Bilirubin gates the TRPM2 channel as a direct agonist to exacerbate ischemic brain damage

Summary Stroke prognosis is negatively associated with an elevation of serum bilirubin, but how bilirubin worsens outcomes remains mysterious. We report that post-, but not pre-, stroke bilirubin levels among inpatients scale with infarct volume. In mouse models, bilirubin increases neuronal excitability and ischemic infarct, whereas ischemic insults induce the release of endogenous bilirubin, all of which are attenuated by knockout of the TRPM2 channel or its antagonist A23. Independent of canonical TRPM2 intracellular agonists, bilirubin and its metabolic derivatives gate the channel opening, whereas A23 antagonizes it by binding to the same cavity. Knocking in a loss of binding point mutation for bilirubin, TRPM2-D1066A, effectively antagonizes ischemic neurotoxicity in mice. These findings suggest a vicious cycle of stroke injury in which initial ischemic insults trigger the release of endogenous bilirubin from injured cells, which potentially acts as a volume neurotransmitter to activate TRPM2 channels, aggravating Ca2+-dependent brain injury.

In brief Liu et al. discover that bilirubin directly gates the opening of TRPM2 channels as an endogenous agonist to aggravate brain damage in ischemic stroke. Molecular perturbation to or blockade of the bilirubin-binding cavity on the TRPM2 channel provides effective neuroprotection in mouse models.

INTRODUCTION
Stroke is one of the leading causes of death and the adult disability worldwide. 1 The most typical pathological cause of stroke is the interruption of cerebral blood perfusion due to atherosclerosis, 2 atrial fibrillation, 3 hypertension, diabetes, 4 etc. Stroke can be divided into ischemic stroke and hemorrhagic stroke, with the former being the most common, constituting around 80% of cerebral infarctions. 5 In ischemic stroke, circulating clots (or atherothrombosis) block the blood vessels in the brain (e.g., middle cerebral artery [MCA]), interrupting blood and oxygen supply and resulting in the necrosis of neurons and eventual damage of brain structure and function. During stroke, cerebral ischemia leads to a massive release of glutamate that activates NMDARs and induces Ca 2+ influx through these iono-tropic channels to overload neurons and destroy their Ca 2+ homeostasis. 6 The Ca 2+ -dependent activation of cell-death signaling immediately downstream of NMDARs triggers a plethora of cascades that work synergistically to induce neuronal death. These may activate other non-selective cation channels such as acid sensing ion channels (ASICs) and transient receptor potential melastatin (TRPM) channels. 7 Despite extensive work on ischemic damage, few drugs offering significant benefits in preclinical studies have successfully advanced to the stage of approved therapeutics for stroke patients, rationalizing the pressing need to identify the mechanisms of stroke and potential new targets for effective treatments of ischemic stroke.
Stroke appears to be associated with an elevation of bilirubin, which is the end product of heme catabolism in mammals. Bilirubin levels are typically low (<1 mg/dL) but when the concentration exceeds certain levels (i.e., 3 mg/dL), it can be visually observed as hyperbilirubinemia (HB) or jaundice, at which point it can be considered a disease state. 8 Unconjugated bilirubin (UCB) is lipophilic and readily exchanges between the brain and periphery, particularly when the integrity of blood-brain barrier (BBB) is compromised in ischemic stroke. UCB can disrupt intracellular Ca 2+ homeostasis and cause neuronal damage by oxidative stress of endoplasmic reticulum (ER), triggering inflammatory responses and apoptosis. [9][10][11][12][13] Conversely, anti-inflammatory treatments can improve survival rate and reduce neurodegeneration in HB animal models. 14,15 Interestingly, increased levels of bilirubin have been found in patients with acute ischemic stroke and is associated with the severity of stroke. 16 However, it is unknown whether this elevated level of bilirubin is a cause or consequence of stroke.
Despite all the pathological changes associated with HB, neither specific targets nor the mechanisms of bilirubin-induced neurotoxicity are known. Extensive work from our group and others in a variety of neurons all converge to the idea that bilirubin-induced hyperexcitation and Ca 2+ overload are the leading causes of neuronal injury. 11,13,17,18 Although many ion channels are modulated by bilirubin to affect neuronal excitability and Ca 2+ levels, it has barely been explored whether bilirubin can directly gate the opening of Ca 2+ permeable ion channels from extracellular space and initiate the Ca 2+ -dependent signaling cascade that precedes ultimate neuronal injury and death.
The TRPM2 channel is one of the several highly Ca 2+ permeable non-selective cation channels and mediates a variety of physiological and pathological processes by regulating multiple intracellular signaling pathways, such as insulin secretion, inflammatory cell migration, and apoptosis. [19][20][21] These channels are expressed in immune cells, such as neutrophils, 22,23 macrophages, 24 monocytes, 25,26 and lymphocytes 27,28 and are deeply involved in their functions, including the release, chemotaxis, and recruitment of cytokines during the inflammatory response. TRPM2 channels are also abundantly expressed in the brain and serve as the receptor for non-painful warm stimulation in the peripheral nervous system. 29,30 TRPM2 is an oxidative stress sensor in vivo and can be activated by ADP-ribose (ADPR) and/or the oxidative stress generated by ADPR. 19,31,32 Canonically, NMDAR-dependent Ca 2+ influx and ADPR rise lead to subsequent activation of multiple TRP channels including TRPM2. These pathways have been viewed as the primary routes for ischemia and reperfusion-induced cell death in stroke. 33,34 However, it remains unknown whether TRPM2 channels can independently serve as the receptors for endogenous agonists from the extracellular environment to activate downstream pathological processes in ischemic brain injury.
In this study, we explored the mechanisms by which bilirubin causes neuronal damage in mouse models and human subjects with stroke. We consistently found a strong correlation between post-stroke serum bilirubin levels and infarct volume in both patients diagnosed with stroke and rodent models. We found that genetic deletion of TRPM2 channels can mitigate bilirubin-induced hyperexcitability in mouse cortical neurons in vitro and reduce ischemia infarction in adult mice with transient middle cerebral occlusion (tMCAO) in vivo. By applying whole-cell and single-channel patch-clamp recordings from primary neurons and cell lines expressing wild type (WT) 35 and mutant TRPM2 channels lacking the canonical binding sites for intracellular agonists, ADPR and Ca 2+ , we demonstrated that bilirubin and its structural derivatives can directly bind to the TRPM2 channel in a deep pocket near the Ca 2+ binding sites and gate the opening of the channel. Supported by experiments and computational modeling of direct interactions between an agonist or antagonist with this cavity as well as the results from a transgenic mouse harboring a mutation in the cavity to prevent bilirubin binding, we propose a noncanonical paradigm, in which bilirubin is released from endogenous sources during ischemia and potentially serves as a volume neurotransmitter or a specific agonist ligand for TRPM2 to aggravate brain injury in stroke.

