Synthesis and In Vitro Evaluation of C-7 and C-8 Luteolin Derivatives as Influenza Endonuclease Inhibitors

The part of the influenza polymerase PA subunit featuring endonuclease activity is a target for anti-influenza therapies, including the FDA-approved drug Xofluza. A general feature of endonuclease inhibitors is their ability to chelate Mg2+ or Mn2+ ions located in the enzyme’s catalytic site. Previously, we screened a panel of flavonoids for PA inhibition and found luteolin and its C-glucoside orientin to be potent inhibitors. Through structural analysis, we identified the presence of a 3′,4′-dihydroxyphenyl moiety as a crucial feature for sub-micromolar inhibitory activity. Here, we report results from a subsequent investigation exploring structural changes at the C-7 and C-8 positions of luteolin. Experimental IC50 values were determined by AlphaScreen technology. The most potent inhibitors were C-8 derivatives with inhibitory potencies comparable to that of luteolin. Bio-isosteric replacement of the C-7 hydroxyl moiety of luteolin led to a series of compounds with one-order-of-magnitude-lower inhibitory potencies. Using X-ray crystallography, we solved structures of the wild-type PA-N-terminal domain and its I38T mutant in complex with orientin at 1.9 Å and 2.2 Å resolution, respectively.


Introduction
Influenza viruses cause illness in a variety of species, including humans. Despite the availability of vaccines and antiviral drugs, influenza remains a serious threat to human health, causing 290,000-650,000 deaths worldwide annually [1]. Influenza virus RNA-dependent RNA polymerase (RdRP) lacks proof-reading activity, which leads to an accumulation of point mutations known as antigenic drift. This is responsible for the emergence of new viral variants causing seasonal flu, which requires the flu vaccine to be reformulated every year. Antigenic drift also contributes to increasing viral resistance against antiviral drugs. Additionally, antigenic shift-the reassortment of viral RNA segments from two or more different influenza strains in animals or humans-could lead to a new pandemic strain. In fact, various zoonotic strains are considered "ticking time bombs", as these pathogens have the potential to mutate to facilitate human transmission given enough time and infected organisms. Generally, vaccination is the best intervention against viral pathogens including influenza. However, the last influenza pandemic in 2009 reminded us that effective vaccines are often not available at the onset of a pandemic. Reformulation of vaccines is time-consuming, and many human lives can be taken by the disease during this process. The intricacies of vaccine development, in combination with influenza's genomic variability, makes the development of novel anti-influenza therapeutics imperative.
Influenza viruses contain a single-stranded, negative-sense RNA genome in complex with RdRP [2][3][4]. RdRP comprises the subunits PA (polymerase acidic protein), PB1 and PB2 (polymerase basic protein 1 and 2). The virus itself is unable to synthesize the 5 -mRNA cap required for eukaryotic translation; this represents the "Achillies heel" of influenza virus. The virus obtains host primers-short oligomers of host pre-mRNA that initiate transcription-by a unique "cap-snatching" mechanism [5][6][7][8][9], which serves as a target for pharmaceutical intervention. The process begins with binding of the PB2 subunit to the 5 -cap (m 7 GTP) of the host pre-mRNA. Subsequently, the PA subunit cleaves the RNA strain approximately 10-13 nucleotides downstream from the 5 -cap to acquire the cap/primer [10]. The PB1 subunit uses this detached RNA segment as a template for viral mRNA synthesis. RdRP is highly conserved across influenza strains, and the three subunits involved in the cap-snatching mechanism have been recognized as attractive targets for drug development in the past decade [11][12][13][14][15][16][17][18].
The PA subunit is a bridged binuclear metalloenzyme with an N-terminal domain (PA-Nter) harboring the endonuclease active site that carries out cleavage of the RNA segment. The active site is a negatively charged pocket that accommodates either Mg 2+ or Mn 2+ ions, with stronger affinity for the latter [19]. These ions are critical for endonuclease activity. Evidence suggests that PA-Nter endonuclease inhibitors must possess a metal-binding pharmacophore with the ability to bind either Mg 2+ or Mn 2+ ions efficiently [9].
