μM dATP substrate. Activity was measured using the MG colorimetric assay. (B) Competitive anisotropy probe displacement assay for A1 site binding of library compounds. Fifty micromolar of each library compound was added to solutions with 2 μM SAMHD1 and 0.5 μM probe (5′ FAM-pdGpdGpdG-3′). (C) Complete binding isotherms for compounds 4, 5a, 5b, and 5c using the competitive anisotropy probe displacement assay.

dNTPase Inhibition Mechanism of 5a

We investigated the binding affinity of 5a using the competition assay and compared it to the affinities of its two fragments 4 and the 3′-bromobiphenyl-3-carboxylic acid (library compound 1023, Chart S1; Figure 4A). Purified 5a (K_d = 5.2 ± 1.2 μM) binds with about 100-fold greater affinity than the amine fragment (4), more than 200-fold tighter than the biphenyl carboxylic acid fragment, and with about 12-fold greater affinity than GTP (Figure 4A; Table 1). Although binding of the isolated carboxylate fragment is a lower limit because it is not known whether this fragment binds to the same site as the linked fragment, the isolated amine fragment binds to the same site as 5a based on its complete displacement of the probe in the A1 site competition assay (Figure 4A).

Figure 4.

A1 site binding by 5a inhibits SAMHD1 by inducing dimerization and preventing tetramer formation. (A) Competitive anisotropy probe displacement assay. SAMHD1 (2 μM) was preincubated with 5′ FAM-pdGpdGpdG-3′ (0.5 μM) prior to the addition of each compound. To facilitate visual comparisons, the red dashed line depicts the binding isotherm for GTP (Figure 2D). (B) Effect of cellular A1 site ligands GTP and dGTP on the inhibitory potency of 5a as determined using the MG colorimetric activity assay. Three-hundred nanomolar SAMHD1 was preincubated with increasing concentrations of 5a before the addition of 50 μM or 200 μM GTP using 50 μM dATP as the substrate. (C) Glutaraldehyde cross-linking assay to determine the effect of 5a on the oligomeric state of SAMHD1. Six-hundred nanomolar SAMHD1 was incubated with 5a for 10 min, cross-linked using glutaraldehyde, and the monomer (M), dimer (D), and tetramer (T) forms of SAMHD1 were resolved by gel electrophoresis. A 50 μM GTP incubation served as the positive control. (D) Glutaraldehyde cross-linking assay to determine the effect of 5a on the oligomeric state of SAMHD1 when in competition with GTP and dATP substrate. Six-hundred nanomolar SAMHD1 was incubated with 5a for 10 min before the addition of 50 μM GTP and 50 μM dATP. Control reactions without 5a were run with the same nucleotide concentrations. (E) SAMHD1 was preincubated with the indicated nucleotides to generate a tetramer and subsequently diluted 100-fold into a solution containing varying concentrations of 5a and 100 μM araCTP as the substrate (see text). All error bars represent the SD of two independent replicate experiments.

In addition to displacing the A1 site probe, further evidence that 5a binds in the same orientation as GTP in the A1 site is provided by the SAMHD1-D137N A1 site mutant.³⁸ This hydrogen bond donor–acceptor reversal mutation changes the A1 site specificity from GTP to XTP and causes a 30-fold increase in the IC₅₀ of 5a (IC₅₀^D137N = 250 ± 160 μM; Figure S4). The finding that fragment 4 binds ∼3 to 5-fold less tightly than dG and dGMP to SAMHD1 indicates that addition of the propylamine linker introduces a modest binding penalty that is masked when the two fragments are linked. Linker optimization will be a key goal in future iterations of this approach.

