In silico and crystallographic studies identify key structural features of biliverdin IXβ reductase inhibitors having nanomolar potency

Heme cytotoxicity is minimized by a two-step catabolic reaction that generates biliverdin (BV) and bilirubin (BR) tetrapyrroles. The second step is regulated by two non-redundant biliverdin reductases (IXα (BLVRA) and IXβ (BLVRB)), which retain isomeric specificity and NAD(P)H-dependent redox coupling linked to BR's antioxidant function. Defective BLVRB enzymatic activity with antioxidant mishandling has been implicated in metabolic consequences of hematopoietic lineage fate and enhanced platelet counts in humans. We now outline an integrated platform of in silico and crystallographic studies for the identification of an initial class of compounds inhibiting BLVRB with potencies in the nanomolar range. We found that the most potent BLVRB inhibitors contain a tricyclic hydrocarbon core structure similar to the isoalloxazine ring of flavin mononucleotide and that both xanthene- and acridine-based compounds inhibit BLVRB's flavin and dichlorophenolindophenol (DCPIP) reductase functions. Crystallographic studies of ternary complexes with BLVRB–NADP+–xanthene-based compounds confirmed inhibitor binding adjacent to the cofactor nicotinamide and interactions with the Ser-111 side chain. This residue previously has been identified as critical for maintaining the enzymatic active site and cellular reductase functions in hematopoietic cells. Both acridine- and xanthene-based compounds caused selective and concentration-dependent loss of redox coupling in BLVRB-overexpressing promyelocytic HL-60 cells. These results provide promising chemical scaffolds for the development of enhanced BLVRB inhibitors and identify chemical probes to better dissect the role of biliverdins, alternative substrates, and BLVRB function in physiologically relevant cellular contexts.

Heme proteins are found in nearly all phyla, and are degraded by oxidative cleavage to release iron and initiate derivatization of the tetrapyrrole backbone for downstream metabolic usage or disposal. The vast majority of heme oxygenases studied to date utilize cytochrome P450 reductase or reduced ferredoxin as an electron source (1), and sequentially cleave heme to biliverdin (BV), 4 which is then reduced to bilirubin (BR) (2). BR is an antioxidant (3), has been associated with lower cardiovascular risk (4 -6), and is implicated in a putative antioxidant BV/BR redox cycle regulated by biliverdin reductase (7). Although heme cleavage would be expected at any of the four meso bridge carbons (yielding BV isomers IX␣, IX␤, IX␥, and IX␦), mammalian BVs are primarily generated by regioselective cleavage at the ␣-meso carbon (BV IX␣), with limited identification of BV (or BR) IX␤, IX␥, or IX␦ in adults. In contrast, isomeric composition appears distinct in the fetus where BR IX␤ is the and project "Institute for Research and Innovation in Health Sciences" Grant POCI-01-0145-FEDER-007274, co-funded by the European Regional Development Fund (FEDER) through the COMPETE 2020-Operational Programme for Competitiveness and Internationalization (POCI), PORTUGAL 2020, the "Structured program on bioengineered therapies for infectious diseases and tissue regeneration" Grant Norte-01-0145-FEDER-000012, supported by Norte Portugal Regional Operational Programme (NORTE 2020), under the PORTUGAL 2020 Partnership Agreement through FEDER. The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article contains Tables S1-S5 and Figs. S1-S3. The atomic coordinates and structure factors ( 1 Both authors contributed equally to this work. 2 To whom correspondence may be addressed. Tel.: 631-216-7425; E-mail: natasha.nesbitt@stonybrook.edu. 3 To whom correspondence may be addressed. Tel.: 631-632-1185; E-mail: jin.wang.1@stonybrook.edu. predominant rubin in neonatal bile (8), although the mechanism for BV IX␤ generation remains enigmatic. Two monomeric pyridine nucleotide oxidoreductases (biliverdin reductases) catalyze the reduction of the four isomers to generate the corresponding bilirubins. Both biliverdin reductases are members of the short chain dehydrogenase/reductase protein family (9) but retain distinct specificities for BV reduction. In humans, biliverdin reductase IX␣ (BLVRA) (10 -12) effectively reduces BV IX␣ with less efficient utilization of BV IX␤, IX␥, or IX␦ (13). In contrast, biliverdin IX␤ reductase (BLVRB) is promiscuous (14,15), catalyzing the NAD(P)H-dependent reduction of non-IX␣ biliverdin isomers (IX␤, IX␥, and IX␦) (16), and several flavins, including flavin mononucleotide (FMN) (17), pyrroloquinoline quinone (18), dichlorophenolindophenol (DCPIP), and ferric ion (Fig. 1a) (18). A single dinucleotide-binding domain accommodates both NADPH (or NADH) and substrate(s) within the verdin/flavin (V/F) binding pocket, with no evidence to date for structural or preferred partitioning for either substrate. BLVRB catalysis likely includes both stepwise (19) and concerted (20) mechanisms, with progressive support for a compulsory two-step mechanism initiated by NADPH binding (19 -21). Protonation of the pyrrolic nitrogen is the likely first step followed by hydride transfer from the nicotinamide for bilirubin formation; bulk solvent is the presumed proton source with no evidence to date for alternative source(s) based on crystallographic structure and mutagenesis studies (9,19).
BLVRB is best characterized as a flavin reductase; however, its physiological function(s) and target(s) remain elusive because of its lack of activity toward the predominant BV IX␣ isomer found in adults (22). It has been speculated that pyrroloquinoline quinone, which has been found in human milk (23), as well as human organs, plasma, and urine (24), may function as a physiological BLVRB substrate due to the affinity of the enzyme for this compound (25), but this has yet to be confirmed. Interestingly, recent data applying a genetic screen in humans have identified a BLVRB loss-of-function redox mutation (BLVRB S111L ) that unmasks a fundamentally important mechanism linking heme degradation to the development of exaggerated thrombopoiesis (platelet production) (26). This mutation affects both flavin and biliverdin coupling, and subsequent experimental and thermodynamic studies provided the foundation for the "encounter" and "equilibrium" states of the binary BLVRB-NADP ϩ complex, and the critical role of the active site Ser-111 in maintaining enzymatic function and thermodynamic stability mediated by the hydrogen-bonded network (27). We have now extended these data by integrating molecular modeling and crystallographic studies to identify an initial class of BLVRB small molecule inhibitors displaying nanomolar activity. In vitro assays using cell-based models establish that BLVRB inhibitors display selective effects on oxidation-reduction (redox) coupling. These data identify critical interactions regulating BLVRB enzymatic function, and outline the initial development of BLVRB inhibitors that can be utilized to probe metabolic and cellular effects of BLVRB function.