Even though metalloenzymes comprise more than one-third of all known enzymes, clinical development of metalloenzyme inhibitors is rather limited and few such inhibitors have been approved by the Food and Drug Administration (FDA) [9,[20][21][22]. Less than 5% of all FDA-approved drugs target metalloenzymes [23]. Unsurprisingly, metal-binding pharmacophores exhibit a lack of structural diversity.
There is currently only one FDA-approved influenza endonuclease inhibitor: baloxavir marboxil (trade name Xofluza), which is administrated as a prodrug [24,25]. Successful antiviral drugs must be able to block at least the existing variant of the target enzyme and should have a high resistance barrier. One key mutation in the influenza A/H1N1 2009 pandemic (A/H1N1pdm) and A/H3N2 viruses is Ile-38 to Thr-38 in PA-Nter. This mutation reduced patient susceptibility to baloxavir marboxil and impaired the virus' replicative fitness in cells [26]. Resistance development could eventually lead to the loss of the clinical relevance of Xofluza. Furthermore, fewer than a dozen PA-Nter endonuclease inhibitor classes have been reported in the literature to date. These classes include diketo acids [27], dopamine derivatives [28], hydroxylated heterocycles [29][30][31], flutimide congeners [32], green tea catechins [33,34], catechol derivatives [35], hydroxylated N-acyl-hydrazones [36] and others [18]. Recently, we identified the molecular mode of action of flavonoids in influenza-infected cells [37]. We developed a screening assay based on AlphaScreen technology and determined the inhibitory potency of 38 flavonoids, of which luteolin (IC 50 of 73 ± 3 nM) and its 8-C-glucoside orientin (IC 50 of 42 ± 2 nM) were the most potent inhibitors (Figure 1). A gel-based endonuclease inhibitory assay confirmed our findings from the competition assay. Finally, we performed structural analysis of complexes of PA-Nter with luteolin and myricetin and described various binding poses of flavonoids in the PA-Nter active site. In the current study, we aimed to pursue structure-assisted drug design guided by our previously reported crystal structure of PA-Nter in complex with luteolin (PDB entry 6YA5, 2.0 Å resolution) [37]. The structure revealed that the hydroxyl group at position C-7 forms a hydrogen bond with the Glu-26 residue of PA-Nter. The surface complementarity and the strong interaction of hydroxyls at the B-ring with metal ions contribute to the high affinity and inhibitory potency of luteolin. Thus, luteolin may serve as a useful scaffold to introduce modifications that improve the inhibitory potency and pharmacokinetic properties. Luteolin itself has low bioavailability [38,39] and is prone to oxidation. Therefore, we replaced the C-7 hydroxyl with other moieties capable of creating hydrogen bonds with Glu-26. Bio-isosteric replacement is a proven tool for modulating the drug-like properties of promising therapeutics [40]. The C-7 hydroxyl group has slightly higher acidity [41] compared to the other hydroxyls of luteolin. This allowed us to selectively modify the C-7 position using Pd-catalyzed cross couplings via the corresponding C-7 triflate.
We then continued to leverage the known structure-activity relationships (SAR) in 3 ,4 -dihydroxyphenyl flavones. Moreover, we expanded our research effort to explore the chemical space around the C-8 position of the luteolin scaffold. Specifically, we set out to investigate whether moieties introduced at the C-8 position by Mannich reaction [42] would be tolerated or even enhance inhibitory potency. Our goal was to find additional points of interaction between luteolin derivatives and PA-Nter and, more importantly, to explain why vitexin exhibits moderate inhibitory potency even though it does not have the 3 ,4 -dihydroxyphenyl motif. This ortho-hydroquinone motif was originally identified as the metal-binding pharmacophore and as such was considered indispensable for inhibition. That paradigm explains the lack of inhibitory potency of apigenin but fails to rationalize vitexin's moderate potency (see Figure 1).