RESULTS
Abnormally elevated bilirubin aggravates the severity of stroke in patients The exact effects of bilirubin are controversial. HB is well established as being harmful to the sensory, motor, and cognitive systems, particularly in the early developing brain. Conversely, bilirubin has also been implicated as an endogenous antioxidant that might be protective against diseases associated with oxidative stress, such as cardiovascular disease (CVD), and even cancer. 36,37 In the case of stroke, clinical data indicate that patients with high serum bilirubin levels seemed to manifest more severe neurological symptoms. 38,39 The NIH stroke scale (NIHSS) widely used in clinical studies is a valid indicator to measure the severity of stroke but is limited by the subjective nature of neurobehavioral indices and may not reflect the timing or extent of brain injury accurately. Magnetic resonance imaging (MRI) can identify the location and size of ischemia accurately and directly; so, we opted to use the MRI results of clinical stroke patients as an objective readout to analyze the relationship between bilirubin and infarct volume ( Figure 1A). We collected and analyzed clinical records of adult stroke inpatients, which were divided into two groups based on their serum total bilirubin (TB) concentrations: normal and HB group ( Figure S1A; Table  S1). The positive diagnosis of HB was made in accordance with the criteria by a clinical biochemistry lab (TB > 18 mmol/L), independent of this study. Post hoc analyses of data from these two groups showed that stroke patients with HB had 2-to 3-fold higher levels of TB and direct bilirubin (DB) levels than the normal group ( Figure 1B, left and middle panel). By quantitatively measuring infarct volume from MRI scan images among the same cohort of stroke patients, we found that the HB group showed a larger infarct volume than the normal (Normal: 18.45 ± 3.25 cm 3 , HB: 76.42 ± 13.38 cm 3 ; p < 0.001; n = 167 vs. 47; Figure 1B, right panel). Similarly, the statistical difference remained robust even if the DB concentration was used to divide the two groups ( Figure S1B). Both TB and DB levels were found to be strongly correlated with the infarct volumes by Spearman correlation analyses ( Figure 1C). By comparing the bilirubin levels from serum tests from patients' medical history before and after stroke in the HB group, we also found that stroke significantly increased TB and DB concentration, suggesting that elevated bilirubin was likely a consequence of stroke ( Figure 1D).
These results indicated that elevated bilirubin levels in human stroke subjects positively scale with the infarct volume.
TRPM2 channel mediates bilirubin-induced brain damage in ischemic mouse models Previous studies have shown that the rise in intracellular Ca 2+ concentration following ischemia leads to the overproduction of reactive oxygen species (ROS) that activates the downstream ion channels and/or cell-death signal pathways. 40 In tMCAO animal models, ROS overproduction predominates during the reperfusion process and can activate TRPM2 channels to exacerbate ischemia-reperfusion injury. 41 Given the known roles of TRPM2 in stroke, we first explored whether this channel was the substrate underlying the heightened effects of HB on infarct volumes. To this end, we examined brain injury 24 h after 30 min tMCAO in adult Trpm2 +/+ and Trpm2 À/À mice from which the relationships between bilirubin and infarct volume were then compared. This relatively short period of tMCAO was deliberately chosen to test if bilirubin can aggravate ischemic infarction while avoiding the ceiling effect of longer tMCAO. Before and after tMCAO surgery, we monitored the blood pressure, body temperature, and serum biochemical indexes of the mice. Although occlusion of the MCA significantly reduced the blood flow of the brain by $60%, these indexes were comparable ( Figures S2A  and 2B). Because mice exhibit relatively lower basal level and post-surgery elevation of bilirubin compared with human patients (likely due to species difference), we injected mice intra-peritoneally with bilirubin (Bil group, 50 mg/g) or saline control (Ctrl group) 30 min prior to surgery. In Trpm2 +/+ mice with tMCAO, bilirubin increased the infarct volume (Ctrl: 34.17% ± 2.61%, Bil: 55.31% ± 3.68%; p < 0.001; n = 14). In contrast, the infarct volume in Trpm2 À/À mice did not change by bilirubin injection (Ctrl: 23.58% ± 3.33%, Bil: 25.69% ± 4.46%; p = 0.35; n = 15, Figures 2A and 2B). Under our mild tMCAO paradigm, this difference appeared to be genotype specific but sex-independent ( Figure S2C). The severity of brain damage in the Bil group remained significantly higher than that in the Ctrl group at day 7 after tMCAO ( Figure S3).
Although the bilirubin levels in the cerebrospinal fluid (CSF) of the human stroke subjects were unknown and assumed to be positively correlated with its levels in the blood, we were able to measure the concentrations of bilirubin in both the blood and CSF from the same mice before their sacrifice and found that mice injected with bilirubin had markedly higher TB levels in both serum and CSF than those injected with saline. Both serum and CSF TB showed significant correlations with the infarct volume in Trpm2 +/+ animals ( Figure 2C). Interestingly, despite similar increases of TB levels in the blood, neither the serum nor CSF bilirubin levels correlated with infarct volume in Trpm2 À/À mice (Figure 2C). By using a highly sensitive fluorescent protein-based (i.e., UnaG) assay of bilirubin, we directly measured concentrations of UCB, which is the most relevant index for free bilirubin. [42][43][44] In Trpm2 +/+ mice, the UCB concentration of the Bil group was also much higher than that of the Ctrl group (both in serum and CSF) 24 h after tMCAO ( Figure 2D), in line with the trend of TB. In contrast, the level of UCB in Trpm2 À/À mice did not change, raising the possibility that ischemic injury elevated the release of endogenous bilirubin as observed in Trpm2 +/+ mice ( Figure 2D). Indeed, when we compared the UCB concentrations in mice before tMCAO surgery, we found no differences in basal UCB concentration in both genotypes. However, a robust elevation of UCB in Trpm2 +/+ , but not in Trpm2 À/À mice, was observed 1 h after surgery, suggesting that ischemic injury alone can increase the level of bilirubin in CSF during the acute phase of reperfusion ( Figures S4A-S4C). UCB concentrations in both Trpm2 +/+ and Trpm2 À/À mice increased significantly during ischemic-reperfusion brain injury, reaching the peak at 3-6 h after surgery. The more drastic elevation of serum UCB levels in Trpm2 +/+ was particularly notable at 3 h ( Figure S4D). In contrast, the serum UCB level in Trpm2 À/À subsided 6 h after surgery. Preinjection of bilirubin may have selectively boosted the level of UCB 24 h after tMCAO surgery at a higher level in Trpm2 +/+ than in Trpm2 À/À mice, likely as a result of elevated local release of endogenous bilirubin due to greater infarct volume and/or red blood cell (RBC) extravasation ( Figures 2E and 2F). However, we cannot exclude the possibility that the metabolism of bilirubin in the periphery and/or its transportation through the BBB may be regulated by TRPM2, indirectly contributing to genotype-specific differences in CSF UCB levels after tMCAO. These differences in ischemic infarct volume and bilirubin levels between Trpm2 +/+ and Trpm2 À/À mice before and after tMCAO led us to conclude that ischemia-reperfusion during stroke aggravates brain injury and increases the concentration of bilirubin in some way that might involve TRPM2.