In a cellular context, inhibitors that target the A1 site must compete with free GTP and dGTP. Of these two, the major intracellular competitor is GTP given its generally higher reported average total intracellular concentration of 230 ± 200 μM for normal cells and 470 ± 230 μM for tumor cells.³⁹ In contrast, the concentrations of dGTP and other dNTPs are only in the 1 to 20 μM range.³⁹ As expected, increasing the concentration of GTP from 50 μM to 200 μM in the dNTPase assay, with 50 μM dATP as the substrate, increased the IC₅₀ of 5a by 8-fold (IC^{50 μM GTP}₅₀ = 8.6 ± 4.1 μM; IC^{200 μM GTP}₅₀ = 70 ± 50 μM; Figure 4B). We are encouraged that 5a can effectively reduce the dNTPase activity with an IC₅₀ of less than 10 μM using conditions that approximate intracellular nucleotide concentrations. The ultimate success of this approach will depend on many factors including (i) a further increases in the binding affinity, (ii) the free GTP concentrations in different cancer cells and the spatiotemporal regulation of GTP concentrations,⁴⁰⁻⁴² and (iii) the possible synergistic therapeutic effects of bifunctional inhibition of the dNTPase and DNA binding activities (see below).

We observed that 5a induced dimerization of SAMHD1 in a concentration-dependent manner (Figure 4C) and that preincubation of 5a with SAMHD1 prevented tetramerization when in direct competition with GTP and dNTPs (Figure 4D). However, 200 μM 5a only weakly inhibited dNTPase activity when the SAMHD1 tetramer was formed first in the presence of A1 and A2 site activators and then diluted 100-fold into a solution containing 100 μM araCTP substrate (Figure 4E). Using araCTP as the substrate in this assay is key because it is unable to activate its own hydrolysis,¹⁷ and therefore, the dNTPase activity arises from the GTP and dATP activators that remain bound to the A1 and A2 sites after dilution. Occlusion of the A1 site from inhibitors after formation of the tetramer has been reported in different contexts.¹ The mechanistic implication is that the A1 site, and possibly the A2 site, are not in equilibrium with free nucleotides during enzyme turnover with a substrate (i.e., the SAMHD1 tetramer exhibits hysteresis). However, it has been recently shown that phosphorylation of Thr592 of SAMHD1 leads to tetramer destabilization, which may facilitate access of small molecules to the activator sites.³⁷ This phosphorylation event catalyzed by CDK2 in early S phase of dividing cells is required for SAMHD1 to function at replication forks.²²

SAMHD1 Nucleic Acid Binding Inhibited by 5a

We tested whether 5a could also disrupt SAMHD1 binding to ssDNA and ssRNA.^37,43,44 SAMHD1 (1 μM) was incubated with either ssDNA or ssRNA 32mers (0.5 μM) comprised of identical base sequences (Table S3). Under these conditions, binding of the RNA and DNA induces dimerization and partial tetramerization of SAMHD1 as determined using the glutaraldehyde cross-linking assay (nucleic acid strands do not cross-link to SAMHD1 under these conditions; Figure 5A,B).^37,43,44 Upon addition of increasing concentrations of 5a, the dimer → tetramer equilibrium shifted toward dimer as 5a displaced the ssDNA and ssRNA (Figure 5A,B). This result differs from the dNTP-bound tetramer, where 5a was unable to disrupt the tetramer (Figure 4E). We surmise that A1 site is more accessible in the DNA- and RNA-bound tetramers than the dNTP induced tetramer. We found that 5a could effectively displace 5′-FAM-labeled ssDNA32 and ssRNA32 (0.5 μM) from SAMHD1 by following the decrease in the fluorescence anisotropy signal (EC^5a_50 ssDNA = 7.7 ± 2.8 μM, EC^5a_50 ssRNA = 14 ± 3.6 μM; Figure 5C,D). Of note, 5a was 12-fold more effective in displacing both ssDNA32 and ssRNA32 than GTP (EC^GTP_50 ssDNA = 89 ± 25 μM, EC^GTP_50 ssRNA 170 ± 32 μM; Figure 5C,D).

Figure 5.

