Bilirubin links heme metabolism to neuroprotection by scavenging superoxide

SUMMARY

Bilirubin is one of the most frequently measured metabolites in medicine, and yet its physiologic roles remain unclear. Bilirubin can act as an antioxidant in vitro, but whether its redox activity is physiologically-relevant is unclear because many other antioxidants are far more abundant in vivo. Here, we report that depleting endogenous bilirubin renders mice hypersensitive to oxidative stress. We find that mice lacking bilirubin are particularly vulnerable to superoxide (O₂^•−) over other reactive oxidants and electrophiles. Whereas major antioxidants such as glutathione and cysteine exhibit little to no reactivity towards O₂^•−, bilirubin readily scavenges O₂^•−. We find that bilirubin’s redox activity is particularly important in the brain, where it prevents excitotoxicity and neuronal death by scavenging O₂^•− during NMDA neurotransmission. Bilirubin’s unique redox activity towards O₂^•− may underlie a prominent physiologic role despite being significantly less abundant than other endogenous and exogenous antioxidants.

eTOC

Bilirubin is routinely measured in clinical blood tests, and yet its physiologic functions remain unclear. Here, Vasavda et al report that bilirubin is a physiologic antioxidant with distinct chemical activity towards superoxide, a reactive oxygen anion. They find that bilirubin is neuroprotective, in part by regulating superoxide signaling during neurotransmission.

Graphical Abstract

INTRODUCTION

Bilirubin is one of the most frequently measured metabolites in clinical medicine, used to diagnose conditions such as hepatobiliary disorders (Osborn et al., 1959), hemolytic anemia (Giblett et al., 1956), and dyserythropoiesis (Berendsohn et al., 1964). It is the terminal product of mammalian heme metabolism, in which heme oxygenase (HO) oxidizes heme to biliverdin, which biliverdin reductase (BVR) then reduces to bilirubin (Kutty and Maines, 1981) (Fig. 1a). Despite being regularly measured in patients, bilirubin is still largely considered a surrogate biomarker with little function. Bilirubin is also neurotoxic at high micromolar concentrations (Claireaux et al., 1953). As bilirubin exhibits toxicity and has no obvious function, it has remained puzzling why diverse organisms evolved and retained the ability to produce it. However, bilirubin synthesis is dynamic, inducible, and tightly regulated, suggesting it may serve important physiologic functions. For example, bilirubin synthesis exhibits circadian periodicity (Kaasik and Lee, 2004) and is both temporally and spatially controlled during development (Falchuk et al., 2002). Bilirubin synthesis also rapidly increases during oxidative stress (Applegate et al., 1991), but why heme metabolism responds in this way is not completely clear.

Figure 1. BVR^−/− mice lack bilirubin and accumulate biliverdin.

a, Schematic of heme catabolism. b, Strategy for deleting exon 3 of mus musculus Blvra gene. c-d, PCR from DNA for exon 3 (c) and RNA for full-length Blvra (d) from WT and BVR^−/− mice. e, Photographs WT and BVR^−/− mouse gallbladders. f-h, HPLC chromatograms of (f) 10 M biliverdin, bilirubin, and hemin standards (λ=405 nm), (g) bile pooled from 6 WT and BVR^−/− mice (λ=405 nm), and (h) plasma from WT and BVR^−/− mice (λ=450 nm). i, Blvra RNA levels in different regions of the brain measured by quantitative PCR. Points represent individual mice. j, Bilirubin immunostaining in cerebellar sections from WT and BVR^−/− mice. Scale bar = 100 μm. k, Coomassie blue gel of purified UnaG. l, Normalized fluorescence intensities of 1 μM apoUnaG incubated with various heme metabolites. m-p, Live-cell images of neurons co-expressing UnaG (m) or IRFP (o) and mCherry. Scale bar = 50 μm. Mean UnaG fluorescence (n) and IRFP fluorescence (p) normalized to mCherry fluorescence as a transfection control. Points represent individual neurons. n = 9–12 neurons per genotype per co-transfection. (i, l, n, p) Mean ± SEM depicted. ** = P < 0.01 by two-way ANOVA followed by a post hoc Tukey test (i) or two-tailed Student’s t-test (n, p).

Some evidence suggests that bilirubin has antioxidant activity. However, whether bilirubin is a physiologic antioxidant is still a point of contention. Earlier conclusions were based on in vitro data (Stocker et al., 1987), experiments with exogenous bilirubin (Baranano et al., 2002; Doré and Snyder, 1999), and correlative findings from hyperbilirubinemic rats (Dennery et al., 1995). Whether bilirubin is a physiologically-relevant antioxidant is also disputed because many other antioxidants are far more abundant in vivo, some present at 10,000-fold higher concentrations (Maghzal et al., 2009; McDonagh, 2010). Despite this, several epidemiological studies suggest that mildly elevated serum bilirubin lowers the risk for ischemic stroke and coronary artery disease (Djousse et al., 2001; Perlstein et al., 2008; Schwertner et al., 1994). The incidence of coronary artery disease in individuals with Gilbert’s Syndrome, a genetic hyperbilirubinemia, is 5-fold lower than the general population (Vitek et al., 2002). While these studies suggest that bilirubin has functional activity, they are correlative and do not concretely identify bilirubin’s physiologic roles. To directly evaluate whether bilirubin is a physiologic antioxidant, we generated a genetic mouse model that selectively and globally lacks endogenous bilirubin.

RESULTS

Depleting endogenous bilirubin renders cells vulnerable to oxidative death

Early work approached studying bilirubin by deleting the enzyme HO-2 (HO-2^−/−) (Doré et al., 2000; 1999), one of the two HO isoenzymes (Maines et al., 1986). However, HO-2^−/− mice are confounded models for studying the redox biology of bilirubin, because they also accumulate heme and free iron (Dennery et al., 1998), both oxidants (Tappel, 1955). Here, we have generated a genetic mouse model that selectively lacks endogenous bilirubin without perturbing heme levels. To eliminate bilirubin, we targeted exon 3 of Blvra, thereby deleting BVR’s obligate NAD(P)H-binding domain (Kikuchi et al., 2001) and ensuring that any splice variant cannot produce bilirubin (Fig. 1b). We confirmed loss of Blvra DNA and mRNA by PCR (Fig. 1c–d) and BVR protein by western blot (Fig. S1a). BVR^−/− tissues do not produce bilirubin even when supplemented with biliverdin, demonstrating a general inability to generate bilirubin (Fig. S1b–c). HPLC analyses of bile from WT and BVR^−/− gallbladders (Fig. 1e–g) indicate a loss of endogenous bilirubin and an accumulation of biliverdin in BVR^−/− mice. Similarly, HPLC analyses of plasma (Fig. 1f and 1h and Fig. S1d–f) indicate loss of endogenous bilirubin. Feces from BVR^−/− mice are still pigmented despite lacking bilirubin, presumably because biliverdin is still readily excreted in bile (Gåfvels et al., 2009; Nytofte et al., 2011). BVR^−/− mice exhibit unaltered plasma heme (Fig. 1f and Fig. S1e and S1g) and HO-2 expression (Fig. S1h–m), establishing BVR^−/− mice as useful tools to specifically assess the redox biology of bilirubin without confounding effects from heme and iron.

To assess bilirubin function in vivo, we first examined BVR expression in WT mice (Fig. S1n–o). BVR protein is variably expressed, with lowest expression in the heart and highest expression in the brain, spinal cord, spleen, and kidney. Within the brain, BVR mRNA is enriched in the cerebellum, double that of the cerebral cortex and almost five times higher than the corpus striatum (Fig. 1i). To determine where bilirubin might function in the brain, we assessed its localization immunohistochemically. Bilirubin exhibits discrete enrichment in neurons, with striking localization to Purkinje and some granule cells in the cerebellum (Fig. 1j). We also assessed the neuronal distribution of bilirubin with UnaG (Fig. 1k), a protein that fluoresces upon binding bilirubin (Kumagai et al., 2013). UnaG is highly specific for bilirubin with an EC₅₀ of 15 ± 1 nM and does not fluoresce with heme or biliverdin (Fig. 1l). UnaG fluoresces when expressed in WT neurons but not BVR^−/− neurons (Fig. 1m–n), suggesting that WT neurons autonomously produce bilirubin. In an analogous approach, we also expressed a phytochrome-based near-infrared fluorescent protein (IRFP) that fluoresces upon binding biliverdin (Filonov et al., 2011). BVR^−/− neurons exhibit over 3 times greater IRFP fluorescence than WT neurons, consistent with an accumulation of biliverdin due to loss of BVR (Fig. 1o–p).