Bilirubin elevates the intrinsic excitability of cortical neurons in adult mice through the TRPM2 channel Previous studies have shown that bilirubin can cause neuronal overexcitation and injury in a Ca 2+ -dependent manner. 10,11,13,18 Combined with the extensive expression of TRPM2 in cortical regions and reduced ischemic injuries in Trpm2 À/À mice, these channels can be rationalized as potential targets of bilirubin. To this end, we made whole-cell current-clamp recordings from slices to investigate the effects of bilirubin on the intrinsic excitability of layer 5 cortical pyramidal neurons from adult mice. Because bilirubin is known to affect neurotransmitter release, 45 we first eliminated excitatory and inhibitory inputs by perfusing a cocktail of synaptic blockers (50 mM D-(-)-2-Amino-5-phosphonopentanoic acid [APV], 10 mM 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo(f)quinoxaline [NBQX], 10 mM bicuculline, and 1 mM strychnine) and measured the intrinsic excitability by injecting a series of current steps (À200 to +400 pA, 50 pA increments). The number of evoked spikes by each current step was counted as a measure of the membrane excitability and plotted against the magnitude of current steps. As reflected by the input-output curves before and after administration of bilirubin in the same cells for 5 and 20 min, bilirubin significantly increased the excitability of cortical neurons in a time-dependent manner without affecting their resting membrane potential (V rest ) in slices from Trpm2 +/+ mice, but these effects were absent in slices from Trpm2 À/À mice ( Figures 3A-3C). Comparison of the spike waveform in cortical neurons evoked by a brief current pulse (2.5 nA current in 0.3 ms from the set membrane potential of À64 mV) showed the half-width did not change but the spike amplitude was significantly attenuated in both groups 20 min after bilirubin (A) Representative images of brain sections by 2,3,5-triphenyl-tetrazolium chloride (TTC) staining of tMCAO Trpm2 +/+ and Trpm2 À/À adult mice with intraperitoneal injection of saline (Ctrl) and bilirubin (Bil). (B) Summary data showing normalized infarct volumes of Ctrl and Bil group in Trpm2 +/+ and Trpm2 À/À mice (n = 14 vs. 15). (C) Spearman correlation analyses of serum/CSF TB concentration and the infarct volume in tMCAO mice (n = 10-17). In Trpm2 +/+ mice, both serum and CSF TB were significantly correlated with the infarct volume, but these correlation were absent in Trpm2 À/À mice. (D) Summary data showing serum and CSF UCB concentration of Ctrl and Bil group in Trpm2 +/+ and Trpm2 À/À animals after tMCAO surgery (n = 14 vs. 15). (E) Comparison of UCB concentration in serum and CSF of Trpm2 +/+ mice in the Ctrl and Bil group before and after ischemia-reperfusion injury (n = 14). (F) Summery data showing UCB concentration in serum and CSF of Trpm2 À/À mice in Ctrl and Bil group before and after ischemia-reperfusion injury (n = 15). Error bars represent means ± SEM; paired Student's t test, one-way ANOVA with the post hoc LSD test. application, likely due to other effects of bilirubin that were not exclusively mediated by TRPM2 ( Figures 3D and 3E). Notably, the basal intrinsic excitability of cortical neurons in Trpm2 À/À mice was slightly elevated, possibly due to compensation after Trpm2 knockout ( Figure 3B). In short, deletion of TRPM2 channels prevented bilirubin from exerting its enhancing effects on the intrinsic excitability, implicating TRPM2 channels as the primary target (or receptor) for bilirubin.
Bilirubin activates TRPM2 currents by directly binding to the channel The TRPM2 channel is classically known for being activated by ROS and requires the binding of both intracellular Ca 2+ and ADPR to reach the open state. Despite being an antioxidant itself with the capacity to neutralize ROS, bilirubin evidently elevated the neuronal excitability and aggravated neuronal death as shown in aforementioned experiments in vitro and in vivo. These intriguing observations led us to hypothesize that bilirubin might act in some unknown way (directly or indirectly) to gate the opening of the TRPM2 channel. To explore this, we used HEK293 cells expressing the human TRPM2 (hTRPM2) channel in a tetracycline-inducible manner (HEK-hTRPM2). We found that bilirubin (9 mM) increased the magnitude of the TRPM2 currents evoked by voltage ramps (À100 to +100 mV), as shown by time-dependent increases in current amplitude at both negative and positive potentials without affecting the reversal potential ($0 mV) of the current-voltage (I-V) curves ( Figure 4A, upper). To distinguish whether bilirubin activated the TRPM2 channel directly or indirectly via elevating ADPR production, we used PJ-34 to block poly (ADPR) polymerase (PARP), the key enzyme for ROS-induced ADPR generation, 46 and found that bilirubin remained effective in activating TRPM2 currents ( Figure 4A, bottom). Bilirubin increased TRPM2 currents when the intracellular solution contained a subthreshold concentration of ADPR (5 mM) but failed to do so when the intracellular concentration of ADPR was increased to 500 mM, a saturated concentration for TRPM2 channel activation ( Figure 4B, upper and middle). These results suggested that bilirubin and ADPR activate the same population of TRPM2 channels.
The activation of TRPM2 channels by ADPR depends on the presence of intracellular Ca 2+ . 47 However, we found that bilirubin remained capable of activating the TRPM2 channel even when fast-acting Ca 2+ buffer 1,2-bis-(2-aminophenoxyethane) ethane-N,N, N 0 , N 0 -tetraacetic acid (BAPTA) (30 mM) was used to replace EGTA to chelate intracellular Ca 2+ , though the current amplitude was noticeably reduced ( Figure 4B, bottom). Figure 4C summarizes the pooled results on bilirubin-induced changes in (B) The mean number of spikes induced by depolarization steps in 50 pA increments to cortical neurons of Trpm2 +/+ and Trpm2 À/À mice (n = 9). Data were fitted with the Boltzmann function, showing an increase in the slope factor and a reduction in the current magnitude to evoke 50% of the maximal spikes in Trpm2 +/+ but not Trpm2 À/À neurons after application of bilirubin for 20 min. (C) Summary plots of the resting membrane potential (Vrest), showing no significant differences between each group (n = 10). (D) Representative recordings of action potential evoked by 2.5 nA current injection in 0.3 ms were compared before and after bilirubin perfusion in Trpm2 +/+ and Trpm2 À/À cortical neurons. (E and F) Summary plots of the peak-amplitude and half-width from the cell in (D) (n = 10). Error bars represent means ± SEM; one-way ANOVA with the post hoc LSD test.