Results
We utilized two parallel approaches to identify BLVRB inhibitors (Fig. 1b), applying both structural similarity and computational approaches for identification of six inhibitors with enhanced potency in comparison to lumichrome, an inhibitor of BLVRB flavin reductase (FR) activity (17) (Table 1).

Halogenated xanthene-based dyes inhibit BLVRB
A manual search was conducted to identify individual compounds that chemically resembled the BLVRB substrate FMN, ranking them according to availability and solubility in aqueous solutions. Among the identified ligand candidates were xanthene-based dyes and acridine-containing compounds (Table 1 and Fig. S1). Measurement of BLVRB FR activity in the presence of these compounds established that halogenated xanthene dyes functioned as potent inhibitors of BLVRB with IC 50 values of 0.70 Ϯ 0.36 M for phloxine B and 0.41 Ϯ 0.09 M for erythrosin extra bluish. The parent compound, xanthene, displayed low activity against BLVRB confirming that this core structure alone is not sufficient for enzyme inhibition. Both phloxine B and erythrosin extra bluish are poor BLVRA inhibitors, demonstrating 77 Ϯ 3 and 89 Ϯ 2% enzyme activity, respectively, at 10 M of either compound. These compounds will exhibit IC 50 values for BLVRA well above concentrations required for BLVRB inhibition.
To explore whether phloxine B and erythrosin extra bluish could inhibit BLVRB using an alternate substrate, compounds were assayed in the presence of the BLVRB substrate DCPIP (28) to determine their effect on its DCPIP reductase activity. Both dyes were shown to inhibit BLVRB activity when DCPIP was used as the substrate (IC 50,phloxineB ϭ 1.1 Ϯ 0.1 M, IC 50,erythrosin ϭ 0.22 Ϯ 0.02 M). This observation indicated that although these compounds are structurally similar to FMN, their inhibitory capabilities are not limited to this substrate.

Two-dimensional virtual screening (2D-VS) hit identification and validation
Phloxine B and erythrosin extra bluish are used as reagents for biological staining, prompting the application of molecular modeling for expanded compound identification; a 2D-VS was modified to incorporate the intrinsic specificity ratio (ISR) as a parameter for ranking hit compounds. In addition to the traditional binding affinity parameter, ISR allows for the discrimination of native binding modes of a protein and its ligand from non-native binding modes (29). Ligands with high ISR scores are predicted to have enhanced selectivity for the protein. We initiated the 2D-VS by re-docking lumichrome into the ligandbinding site of the BLVRB-NADP ϩ -lumichrome crystal structure (PDB entry 1HE5 (22)). The root mean square deviation (r.m.s. deviation) for the ligand between the modeled and experimental lumichrome complex structures was 1.15 Å, thus validating the model. This structure was then used as the template for a 2D-VS against the NCI Diversity Set IV database (Developmental Therapeutics Program), which contains ϳ1,600 structurally diverse compounds covering the broad chemical space of ϳ140,000 compounds. This screen led to the identification of 10 compounds that were predicted to have Small molecule inhibitors of BLVRB enhanced selectivity for BLVRB (Table 1 and Table S2 and Fig.  S1), nine of which were soluble and tested for their ability to inhibit BLVRB FR activity. Seven of the identified compounds displayed modest activity (Ͻ65%) toward BLVRB, whereas one compound (NSC130813) inhibited Ͼ90% of the enzymatic activity (Table S2). As demonstrated using phloxine B and erythrosin extra bluish, NSC130813 inhibited BLVRB enzymatic activity when DCPIP was used as the substrate, displaying activity comparable with that observed using the FR assay (IC 50 ϭ 30 Ϯ 5 M).

Expansion of BLVRB inhibitor library
A focused or close analog library was built using similarity and pharmacophore search methods to identify compounds within the ZINC database (30) encompassing scaffolds similar to NSC130813. These in silico screens led to the identification of nearly 20,000 "similar" compounds that were further filtered to eliminate non-drug-like compounds (31). Iterative computational studies identified 13 additional compounds based on ISR and predicted affinity, nine of which were commercially available and tested for their ability to inhibit enzymatic activity. ZINC4366439 (NSC12516) was identified as the most potent inhibitor within this second group of compounds with activity toward BLVRB comparable with that of NSC130813. ZINC0977089, ZINC27528243, ZINC21093196, ZINC0977107, and PH006888 contain acridine (or similar polycyclic aromatic) core structures, but have different substitution patterns on their core and/or aniline moieties, and displayed modest (or no) activity toward BLVRB (Table 1 and Fig. S3). Similarly, ZINC71767097 and ZINC71767103, which contain similar substitution patterns on their aniline moieties but have quinolone core structures, displayed minimal activity toward BLVRB (Table S3). These observations indicate that within this group of inhibitors both the acridine and aniline moieties are important for inhibiting BLVRB.