BVR^−/− cells are specifically hypersensitive to O₂^•−

To assess whether removing endogenous bilirubin renders cells sensitive to oxidative stress, we challenged fibroblasts (MEFs) from WT and BVR^−/− littermate mice with various oxidants and electrophiles. BVR^−/− MEFs are highly vulnerable to oxidative death (Fig. 2a) and are particularly sensitive to the superoxide (O₂^•−₂) donors pyrogallol, paraquat, and menadione. BVR^−/− MEFs are comparatively mildly sensitive to the oxidant hydrogen peroxide (H₂O₂) and the electrophile 4-hydroxynonenal (4-HNE) (Table S1). BVR^−/− MEFs are also unable to clear reactive oxygen species (ROS) as efficiently as WT MEFs after exposure to menadione, as measured by the fluorescent redox probe dihydroethidium (DHE) (Fig. 2b). Virally reintroducing BLVRA into the genome of BVR^−/− MEFs reverts their vulnerability to pyrogallol (Fig. S2a), indicating that BVR^−/− cells’ sensitivity to oxidants is due to loss of BVR. Because BVR^−/− cells also accumulate biliverdin, we tested whether excess biliverdin biases cells towards oxidative stress. However, biliverdin does not alter WT cells’ susceptibility to oxidative stress (Fig. S2b), suggesting that biliverdin itself is not toxic.

Figure 2. BVR^−/− cells are hypersensitive to exogenous and endogenous O2•− over other ROS and electrophiles.

a, WT and BVR^−/− MEF viability after 8 h exposure to varying concentrations of oxidants/electrophiles. n = 2–4 in triplicate. b, Live-cell images of WT and BVR^−/− MEFs loaded with dihydroethidium (DHE) after exposure to 25 μM menadione for 1 h. Scale bar = 100 μm. c, Representative HPLC chromatograms of DHE alone, DHE + KO₂, authentic ethidium (E⁺), and DHE + KO₂ spiked with authentic E⁺. d-f, Representative normalized HPLC chromatograms (d) and quantification of E⁺ and 2-OH-E⁺ (e-f) of organic extracts from WT and BVR^−/− MEFs loaded with DHE at baseline or after 1 h treatment with 25 μM menadione with or without bilirubin pre-treatment. g-h, Live-cell images (g) and quantification (h) of MitoSOX Red fluorescence WT and BVR^−/− MEFs after 4 h exposure to 25 μM rotenone with or without bilirubin pre-treatment. Scale bar = 25 μm. (a, e, f, h) Mean ± SEM depicted. * = P < 0.05 and ** = P < 0.01 by two-way ANOVA followed by a post hoc Tukey test (e, f, h).

BVR is a remarkably pleiotropic protein. Apart from reducing biliverdin to bilirubin, BVR functions as an enzyme-linked receptor at the plasma membrane (Wegiel et al., 2009) and as a kinase (Lerner-Marmarosh et al., 2008; 2005). These alternative functions lead to AKT and MAPK signaling (Hinds et al., 2016; Lerner-Marmarosh et al., 2008; Wegiel et al., 2009), both of which promote cell survival during oxidative stress. However, we observed no difference in the expression, phosphorylation, or nuclear translocation of AKT or ERK1/2 between WT and BVR^−/− MEFs (Fig. S2c–d), persuading us to focus on BVR’s canonical enzymatic activity instead.

BVR^−/− cells are markedly more sensitive to the three different O₂^•− cyclers over other oxidants and electrophiles (Fig. 2a), suggesting that bilirubin may physiologically interact with O₂^•−. Regulating O₂^•− is critical, because it is the immediate product of reduced O₂ and is the precursor to several other ROS. Despite its name, O₂^•− is not particularly “super” reactive: its doublet electronic configuration results in a kinetic spin barrier, whereas its anionic character limits reactivity with other electron-rich centers (Bielski and Richter, 1977; Bielski et al., 1985; Fee, 1982; Sawyer and Valentine, 2002). Accordingly, O₂^•− is not readily scavenged by the major endogenous small molecule antioxidants in humans (such as glutathione, cysteine, uric acid, melatonin, and lipoic acid) at physiologically-relevant rates (Jones et al., 2002; Kuzkaya et al., 2005; Winterbourn and Metodiewa, 1999). Superoxide dismutase enzymes (SODs) scavenge O₂^•− (McCord and Fridovich, 1969), but they are restricted to soluble cellular compartments. However, O₂^•− is often produced within aprotic intramembranous compartments, which are more accessible to lipophilic molecules like bilirubin than to SODs. SODs are also imperfect antioxidants because they also convert O₂^•− to H₂O₂, another oxidant.

We leveraged DHE to measure whether BVR^−/− cells accumulate O ^•−₂, as DHE reacts with O ^•−₂ to form the fluorescent product 2-hydroxyethidium (2-OH-E⁺). However, DHE also reacts with other oxidants to form ethidium (E⁺). Both 2-OH-E⁺ and E⁺ exhibit similar fluorescence spectra and thus must be separated by HPLC in order to specifically quantify intracellular O₂^•− (Zhao et al., 2005; 2003). Using the method established by Zhao and colleagues to resolve 2-OH-E⁺ and E⁺ (Fig. 2c), we measured formation of 2-OH-E⁺ and E⁺ within WT and BVR^−/− cells at baseline and after exposure to menadione (Fig. 2d–f). As expected, treating both WT and BVR^−/− cells with menadione results in more 2-OH-E⁺. However, BVR^−/− extracts have almost two-fold more 2-OH-E⁺ than WT extracts, suggesting that removing endogenous bilirubin leads to an accumulation of intracellular O₂^•−. Conversely, loading WT and BVR^−/− cells with exogenous bilirubin reduces 2-OH-E⁺ formation by approximately one-third. To confirm that exogenous bilirubin can enter cells, we tested whether applying bilirubin to bilirubin-null neurons (i.e., BVR^−/− neurons) can trigger UnaG fluorescence. UnaG fluorescence increases within seconds of applying bilirubin to BVR^−/− neurons (Fig. S3a–b), evidence that extracellular bilirubin can enter cells.

Although pyrogallol, menadione, and paraquat auto-oxidize to produce O₂^•−, they are only crude stereotypes of pathologic or physiologic O₂^•−. O₂^•− is generated by various processes in vivo, but most commonly forms in mitochondria when electrons leak from Complexes I – III of the electron transport chain to O₂ (Miwa et al., 2003; Murphy, 2009). O₂^•− is often retained in the matrix or intramembranous spaces because its negative charge prevents passive diffusion outward. We examined whether bilirubin localizes to mitochondria by heterologously expressing mitochondrial-targeted UnaG (mitoUnaG) (Kumagai et al., 2013; Park et al., 2016). MitoUnaG fluoresces and localizes with mitochondrial-targeted blue fluorescent protein (mitoBFP) (Fig. S3c), indicating that endogenous bilirubin localizes to mitochondria as demonstrated previously (Park et al., 2016).

To determine if bilirubin scavenges physiologic mitochondrial O₂^•−, we treated WT and BVR^−/− MEFs with rotenone and measured O₂^•− with the mitochondrial O₂^•− probe MitoSOX Red. Rotenone inhibits Complex I of the electron transport chain, thereby driving electron transfer to O₂ to form O₂^•− (Oberg, 1961). BVR^−/− MEFs exhibit more than double the MitoSOX Red fluorescence of WT MEFs after exposure to rotenone (Fig. 2g–h) despite expressing similar levels of Sod1, Sod2, Sod3 and respiratory chain complex proteins (Fig. S3g–i). Pre-treating both WT and BVR^−/− MEFs with bilirubin reduces MitoSOX Red fluorescence (Fig. 2g–h). In the brain, mitochondrial O₂^•− is a major driver of neuronal death in instances like ischemic stroke (Chouchani et al., 2014; Kontos et al., 1992). To determine whether bilirubin counteracts mitochondrial stress in the brain, we stereotactically injected rotenone into the cerebellum of WT and BVR^−/− mice. We injected vehicle contralaterally to control for surgical injury. BVR^−/− mice exhibit substantially larger lesions after exposure to rotenone, suggesting that bilirubin may protect against mitochondrial oxidative stress in the brain (Fig. S3d–f).