Figure 4. Bilirubin activates TRPM2 currents independent of intracellular ADPR
(A and B) Representative time course of the currents activated by voltage ramps from À100 to +100 mV (500 ms) and the amplitudes measured at +80 mV (strawberry circle) and À80 mV (black circle) are plotted against time after membrane breakthrough. Example ramp current traces before and after bilirubin (legend continued on next page) ll OPEN ACCESS Article the amplitude of TRPM2 currents under these conditions. Furthermore, we examined that the effect of bilirubin in HEK293 cells, transiently expressing hTRPM2 channels with mutations, is known to disable binding sites for Ca 2+ and found that it did not prevent the activation of TRPM2 currents by bilirubin ( Figure S5), despite some differences in the magnitude of TRPM2 currents under these experimental conditions. Overall, these results supported the notion that bilirubin can activate TRPM2 channels independently of the canonical channel activators, namely ADPR and Ca 2+ .
To provide direct evidence for TRPM2 channel activation by bilirubin, we performed single-channel current recordings in the outside-out configuration, under which pipette solution dialysis was expected to remove any residual ADPR that remained present in whole-cell recordings. We found that bilirubin (300 nM) elicited single-channel openings in membrane patches from induced HEK-hTRPM2 cells: these channels exhibited biophysical properties and pharmacological sensitivity to the blocker flufenamic acid (FFA) characteristic of the TRPM2 channel ( Figures 4D-4G). As shown in Figure 4E, the all-point histograms of the current amplitudes before and after perfusing bilirubin revealed that it generated single-channel currents of À2.02 ± 0.05 pA at À40 mV. Further analyses of the dwell times in open and closed states showed that the opening probability (Po) was significantly increased by bilirubin, which was reversed by FFA ( Figure 4F). By plotting the amplitude of single-channel currents at different holding potentials ( Figure 4G), we derived a slope conductance of 76.29 ± 7.85 pS from the linear fit to their bilirubin-driven current-voltage relationship, which is identical to that of its intracellular agonist ADPR ( Figure S6) and consistent with previously reported values (i.e., 60-74 pS). [48][49][50][51] Taken together, these results strongly support the notion that bilirubin directly gates the TRPM2 channel.

Derivatives of bilirubin activate the TRPM2 channel
Bilirubin is an endogenous substance that normally exists in vivo. When its concentration is abnormally high, the levels of related metabolic derivatives also increase. [52][53][54] We explored whether these derivatives with a similar structure can activate the TRPM2 channel. The chemical structures of bilirubin and three selected derivatives are shown in Figure 5A. Biliverdin, a linear tetrapyrrole intermediate, is degraded from heme and rapidly reduced by biliverdin reductase (BVR) to bilirubin. As showed in Figure 5B, biliverdin hydrochloride (Bild, 9 mM) induced TRPM2 currents in HEK-hTRPM2 cells, with comparable amplitudes to those induced by bilirubin. Bilirubin conjugate, ditaurate, disodium salt (BDS), is a water-soluble analog but is not membrane permeable. Similar effects to bilirubin application were observed in extracellular perfusion of BDS (9 mM) (Figure 5C), implying that the TRPM2 binding site is accessible from the outside. Finally, we found that xanthobilirubic acid methyl ester (XAME, 9 mM), a much smaller molecule that retains only two pyrrole backbones of bilirubin, was the most powerful activator of TRPM2 channels, not only showing shorter activation latency, but also generating a much greater current amplitude ( Figure 5D). The effects of bilirubin and these three derivatives on the amplitude of the activated TRPM2 currents are summarized in Figure 5E. In fact, when the dose-dependent curve was constructed for XAME, we found that it had a much higher potency than bilirubin ( Figure 5F), implying that XAME preserved the minimum structure required to be a direct agonist for TRPM2 channels and likely fit more readily than bilirubin into the same binding pocket to activate TRPM2 channels due to the smaller size of the molecule. These results raised the possibility that bilirubin and its metabolic derivatives associated with HB can all act as agonists to activate TRPM2 channels and exacerbate brain damage under ischemia conditions.
Bilirubin activates the TRPM2 channel through a new gating mechanism Because both bilirubin and its structural derivatives directly activated TRPM2 channels, we postulated that there must be a common binding site such that these substances can access the TRPM2 channel from the extracellular side. Recent structural studies have established that the endogenous agonist ADPR binds to two intracellular sites, mediated by the N-terminal MHR1/2 and the C-terminal NUDT9-H domains, respectively, and cooperates with Ca 2+ binding to their sites in the transmembrane domain to control the opening of the TRPM2 channel. 55 Because bilirubin is highly hydrophobic and membrane permeable, it was necessary to rule in or out the possibility that bilirubin binds to the ADPR-binding pockets to exert its agonist effects. We introduced point mutations at the N-terminal ADPR-binding site (R302A/R358A) and the C-terminal NUDT9-H domain (R1433A), which are known to block the binding of ADPR. We found that these mutant TRPM2 channels remained responsive to bilirubin, albeit with a reduced magnitude of activation ( Figures 6A and  6B). In contrast, ADPR failed to evoke currents from R302A/ R358A mutants ( Figure S7). ADPR binding is coupled to the conformational change of the pore region, where specific mutations in the pore (i.e., top: P983A and D987A; bottom: A1046C and N1049C) ( Figure S8) can prevent the pore opening and ion conduction of TRPM2 channels. Strikingly, bilirubin also effectively activated these pore mutants. To reinforce these application and co-application with FFA were transformed into I-V relationships and overlaid on the right panels observations, we made recordings from cells expressing another TRPM2 mutant, in which both N-and C-terminal ADPR-binding domains were truncated (TRPM2-DN/DC, with N terminus residues 535-555 and C terminus residues 1,291-1,329). 56 Such truncations eliminated the activation of the TRPM2 channel by ADPR but not by bilirubin ( Figures 6C and 6D). These findings demonstrate that bilirubin gates the opening of the TRPM2 channel differently from intracellular activators.