Proposed mode of binding of active inhibitors
Binding modes for top inhibitors were predicted using the AutoDock (32) package (Fig. 2). All four compounds identified from the 2D-VS are located within the substrate-binding site, adjacent to the nicotinamide moiety of the NADP ϩ cofactor (Fig. 2, top panel). As previously described for structures of BLVRB complexed with substrates and inhibitors (22), these compounds are predicted to bind primarily through hydrophobic interactions with the protein. NSC130813 and NSC12516 are docked such that their hydrophobic acridine moieties adopt similar conformations by projecting into the hydrophobic subpocket composed of Gly-76, Ser-111, and Leu-125 (Fig. 2, middle panel). The hydroxyl groups located on the phenyl moiety of NSC130813 and NSC12516 form hydrogen bonds (about 2.8 Å) with the main chain nitrogen of Arg-78 and Thr-77 (Fig. 2, a  and b). Different from the interactions of NSC130813 and NSC12516, the acridine nitrogen of PH001924 is predicted to form a hydrogen bond (about 3.2 Å) with the oxygen atom of Ser-111 (Fig. 2d). Additionally, its chloro group interacts with the main chain oxygen atom of Thr-129. PH006888 displays a

Small molecule inhibitors of BLVRB
different binding mode with its acridine moiety being rotated and the acridine nitrogen forming a hydrogen bond with the nitrogen of His-153 (2.82 Å) (Fig. 2c). Additionally, the methane sulfonic acid moiety of PH006888 forms two hydrogen bonds (ϳ2.60 Å) with the NADP ϩ cofactor and the ND1 nitrogen of Thr-77.

Small molecule inhibitors of BLVRB
Favorable and unfavorable interactions between BLVRB and inhibitors were identified based on pair potential analyses (Fig.  2, bottom panel). The phenyl moiety of NSC130813 and NSC12516 is predicted to make similar contributions (e.g. attractive interactions) with BLVRB ( Fig. 2, a and b). In contrast, the acridine moiety of NSC130813 and NSC12516 interacts more favorably through hydrophobic interactions, whereas the same fragment of PH006888 and PH001924 results in less favorable contacts (e.g. repulsive interactions; red lines in Fig. 2c and 2d) and fewer contributions (smaller blue spheres). These differences account for the lower binding affinity of PH006888 and PH001924 predicted from docking calculations ( Table 1). Pair potential analysis predicts that PH006888 is a weak inhibitor of BLVRB due to the large number of unfavorable contacts with the protein.
We assessed the ability of AutoDock (32) to accurately dock phloxine B and erythrosin extra bluish into the ligand-binding site of the BLVRB-NADP ϩ -xanthene-based inhibitor complex The inhibitor (purple) and residues predicted to be involved in hydrogen bond interactions (orange) are shown in ball-and-stick representation. Hydrogen bonds are indicated by green dotted lines, whereas the spoked arcs represent protein residues making non-bonded contacts with the inhibitor. Bottom panel: pair potential analysis results for the six ligands. Blue spheres indicate "good" pair potential per-atom score with the size of the sphere reflecting the strength of the interaction (i.e. the larger the sphere the more attractive the interaction). Blue and red lines represent good and bad contacts with binding site residues, respectively. Residues within the binding site are not shown for the sake of clarity. (Fig. 2, e and f). The r.m.s. deviation for the ligand between the modeled and experimental complex structure (see below) was 0.47 and 1.67 Å (heavy atoms) for phloxine B and erythrosin extra bluish, respectively. Of note, molecular docking studies were performed in the absence of water molecules. Hydrogen bonds between the carbonyl oxygen of phloxine B and erythrosin extra bluish and hydroxyl oxygen O2D from the ribose of NADP ϩ are formed with an r.m.s. deviation of 3.32 and 2.50 Å, respectively. Additionally, our model predicts that these compounds will bind to BLVRB primarily through hydrophobic interactions. Phloxine B is predicted to establish contacts with Phe-113, Trp-116, Arg-124, Leu-125, Pro-152, and His-153 of BLVRB (Fig. 2e, middle panel). The chlorine atom of the tetrachlorobenzoate group of phloxine B is predicted to form a hydrogen bond (3.23 Å) with the nitrogen of the carboxamide moiety of the nicotinamide ring of NADP ϩ . The binding pocket for erythrosin extra bluish is predicted to be composed of Asn-79, Arg-124 Leu-125, Val-128, His-132, and His-153 (Fig. 2f, middle panel). The carboxylate group of the benzoate moiety of erythrosin extra bluish is predicted to form a hydrogen bond (3.33 Å) with the side chain of His-153. As expected, based on their inhibitory activity, both phloxine B and erythrosin extra bluish are predicted to make several favorable interactions with BLVRB ( Fig. 2, e and f, bottom panel).

BLVRB inhibitors display non-competitive inhibition toward FMN
Previous studies conducted to elucidate the inhibitory properties of lumichrome showed that this compound is a competitive inhibitor of FMN (17). To investigate the mode of inhibition by erythrosine extra bluish, phloxine B, and NSC130813 we assayed BLVRB using FMN as the variable substrate in the presence of various concentrations of each inhibitor. All three compounds displayed non-competitive inhibition with inhibition constants (K i ) of 0.70 Ϯ 0.032, 1.8 Ϯ 0.096, and 25.1 Ϯ 2.4 M for erythrosin extra bluish, phloxine B, and NSC130813, respectively (Fig. 3).