Bilirubin scavenges O₂^•−

To determine whether bilirubin directly scavenges O₂^•−, we added one equivalent of bilirubin to O₂^•− in an aprotic solvent and monitored the spectroscopic signal of the unpaired electron on O₂^•− by electron paramagnetic resonance (EPR). At 100 K, O₂^•− alone exhibits its characteristic EPR spectrum (Bagchi et al., 1989). Adding bilirubin immediately results in loss of EPR signal (Fig. 3a), suggesting that bilirubin reacts directly with O₂^•−. We assessed whether other heme metabolites, biliverdin and ditaurated bilirubin, can scavenge O₂^•−. Using the spin trap 5,5-dimethyl-1-pyrroline N-oxide (DMPO) at 298 K, we observe that ditaurated bilirubin is almost as effective as bilirubin in scavenging O₂^•− (Fig. 3b). Biliverdin also scavenges O₂^•−, but more slowly than bilirubin. In contrast, glutathione exhibits no effect even at 100-fold excess.

Figure 3. Bilirubin selectively scavenges O2•− at physiologically-relevant rates.

a, EPR spectrum KO₂ without (blue) and with (green) equimolar bilirubin in 0.2 M 18-crown-6 in DMSO at 100 K. b, EPR spectra of DMPO-spin adducts of radicals from solutions of KO₂ with or without 1 equivalent of bilirubin, bilirubin ditaurate, or biliverdin or 100 equivalents of glutathione at 298 K. c, Change in pmoles of bilirubin per second with xanthine/xanthine oxidase (XO) with or without excess 100 nM MnSOD or 1 μM catalase. d, Normalized change in UnaG fluorescence from 1 μM bilirubin and 1 μM apoUnaG with either 1 mM H₂O₂ (red), 100 μM pyrogallol (green), or 100 μM pyrogallol and 100 nM MnSOD (orange). e, Representative HPLC chromatograms of bilirubin or bilirubin + KO₂ with the detector set to (left) 440 nm or (right) 320 nm. f, UV-visible difference spectrum of bilirubin subtracted from bilirubin + KO₂. g, Average fold change in chemiluminescence upon mixing 5 μM bilirubin with varying equivalents of KO₂. h, Average competitive kinetic inhibition curves for bilirubin, glutathione, cysteine, and ascorbic acid against p-nitro blue tetrazolium (NBT) and XO-derived O₂^•−. i, LUMO densities (|LUMO|) and change in Gibbs free energy (ΔG) for reducing indicated metabolites as calculated by the B3LYP functional and 6–31+G* diffuse basis set. Blue illustrates greater LUMO density, whereas red indicates lower density. (c, g, h) Mean ± SEM depicted. ** = P < 0.01 and ns = P > 0.05 by two-way ANOVA followed by a post hoc Tukey test (c).

Mitochondrial Complexes II and III produce O₂^•− at membranous sites (Guo and Lemire, 2003; Husen and Solov’yov, 2016), compartments more accessible to bilirubin than SODs. Bilirubin’s reactivity towards O₂^•− in aprotic solvents suggests that it may function as an important membranous O₂^•− scavenger. To test whether bilirubin can scavenge O₂^•− in cytosolic compartments as well, we assessed whether bilirubin can scavenge O₂^•− produced by xanthine oxidase in aqueous buffer. Bilirubin reacts when incubated with xanthine oxidase (Fig. 3c), but could be reacting here with either O₂^•− or H₂O₂, since O₂^•− rapidly disproportionates to H₂O₂ in aqueous solvents. However, adding catalase to remove H₂O₂ does not prevent bilirubin from reacting whereas excess SOD does (Fig. 3c). Similarly, incubating UnaG-bilirubin complexes with pyrogallol quenches UnaG fluorescence whereas H₂O₂ has no effect (Fig. 3d). This suggests that bilirubin reacts directly with O₂^•− but is inert towards H₂O₂, consistent with BVR^−/− cells heightened vulnerability to O₂^•− over H₂O₂.

To more fully understand how bilirubin reacts with O₂^•−, we sought to better characterize the product and mechanism of the reaction. We find that the product elutes earlier than bilirubin on a reverse-phase column, indicating that it is likely more polar than bilirubin (Fig. 3e). The product also exhibits a red-shifted UV-visible spectrum with a local λ_max at 545 nm (Fig. 3f), which suggests expansion of bilirubin’s conjugate π network. Formation of a red-shifted electronic network could proceed through several mechanisms, the simplest being a single electron transfer (SET) from O₂^•− to bilirubin. SET reactions between O₂^•− and heterocyclic compounds, such as lucigenin (Legg and Hercules, 1969) and N-methylacridone (Rosenthal and Bercovici, 1973), often result in weak chemiluminescence. As with these heterocyclics, incubating O₂^•− with bilirubin results in chemiluminescence (Fig. 3g). However, the exact molecular identity of the product remains unclear and would require additional electrochemical and spectroscopic data.

While bilirubin appears to scavenge O₂^•−, we wondered whether the reaction is rapid enough to be physiologically relevant. To determine the rate of the reaction, we employed a steady-state competitive kinetic analysis against p-nitro blue tetrazolium (Fig. 3h), an established O₂^•− scavenger (Bielski and Richter, 1977). We determined that bilirubin scavenges O₂^•− with a second-order rate constant of (4.6 ± 0.8) × 10⁵ M⁻¹ s⁻¹, a rate similar to other physiologic rate constants like those of urease (4.0 × 10⁵ M⁻¹ s⁻¹) and chymotrypsin (2.9 × 10⁵ M⁻¹ s⁻¹) (Voet et al., 2016). Two abundant antioxidants in humans, glutathione and cysteine, exhibit little to no reactivity towards O₂^•− (Table S2). Bilirubin scavenges O₂^•− faster than biliverdin, highlighting a potential evolutionary and functional distinction between the two metabolites (Fig. 3h). Although the second-order rate constant for the reaction between bilirubin and O₂^•− is larger than the other antioxidants we evaluated, O₂^•− also reacts with nitric oxide and tyrosine radicals at near-diffusion controlled rates (k > 10⁹ M⁻¹ s⁻¹) (d’Alessandro et al., 2000; Huie and Padmaja, 1993). O₂^•− reacts with nitric oxide and tyrosine radicals to form peroxynitrite and tyrosine hydroperoxides, respectively, both of which contribute to O₂^•− toxicity (Nagy et al., 2009; Winterbourn et al., 2004). It is unlikely that bilirubin outcompetes nitric oxide or tyrosine radicals for O₂^•−, which suggests that bilirubin instead prevents other modes of O₂^•− toxicity.

To identify kinetic and thermodynamic factors that influence bilirubin reactivity towards O₂^•−, we calculated and compared the electronic structures of bilirubin, biliverdin, cysteine, glutathione, uric acid, melatonin, and ascorbic acid by density functional theory (DFT) (Becke, 1993; Hohenberg and Kohn, 1964; Kohn and Sham, 1965; Stephens et al., 1994). DFT calculations predict that bilirubin’s lowest unoccupied orbital (LUMO) is extensively delocalized, which may impart unusual electron-accepting properties and kinetic reactivity towards O₂^•− (Fig. 3i). Biliverdin also exhibits a delocalized LUMO similar to bilirubin, but one that is more densely concentrated at the α-meso carbon bridge. A more localized electron density could slow SET reactions and may explain the different reactivities of biliverdin and bilirubin to towards O₂^•−. In contrast to the heme metabolites, cysteine, glutathione, uric acid, melatonin, and ascorbic acid exhibit much more densely localized LUMOs, rendering them less effective electron-acceptors. DFT calculations also predict that reducing bilirubin and biliverdin is exergonic. In clear contrast, the Gibbs free energy requirements for reducing cysteine, glutathione, uric acid, and melatonin are thermodynamically unfavorable. Reducing ascorbic acid is approximately 32 kcal/mol less thermodynamically favorable than reducing bilirubin (Table S3).

NMDA receptors signal via O₂^•−, and bilirubin prevents NMDA excitotoxicity by scavenging O₂^•−

BVR is highly expressed in nervous tissues (Fig. S1n–o), suggesting that bilirubin may exert physiologic activity in the brain despite being neurotoxic at higher concentrations. As with MEFs, removing endogenous bilirubin renders neurons vulnerable to oxidative death. BVR^−/− neurons are also hypersensitive to O₂^•− over H₂O₂ (Fig. S4a). Ex vivo, DHE fluorescence suggests that BVR^−/− neurons similarly cannot clear ROS as efficiently as WT when challenged with pyrogallol (Fig. S4b–d).