5a disrupts nucleic acid complexes with SAMHD1. (A, B) SAMHD1 (1 μM) was added to a solution containing 5′ FAM-labeled 32mer single-stranded DNA or RNA (0.5 μM), and the indicated [5a] was added. Protein was then cross-linked using glutaraldehyde and resolved by denaturing gel electrophoresis. Tetramer (T), dimer (D), and monomer (M) forms are indicated. (C, D) One micromolar SAMHD1 and 0.5 μM 5′ FAM-labeled 32mer ssDNA and ssRNA were preincubated prior to the addition of increasing concentrations of 5a or GTP (positive control for A1 site binding). Error bars in panels C and D represent SD from triplicate experiments.

Assessing Cellular Activity of 5a

We anticipated that the cellular activity of 5a might be compromised by the negatively charged phosphate group in the fragment linker. This was confirmed with a widely used “AraC rescue” assay where a THP1 cell line expressing high levels of SAMHD1 is cultured with AraC in the presence and absence of 5a. In this assay, the cytotoxicity IC₅₀ of AraC (as measured using a MTS cell viability assay) is expected to decrease when SAMHD1 dNTPase is depleted or inhibited. Accordingly, the IC₅₀ of AraC decreased 20-fold in an isogenic THP1 cell line that was stably transduced with a lentiviral vector expressing SAMHD1 shRNA, as compared to the same line transduced with a shRNA control (Figure S6). In contrast, treatment of the THP1 cell line with a nontoxic concentration of 5a (50 μM) resulted in no change in the IC₅₀ value of AraC (Figure S7). Of course, we cannot exclude the possibility that nucleotide competition also contributes to the absence of cellular activity of 5a. Accordingly, future iterations of the guanine tethering approach will use uncharged linkers to increase membrane permeability and optimized functional groups that enhance binding relative to the current bromobiphenyl substituent.

Structural Model for SAMHD1 and 5a Complex

Crystals of SAMHD1 in complex with 5a were obtained by soaking 5a into an apo-SAMHD1 crystal that was grown in the absence of activating nucleotides (Figure S8A, Table S6). The apo-SAMHD1 crystals produced crystallographic dimers with empty A1 binding pockets, which allowed A1 site occupancy by 5a upon soaking (Figure 6A, Figure S8B). As expected, the guanine base of 5a aligned exactly with the previously determined binding mode of GTP in the A1 site with excellent electron density and in a one-to-one stoichiometry with each SAMHD1 monomer (Figure S8B,C). The key hydrogen bonds between Asp137, Gln142, and Arg145 to the N1, N2, O6, and N7 groups of guanine, respectively, were perfectly conserved. In addition, the 5′ phosphate of 5a interacted with the Arg451′ of a second monomer in the same manner as observed with bound GTP. However, despite the observation of a charged hydrogen bond, the phosphate group is not required for binding or inducing dimerization because dG and dGMP bind with similar affinities (Figure 2, Table 1). We note only very small structural rearrangements between the apo-SAMHD1 structure and the complex with 5a but marked differences with the fully activated tetramer structure (see below).

Figure 6.

Crystal structures of apo-SAMHD1 and its complex with 5a. (A) SAMHD1 dimer in complex with two molecules of 5a (PDB: 8GB1; 2.46 Å resolution). The guanine base of 5a binds to the A1 site, and the flexible C-terminal domains (CTD) are shown in dark cyan and dark green. (B) Apo-SAMHD1 (PDB: 8GB2) shows no electron density in the A1 site pocket. However, the dG fragment of 5a shows strong A1 site density, and the bromobiphenyl group of 5a binds in a crevice between the CTD hinge and the core of the protein. (C) Detailed binding mode of 5a showing the complete network of interacting side-chains. Side chains from the blue SAMHD1 monomer orient the guanine base, while a second monomer (green) interacts with the linker and bromobiphenyl group. These interactions appear to involve primarily hydrophobic and van der Waals forces. (D) Movement of CTD required for tetramerization is hindered by binding of 5a (yellow tetramer, PDB: 6TXC).