Model validation using X-ray crystallography
Model validation for the BLVRB-NADP ϩ -NSC130813 complex was not solvable because of low aqueous solubility of the ligand, although crystal structures of the xanthene-based inhibitors erythrosin extra bluish and phloxine B were determined at 1.33-and 1.20-Å resolution, respectively. The overall structures of phloxine B-bound and erythrosin-bound BLVRB are almost identical to that of ligand-free BLVRB (PDB entry 1HDO (22)), with a r.m.s. deviation of 0.13 and 0.16 Å (204 aligned C␣ atoms) for phloxine B-bound and erythrosin-bound BLVRB, respectively. Electron density for the halogenated xanthene moiety of phloxine B was clear in the molecular replacement maps and allowed the straightforward placement of the ligand (Fig. 4a). The tetrachlorobenzoate group of phloxine B is disordered and could not be modeled in the electron density maps. The strong anomalous difference peaks observed for the BLVRB-erythrosin extra bluish complex allowed pinpointing the positions of the iodine atoms of the ligand and further facilitated its placement in the electron density maps (Fig. 4b). Both substrates are found in the ample binding site of BLVRB, which is adjacent to the nicotinamide moiety of the NADP ϩ cofactor and is bordered by flexible loops 80, 120, and the N-terminal half of helix ␣E (Fig. 4). Both phloxine B and erythrosin extra bluish bind to BLVRB in a way resembling that described for FMN and lumichrome (22), where a single hydrogen bond is

Small molecule inhibitors of BLVRB
formed between a carbonyl oxygen of the ligand and hydroxyl oxygen O2D from the ribose of the NADP ϩ cofactor (Figs. 4 and 5a). This interaction is preserved in phloxine B and erythrosin extra bluish, involving the equivalent oxygen of each ligand.
Similar to FMN and lumichrome, besides the stacking interaction with the nicotinamide moiety of NADP ϩ , several hydrophobic contacts are established between BLVRB and phloxine B (with the side chains of Phe-113, Trp-116, Leu-125, and His-153) or erythrosin (with the main chain of Asn-79 and the side chains of Leu-81, Phe-113, Arg-124, Leu-125, Val-128, and His-153) (Fig. 4, d and e). However, whereas the isoalloxazine ring system of FMN and alloxazine ring system of lumichrome are almost perfectly coincident in their complexes with BLVRB (PDB entries 1HE4 and 1HE5 (22)), the xanthone moieties of phloxine B and erythrosin extra bluish deviate significantly from this position (Fig. 5b). Although the planes of the isoalloxazine and xanthone moieties are coincident, rotation of

Small molecule inhibitors of BLVRB
the latter around an axis perpendicular to the plane of the ring sitting deeper in the active site of BLVRB, originates a deviation of 25 degrees for phloxine B and 45 degrees for erythrosin extra bluish (Fig. 5b). In consequence, phloxine B and erythrosin extra bluish bind less deeply than FMN or lumichrome in the active site of BLVRB, and phloxine O1 and erythrosin O (equivalent to N5 in FMN and lumichrome) are displaced 1.7 or 2.4 Å away from the back wall of the binding pocket, respectively. In consequence, the hydrophobic contact with the side chain of Trp-116, conserved in FMN, lumichrome, and phloxine B, is absent in erythrosin extra bluish.
Contrary to FMN and lumichrome, phloxine B and erythrosin extra bluish establish several additional direct and watermediated polar interactions with BLVRB. Bromine atoms BR1 and BR2 of phloxine B are almost equidistant (4.3 and 4.0 Å, respectively) to the hydroxyl oxygen of Ser-111. The latter bromine establishes a water-mediated interaction with the side chains of Ser-111 and His-132, whereas bromine BR3 interacts with the carbonyl oxygen of Arg-78. In accordance to its seemingly higher affinity for BLVRB, erythrosin extra bluish displays even more interactions with the enzyme. Unlike the disordered tetrachlorobenzoate moiety of phloxine B, the equivalent benzoate moiety of erythrosin extra bluish is well defined, with its carboxylate group establishing a bidentate interaction with the nicotinamide nitrogen of NADP ϩ and the side chain of His-153, in addition to water-mediated contacts with the carbonyl oxygen of Ile-154, the side chain nitrogen of Gln-14, the side chain oxygen of Asn-79, and the NADP ϩ pyrophosphate. Additionally, the four iodine substituents of erythrosin extra bluish engage in direct or solvent-mediated interactions. Iodine atom I establishes a bond with the hydroxyl of Thr-77, as well as water-mediated contacts with the main chain nitrogen of Asn-79, and the pyrophosphate group and the nicotinamide nitrogen of NADP ϩ . Atoms I1 and I2 establish water-mediated bonds with the carbonyl oxygen of Arg-174 and erythrosin oxygen O4, respectively. Finally, I3 is within direct bonding distance of the hydroxyl group of Ser-111, and establishes solventmediated contacts with the side chains of Ser-111, Thr-129, and His-132.

Cellular effects of BLVRB inhibitors
In vitro effects of BLVRB inhibitors were subsequently studied in promyelocytic HL-60 cells, using resazurin reduction as a live probe of BLVRB functional redox activity. HL-60 cells express endogenous BLVRB, although overexpression caused time-dependent enhancement of redox activity compared with mock-transfected controls (Fig. 6, a and b). Exaggerated redox coupling was not due to cellular expansion (Fig. 6c), establish-