Compound Synthesis
Based on the SAR outlined in Figure 1, our initial efforts focused on exploring bioisosterism at the C-7 moiety of luteolin. Guided by general knowledge of hydrogen bond formation and prior crystallographic study, we proposed that the hydroxyl group could be replaced with a small group featuring either N-H or O-H bonds. We also decided to introduce heterocyclic moieties that are not necessarily associated with textbook OH bio-isosteric replacement. Preparation of this C-7 series started from luteolin, which was per-acetylated. Compound 1 was then selectively mono-deacetylated at C-7 using previously reported conditions [43] (Scheme 1). This regioselective deprotection takes advantage of the higher acidity of the C-7 hydroxyl group in the luteolin scaffold caused by the electron-withdrawing pyrone carbonyl moiety in the para position. Therefore, thiophenolate-mediated O-deacetylation selectively generates the corresponding phenolate (2), which was next treated with triflic anhydride to provide intermediate 3 in a decent yield. Scheme 1. Reagents and conditions: (i) Ac 2 O, pyridine, 145 • C, 3 h, 77%; (ii) thiophenol, imidazole, NMP/THF (1:3), 0 • C to r.t., 5 h, 68%; (iii) Tf 2 O, pyridine, DCM, 0 • C, 3.5 h, 41%.
Triflate 3 was subjected to a wide range of palladium-catalyzed cross-couplings. In addition to the desired products, certain reactions also yielded C-5 O-deacetylated analogues (Scheme 2). However, formation of such byproducts was not an obstacle, because we subsequently performed global deprotections of the flavone scaffolds. Hirao coupling [44] of 3 resulted in a high yield of diethyl phosphonate 4. Subsequent standard dealkylation using trimethylsilyl bromide [45] resulted in phosphonic acid 5. Molybdenum hexacarbonyl was used as a source of carbon monoxide for Pd-catalyzed methoxy-carbonylation. This approach was superior to methoxycarbonylation with dicobaltoctacarbonyl [46] in terms of yield and subsequent cleaning of the reaction apparatus. The pre-purified mixture of methyl esters 6 and 7 was subjected to hydrolysis and to different aminolysis reactions, which produced 8, 9 and 10. One-pot Curtius rearrangement converted 3,4,5-trihydroxyflavone-7-carboxylic acid (8) to amino-derivative 11 in an acceptable yield of 30%. In general, electron rich anilines are prone to quick degradation, and thus the corresponding acetamide 12 was prepared from 11 using AcOSu. Palladium-catalyzed cyanation of triflate 3 by a modified Buchwald's procedure [47] gave rise to tetrazole derivative 14 and nitrile 15. Amino-methyl transfer to triflate 3 was accomplished by protected potassium aminomethyltrifluoroborate under slightly modified conditions, reported by Molander [48]. The crude product of crosscoupling was then subjected to trifluoroacetic-acid-mediated de-tert-butylation, affording 17 as a homolog of amine 11.
Our well-established method for preparation of the key intermediate 3 enabled us to execute heteroarylations of the flavone scaffold. We hypothesized that replacement of the hydroxyl group with pyrazole and azaindole could, in principle, ensure engagement of Glu-26 while improving the metabolic stability of the flavone. Synthesis of 18-20 was mediated by Pd(PPh 3 ) 4 or by Buchwald 2nd generation pre-catalysts. Regardless of which catalyst was used, 18-20 were obtained in low yields (Scheme 3). Although efficient Mannich reactions with luteolin [49] and quercetin [50,51] have been previously described, we found this type of C-C formation reaction very challenging. In our hands, formation of the desired products was in all cases accompanied by formation of C-6 regioisomers, contrary to reports in the literature. We also observed formation of other byproducts with a methylendioxy group on the B-ring serving as a 2:1 adduct of formalin to luteolin. The very limited solubility of luteolin further complicated our effort to optimize the Mannich reaction. Methanol worked least poorly among a variety of solvents screened (methanol, ethanol, propan-2-ol, trifluoroethanol, 1,4-dioxane and DMF). Room temperature and equimolar loading of formalin and the secondary amine with respect to luteolin efficiently suppressed the double Mannich reaction at both the C-6 and C-8 positions. To avoid issues with low regioselectivity and to exclude methylendioxy formation, we attempted Mannich reactions with tetraacetate 1 and triacetate 2. These reactions led to complex reaction mixtures due to low conversions and the lability of acetates in the presence of secondary amines. According to HPLC analysis, subsequent methanolysis resulted in a simplified mixture of compounds, but yields of products were low (~15%). Considering the need to prepare triacetate 2, this approach is clearly not advantageous to the Mannich reaction of parental luteolin in methanol. Data resulting from screening of the Mannich reaction with 1 and 2 are provided in the Supplementary Materials.