The observations that bilirubin and its derivatives, particularly with BDS being membrane impermeable, all had relatively long time courses for activation of TRPM2 channels, and that channel activation by a smaller molecule, such as XAME, was stronger and faster, implied that the binding pocket could be accessed from the extracellular space. However, such a binding pocket must be located rather deep into the TRPM2 channel complex with physical hindrances present that would slow the access and binding of bilirubin. In search of the binding pocket, we performed molecular docking in silico of bilirubin onto the extracellular and transmembrane parts of the TRPM2 channels in the inactive (apo) and closed states versus ADPR and/or Ca 2+ bound open states. The possible conformations of the TRPM2-bilirubin complex were rank-ordered by final docking scores from calculations of internal space constraints and molecular energetics of chemical bonds between different chemical moieties of bilirubin and the amino acid side chains of the TRPM2 protein (Figures 6E   and 6F; Table S2). We found an optimal cavity near the Ca 2+ binding sites, where bilirubin was surrounded by the S3, S5, and TRP helix ( Figure 6F, left panel). Amino-and carboxyl moieties of bilirubin make four hydrogen bonds (H-bonds) with D866, W868 in S3, and D1069 in the TRP helix and two additional salt bridges with K928 in S5, whereas its pyrrole moiety forms a strong p-p interaction with W868 ( Figure 6F, right panel).
Among these hypothetical interaction sites, K928 and D1069 represent the strongest binding sites based on bond energy analysis. To gain further insights into the ligand-channel interactions as predicted by in silico modeling, we used site-directed mutagenesis to replace the positively charged lysine at 928 and negatively charged aspartate at 1069 with nonpolar alanine. We found that both intracellular ADPR and Ca 2+ perfectly retained their agonist capacity to activate this K928A/D1069A double mutant TRPM2 channel with the current amplitude being indistinguishable from that of the WT hTRPM2 channel ( Figures  6G-6I). These indicated that the double mutations did not perturb the gating and permeation properties of TRPM2 channels via canonical intracellular agonists. In contrast, bilirubin and its derivatives completely lost its ability to activate this double mutant ( Figures 6J and 6K), except for XAME, which induced very small currents ( Figure S9). Molecular docking simulations showed that the binding pocket of XAME is not identical to that of bilirubin, engaging partially overlapping amino acid residues  (Figures S10A and 10B). Further analysis of their binding pockets showed that either ligand can be perfectly accommodated at equilibrium (Table S3).
Given the difference in the binding affinity and activation time course between XAME and bilirubin, we subsequently performed molecular dynamics (MD) simulations to model the state transition of the ligand-TRPM2 complex embedded in the phospholipid bilayer ( Figure S10C). Statistical analysis of root mean square deviation (RMSD) for the first 50 ns upon the formation of the channel-ligand complex showed that both ligands can associate with TRPM2 into stable complexes. XAME appeared to approach a stable association within the first 10 ns, faster than bilirubin, which continued to display large fluctuations in binding and unbinding throughout the 50 ns bouts of simulation, likely because of the lower polarity and greater flexibility of XAME compared with bilirubin ( Figures S10D and 10E). Molecular mechanics-Poisson Bolzmann surface area (MM-PBSA) calculations further validated that XAME has a higher affinity for the TRPM2 channel than bilirubin ( Figure S10F). The binding energies contributed by different residues in the binding cavity indicated that the main residues that stabilized bilirubin binding to the TRPM2 channel are D866, K928, and D1069, which partially overlap with or are in the vicinity of the residues that interact strongly with XAME ( Figure S10G). These compelling results demonstrated that bilirubin activates the TRPM2 channel via a different gating mechanism from canonical intracellular ligands.
The bilirubin-binding cavity in the TRPM2 channel is an ideal druggable site for antagonizing neurotoxicity TRPM2 has been well established as a target for stroke therapy with several structurally distinct blockers, such as clotrimazole, tatM2NX, and A23, considered neuroprotective in preclinical studies with animal models. [57][58][59] Using the identified binding pocket for bilirubin as a 3D docking template, we performed extensive in silico screening of these known compounds. We found A23, a newly published TRPM2 blocker, formed a highly stable association with the cavity, thus providing a structural explanation for its nanomolar half-maximal inhibitory concentration in vitro and effectiveness against ischemic injury in vivo. 59 As illustrated in Figures 7A and 7B, unlike BDS, an agonist that required multidimensional interactions with the K725, L862, N869, K870, K928, and K932 residues through H-bonds and salt bridges, A23 formed H-bonds and salt bridges with K928 to optimally fit into the cavity. It exhibited the lowest docking score for the Apo State (6PUO) of TRPM2 (bilirubin À5.04, BDS À4.47, and A23 À8.57) and, therefore, the strongest binding. Patch-clamp experiments reinforced these simulation results. To avoid any off-target effects of bilirubin on intracellular gating pathways, we used its water-soluble and membraneimpermeable derivative BDS as an agonist to activate the TRPM2 channel ( Figure 7C). A23 not only completely suppressed TRPM2 currents preactivated by BDS but also prevented its activation when A23 was first applied (Figure 7D), demonstrating its high affinity. Further experiments showed that A23 blocked currents activated by BDS, bilirubin, and ADPR with their IC 50 being 0.72, 0.92, and 0.83 mM for each ligand, respectively ( Figure 7E). Given its high affinity for the bilirubin-binding cavity in TRPM2, we next tested if A23 can attenuate bilirubin-driven neurotoxicity with an in vitro cell-death assay. We used calcein-am and propidium-iodide (PI) to label surviving and dead cortical neurons, respectively ( Figure 7F). In adult Trpm2 +/+ brain slices, bilirubin significantly increased the mortality of cortical neurons after 1 h of incubation, and A23 fully antagonized this effect of bilirubin ( Figure 7G Ctrl: 35.01% ± 1.49%, n = 6; Bil: 60.84% ± 2.13%, n = 6; Bil+A23: 35.27% ± 1.85%, n = 11).
Although we demonstrated that tMCAO paired with bilirubin injection leads to an increase in infract volume and elevated levels of both serum and CSF bilirubin (Figures 2A, 2B, and S4D), it remained unknown whether bilirubin can originate from the brain itself. To address this question, we first performed immunofluorescence staining to investigate the co-localization of the TRPM2 channel and key metabolic enzymes BVR and heme oxygenase-1 (HO-1) upstream of bilirubin. We found that both BVR and HO-1 were exclusively expressed in mouse neurons (i.e., not astrocytes) as shown by their strong co-staining with the neuronal marker (NeuN) in the brain of Trpm2 +/+ and Trpm2 À/À mice ( Figures 7H and S11). These observations suggested that bilirubin must be primarily produced in and released from neurons during stroke. To simulate the pathological environment of ischemia and hypoxia, we performed oxygenglucose-deprivation (OGD) experiments on acutely isolated brain slices and found that OGD increased the UCB concentration much more drastically in the supernatant of brain tissue from Trpm2 +/+ mice than from Trpm2 À/À mice over the period of 6 h. This increase in Trpm2 +/+ was blocked by application of A23 either at the onset of OGD or 15 min after. In contrast, A23 had no effect in Trpm2 À/À mice ( Figure 7I). These results indicated that ischemic insults can directly induce the release of endogenous bilirubin from the brain to exacerbate neurotoxicity in stroke.