In the brain, O₂^•− is an important neuromodulator. Activated glutamatergic NMDA receptors (NMDARs) stimulate NADPH oxidase to generate O₂^•− (Brennan et al., 2009; Lafon-Cazal et al., 1993), which functions as a second messenger in long-term potentiation and other redox signaling axes (Gao et al., 2007; Heusler and Boehmer, 2004; Klann, 1998; Wang et al., 1996). BVR^−/− mice exhibit exaggerated locomotor activity when injected with NMDA or MK-801 (an NMDAR agonist and antagonist, respectively) (Fig. 4a–c), indicating that BVR^−/− mice exhibit irregular NMDAR activity. Because bilirubin readily scavenges O₂^•−, we considered whether this aberrant activity stems from disrupted NMDAR O₂^•− signaling.

Figure 4. NMDA receptors signal via O2•−, and bilirubin prevents NMDA excitotoxicity by scavenging O2•−.

a-c, WT and BVR^−/− male mouse open field locomotor activity over time at (a) baseline or after treatment with either (b) NMDA or (c) MK-801. n = 5 WT and 6 BVR^−/− mice (baseline and NMDA) and 9 WT and 5 BVR^−/− mice (MK-801). d-f, Normalized DHE fluorescence (d) or live images (e) and traces (f) of UnaG fluorescence from primary cortical neurons over time. Neurons were treated (arrow) with 100 μM NMDA + 10 μM glycine, 50 μM (S)-AMPA, or 50 μM kainate. Scale bar = 25 μm. n = 6 neurons in (d) and n = 5 (NMDA), 3 (S-AMPA), and 3 (kainate) neurons in (f). g-i, Representative normalized HPLC chromatograms (g) and quantification of E⁺ and 2-OH-E⁺ (h-i) of organic extracts from WT and BVR^−/− ex vivo brain slices loaded with DHE at baseline or after treatment with 100 μM NMDA + 10 μM glycine for 30 min. Points represent normalized average of 2–3 slices from individual mice. j, Average thiobarbituric acid-reactive substances (TBARS) in WT and BVR^−/− neurons 24 h after treatment with vehicle or 100 μM NMDA + 10 μM glycine for 20 min with or without bilirubin pre-treatment. k, Schematic illustrating NMDA microinjections. l-m, Nissl stained coronal sections (l) and lesion volumes (m) from WT and BVR^−/− mice 48 h after injection of vehicle or 20 nmoles NMDA. Scale bar = 2 mm. Points represent individual mice. (a-d, f, h-j, m) Mean ± SEM depicted. * = P < 0.05, ** = P < 0.01, and ns = P > 0.05 by one-way (d, f) or two-way ANOVA (h-j) followed by a post hoc Tukey test or two-tailed Student’s t-test (a-c, m).

Gauged by real-time DHE fluorescence, we observe that treating neurons with NMDA stimulates O₂^•− signaling, consistent with previous reports (Fig. 4d) (Bindokas et al., 1996; Brennan et al., 2009; Lafon-Cazal et al., 1993). Notably, activating glutamatergic AMPA or kainate receptors does not lead to O₂^•− signaling in the presence of excess Mg²⁺ or APV, both of which inhibit NMDARs. To determine whether endogenous bilirubin scavenges NMDAR-associated O₂^•−, we treated WT neurons expressing UnaG with NMDA and monitored changes in UnaG fluorescence. As with pyrogallol in vitro (Fig. 3e), neuronal UnaG fluorescence decays after treatment with NMDA on the same time scale that DHE fluorescence increases (Fig. 4e–f). Applying the spin trap DMPO to the bath beforehand preserves UnaG fluorescence, suggesting that changes in UnaG are due to NMDAR-evoked O₂^•−. Treating neurons with (S)-AMPA or kainate with excess Mg²⁺ or APV also does not alter UnaG fluorescence, presumably because AMPA and kainate receptors do not trigger O₂^•− production.

Previous studies of whether NMDARs initiate O₂^•− signaling relied on experiments that measured DHE fluorescence rather than formation of 2-OH-E⁺ (Bindokas et al., 1996; Brennan et al., 2009; Reyes et al., 2012). To establish whether NMDARs drive O₂^•− signaling, we measured neural 2-OH-E⁺ by HPLC before and after treating brain slices ex vivo with NMDA. Treating WT slices with NMDA leads to significant 2-OH-E⁺ formation, indicating that NMDARs indeed signal in part through O₂^•− (Fig. 4g–i). By contrast, BVR^−/− slices accumulate almost two times more 2-OH-E⁺ than WT slices, suggesting that neural bilirubin scavenges NMDAR-associated O₂^•− (Fig. 4g–i).

Because NMDARs signal in part through O₂^•−, pathologic NMDAR activity can also lead to uncontrolled lipid peroxidation and neuronal death (Reyes et al., 2012). Persistent NMDAR activity, or excitotoxicity, is thought to mediate several neurodegenerative diseases (Dexter et al., 1994; Ferrante et al., 1997; Kellogg and Fridovich, 1975; Subbarao et al., 1990). As expected, treating WT and BVR^−/− neurons with high concentrations of NMDA results in accumulation of malondialdehyde, indicative of lipid peroxidation (Fig. 4j). However, BVR^−/− neurons exhibit greater lipid peroxidation than WT neurons, suggesting that endogenous bilirubin protects neurons from excitotoxic oxidative stress. Pre-treating both WT and BVR^−/− neurons with bilirubin reduces peroxidation (Fig. 4j). To establish whether bilirubin prevents NMDA-driven excitotoxicity in the brain, we stereotactically injected NMDA into the cortex of WT and BVR^−/− mice (Fig. 4k). BVR^−/− mice exhibit substantially larger lesions after exposure to NMDA, suggesting that endogenous bilirubin exerts neuroprotection in part by preventing excitotoxicity (Fig. 4lm).

DISCUSSION

O₂^•− is a key messenger in mammals, but its redox activity involves chemical specificity. The present study establishes a prominent role for endogenous bilirubin as a physiologic O₂^•− scavenger and establishes an activity specific to bilirubin over other major antioxidants. We also find that bilirubin is substantially more reactive towards O₂^•− than biliverdin. Bilirubin’s unique antioxidant profile may also explain why it is a major physiologic antioxidant despite being many-fold lower in concentration than other antioxidants. Intriguingly, bilirubin levels inversely correlate with disease severity in several disorders in which O₂^•− scavenging is compromised, including Parkinson’s (Hatano et al., 2016; Moccia et al., 2015; Scigliano et al., 1997) and coronary artery diseases (Breimer et al., 1995; Levinson, 1997).

Bilirubin’s distinct redox activity may mediate unique redox signaling that underlies larger cellular processes recently linked to BVR. By selectively ablating Blvra in myeloid cells, Bisht and colleagues recently discovered that BVR modulates C5aR1 expression, thereby regulating macrophage chemotaxis (Bisht et al., 2019). Just prior, Hinds and colleagues reported that eliminating BVR in hepatocytes leads to hepatic steatosis (Hinds et al., 2016). Bilirubin could signal in both processes by scavenging O₂^•−, which regulates both macrophage chemotaxis and hepatic steatosis (Browning and Horton, 2004; Petrone et al., 1980). Additionally, Chen and colleagues independently generated a mouse that also globally lacks BVR (Chen et al., 2018). By rigorously examining plasma lipids and peroxiredoxin-2, Chen et al. observed that deleting BVR leads to endogenous oxidative stress even in naïve mice. Both of our studies conclude that bilirubin exerts physiologic redox activity despite being significantly less abundant than other antioxidants. Together with the redox mechanisms identified in the present study, these studies may inform future work that discovers redox mechanisms specific to bilirubin.

Earlier studies also support our findings that bilirubin is a physiologic antioxidant, but also provide intriguing evidence that bilirubin functions indirectly. Some groups have demonstrated that bilirubin and related tetrapyrroles attenuate O₂^•− signaling by inhibiting NADPH oxidase activity (Iwanaga et al., 1987; Kwak et al., 1991; Nakamura et al., 1987). Other studies have observed that moderately hyperbilirubinemic mice exhibit less nitrositive/oxidative stress, again suggesting that bilirubin acts by inhibiting NADPH oxidase (Stec et al., 2013; Vera et al., 2009). However, these studies monitored the effect of bilirubin on NADPH oxidase activity by measuring O₂^•− and therefore may not disambiguate whether bilirubin acts by inhibiting NADPH oxidase, directly scavenging O₂^•− as identified here, or a mix of both.

Our findings also suggest that diverse metabolic pathways outside of oxidative phosphorylation and glycolysis are linked to neural activity. Although neural activity triggers changes in cellular respiration to adapt to energy demands, less is known about the influence of neural activity on other processes. We propose that heme metabolism influences NMDAR redox signaling, and speculate that other metabolites may also shape neural function in yet unidentified ways.