To explore the chemical space around C-8 as much as possible, we used structurally different secondary amines possessing additional functional groups (Scheme 4). Mannich reactions were stirred at room temperature overnight until almost all luteolin was consumed. The ratio between C-8 Mannich adduct and unwanted C-6 regioisomer ranged from 1.7:1 to 3.0:1. Since both regioisomers have similar retention factors, the purification of 21-32 was very laborious. Products were purified by multiple HPLC preparations that further negatively influenced the yields. Alkyl esters of secondary amines were used for preparation of 30-32, and after the Mannich reaction, these ester moieties were trans-formed into carboxylic acids.

Relationship between Chemical Structure and Inhibitory Potency
The inhibition potencies of the prepared compounds were assessed by an assay that we recently developed for screening PA-Nter inhibitors based on the amplified luminescent proximity assay system (AlphaScreen) [37]. Examples of titration curves are shown in Figure 2A and Figure S1. The series of C-7 luteolin congeners displayed moderate inhibitory potencies with one-to two-order-of-magnitude higher IC 50 values compared to luteolin (Table 1).   18 3.4 ± 0.6

19
0.81 ± 0.12 Phosphonic acid 5 inhibited PA-Nter with an IC 50 of 9.2 µM, which was one of the weakest inhibitory potencies of the whole series. On the other hand, carboxylic acid 8 was a sub-micromolar inhibitor (IC 50 = 0.98 µM). Related amides 9 and 10 were less potent than 8 (IC 50 = 2.0 and 7.6 µM, respectively), which might indicate relatively confined C-7 proximal space. The amino derivative 11 had approximately 20-times weaker binding potency than luteolin. Clearly, the anticipated interaction between the C-7 amino group and side chain of Glu-26 did not result in superior binding. We speculate that solvation of the protonated amino group disrupted the effective formation of a salt bridge with Glu-26. Acetamide 12, tetrazole derivative 14, and nitrile 15 exhibited relatively flat SAR with IC 50 values of 3.0, 1.4 and 1.3 µM, respectively. Amino-methyl derivative 17 exhibited a significantly decreased inhibitory activity compared to luteolin (IC 50 = 6.7 µM). Apparently, neither strongly acidic nor basic moieties at C-7 are tolerated (see the inhibitory potencies of phosphonic acid 5 and aminomethyl derivative 17). The relatively bulky azaindole congener 18 exhibited appreciable inhibitory potency with an IC 50 value of 3.4 µM. Pyrazoles 19 and 20 differed slightly in inhibitory potency; the former was more potent than its methylated analogue 20, but both exhibited IC 50 values comparable with that of tetrazole 14 (0.81 and 2.6 versus 1.4 µM). Considering the almost identical half-maximal inhibitory concentrations of weakly basic pyrazoles 19 and 20 and acidic tetrazole 14, we surmise that interaction with PA-Nter is not susceptible to the "proton affinity" of the moiety introduced at C-7. The inhibitory potencies of 5 and 17 further strengthen our hypothesis. On the other hand, it seems that steric effects play a key role in this specific point of interaction with the endonuclease. To confirm this, we screened cynaroside, a commercially available luteolin 7-O-β-D-glucoside. The introduction of a bulky glucose residue at C-7 led to an almost complete loss of inhibition (IC 50 = 32 ± 3 µM; three orders of magnitude worse than luteolin). We conclude that relocation of amino acids surrounding Glu-26 within PA-Nter and/or a clash with the Glu-26 side chain leads to a significant drop of inhibitory potency.