Molecular perturbation of the bilirubin-binding cavity on TRPM2 abolishes bilirubin-induced upregulation of excitability and neurotoxicity
Our experimental and simulation results converge to the conclusion that bilirubin and its structurally similar derivatives can activate TRPM2 channels by binding to a cavity near K928/D1069 where A23 exerts a competitive antagonism to block the binding of these agonists. To acquire unequivocal evidence that this cavity in the TRPM2 enables the binding of bilirubin to mediate its neurotoxicity, we made a knockin (KI) mouse line in which aspartic acid at the site 1066 of the TRPM2 (i.e., the analogous D1069 site of the human TRPM2) was replaced by a neutral amino acid residue alanine (D1066A). Our simulations and patch-clamp experiments in our cell line showed that this site is of critical importance for bilirubin, but not Ca 2+ /ADPR, to function as a potent agonist for the TRPM2 channel. The KI mouse line thus serves as an ideal model to differentiate bilirubindependent and -independent actions of TRPM2. The D1066A mutant mice were viable and showed no obvious anatomical or behavioral abnormalities. Electrophysiological experiments revealed that in D1066A mice, bilirubin did not induce the hyperexcitability in cortical pyramidal neurons seen in WT even over a prolonged perfusion time ( Figure 8A), nor did it cause any significant changes in the spike waveform ( Figure 8B). Strikingly, exacerbated ischemic brain injury by bilirubin in the tMCAO model was completely ablated in D1066A KI mice, as was the increase in the UCB concentration before and after tMCAO surgery ( Figures 8C-8E). In vitro OGD experiments on D1066A brain slices showed that A23 no longer had any effect on the release of endogenous bilirubin ( Figure 8F). Because the TRPM2 channel with D1066A mutation remains functionally intact (with other intracellular agonists), D1066A KI mice mirror TRPM2 KO mice in attenuating bilirubin-induced upregulation of excitability and infarct volume, leading us to the conclusion that the binding cavity in the TRPM2 channel is indispensable for bilirubin to exert its actions in stroke.

DISCUSSION
Bilirubin is susceptible to deposition in brain tissue and results in damage to a subset of highly sensitive brain regions. 60 An increasing number of studies have shown that the bilirubin neurotoxicity is related to the disruption of the oxidative stress state of neurons and the increased production of oxidative stress products caused by bilirubin. 9,12,61,62 Gurses et al. 63 showed that HB affects the electrical activity in the cerebral cortex of human neonates. Our previous studies demonstrated that bilirubin damages mouse sensory neurons by affecting the function and/or level of ion channels and excitatory neurotransmission to cause neuronal overexcitation, 10,11,18,64 and by causing Ca 2+ overload due to Ca 2+ influx via voltage-gated calcium channels (VGCCs, P/Q-type in particular) and mobilization from internal Ca 2+ stores. However, it remains ambiguous whether bilirubin exerts its multifaceted actions by binding to a broad spectrum of subcellular substrates or one specific receptor to cause neuronal overexcitation, Ca 2+ overload and cell death. In this work, we demonstrated that bilirubin binds directly to a specific cavity of TRPM2 channels

OPEN ACCESS
Article to gate channel opening and that knockout or blockade of TRPM2 channels and KI of TRPM2 D1066A devoid of the bilirubin-binding site can all antagonize bilirubin-induced hyperexcitation in vitro and attenuate neurotoxicity in vivo. These results provided compelling evidence that the TRPM2 channel serves as the core receptor on the plasma membrane of central neurons to sense bilirubin or its metabolites in CSF at clinically relevant concentrations. These channels directly mediate Ca 2+ inflow to initiate divergent signaling cascades underpinning major phenotypical changes in neuronal excitability and injury associated with HB pathology. Our findings also implicated the bilirubin-binding cavity in the TRPM2 channel as a potential target for developing therapeutics to alleviate ischemic injury in stroke and other HBrelated neuronal injuries.
The actions of bilirubin are intricate and may engage both direct and indirect signaling pathways. Previous studies have shown that HB can create an oxidative stress microenvironment by promoting the production of ROS 32 and that oxidative or nitrosative stress induces the production of free ADPR from nicotinamide adenine dinucleotide (NAD) catalyzed by NADase in mitochondria and by PARP in the nucleus. 65,66 ADPR generated by these pathways can serve as intracellular agonists to activate the TRPM2 channel. Indeed, we found a significant reduction in the amplitude of bilirubin-induced TRPM2 channel currents by PJ-34, which blocks PARP-dependent ADPR production in the nucleus, suggesting that a considerable portion of bilirubin action was synergistic with intracellular ADPR production. Similarly, after chelation of intracellular Ca 2+ by BAPTA, we found that the amplitude of TRPM2 channel currents induced by bilirubin was also significantly reduced. Thus, the action of bilirubin on TRPM2 channels is convergent with that of intracellular ADPR and Ca 2+ . However, we would argue that these were largely indirect consequences following direct activation of Ca 2+ -permeable TRPM2 channels by bilirubin. Multiple lines of evidence provide strong support of this view: (1) bilirubin directly gates the opening of TRPM2 channels at the single-channel level; (2) bilirubin activates the TRPM2 channels despite mutational disruption of their sensitivity to ADPR and Ca 2+ ; (3) structurally related derivatives of bilirubin activate TRPM2 channels with varying potency, and are competitively antagonized by A23, which possesses much higher affinity for the same binding site; and (4) bilirubin actions on both neuronal excitability and neurotoxicity are ablated in TRPM2 D1066A KI mice. Therefore, we concluded that bilirubin is the first messenger molecule directly acting upon the plasma membrane TRPM2 channel to initiate downstream signaling pathways that amplify a vicious cycle of overexcitation, Ca 2+ overload, and ROS and ADPR production, ultimately driving cell death during ischemia.