STAR★METHODS

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Solomon Snyder (ssnyder@jhmi.edu).

Experimental Model and Subject Details

Mice

Animal care and use

All experiments were performed in accordance with protocols approved by the Animal Care and Use Committee at the Johns Hopkins University School of Medicine.

Generation of BVR knock-out (BVR^−/−) mice

BVR^−/− mice were generated at Ozgene (Australia) on a C57BL/6J background. To ensure loss of bilirubin biosynthesis, exon 3 of Blvra was flanked by two LoxP sequences for deletion via the Cre recombinase. BVR^−/− mice were then backcrossed to wild-type C57BL/6J mice for four or more generations before these studies. Age-matched wild-type C57BL/6J mice were used as controls.

Cell culture

MEFs

To isolate WT and BVR^−/− MEFs, embryonic day 14.5 (E14.5) embryos were obtained from timed BVR^+/− matings. The pups were decapitated and eviscerated, after which the remaining portion was trypsinized and sheared. Isolated MEFs were plated in 2 wells of a 6-well plate and cultured overnight in DMEM (Gibco) supplemented with 10% FBS, 100 U/mL penicillin and streptomycin, and 2 mM glutamine at 37°C with 5% CO₂. MEFs were then expanded to 6-well plates and transiently transfected with SV40T antigen (Addgene) and maintained until stably proliferative. DNA was isolated from the heads of the pups and genotyped to confirm the genotype of the corresponding MEF cell line. Data for WT and BVR^−/− cells were drawn from experiments on at least two independent cell lines of each genotype.

Primary neurons

To isolate WT and BVR^−/− neurons, cortices of embryonic day 14.5 (E14.5) embryos were isolated and washed with HBSS. Tissue was then incubated with 0.25% trypsin in HBSS for 18 min at 37°C in a conical tube, after which trypsin was inactivated with the addition of FBS. Tissue were then pelleted at 2000g for 5 min at 25°C, after which the supernatant try psin/FBS was aspirated off. Tissue were then rinsed twice with HBSS, resuspended in HBSS, and then gently triturated through a fire-polished glass pipette to promote additional dissociation. Tissue was then filtered through a 70 μm nylon cell strainer to remove debris. The dissociated cells were pelleted at 2000g for 5 min at 25°C and then resuspended in neuronal culture media (composed of Neurobasal (Gibco) supplemented with 2% B-27 supplement and 2 mM glutamine). Cells were filtered once more through a 70 μm nylon cell strainer. Dissociated cells were then plated onto rinsed, poly-D-lysine-coated dishes. Four days later, the conditioned media was supplemented with fresh neuronal culture media. The day before dissections, plates were prepared and coated in 0.25 mg/mL poly-D-lysine overnight at 37°C. Just prior to plating, plates were rinsed three times with sterile water and once with neuronal culture media.

Method Details

Materials and preparation

Solvents and compounds were obtained as follows:

potassium superoxide (KO₂) (Sigma), dimethyl sulfoxide (DMSO) (Sigma), 18-crown-6 (Sigma), bilirubin IXα (Frontier Scientific), biliverdin IXα (Sigma), bilirubin ditaurate (Lee Biosciences), hemin (Frontier Scientific), pyrogallol (Sigma), paraquat (Santa Cruz Biotech), menadione (Sigma), hydrogen peroxide (H₂O₂) (Sigma), 4-hydroxynonenal (4-HNE) (Cayman Chemical), methylthiazoletetrazolium (MTT) (Sigma), rotenone (Sigma), dihydroethidium (DHE) (Thermo Fisher Scientific), MitoSOX Red (Thermo Fisher Scientific), Hoechst 33258 (Abcam), p-nitro blue tetrazolium (Sigma), 5,5-dimethyl-1-pyrroline N-oxide (DMPO) (Cayman Chemical), L-glutathione (Sigma), L-cysteine (Sigma), L-ascorbate (Sigma), (S)-α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (S-AMPA) (Tocris Bioscience), kainic acid (Tocris Bioscience), N-Methyl-D-aspartic acid (NMDA) (Sigma), (+)-MK-801 (Sigma), and glycine (Sigma)

Material preparation:

Bilirubin is highly susceptible to oxidation and photolysis. Accordingly, bilirubin was freshly prepared in DMSO immediately prior to each experiment and then maintained in the dark. O₂^•− was freshly prepared by dissolving finely ground KO₂ in new 0.2 M 18-crown-6 in DMSO. KO₂ was prepared and maintained in a glove-bag under N₂. Glutathione, cysteine, and ascorbate were all prepared freshly before experimentation. Both bilirubin and KO₂ stock solutions were calibrated by UV-light spectroscopy. All other compounds were prepared as 100 μl – 1,000 μl aliquots and stored at −20°C before thawing at 4° C. Freeze/thaw cycles were avoided whenever possible.

Plasmids/cDNA, cloning, and lentivirus

Plasmids/cDNA

cDNA encoding UnaG was a gift from Atsushi Miyawaki, PhD (RIKEN, Japan). cDNA encoding mitochondrial-targeted UnaG was a gift from Jong-Seok Park, PhD and Hyun-Woo Rhee, PhD (Ulsan National Institute of Science and Technology, South Korea). cDNA encoding IRFP was a gift from Vladislav Verkhusha (Addgene plasmid #31857). psPAX2 and pMD2.G were gifts from Didier Trono (Addgene plasmids #12260 and #12259, respectively).

Lentivirus

Lentiviruses were generated as previously described with a few modifications (Jenkins et al., 2013). Briefly, cDNA encoding BVR and GFP separated by a self-cleaving 2A peptide was cloned into pLenti6 and transfected HEK293 cells with psPAX2 and pMD2.G. Virus was harvested from the media and collected by centrifugation at 100,000g for 2 h. BVR^−/− MEFs were infected with lentivirus in the presence of 8 g/ml Polybrene overnight. After 48 h, cells were sorted by GFP fluorescence using fluorescence-activated cell sorting (FACS).

Spectroscopic analyses

Electron paramagnetic resonance (EPR) spectroscopy

EPR spectra were recorded with a Bruker EMXmicro spectrometer equipped with a ER4119HS resonator, and ER4141VT temperature control unit (Bruker Biospin, Billerica, MA). For analysis of naked radicals, the spectrophotometer was set at the following parameters unless otherwise stated: temperature, 100 K; sweep width, 300 G; modulation amplitude, 1 G; modulation frequency, 100 KHz; microwave power, 10 mW. For analysis of spin-trapped radicals (100 μM DMPO), the spectrophotometer was set at the following parameters unless otherwise stated: temperature, 298 K; sweep width, 150 G; modulation amplitude, 1 G; modulation frequency, 100 KHz; microwave power, 10 mW. Samples were equilibrated for 3 min before recording spectra.

UV-visible spectroscopy

UV-visible spectra were recorded from 300 to 750 nm in quartz cuvettes with a Beckman Coulter DU 600 spectrophotometer.

Chemiluminescence spectroscopy

Chemiluminescence measurements were obtained with a Turner Designs 20/20 luminometer without filters. Reactions were initiated by adding various equivalents of KO₂ to a fixed quantity of bilirubin. Chemiluminescence was quantified by integrating total counts over 10 seconds. Any chemiluminescence from dilution of KO₂ or bilirubin was subtracted as background.

Kinetic analyses

To determine second order rate constants of reactions with O₂^•−, each reactant was individually mixed with 50 μM p-nitro blue tetrazolium (NBT), an established O₂^•− scavenger with a second order rate constant of k_NBT = (5.94 ± 0.5) × 10⁴ M⁻¹ s⁻¹ (Bielski and Richter, 1977). The reactions were initiated by adding xanthine/xanthine oxidase as a source of O₂^•−. The ultimate reactions conditions were as follows: 0.1 M Na₄P₂O₇, 100 nM deferoxamine (Sigma), 100 μM xanthine (Sigma), 10 mU xanthine oxidase (Sigma) at pH 8.3. The reaction between O₂^•− and NBT was monitored spectrophotometrically at 560 nm. The second order rate constants of each reactant was determined as a steady-state approximation by evaluating the reaction when r_NBT = r_reactant ≡ k_NBT[NBT][O₂^•−] = k_reacant[reactant][O₂^•−]. Therefore, kreacant = k_NBT[NBT]/[reactant].

Xanthine/xanthine oxidase-derived O₂^•−

To determine rate at which bilirubin reacts with xanthine/xanthine oxidase-derived O₂^•−, 5 μM bilirubin was incubated with 100 μM xanthine and 100 nM desferrioxamine and various amounts of xanthine oxidase in 0.1 M sodium pyrophosphate at pH 8.5 at 25°C. React ions were sometimes incubated with 100 nM MnSOD, 100 nM boiled MnSOD, or 1 μM catalase. Reaction was monitored and measured as a change in absorbance at 442 nm.