Small molecule inhibitors of BLVRB
ing the BLVRB-dependent nature of the resazurin coupling reaction. Incubation of HL-60/BLVRB cells with either xanthene (phloxine B)-or acridine (NSC130813)-based BLVRB inhibitors caused dose-dependent loss of redox activity with no inhibitory effects using the modest BLVRB inhibitor NSC109836 (Fig. 6d, refer to Table S1). The cellular EC 50 of both compounds was comparable (5.2 Ϯ 1 M (NSC130813) and 3.1 Ϯ 1 M (phloxine B)) despite their distinct biochemical properties (Table 1), presumably explained by differential cellular bioavailability or target binding affinities. For phloxine B, we exploited its fluorescence properties to validate a direct relationship between intracellular accumulation and cellular redox activity. As shown, flow cytometric quantification confirmed a clear dose-dependent increase in cellular uptake and intracellular phloxine B accumulation as quantified by the geometric fluorescent mean intensity (Fig. 6e). Phloxine B at a concentration (10 M) resulting in Ͼ99% cellular uptake resulted in ϳ70% cellular redox inhibition, with nearly complete inhibition evident with progressive intracellular accumulation (100 M). These results confirm that biochemical inhibition has selective effects on BLVRB-dependent intracellular redox coupling.

Discussion
We have used both in silico and crystallographic studies to identify and characterize BLVRB small molecule inhibitors retaining potencies nearly ϳ400 times greater than the previously described inhibitor lumichrome (22). The most potent inhibitors contain a tricyclic hydrocarbon core structure that is similar to the isoalloxazine ring of the BLVRB substrate FMN and to the alloxazine ring of lumichrome. Acridine-based compounds were computationally located within the substratebinding site adjacent to the nicotinamide moiety of the NADP ϩ cofactor and predicted to bind primarily through hydrophobic interactions with BLVRB. Attempts to obtain an experimental structure of the BLVRB-NADP ϩ -NSC130813 complex have been hampered by the low aqueous solubility of the ligand, although crystal structures of the xanthene-based inhibitors erythrosin extra bluish and phloxine B were solvable. Structures were nearly identical to ligand-free BLVRB (22), with ligands found within the ample BLVRB-binding site adjacent to the nicotinamide moiety of the cofactor and bordered by the flexible loops 80, 120, and the N-terminal half of the helix ␣E (27). Although erythrosin extra bluish and phloxine B bind less deeply than FMN or lumichrome into the BLVRB active site, both compounds interact with the side chain of Ser-111, a residue previously identified as critical for maintenance of the enzymatic active site (27) and cellular reductase functions that maintain redox balance and antioxidant handling in hematopoietic cells (26). Consistent with what is observed in the crystal structures for phloxine B-and erythrosin-bound BLVRB, the halogen located on the tricyclic hydrocarbon of NSC130813 and NSC12516 is predicted to be in close proximity to Ser-111 (Fig. 2, a and b). Although the precise mechanism(s) by which these compounds inhibit BLVRB remains under investigation, this common protein-ligand interaction highlights critical requirements for enzymatic activity.
Kinetic studies using our top BLVRB inhibitors resulted in double-reciprocal plots that indicated non-competitive inhibi-tion against FMN, although it is clear from crystallographic studies that phloxine B and erythrosin extra bluish occupy the same binding site as the substrate. These findings contrast what was observed for lumichrome, a competitive inhibitor with respect to FMN with a K i of 76 M (17). The discrepancy between our crystallographic and kinetic experiments may indicate a complex mode of inhibition for these inhibitors (e.g. tight binding (33)). Additionally, results reported herein indicate that enzyme-inhibitor interactions between BLVRB and erythrosin extra bluish and phloxine B display Goldstein's zone B behavior (34,35) versus the zone A behavior that is observed for lumichrome.
Enzymatic assays confirm inhibitory effects using two distinct BLVRB substrates, FMN and DCPIP. Because both verdins and flavins bind within the single Rossmann-fold, we characterized inhibition of biliverdin reductase activity using a BV isomeric mixture generated by coupled oxidation from a heme-heme oxygenase (HemO) complex previously known to selectively generate BVIX␤ and BVIX␦ (36), both of which serve as substrates for BLVRB (37). In contrast to what was observed when FMN and DCPIP were used as substrates, we did not observe a significant inhibition of biliverdin reductase activity in the presence of our inhibitors at concentrations suitable for inhibition of flavin reductase and DCPIP reductase activities (10 M phloxine B and erythrosin extra bluish and 200 M NSC130813). This is not unexpected given the known technical limitations arising from the use of tetrapyrrole substrates, which variably demonstrate inhibition (at low concentrations) or activation (at higher doses) of BLVRB's biliverdin reductase activity dependent on the nature of the tetrapyrrole (37,38). Indeed protohemin, a compound that inhibits BLVRB's flavin reductase activity, was shown to activate the enzyme's biliverdin reductase activity (16). Additionally, lumichrome (a wellcharacterized inhibitor of BLVRB's flavin reductase activity) does not inhibit the biliverdin reductase activity of the enzyme (16). These observations highlight the technical difficulties associated with characterizing BLVRB's biliverdin reductase activity, although flavin inhibitory effects remain evident.
The most potent acridine derivative (NSC130813) is predicted to have the greatest affinity and specificity for BLVRB based on our computational models, although previous functions as a topoisomerase inhibitor (39), a ligand for bacterial RNA (40) and inhibitor of ERCC1-XPF protein-protein interactions (41) have been described. This is not unexpected given the known cancer-modulatory characteristics of the starting NCI Diversity Set used for the initial in silico screen. In contrast, erythrosin extra bluish and phloxine B have less well-characterized functions, although effects on protein-protein interactions within the tumor necrosis superfamily have been delineated (42). Interestingly, both phloxine B and erythrosin extra bluish display inhibitor effects on uridine 5Ј-diphospate (UDP) glucuronosyltransferase 1A6 (UGT1A6) (43, 44), the enzyme responsible for bilirubin glucuronidation and hereditary unconjugated hyperbilirubinemic disorders (Gilbert's and Crigler-Najjar syndromes) (45). Detailed biochemical mechanism(s) of this effect are not well established, and direct enzymatic inhibition with structural interactions as outlined here have not been described.