Our results from TRPM2 mutants led us to suggest that bilirubin must gate TRPM2 channels through a noncanonical binding site. Huang et al. 55 suggested two distinctive ADPR-binding sites in the hTRPM2 channel, one formed by the N-terminal MHR1/2 domains and the other by the C-terminal NUDT9-H domain, both of which work in concert with Ca 2+ binding to the sites in the transmembrane domain to gate the channel opening. However, our results showed that introduction of point mutations in the MHR1/2 domain (i.e., R302A/R358A), NUDT9-H domain (i.e., R1433A), or truncations of both domains (i.e., TRPM2-DN/ DC) prevent activation of the TRPM2 channel by ADPR as antic-ipated, but not by bilirubin. The binding site for bilirubin to activate the TRPM2 channel is, therefore, independent of that for ADPR ( Figure 6). Our in silico computational simulations complemented by site-directed mutagenesis guided us to demonstrate that introduction of K928A/D1069A mutation eliminates the activation of the TRPM2 channel by bilirubin but not ADPR or Ca 2+ , delineating the structural basis of bilirubin binding within the cavity surrounded by the S3, S5, and TRP helix. It should be noted that bilirubin might also directly interact with Ca 2+ binding sites, as inferred by both H-bond and p-p interactions of its pyrrole moiety with D866 and W868 in the vicinity of the Ca 2+ -binding pocket, reducing the requirement of intracellular Ca 2+ for channel gating. Interestingly, the location of the bilirubin-binding pocket in the TRPM2 channel is proximal to that for icilin in the TRPM8 channel. 67 The structural similarity between these two channels suggests a conserved cavity for agonist binding of TRP channels. The bilirubin-binding site is located deep in the transmembrane region and access to this binding cavity can take time, as elucidated by our MD simulations of the ligand-TRMP2 complex ( Figure S10), explaining the relatively slower action of bilirubin compared with its structural derivative XAME, which has a smaller molecular size and can more readily reach the binding site. Our findings indicated that the TRPM2 channel possesses a novel binding pocket for bilirubin and its derivatives that gate the channel extracellularly via a mechanism distinct from that of intracellular APDR and Ca 2+ .
When HB occurs in adults, non-histaminergic pruritus caused by abnormal bilirubin levels (>5 mg/dL) is sometimes observed in patients with jaundice. 68 Bilirubin, as a pruritogen, can produce itching by acting on the Mas-related G-protein coupled receptor X4 (MRGPRX4) receptor in humans and MRGPRA1 in mice. 69 MRGPRX4 and MRGPRA1 are mainly distributed in the periphery, specifically in sensory neurons within the dorsal root ganglia (DRG). 70 Considering that MRGPRs serve as G-protein-coupled metabotropic receptors for bilirubin, 4 our current work raises the possibility that bilirubin may also act on TRPM2 channels, which are known to be expressed in DRG neurons, contributing to the phenomenon of pruritus. Given the fact that bilirubin can act as an agonist for both ligand-gated channels (i.e., TRPM2) and metabotropic receptors (i.e., MRGPRA1 and MRGPRX4), we propose that bilirubin largely fulfills the criteria as a volume neurotransmitter involved in physiological and pathological functions in both central and peripheral nervous systems. Although further studies are needed to firmly establish bilirubin as a volume transmitter, our proposal is supported by the findings from this study, complemented by the observations that TRPM2 antagonist A23 specifically competes against the same site with bilirubin, and that specific pathways for endogenous synthesis and metabolism of bilirubin exist in neurons, independent of circulating bilirubin produced by the liver and bile. 71,72 Although the properties of bilirubin storage, degradation, and release modes from neurons remain unknown, we suggest that TRPM2 can function as a ligand-gated non-selective ion channel and/or a metabolic sensing channel for endogenously released bilirubin and its structural derivatives under physiological conditions, potentially playing an important role in regulating intrinsic excitability, synaptic transmission, and plasticity, aside from its pathological roles in ischemic stroke as demonstrated in this study.

Article
Ischemic neuronal death is mainly caused by activation of NMDA receptors and TRPM channels (e.g., TRPM2, 4, and 7) and their reciprocal cross talk via intracellular ROS, ADPR, and/or Ca 2+ . [73][74][75][76] However, antagonists for NMDARs have largely been proven to be clinically ineffective for neuroprotection against stroke, raising the possibility that other mechanisms exist to activate TRPM channels and signaling pathways that cause cell death. Although clinical studies have shown a possible correlation between bilirubin levels and symptomatic severity in stroke patients, it remains elusive how stroke leads to an elevated bilirubin level. Previous studies suggested that excessive heme could be released from cytoplasmic proteins and mitochondrial cytochromes from cells injured or undergoing necrosis in the core of infarct region during ischemia. 77 Moreover, RBC extravasation provides another source of heme that could then be converted to bilirubin in the brain and CSF. 78 As shown by our immunohistochemistry (IHC) co-localization of endogenous enzymes for bilirubin synthesis in neurons, we suggest that in WT mice, ischemia induced elevation in the level of endogenous heme from neurons and/or RBC extravasation from the infarct as well as in the activity of HO in a TRPM2dependent manner in the infarct area. 79 These converge to boosting levels of UCB (or bilirubin) and/or other more metabolic derivatives in CSF, as modeled by our in vitro OGD experiments ( Figure 7I). Under stroke circumstances, local UCB concentrations in the infarct area may greatly exceed the global level in CSF: local TB levels sometimes increase to a staggering 25 mmol/L. Such a high concentration will effectively activate TRPM2 channels to exacerbate the severity of cell death. This is evidenced by doubling of infarct volume in stroke patients from MRI scan images or in mouse tMCAO models with a 1-fold increase in DB levels ( Figure 1).
Using an in vitro cell-death assay with/without TRPM2 antagonist A23, tMCAO models from WT, and TRPM2 knockout mice, we have demonstrated that HB aggravates brain damage primarily through the TRPM2 channel. Using the D1066A KI transgenic mice, we showed that bilirubin indeed binds to the transmembrane binding pocket of the TRPM2 channel via a key interaction with the residue D1066 (or amino acid residue D1069 in hTRPM2) and initiates and aggravates excitotoxicity in vivo. Although it remains unknown how TRPM2 regulates the release of endogenous bilirubin from neurons and/or other peripheral cells and/or its clearance during stroke, the observation that D1066A mutation abolishes the neurotoxic component of the TRPM2 channel in ischemia-hypoxic brain injury raises the conceptual possibility of differentiating the physiological and pathological signaling embedded in the channel itself ( Figure 8). These findings highlight the unexpected roles of bilirubin and its metabolites in activating TRPM2 channels to underlie neurotoxicity independently from canonical signaling pathways.
In this study, we boosted the level of bilirubin via acute injections into mice before tMCAO surgery to closely model the subset of the human cohort with high levels of serum bilirubin after stroke. Administration of exogenous bilirubin limits the extrapolation of our findings from mice to the reciprocal relationship between endogenous bilirubin and brain injury as seen in stroke patients. Given that TRPM2 is widely expressed in both central and peripheral nervous systems and that non-neuronal cells such as microglia also express TRPM2, 80 we envision that deleting TRPM2 in a system-, region-, and cell-specific manner with the Cre-loxP system from TRPM2 flox/flox mice is needed in future studies to clearly define the origin and sites of actions of bilirubin and its regulation in healthy and diseased brains.