Next, we assessed the effect of C-8 substituents. As shown in Table 2, almost all aminomethylene moieties were well-tolerated. Compounds 21, 23-25 and 29-32 exhibited inhibitory potencies roughly comparable to that of orientin (IC 50 = 0.042 µM). Moieties with additional basic (25) or acidic residues (30-32) had IC 50 values within the same range. This suggests the lack of a PA-Nter amino acid featuring proton affinity in the chemical space around C-8. The bulky nor-tropine (27, IC 50 = 0.075 µM) and dihydroxytetrahydroisoquinoline (28, IC 50 = 0.12 µM) scaffolds were not found to be significantly unfavorable. It is likely that both bulky residues are oriented away from the PA-Nter active site and into the solvent. This hypothesis fits well with the findings presented in Section 2.3. 4-Hydroxypiperidine derivative 22 showed a slightly decreased potency in comparison with 3-hydroxy analogue 21 (0.14 versus 0.083 µM). However, the micromolar inhibitory potency of alcohol 26 (IC 50 = 1.2 µM) has no rational explanation based on the acquired SAR. Rather, we anticipated that 26, as a cyclic analogue of 21, would be a sub-micromolar inhibitor. To assess the inhibition of endonuclease activity by selected compounds by a direct mechanism-based method, we applied a gel-based endonuclease inhibitory assay ( Figure 2B). This assay was performed for five selected ligands (orientin, baloxavir acid, luteolin, 21 and 30) with wild-type PA-Nter (see Supplementary Materials, Figure S2). The analysis showed that orientin had a higher inhibitory potency than luteolin, in agreement with the AlphaScreen assay. The gel-based assay also confirmed similar inhibition activities for baloxavir acid, orientin, 21 and 30. We also attempted this assay with the PA-Nter I38T variant, but the mutation of Ile-38 to Thr-38 led to complete loss of ssDNA cleavage ability (see Figure S3). Thus, we used a fluorescent-labelled ssRNA substrate and performed a FRET-based endonuclease assay with this variant. Compared to wild-type PA-Nter, the I38T variant had only 1.9% activity (see Figure S4) toward the ssRNA substrate and none for ssDNA. This result is in line with a recently reported observation of significantly reduced fitness and ssRNA nuclease activity of a virus harboring the I38T variant [26].

Crystal Structures of Wild-Type and I38T PA-Nter Domains in Complex with Orientin
To reveal the structure of orientin bound to wild-type PA-Nter and the I38T mutant, we prepared both proteins for X-ray crystallographic studies. Thanks to its high inhibitory potency and aqueous solubility, orientin (PDBe ligand USE) was soaked into the unoccupied protein crystals. The structure of wild-type PA-Nter in complex with orientin was refined to 1.9 Å resolution (PDB ID 7NUG), and the I38T mutant complex (PDB ID 7NUH) was refined to 2.2 Å resolution. Both crystallographic models consisted of one protein molecule per asymmetric unit. Two metal ions were embedded in the endonuclease active site. Based on the strong anomalous signal observed, we speculate a mixed occupancy of Mn 2+ and Mg 2+ cations. The majority (0.8 occupancy) proximal ion was Mn 2+ , coordinated by four protein atoms (N ε2 His-41, O δ2 Asp-108, O ε2 Glu-119, O Ile-120) and two hydroxyl groups from the 3 ,4 -dihydroxyphenyl moiety of orientin. The distal octahedrally coordinated sphere was partially assigned as the central Mg 2+ cation (0.4 occupancy), which corresponds to lower anomalous scattering and is in agreement with previously reported PA-Nter complexes ( Figure 3) [19,37]. The distal ion was coordinated by O ε2 Glu-80, O δ2 Asp-108, the 4 -hydroxyl group of orientin, and three water molecules (W1, W2, W3). Unoccupied PA-Nter (PDB entry 5DES, not shown) harbors metal ions coordinated by two additional water molecules, which are replaced by the two hydroxyl groups from the flavonoid's B ring in our wild-type and I38T variant structures. Orientin adopts a similar position as luteolin in a previously described PA-Nter structure (PDB ID 6YA5) [37]. Ligands in both structures form a hydrogen bond with O ε2 Glu-26 at the C-7 position. The previously observed high affinity of luteolin for PA-Nter is likely to be enhanced by the additional hydrogen bonding network surrounding orientin's glucosyl moiety. This was observed in both protein variants, as most of the water molecules in the first solvation shell are located at similar positions ( Figure 4  The side chain of Tyr-24 in the mutant variant approaches the active site pocket, whereas in the wild-type, Tyr-24 is pushed away from the cavity (RMSD of 0.029 Å for side chain atoms). Thr-38, which is one atom shorter than Ile-38, helps accommodate orientin. O γ1 Thr-38 forms a hydrogen bond through W21 to the glucosyl moiety of the ligand (O6) ( Figure 5). This is not observed in the wild-type, as Ile-38 does not contain a hydroxyl group capable of such interaction.