Quantum mechanical calculations

Density Field Theory calculations were performed and modeled with Spartan 16 (Wavefunction). Full geometry optimizations and frequency calculations were carried out with DFT-Hartree Fock hybrid B3LYP theory with the 6–31+G* basis set. Energies were calculated at ground state in the gas phase at 298 K.

Western blotting

For western blot analysis, tissues were dounce homogenized at 4°C in lysis buffer (pH 7.4 solution of 50 mM Tris-HCl, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, and 1% Triton X-100) supplemented with protease inhibitors (Sigma). Lysates were then pulse sonicated and centrifuged at 16,000g for 15 min at 4°C. Fifteen micrograms of cleared lysate we re run on a 4–12% polyacrylamide Bis-Tris gradient gel in running buffer (pH 7.3 solution of 50 mM MES, 50 mM Tris Base, 0.1% SDS, 1 mM EDTA) and then transferred to a PVDF membrane. Membranes were blocked with 5% milk in TBS-T (pH 7.6 solution of 16 mM Tris-HCl, 140 mM NaCl, 0.1% Tween-20) and incubated with primary antibodies in 3% bovine serum albumin (BSA) (w/v) overnight at 4°C. The following day, membranes were washed with TBS-T, and then incubated with secondary antibodies in 3% BSA for 1 h at 25°C. The following primary antibodies were used: rabbit anti-BVR (Abcam ab180208; 1:1,000), rabbit anti-HO-2 (generated as previously described (Zakhary et al., 1996); 1:1,000), mouse anti-total OXPHOS cocktail (Abcam ab110413; 1:5,000), mouse anti-pS473 AKT (Cell Signaling 587F11; 1:1000), rabbit anti-AKT (Cell Signaling 11E7; 1:1000), rabbit anti-pERK (Cell Signaling 9101; 1:1000), rabbit anti-tERK (Cell Signaling 9102; 1:1000), and mouse anti-β-actin (Santa Cruz Biotech sc-47778 HRP; 1:10,000). The following secondary antibodies were used: sheep anti-mouse IgG (GE Healthcare NA931; 1:10,000), and donkey anti-rabbit IgG (GE Healthcare NA934; 1:10,000)

RNA isolation, reverse transcriptase PCR (RT-PCR), and quantitative-PCR (q-PCR)

RNA isolation and qPCR were performed as previously described (Paul et al., 2014). In brief, total cellular or tissue RNA was extracted using the RNeasy Plus Universal Kit (Qiagen) per the manufacturer’s instructions. RT-PCR was performed with the SuperScript III One-Step RT-PCR System (Invitrogen), whereas q-PCR was performed with the TaqMan RNA-to-C_t 1-Step Kit (Life Technologies).

Brain slice preparations

Ex vivo slices were prepared as described (Wang and Kass, 1997). In brief, brains from WT and BVR^−/− mice were sectioned in 300 μm intervals with a vibratome in the sagittal plane. Slices were cut in ice-cold sectioning artificial cerebrospinal fluid (S-aCSF) (pH 7.4 solution of 126 mM NaCl, 3 mM KCl, 1.4 mM NaH₂PO₄, 10 mM MgSO₄, 0.5 mM CaCl₂, 26 mM NaHCO₃ and 10 mM glucose), equilibrated with 95% O₂ and 5% CO₂. Sections were equilibrated in room temperature experimental aCSF (E-aCSF) (pH 7.4 solution of 126 mM NaCl, 3 mM KCl, 1.4 mM NaH₂PO₄, 0 mM MgSO₄, 1.4 mM CaCl₂, 26 mM NaHCO₃ and 4 mM glucose) for 1 h before assaying.

High pressure liquid chromatography (HPLC)

Bile and plasma heme metabolites

Bile and plasma bilirubin, biliverdin, and heme were separated by HPLC with an analytical LC-18 column, 250 × 4.6 mm (Xterra, Waters Corporation) with the following mobile phases: 0.1 M ammonium acetate in 60% methanol / 40% water (v/v) (pH 5.2) (Solvent A) and 100% methanol (Solvent B) at 1 mL/min. Metabolites were eluted as follows: 0 to 14 min: linear gradient from 100% A to 100% B; 14 to 19 min: linear gradient from 100% A to 100% B; 19–24 min: isocratic elution at 100% A. The metabolites were detected by measuring absorbance at 405 nm or 450 nm, where indicated. The bilirubin standard exhibited a retention time of approximately 14–15 min, whereas biliverdin and heme standards exhibited retention times of 10–11 and 17–18 min, respectively.

DHE, E⁺, and 2-OH-E⁺

DHE, E⁺, and 2-OH-E⁺ were extracted and analyzed by HPLC as previously established with a few modifications (Zhao et al., 2003; 2005). To extract intracellular DHE, E⁺, and 2-OH-E⁺, cells were lysed in lysis buffer (pH 7.4 solution of 50 mM Tris-HCl, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, and 1% Triton X-100) supplemented with protease inhibitors (Sigma), mixed with 1 equivalent of 1-butanol, vortexed for 1 min, and then centrifuged at 16,000g for 15 min at 4°C. The organic phase was separated and dried under nitrogen gas. Treatments and extractions were performed under dim light to minimize photooxidation of DHE (Zielonka et al., 2006). DHE, E⁺, and 2-OH-E⁺ were separated by HPLC with an analytical LC-18 column, 150 × 3.9 mm (Xterra, Waters Corporation) with the following mobile phases: 10% acetonitrile in in 0.1% trifluoroacetic acid (Solvent A) and 70% acetonitrile in in 0.1% trifluoroacetic acid (Solvent B) at 1 mL/min. The samples were eluted with a linear gradient from 100% A / 0% B to 80% A / 20% B from 0 to 40 min. DHE, E⁺, and 2-OH-E⁺ were detected with a fluorescence detector with excitation and emission at 510 and 595 nm, respectively.

1. MEFs: MEFs were treated with 25 M menadione for 1 h, washed with HBSS, loaded with 25 μM dihydroethidium (DHE) for 45 min, and then rinsed twice more in HBSS before lysis.

2. Brain slices: Slices were loaded with 25 μM dihydroethidium (DHE) for 30 min, washed twice with E-aCSF, and then treated in separate chambers with vehicle or 100 μM NMDA + 10 μM glycine for 30 min. Brain slices were assayed at room temperature in E-aCSF continuously bubbled with 95% O₂ and 5% CO₂.

Cell viability assays

Clonogenic assays

Clonogenic assays were performed as previously described (Weyemi et al., 2015). In brief, equivalent numbers of WT and BVR^−/− MEFs were plated on 35 mm dishes. The cells were treated the following day with an oxidant and then fed fresh media the subsequent day. Six days later, the cells were fixed with 100% methanol for 10 min, stained with 1% methylene blue, and then destained for 5 min with a solution of 40% methanol / 10% acetic acid / 50% water (v/v/v). Colonies were quantified under a dissection microscope.

Methylthiazoletetrazolium (MTT) assays

MTT assays were performed as previously described (Baranano et al., 2002). In brief, MEFs were plated at a density of 75,000 cells/well into a 24-well plate. The following day, the culture media was supplemented with pyrogallol, paraquat, menadione, H₂O₂, or 4-HNE for 8 h, after which the media was replaced with fresh media containing 0.5 mg/mL MTT for 1 h. The media was then removed before dissolving the reduced cellular MTT in DMSO. Cell viability was determined by measuring the absorbance at 570 nm and normalizing absorbances to the vehicle condition.

BVR enzymatic activity assays

BVR activity was assayed at pH 8.5 at 37°C. Tissues were dounce homogenized at 4°C in lysis buffer (pH 7.4 solution of 50 mM Tris-HCl, 150 mM NaCl, 1% Triton X-100, and 5% glycerol) supplemented with protease inhibitors (Sigma). Lysates were centrifuged at 16,000g for 15 min at 4°C. 50 micrograms of clarified lysate was incubated for 5 min at room temperature in a pH 8.5 solution of 50 mM Tris, NADP+, glucose-6-phosphate, and glucose-6-phosphate dehydrogenase (NADPH Regeneration System, Promega). The samples were then placed at 37°C and spiked with 10 μM biliverdin IXα. Reaction rates were determined by monitoring the change in absorbance at 442 nm over time.