Small molecule inhibitors of BLVRB
Of the two classes of inhibitors identified in this study, the acridine-based compounds possess the most drug-like properties (i.e. molecular weight, octanol-water partition coefficient (logP), hydrogen bond donors/acceptors, etc.) (Fig. S2). Based on a BOILED-Egg analysis of the acridine-based compounds, there is high probability that NSC130813 and NSC12516 will passively cross the blood-brain barrier (BBB) (Fig. S3). In contrast, PH006888 and PH001924 are not predicted to cross the BBB. All four acridine-based compounds are predicted to be highly absorbed by the gastrointestinal tract. Although the drug-like properties of phloxine B and erythrosin extra bluish are inferior to the acridine-based compounds, xanthene-based compounds have been shown to have promising therapeutic indications (46,47).
BLVRB induction (nearly 40-fold) has been identified during terminal erythropoiesis (26), and BLVRB overexpression has been identified in multiple tumor subtypes, including but not limited to cervical cancer, hepatocellular carcinoma, and triple-negative breast cancer (Protein-Atlas, proteinatlas.org/ ENSG00000090013-BLVRB/pathology), 5 furthermore, targeted BLVRB down-regulation suppresses hepatocellular carcinoma cell growth (49). Unlike BLVRA whose pleiotropic cellular effects are mediated by multidomain features (50), the BLVRB structure is generally featureless with a function presumably limited to its substrate/NAD(P)H redox activity (22). Indeed, BLVRB inhibition caused selective loss of redox coupling in vitro, suggesting that these compounds can now be used to address target validity and further refine differences in substrate utilization relevant to various human models linked to hematopoiesis and/or cancer proliferation. Ongoing studies designed to characterize BLVRB-null cells and BLVRB knockout mice will provide further insight into target selectivity and physiological role(s) of this enzyme in development and redoxregulated cellular functions.
The identification of first-generation BLVRB inhibitors may now be applied to better dissect the role of biliverdins, alternative substrates, and BLVRB expression in hematopoietic and cancer subtypes. Indeed, the recent development of phytochrome-optimized infrared probes retaining distinct biliverdin-dependent red shifted fluorescence (59) document that cells accumulate biliverdins. Although there are limited data on BV quantification (and/or isomeric distribution) across cell types, the availability of these and optimized structure-based inhibitory compounds provide a critical framework for further dissection of BLVRB cellular effects.

Experimental procedures
Materials FMN, NADPH, DCPIP, biliverdin, HEPES, PH002274, PH002289, PH006888, PH001924, NSC130813, lumichrome, phloxine B, erythrosin extra bluish, xanthene, proflavine, safranin O, and Alizarin Red S were purchased from Sigma. Cayman Chemical Company and Santa Cruz Biotechnology served as secondary sources for phloxine B and erythrosin extra bluish, respectively. NSC compounds were obtained from the National Cancer Institute Developmental Therapeutic Program Repository. Commercially available ZINC compounds were purchased from ChemDiv with the exception of ZINC4366439 (NSC12516), which was obtained from the Developmental Therapeutic Program Repository.

Protein expression and purification
Recombinant BLVRB was overexpressed and purified as a glutathione S-transferase (GST) fusion protein in BL21(DE3) cells as previously described (51). Briefly, protein expression was induced using 0.1 mM isopropyl ␤-D-1-thiogalactopyranoside and allowed to proceed for 3 h at 37°C. GST-tagged BLVRB was purified by affinity chromatography using glutathione-Sepharose 4B, and the fusion protein was cleaved overnight at 4°C with 1 nM thrombin to remove the GST tag, and purified BLVRB lacking the GST tag was eluted from the column with phosphate-buffered saline (PBS). BLVRB concentration was determined using the bicinchoninic acid assay (52) with BSA as the standard.

Enzymatic activity assays
Inhibition assays were performed using a Synergy H1 hybrid microplate reader (BioTek). Flavin reductase activity was determined spectrophotometrically by monitoring the rate of oxidation of NADPH to NADP ϩ at 340 nm (⑀ ϭ 6200 M Ϫ1 cm Ϫ1 ). Enzymatic activity was measured at 25°C in 100 mM HEPES, pH 7.0, containing 200 M FMN, 100 M NADPH, and 200 M inhibitor (except where noted). Dye-based inhibitors were assayed at a final concentration of 10 M. Reactions were initiated by the addition of 450 nM enzyme. Assays lacking inhibitor (solvent only) and enzyme served as positive and negative controls, respectively. Secondary inhibition assays were run using DCPIP (50 M) as the substrate in a reaction containing 100 mM HEPES, pH 7.0, and 100 M NADPH at 25°C. Reactions were initiated by the addition of 100 nM BLVRB and reductase activity was measured by following the decrease in absorbance at 600 nm corresponding to the reduction of DCPIP (⑀ ϭ 20.7 mM Ϫ1 cm Ϫ1 ) (53).
The concentration of inhibitor required to inhibit enzymatic activity by 50% (IC 50 ) was determined by adding varying concentrations of inhibitor to reaction mixtures at fixed concentrations of substrate FMN (200 M) or DCPIP (50 M) and NADPH (100 M). The total amount of organic solvent added was kept constant at a final concentration of 2% (v/v). IC 50 values were determined by non-liner regression fits of data to Equation 1 using GraFit 4.0 (Erithacus Software), where y is the percent activity, [I] is the concentration of inhibitor, and s is the slope factor. Controls lacking enzyme were run at each concentration of inhibitor to determine any background caused by the inhibitor. The mode of inhibition was determined by varying the concentration of inhibitor at fixed concentrations of FMN (25,50,5 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party hosted site.