AlphaScreen Assay
AlphaScreen experiments were performed using a Perkin Elmer Enspire plate reader in 96-well ProxiPlates. Biotinylated L-742.001 derivative [37] was captured on Streptavidincoated donor beads (Perkin Elmer). Separately, GST-PA-Nter fusion protein was bound to GSH-coated acceptor beads (Perkin Elmer). These solutions were incubated for 60 min at room temperature in the dark and subsequently mixed and incubated for an additional 120 min. In experiments screening for endonuclease inhibitors, compounds were mixed with both types of beads prior to the 120-min incubation. The optimal concentrations of biotinylated L-742.001 derivative and GST-PA-Nter were 15 nM and 50 nM, respectively. The concentrations of donor and acceptor beads were 5 µg/mL each in a 50 µL reaction volume. All experiments were performed in 25 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.05% Tween20, 1 mM MnCl 2 , 10 mM MgCl 2 , and 1 mM 2-mercaptoethanol.

Gel-Based Endonuclease Inhibitory Assay
To compare the endonuclease activities of wild-type and I38T PA-Nter in the presence of selected compounds (baloxavir acid, luteolin, orientin), we used a gel-based endonuclease inhibitory assay.

Crystallization and Diffraction Data Collection
Hexagonal bifrustum crystals of empty wild-type and I38T PA-Nter subunits were obtained by the hanging-drop vapor diffusion method. Protein solution (12 mg/mL) was mixed with crystallization reservoir solution (12.5% w/v PEG 1000, 12.5% w/v PEG 3350, 0.1% M MOPS/HEPES-Na pH 7.5, 0.06 M magnesium chloride, 0.06 M calcium chloride) and PA-Nter seed in a 1:1:0.2 ratio. Crystals grew at 18 • C until they reached approximately 0.2 mm in diameter. Ligands (100 mM solution in DMSO) were soaked in for 15 min (the final DMSO concentration did not exceed 5%). Crystals were harvested, flash-cooled by plunging into liquid nitrogen and stored at −196 • C.
Diffraction qualities were tested at BESSY II and data were collected at −173 • C on a home diffractometer (MicroMax-007 HF microfocus equipped with a PILATUS 300 K detector, Rigaku). The crystal of wild-type PA-Nter soaked with orientin diffracted to a resolution of up to 1.87 Å, and the I38T PA-Nter/orientin crystal diffracted up to 2.15 Å. Diffraction data were processed, integrated, and reduced using XDS [53] and scaled using XSCALE from the XDS suite [54]. Both crystals belonged to the P6 4 22 space group and contained one molecule per asymmetric unit, with a solvent content of 47.5% (wild-type PA-Nter), and 48.2% (I38T PA-Nter). We observed anomalous signals up to a resolution of 3.0 Å for wild-type and 3.5 Å for the I38T variant. Therefore, the data were processed with unmerged Friedel pairs. Detailed crystal parameters and data collection statistics are given in Table S1.

Structure Determination and Analyses
Structures of wild-type and I38T PA-Nter were determined by molecular replacement with MOLREP [55] from the CCP4 package [56] using a previously reported structure of PA-Nter as a template (PDB entry 6YA5 [37]). The final step of complex structure polishing was carried out by cycles of manual adjustments using Coot software [57] followed by refinement in REFMAC 5.8.0103 [58]. MolProbity [59] was used to validate the quality of the final models. Refinement statistics are given in Table S1. All figures illustrating structural representations were prepared with PyMOL (The PyMOL Molecular Graphics System, Version 2.4.2 accessed on 10 March 2020; Schrödinger, LLC., New York, NY, USA). Atomic coordinates and experimental structure factors are deposited in the Protein Data Bank under codes 7NUG for wild-type PA-Nter in complex with orientin and 7NUH for I38T PA-Nter in complex with orientin.
Compounds 1 and 2 were prepared according to literature procedures [43]. Analytical data for these compounds were in agreement with published data.