Live-cell imaging and analysis

DHE ROS imaging

MEFs:

To compare the ROS scavenging capacities of WT and BVR^−/− MEFs, cells were treated with 25 μM menadione for 1 h. Cells treated with bilirubin were pre-treated with 500 nM bilirubin for 30 min prior to rotenone. MEFs were then loaded with 25 μM dihydroethidium (DHE) for 45 min, after which they were washed and maintained in HBSS for imaging. Cells were immediately photographed with a confocal microscope.

Brain slices:

To assess ROS levels ex vivo, cerebellar slices were treated with 100 μM pyrogallol for 30 min in E-aCSF. After washing with E-aCSF, slices were loaded with 10 M DHE in aCSF for 45 min. After washing with E-aCSF, slices were immediately photographed with a fluorescent confocal microscope. Sections were assayed at room temperature in E-aCSF continuously bubbled with 95% O₂ and 5% CO₂.

MitoSOX Red ROS imaging

MEFs:

To specifically quantify mitochondrial ROS, MEFs were plated at a density of 10,000 cells/well in a 96 well plate. The following day, the cells were treated with 25 M rotenone (Sigma) for 4 h in fresh media. Cells treated with bilirubin were pre-treated with 500 nM bilirubin for 30 min prior to rotenone. The MEFs were then loaded with 5 μM MitoSOX Red in HBSS for 15 min, after which they were washed and maintained in HBSS for imaging. Cells were immediately photographed with a fluorescent confocal microscope.

Mitochondrial bilirubin imaging

To observe mitochondrial bilirubin, MEFs were plated at a density of 60,000 cells/well in a 35 mm glass-bottom plate (MatTEK). The following day, MEFs were transfected with mitochondria-targeted UnaG (mitoUnaG, Dr. Park Lab) and mitochondria-targeted BFP (mitoBFP, Addgene) using Lipofectamine 3000 (Invitrogen) per the manufacturer’s instructions. Two days later, the media was replaced with media lacking phenol red and imaged with a fluorescent confocal microscope. To assess bilirubin trafficking to the mitochondria, exogenous bilirubin was added to the dish to a final concentration of 1 μM. UnaG fluorescence was imaged at several time points.

Neuronal DHE and UnaG imaging

To observe neuronal DHE or UnaG fluorescence, mouse primary cortical neurons were isolated from E15.5-E17.5 mice and plated at a density of 1 million cells/well in a 35 mm glass-bottom plate (MatTEK). Neurons were maintained in Neurobasal medium for at least 14 days in vitro (DIV) before imaging. Neurons were imaged at 25°C in neuronal extracellul ar solution (NES) (pH 7.3 solution of 5 mM HEPES, 125 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, and 33 mM glucose). After establishing baseline fluorescence over 2 min, neurons were treated with either 100 μM NMDA + 10 μM glycine, 50 μM (S)-AMPA, or 50 μM kainate. Neurons treated with (S)-AMPA or kainate were pre-equilibrated with NES supplemented with 2 mM MgCl₂ and/or 50 μM APV. Neurons treated with DMPO were pre-equilibrated with 10 mM DMPO 15 min prior to imaging.

1. DHE fluorescence imaging: Neurons were loaded with 5 μM DHE in NES for 15 min at 25°C, after which neurons were washed twice in NES.

2. UnaG fluorescence imaging: Neurons were transfected with UnaG cDNA at two days before imaging, typically DIV12.

Image analysis

Digitized images were quantified with FIJI (ImageJ). The mean fluorescence signal intensity was measured as the mean of the frequency distribution of all collected pixels. When imaging MEFs, frequency distributions were pooled from 3–4 replicates per sample condition. When imaging tissues ex vivo, frequency distributions were pooled from 2 sections per brain for each ‘n’.

UnaG purification

UnaG was expressed in pMAL-6P2–6×HIS in BL21(DE3) cells. Starter cultures were grown to saturation overnight in Luria Broth (LB) at 37°C and then dilu ted 10-fold in LB and grown to an OD₆₀₀ of 0.3 at 37°C, after which cultures were moved to 18°C. At OD₆₀₀ of 0.6, UnaG expression was induced by the addition of 400 μM isopropyl β-D-1-thiogalactopyranoside for 16 h at 18°C. Cells were harvested by centrifugation, resuspended in 15 ml of resuspension buffer (50 mM HEPES, 300 mM NaCl, 0.5 mM TCEP, 10% glycerol, 1 mM PMSF, 2.34 M leupeptin, 1.45 M pepstatin at pH 7.4) per liter of culture, and flash frozen in liquid nitrogen. Pellets were thawed on ice and sonicated to lyse. Lysate was clarified by centrifugation at 26,000g at 4°C for 30 min and sub sequently loaded onto an amylose column. Protein was eluted with 20 mM maltose in protease buffer (50 mM Tris, 150 mM NaCl, 0.5 mM TCEP, 10% glycerol, 0.01% TritonX-100 at pH 7.4), after which the MBP was were removed by the addition of Prescission Protease (GE Healthcare) for 16 h at 4°C. UnaG was further purified via a nickel column to remove the cleaved tag and Prescission Protease. Protein was further purified on a gel-filtration column (S-200, GE Healthcare) in 50 mM HEPES, 300 mM NaCl, 0.5 mM TCEP, 10% glycerol at pH 7.4, concentrated, and flash frozen in gel filtration buffer at 30% glycerol for storage at −80 °C.

Bilirubin immunostaining

8–12 weeks old mice were anesthetized with 50 mg/kg pentobarbital and transcardially-perfused with 20 mL cold 0.1 M heparin/PBS (pH 7.4), followed by 25 mL of cold fixative (4% formaldehyde (v/v)). Tissues were post-fixed in 4% paraformaldehyde overnight at 4°C and then cryoprotected through a series of 10%, 20%, and 30% sucrose (w/v) gradients for 48 h at 4°C. Tissues were then embedded in Optimal Cutting Temperature compound (OCT) and sectioned in 30 μm intervals with a cryostat, after which the sections were dried on slides for at least 1 h at 25°C. Sections were fixed with 4% paraformaldehyde f or 15 min at 25°C and then permeabilized with 100% met hanol for 7 min at −20°C. The slides were then pre-incubated in blocking solution (50% Seablock (ThermoFisher Scientific) (v/v) in TBS-T, pH 7.4) for 30 min at 25°C. Sections were incubated overnight at 4°C w ith a rabbit polyclonal anti-bilirubin antibody (1:100) in blocking solution. Sections were washed and incubated with goat anti-rabbit secondary antibody (Alexa 488, A32723 Invitrogen) diluted 1:250 in blocking solution for 1 h at 25°C. Tissues were then mounted with ProLong Gold Antifade Mountant with DAPI (Invitrogen). Primary bilirubin antibody was generated as previously described (Doré et al., 1999).

Lipid peroxidation quantification

DIV10–14 primary WT and BVR^−/− cortical neurons were exposed to vehicle or 10 μM glycine/100 μM NMDA for 20 min in control salt solution (CSS) (pH 7.4 solution of 120 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl₂, 25 mM Tris HCl, 15 mM glucose). Neurons were washed twice in CSS and then re-fed conditioned media for another 24 h, after which thiobarbituric acid-reactive substances (TBARS) were measured by absorbance at 532 nm. TBARS were quantified using the OxiSelect TBARS Assay Kit (Cell Biolabs) per the manufactures instructions. Neurons treated with bilirubin were pre-treated with 100 nM bilirubin for 1 h prior to glycine/NMDA.

Stereotactic microinjections

Adult male mice (2–3 month old) were anesthetized with ketamine (80–100 mg/kg) and xylazine (5–10 mg/kg) and subsequently immobilized to a stereotaxic frame (Stoelting Digital Mouse Stereotaxic Instrument) with ear bars. Mice were maintained at 37°C with a heating pad during surgery. The needle was inserted at a rate of 0.1 mm/s and extracted at a rate of 0.5 mm/min, with a 5 min needle dwell time before extraction.

NMDA

Mice treated with NMDA received unilateral intracortical microinjections of 20 nmoles NMDA with a contralateral microinjection of PBS as a control. NMDA was delivered in a final volume of 0.2 μL over 2 min. NMDA was injected at the following coordinates: AP = 0.5, ML = −2.5, DV = −0.5.

Rotenone

Mice treated with rotenone received unilateral intracerebellar microinjections of 2.5 nmoles rotenone with a contralateral microinjection of DMSO as a control. Rotenone was delivered in a final volume of 1 μL at a rate of 0.2 μL/min. Rotenone was injected at the following coordinates: AP = −6.5, ML = −2.5, DV = −2.0.