Small molecule inhibitors of BLVRB
100, 150, and 200 M) and a fixed concentration of 100 M NADPH in 100 mM HEPES, pH 7.0, at 25°C. Reactions were initiated by the addition of 360 nM BLVRB with the exception 600 nM BLVRB being used for NSC130813 experiments. Lineweaver-Burk plots were generated from transformed data using the program SigmaPlot 12.5 (Systat Software Inc.) and inhibition constants were determined by fitting data to standard equations for competitive and non-competitive inhibition (Equations 2 and 3, respectively), where V max is the maximal velocity, [S] is the substrate concentration, K M is the Michaelis-Menten constant for the substrate, [I] is the concentration of inhibitor added to the reaction, and K i is the inhibition constant. BLVRA assays were performed in the scanning kinetics mode (800 -320 nm) using a Cary 60 UV-visible spectrophotometer. Reactions were run in 50 mM Tris-HCl, pH 8.7, containing 100 M NADPH and 10 M biliverdin at 37°C. Reactions were initiated by the addition of 50 ng of enzyme (Fitzgerald, 30R-137) and enzymatic activity was monitored following the conversion of biliverdin (⑀ 650nm ϭ 14.3 mM Ϫ1 cm Ϫ1 ) (37) to bilirubin (⑀ 450nm ϭ 53 mM Ϫ1 cm Ϫ1 ) (37). BLVRB biliverdin reductase assays were performed as described for BLVRA with the following modifications: reactions were run in 100 mM HEPES, pH 7.0, containing 100 M NADPH, 37 M bovine serum albumin, and 10 M biliverdin (IX␦/IX␤). Reactions were initiated by the addition of 200 nM BLVRB. Reductase activity was monitored following the conversion of biliverdin (⑀ 650nm ϭ 15.4 mM Ϫ1 cm Ϫ1 ) (37) to bilirubin (⑀ 450nm ϭ 20.5 mM Ϫ1 cm Ϫ1 ) (37).

Ligand docking validation
All docking experiments were carried out using AutoDock 4.26 (54) unless otherwise noted. Molecular docking was carried out to predict binding modes of compounds to BLVRB. Docking experiments were validated by docking lumichrome into the ligand-binding site of the BLVRB-NADP ϩ -lumichrome X-ray crystal structure (PDB entry 1HE5 (22)). A r.m.s. deviation of Յ2 Å of docked poses relative to the X-ray crystal structure ligand pose was used as the threshold for determining docking success. Calculations obtained from docking lumichrome into the ligand-binding pocket of BLVRB were used to determine the docking parameters for inhibitors.

Two-dimensional virtual screening
A 2D-VS was performed by targeting the ligand-binding site of BLVRB against the Developmental Therapeutics Program NCI Diversity Set IV database. NADP ϩ as a cofactor was present within the binding site (other ligands and water were deleted) and side chains of residues and termini protons were assigned according to their corresponding protonation state at pH 7.0. Molecular docking calculations were run to explore the conformation space of ligands for further ISR calculations. ISR was calculated using Equation 4, where ␦E is the energy gap, ⌬E is the energy roughness, and the size of the binding funnel measured by the entropy S. Docking simulations were carried out using a grid size of 80 ϫ 80 ϫ 80 and spacing of 0.375 Å, with respect to the binding pocket of BLVRB, and grid maps for scoring were created using AutoGrid. Dockings (10,000 runs) were performed using the AutoDock score function for evaluation and the Lamarckian Genetic Algorithm (LGA) for sampling poses. Molecular modeling was carried out using default parameters with the following exceptions: population size of 100 and the energy evaluation of 100,000 per run with the maximum number of generations of 27,000. Docking results were evaluated according to predicted binding affinities and the subsequent cluster analysis was based on a r.m.s. deviation cutoff of 2.0 Å.

Similar compound library preparation
Compounds structurally similar to the hit compound NSC130813 were identified using five complementary approaches. The first approach employed a similarity search against the ZINC database (55) using a Tanimoto score of 0.8 as the threshold for compound selection. The 123 compounds identified from this screen were further filtered by 3D atom pair fingerprint (3D-APfp) (56) with a Max Count set to 50 to identify compounds with 3D molecular shape similar to NSC130813 and the top 50 hits were selected as potential inhibitors of BLVRB. Second, a LigCSRre similarity search (57) was run using NSC130813 as the query to identify compounds within the program's 13 ligand databases that have maximum common substructure similarity to NSC130813 based on the CSR algorithm. This search resulted in the identification of 400 potential BLVRB inhibitors with a minimal r.m.s. deviation of 1.0 Å. Third, a SHApe-FeaTure Similarity (SHAFTS) search (58) was run against the ZINC database (30). This hybrid 3D (rigid superposition) similarity calculation method integrates pharmacophore match and shape similarity. Using a similarity score of 0.8, 15,900 compounds were identified as potential inhibitors of BLVRB. The fourth approach, Multi-Fingerprint browser for ZINC (MCSS) (59), was utilized to rapidly identify close analogs of NSC130813 in the ZINC database (30). The use of MCSS' four fingerprints, sFP (substructure fingerprint), ECFP4 (extended connectivity fingerprint), MQN (Molecular Quantum Numbers), and SMIfp (SMILES fingerprint), allowed for the identification of 2,477 compounds that were determined to be nearest neighbors of NSC130813 based on city-block distance. The Max Count was set to 500 for this screen. Last, a pharmacophore similarity search against the ZINC database (30) using ZINCPharmer (60) was performed resulting in 932 similar compounds that were further filtered by 3D extended atom pair fingerprint (3D-Xfp) (56), using a minimal r.m.s. deviation of 1.0 Å, to identify the top 50 compounds.