The structure shows that the bromobiphenyl group occupies a hydrophobic cleft formed by a single SAMHD1 monomer that has geometric and size properties well-matched for binding 5a (Figure 6B). Compared to the deoxyguanine nucleoside, the biphenyl ring system appears to experience dynamics as suggested by its decreased electron density (Figure S8D). Binding of the biphenyl ring system is partially stabilized by insertion of the bromine atom in a pocket formed by the β-methylene of Asn452′ and the hydrophobic side chain of Val557′ (Figure 6C). We infer that the bromobiphenyl group inhibits tetramer formation by preventing rotation of the adjacent carboxyl terminal lobe containing helix α19 and strand β5. This conformational aspect of the inhibition mechanism is depicted in an overlay of the 5a complex with the dNTP activated tetramer (Figure 6D). Upon formation of the tetramer, both helix α19 and strand β5 move toward the activator sites to promote a productive conformation of the active site, which would be sterically prevented in the presence of 5a. An additional component of the inhibition by 5a likely involves preventing dNTP binding to the proximal A2 site. Under normal activation conditions, binding of the A2 site dNTP brings its triphosphate group in such close contact with the triphosphate group of GTP that they share a bridging magnesium ion for charge stabilization.⁴⁵ Thus, these structural and biochemical studies point to an inhibition mechanism where 5a stabilizes the SAMHD1 dimer and then blocks binding of the A2 site dNTP and prevents CTD rearrangement required for tetramerization. Inhibition of ssDNA and ssRNA binding also likely arises from 5a blocking the A1 site that is used to bind to guanine nucleotides within nucleic acid sequences.^23,37

Conclusions

Although several unsuccessful library screens have been reported with SAMHD1,^31,46,47 and a nucleotide-based inhibitor with a methylene bridge connecting the α-phosphate and 5′-carbon has been described,³⁵ a general strategy with the potential for discovery of biologically active inhibitors is needed. In this regard, we have described an efficient strategy that has led to the first bifunctional small molecule ligand for SAMHD1 that inhibits both its dNTPase and nucleic acid binding activities. The dNTPase inhibition mechanism is complex and involves induced changes in the oligomeric state of apo-SAMHD1 to promote stable dimers that are refractory to forming tetramers. An additional virtue of targeting the A1 site is the key role this site plays in nucleic acid binding. Thus, these findings provide a strong proof-of-principle that bifunctional small molecule inhibitors that inhibit both dNTPase and nucleic acid binding is a viable drug design strategy. Such molecules have the opportunity to synergistically increase the incorporation of chain terminating nucleosides that promote fork arrest and also prevent fork restart through perturbation of the replication fork restart activity of SAMHD1. It is appealing to envision a new immunotherapy approach where inhibition of SAMHD1, in conjunction with widely used nucleoside therapeutics, could enhance tumor killing by direct chemical toxicity and through activation of an innate immune response against the tumor.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.3c00118.

• Experimental methods, supplementary tables (Tables S1–S6), supplementary figures (Figures S1–S12), supplementary chart (Chart S1), synthetic procedures (Schemes S1–S6), and spectroscopic characterizations of 5a and synthetic precursors (PDF)

Accession Codes

Atomic coordinates and structure factor amplitudes of the Apo-SAMHD1 and the SAMHD1-5a complex have been deposited in the Protein Data Bank (www.rcsb.org) with the PDB accessions codes 8GB2 and 8GB1, respectively. The coordinates are designated for immediate release upon publication.

Author Contributions

M.E. performed library synthesis, high-throughput screening, and crystallization experiments. L.D. performed binding measurements in Table 1. A.H.H. assisted in library synthesis. S.B. assisted in chemical synthesis and performed analytical chemistry on compounds. B.O. performed protein purification and nucleic acid binding measurements. M.B. collected and analyzed diffraction data and solved the crystallographic models. J.T.S. and M.G. conceived the concept and supervised all of the experimental work. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.

Special Issue

Published as part of ACS Chemical Biologyvirtual special issue “Nucleic Acid Regulation”.