Histology and lesion quantification

Histology and quantification were performed as previously described (Paul et al., 2014). In brief, mice were euthanized 2 days post-injection, after which brains were fixed, cryopreserved, and sectioned in the same manner as with bilirubin immunostaining. Dry sections were then stained with Nissl stain. Lesional volumes were calculated blinded to both genotype and treatment as a sum of 10 sections for NMDA-induced lesions and 6 sections for rotenone-induced lesions, using Cavalieri’s principle (Cyr et al., 2005) (V = s₁d₁ + s₂d₂ + … + s_nd_n, where s is surface area and d is the distance between two sections).

Behavioral tests

Behavioral tests were performed on 2–4 months old WT and BVR^−/− littermate male mice. Baseline, NMDA- and MK-801-induced activity in the open field was assessed using activity chambers equipped with infrared beams (Photobeam Activity System; San Diego Instruments, San Diego, CA, USA). Behavioral assays were initiated by placing mice were placed in the center of an enclosed acrylic chamber. Mice were initially habituated to the chambers for 30 min followed by a single injection of either NMDA (100 mg/kg, i.p.) or MK-801 (0.3 mg/kg, i.p.). Total locomotor activity was automatically recorded and analyzed via photobeams in the x and y directions.

Quantification and Statistical Analysis

Group data were expressed as mean ± SEM unless otherwise noted. Statistical comparisons were performed using two-tailed unpaired Student’s t-tests and ANOVA analyses, as noted. Differences were considered significant at P < 0.05. All ex vivo or in vivo experiments were performed concomitantly or in a blocked manner with consideration for both genotype and treatment.

Data and Code Availability

Cartesian coordinates for ground state and reduced metabolites are provided in Table S3. The corresponding code for our computations is available at Mendeley Data (DOI 10.17632/6cbnvr4fwk.1).

Supplementary Material

1

Table S1. LD₅₀s of oxidants and electrophiles (related to Fig. 2). n = 2–4 experiments in triplicate.

2

Table S3. Cartesian coordinates for ground state and reduced metabolites (related to Fig. 3).

3

Table S4. Primer sequences used for genotyping and cloning (related to STAR Methods).

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
rabbit anti-BVR	Abcam	ab180208
rabbit anti-HO-2	Zakhary et al., Proc. Natl. Acad. Sci. U.S.A. 93, 795–798.	N/A
mouse anti-total OXPHOS cocktail	Abcam	ab110413
mouse anti-pS473 AKT	Cell Signaling	587F11
rabbit anti-AKT	Cell Signaling	11E7
rabbit anti-pERK	Cell Signaling	9101
rabbit anti-tERK	Cell Signaling	9102
mouse anti-β-actin	Santa Cruz Biotech	sc-47778 HRP
sheep anti-mouse IgG	GE Healthcare	NA931
donkey anti-rabbit IgG	GE Healthcare	NA934
rabbit anti-bilirubin	Doré, S., et al. (1999) Proc. Natl. Acad. Sci. U.S.A. 96, 2445–2450.	N/A
goat anti-rabbit Alexa 488	Thermo Fisher Scientific	A32723
Bacterial and Virus Strains
pLenti6-V5-BLVRA-2A-GFP	This paper	N/A
Chemicals, Peptides, and Recombinant Proteins
biliverdin IXα	Sigma	30891
bilirubin ditaurate	Lee Biosciences	910–12
hemin	Frontier Scientific	H651–9
pyrogallol	Sigma	83130
paraquat	Santa Cruz Biotech	sc-257968
menadione	Sigma	M5625
hydrogen peroxide (H2O2)	Sigma	H1009
4-hydroxynonenal (4-HNE)	Cayman Chemical	32100
methylthiazoletetrazolium (MTT)	Thermo Fisher Scientific	M6494
rotenone	Sigma	R8875
dihydroethidium (DHE)	Thermo Fisher Scientific	D1168
MitoSOX Red	Thermo Fisher Scientific	M36008
Hoechst 33258	Abcam	ab228550
p-nitro blue tetrazolium	Sigma	N5514
5,5-dimethyl-1-pyrroline N-oxide (DMPO)	Cayman Chemical	10006436
L-glutathione	Sigma	G4251
L-cysteine	Sigma	30089
L-ascorbate	Sigma	A7631
S)-α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (S-AMPA)	Tocris Bioscience	0254
kainic acid	Tocris Bioscience	0222
N-Methyl-D-aspartic acid (NMDA)	Sigma	M3262
(+)-MK-801	Sigma	475878
potassium superoxide (KO2)	Sigma	310492
dimethyl sulfoxide (DMSO)	Sigma	D5897
18-crown-6	Sigma	186651
bilirubin IXα	Frontier Scientific	B584–9
glycine	Sigma	410225
xanthine oxidase protein	Sigma	X4500
xanthine	Sigma	X7375
desferrioxamine	Santa Cruz Biotech	sc-203331
recombinant human MnSOD protein	Abcam	ab82656
catalase	Sigma	C40
Prescission Protease	GE Healthcare	27084301
Critical Commercial Assays
RNeasy Plus Universal Kit	Qiagen	73404
Superscript One-Step RT-PCR Kit	Thermo Fisher Scientific	12574026
TaqMan RNA-to-Ct 1-Step Kit	Thermo Fisher Scientific	4392653
OxiSelect TBARS Assay Kit	Cell Biolabs, Inc	STA-330
Deposited Data
Computational chemistry code	This paper	Mendeley Data (DOI 10.17632/6cbnvr4fwk.1)
Experimental Models: Cell Lines
Primary fibroblasts from WT mice	This paper	N/A
Primary fibroblasts from BVR−/− mice	This paper	N/A
Primary neurons from WT mice	This paper	N/A
Primary neurons from BVR−/− mice	This paper	N/A
HEK293	American Type Culture Collection	CRL-1573
Experimental Models: Organisms/Strains
Mouse: BVR−/−: C57BL/6J-Blvratm1Ozg	This paper	N/A
Mouse: WT: C57BL/6J	Jackson Labs	000664
Oligonucleotides
Primer sequences used for genotyping WT and BVR−/− mice	See Table S4	N/A
Primer sequences used for cloning BVR constructs	See Table S4	N/A
Primer sequences used for cloning UnaG constructs	See Table S4	N/A
Primer sequences used for cloning IRFP constructs	See Table S4	N/A
Recombinant DNA
Plasmid: FLAG-UnaG	Kumagai, et al. (2013) Cell 153, 1602–1611	N/A
Plasmid: FLAG-mitoUnaG	Park, J., et al. (2016) ACS Chem. Biol. 11, 2177–2185.	N/A
Plasmid: pIRFP	Filonov, G.S., et al. (2011) Nat. Biotechnol. 29, 757–761.	Addgene plasmid #31857
Plasmid: pmCherry-N1	Clontech	632523
Plasmid: pENTR/D-TOPO	Thermo Fisher Scientific	K240020
Plasmid: pLenti6-V5-DEST	Thermo Fisher Scientific	V49610
Plasmid: UnaG pMAL-6P2–6×HIS	This paper	N/A
Plasmid: IRFP pLenti6-V5	This paper	N/A
Plasmid: psPAX2	Addgene	Addgene plasmid #12260
Plasmid: pMD2.G	Addgene	Addgene plasmid #12259
Software and Algorithms
GraphPad Prism	GraphPad Software, Inc	graphpad.com
Spartan 18	Wavefunction, Inc	wavefun.com
Gaussian 09	Gaussian, Inc	gaussian.com
FIJI	NIH	imagej.net

SIGNIFICANCE.

Reactive oxygen species (ROS) are redox-active chemical species that underlie a broad range of cellular processes. ROS are a family of molecules, and each member displays discrete reactivities and biological roles. Here, we report that the metabolite bilirubin physiologically scavenges superoxide (O₂^•−). Unlike other endogenous antioxidants, bilirubin readily scavenges O₂^•−. Moreover, bilirubin trafficks and localizes to mitochondria, where O₂^•− is predominantly produced. We find that bilirubin’s redox activity is particularly important in the brain, where it scavenges O₂^•− during NMDAR neurotransmission. Bilirubin’s unique redox activity towards O₂^•− may underlie a prominent physiologic redox role despite being significantly less abundant than other endogenous and exogenous antioxidants.

Highlights.

• Eliminating endogenous bilirubin renders cells vulnerable to oxidative stress

• Bilirubin exhibits distinct redox activity among endogenous small molecules

• Bilirubin scavenges O₂•−

• Neural bilirubin prevents NMDA receptor excitotoxicity by scavenging O₂•−