Analysis of inhibitor binding mode
Interactions between BLVRB and inhibitors were initially established based on AutoDock scores. Docking modes for top hits were further evaluated using knowledge-based DSX pair potentials derived from the DrugScore (61, 62) available from DSX online (63). Analyses for binding modes obtained from molecular docking and the detailed contribution (per-atom scores) of each interaction were displayed in PyMOL. Twodimensional schematic representations of predicted interactions between inhibitors and residues within the binding site of BLVRB were generated using Ligplotϩ (64).

Drug-likeness filtering and ADME prediction
Potential inhibitors of BLVRB were filtered to determine their absorption, distribution, metabolism, excretion, and toxicity (ADME/Tox) and drug-likeness properties using FAF-Drugs3 (65) and Molsoft 5 program, respectively. Detailed ADME/Tox profiles of validated hit compounds were evaluated using SwissADME (66). This program allowed for in silico prediction of key ADME parameters including gastrointestinal absorption, BBB permeation, cytochrome P450 inhibition, and whether these hits served as substrates for P-glycoprotein. Ligand efficiencies (LE) for top hits were determined using Equation 5, where IC 50 is the concentration of inhibitor required to inhibit the enzyme by 50% and N is the number of non-hydrogen bonds. Compounds within the similar compounds library that were drug-like and had acceptable (predicted) ADME/Tox properties were subjected to two-dimensional virtual screening as described above.

Oral toxicity analysis
Rodent oral toxicities of hit compounds were predicted in silico using ProTox (67). The median oral lethal dose (LD 50 ) of hit compounds was determined based on known toxicities of structurally related compounds.

Crystallization of BLVRB
Recombinant human BLVRB was crystallized as previously described (22). The phloxine B and erythrosin extra bluish complexes were prepared by soaking the BLVRB-NADP crystals in mother liquor containing 50 mM ligand for 24 h at 20°C. The crystals of the resulting ternary complexes were cryoprotected by brief (ϳ5 s) sequential transfers to mother liquor supplemented with increasing concentrations (up to 15% (v/v)) of glycerol, plunged in liquid nitrogen, and stored under cryogenic conditions until measured.

Data collection and processing
X-ray diffraction data were collected from single, cryocooled (100 K) crystals at beamline ID23-1 (68) of the European Synchrotron Radiation Facility (ESRF, Grenoble, France). For each complex, a low-resolution pass with 1 degree rotation and 0.1-s exposure per image, and a high-resolution pass with 0.5 degree rotation and 0.1-s exposure per image were collected, using radiation of 0.9763-Å wavelength. The low-resolution pass was composed of 140 or 200 images and the high-resolution pass comprised 300 or 400 images, for the phloxine B or the erythrosin extra bluish complex crystal, respectively.

Structure determination and refinement
The structure of ternary BLVRB complexes were solved by molecular replacement with PHASER (73) using the coordinates of BLVRB (PDB entry 1HDO (22)) stripped from ligands and solvent as search model. To reduce model bias, the protein moiety was automatically rebuilt using ARP/wARP (74) and the resulting models were subjected to alternating cycles of manual building with COOT (75) and refinement with PHENIX (76) until model completion. The refinement statistics are summarized in Table S5. All crystallographic software was supported by SBGrid (77). The refined three-dimensional coordinates and structure factors were deposited at the Protein Data Bank with accession codes 5OOG (phloxine B) and 5OOH (erythrosin extra bluish).

Analysis of atomic structures
Superposition of atomic structures was performed with the program Theseus (78). The molecular structures shown in Figs. 4 and 5 were depicted with PyMOL, version 1.6.0 (Schrödinger).

Cell-based assays
Promyelocytic HL-60 cells (ATCC CCL-240) were maintained as suspension cultures in Iscove's modified Dulbecco's medium supplemented with 20% FBS. Stable BLVRB-expressing cell lines HL-60/BLVRB were generated by lentivirus (Lv) infection (or empty control) using multiplicity of infection of 5, followed by puromycin (2 g/ml) selection as previously described (26); Lv/BLVRB was expressed with chimeric C-terminal His 6 and V5 tags (79) to distinguish endogenous from exogenous proteins. Uninfected cells (i.e. virus-deficient) supplemented with puromcycin demonstrated complete cell death by 24 h. Cellular oxidation/reduction activity was quantified in vitro using the NADPH-dependent redox coupler resazurin; cells (1 ϫ 10 4 /well) were plated and propagated in complete medium with various concentrations (or not) of inhibitor compounds, followed by addition of resazurin (0.1% (v/v)) for fluorimetric detection of reduced resorufin (Ex 530 nm /Em 590 nm ); cellular counts were quantified in parallel using trypan blue exclusion on an automated cell counter (Countess, Thermo Fisher Scientific).
Intracellular accumulation of phloxine B was quantified using a BD LSR Fortessa flow cytometer (BD Biosciences) as previously described (48,80,81). HL-60 cells (1 ϫ 10 5 /well) were plated overnight, followed by incubation with various concentrations (or not) of phloxine B for 24 h, at which point Small molecule inhibitors of BLVRB cells were harvested, washed twice with PBS, and passed through a cell strainer for flow cytometric analyses. Unstained cells were used for definition of background gates, followed by data acquisition of 20,000 gated events using logarithmic gain settings for light scatter and phloxine B (Ex 540 nm /Em 564 nm ) fluorescence detection (26,81). Flow cytometric data were exported and analyzed using Kaluza software (Beckman Coulter).

Statistics
All statistical comparisons were established using unpaired t test, defining p Ͻ 0.05 for statistical significance.