Conclusions
In conclusion, our study demonstrates that bilirubin directly binds to and activates the TRPM2 channel via a specific binding cavity in the transmembrane domain and that this cavity can be targeted by competitive blockers, such as A23. Being the first messenger for TRPM2 channels, bilirubin triggers and exacerbates neurotoxicity associated with brain ischemia by directly driving Ca 2+ influx through these channels to elevate neuronal excitability and activate Ca 2+ -dependent cell-death pathways. Our findings provide mechanistic insights and a proof of principle for developing new strategies targeting the bilirubin-binding pocket in TRPM2 channels to alleviate and prevent brain injury associated with stroke and jaundice in patients.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
We have filed a patent application for the bilirubin-binding cavity on the TRPM2 channel for developing therapeutics to treat brain injury in stroke and jaundice. d Any additional information required to reanalyze the data reported in this work paper is available from the lead contact up request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
Ethical approval for data analysis of stroke patient blood and MRI For hospitalized stroke patients, blood and MRI were used for diagnosis purpose. The levels of blood total bilirubin (TB), direct bilirubin (DB), alanine aminotransferase (ALT), and aspartate aminotransferase (AST) concentration were measured in clinical biochemistry laboratory, The Sixth People's Hospital of Shanghai. All enrolled participants underwent a head MRI (Achieva 3.0 T MRI system, Philips Healthcare, Amsterdam, The Netherlands) scan within 72 hours after inpatient at the Department of Radiology. Clinical data collection and analyses from human subjects were approved by the Institutional Ethic Committee of the Sixth People's Hospital of Shanghai (Approval NO: 2020-232).

Animal and ethical approval
The Trpm2 -/-C57BL/6 mice were originally generated in the University of Leeds as described previously. 81 The D1069 site of the hTRPM2 channel is conserved, which correspond to D1066 in mice ( Cell culture and transfection of human embryonic kidney 293 cell line Human embryonic kidney 293 (HEK293) cells with tetracycline-inducible expression of human TRPM2 channel (hTRPM2) were cultured in a mixed medium containing DMEM/ F-12 (Gibco, USA) and 10% fetal bovine serum (Gibco, USA) as well as blasticidin (50 mg/ml, Invivogen, USA) and zeocin (0.4 mg/ml, Invivogen, USA). The expression of hTRPM2 was induced by substituting blasticin and zeocin for tetracycline (1 mg/ml, Sigma, USA) 24-48 hours before use. TRPM2 channel mutants were transiently expressed in HEK293T cells by transfection with using Lipofectamine 3000. Briefly, HEK293T cells were transiently transfected with cDNAs encoding the mutant hTRPM2 channel. The cDNA for GFP was co-transfected as a marker for identification of the transfected cells for electrophysiological experiments 16-24 hours after transfection. All the cells were seeded on 96-well coverslips (3 x 3 mm, WHB) and cultured at 37 C under a humidified atmosphere containing 5% CO 2 .

METHOD DETAILS
Enrolled clinical data Clinical data were collected from the department of neurology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, which was composed of 944 stroke patients hospitalized from November 2019 to May 2020 (age: 25 to 95, both man and woman). MRI examination and bilirubin concentration were the results of the first detection of stroke patients within 72 hours after hospitalization. The exclusive criteria were stroke patients (1) with hemorrhagic etiology, (2) without brain MRI, (3) with brainstem or cerebellum ischemic/infarction, (4) with cerebrovascular obstructed, but DWI MRI images did not show abnormalities, (5) with abnormal liver function by laboratory biochemical measurements (6) incomplete and missing clinical data ( Figure S1). In total, clinical data from 214 subjects were obtained and divided into two groups based on the serum total bilirubin levels, including stroke patients with normal bilirubin concentration (Normal group, n = 167; Male: 106 and Female: 61; Average age 71), stroke patients with abnormal bilirubin concentration (HB group, n = 47; Male: 36 and Female: 11; Average age 68). Patients in the HB group with traceable serum biochemistry in their previous hospitalization history were used to compare changes in bilirubin levels before and after stroke. Infarct volumes were calculated by Osirix 12.0 software in diffusion weighted imaging (DWI) sequence and clinical data were analyzed by two individuals independently.
Preparation of brain slices Brain slices containing cortical neurons from mice of both sexes were prepared for electrophysiology as previously described. 82 Briefly, all the adult mice aged 6-8 weeks were first anesthetized with sodium pentobarbital (55 mg/kg, i.p.) and then decapitated. The brains of Trpm2 +/+ and Trpm2 -/mice were quickly but carefully removed and immersed into ice-cold oxygenated cutting solution (95% O 2 and 5% CO 2 ), dissected, and sectioned at a thickness of 300 mM using a micro slicer (VT-1000S, Leica Microsystems, Nussloch, Germany). Slices were transferred to the incubation solution to recovery at 37 C for 40 minutes and then transferred to a recording chamber at room temperature (21-26 C) before use.
Transient middle cerebral occlusion (tMCAO) animal model tMCAO was introduced as described previously. 41 In brief, C57BL/6 mice, either Trpm2 +/+ , Trpm2 -/and D1066A (age: 6-8 weeks, weight: 20-25 g, both male and female), were anesthetized using a 2% isoflurane-oxygen mixture for induction and 1.5% for maintenance. Bilirubin was first dissolved in 1 M NaOH solution, and its pH value was adjusted back to 7.4-8.0 by titrations with 1 M HCl. Before the operation, bilirubin (50 mg/g) and saline were injected intraperitoneally. tMCAO model was achieved by inserting a monofilament suture (RWD Life Science) into the right MCA via the internal carotid artery. MCA embolization lasted for 30 min, and body temperature was maintained at 37 C using a heated blanket. Adequate ischemia was confirmed by continuous laser Doppler flowmetry (moor FLPI-2). Animals that did not have a significant reduction of blood flow less than 30% baseline values during MCAO were excluded. When the surgery was finished, mice were placed on another 37 C heating blanket till they regained consciousness and then returned to the cage. The blood pressure, body temperature and serum biochemical markers were monitored in mice before and after tMCAO surgery ( Figure S2) Laser speckle imaging Mice were anaesthetized by 1% isoflurane and their head were restrained in a stereotaxic cylinder frame to minimize breathing motion. The scalp and the skull fascia were gently incised down the midline and peeled to the side. Saline was titrated onto the skull to maintain moist. Laser speckle images were recorded with a CMOS camera before MCAO, 15 min after occlusion and 15 min after reperfusion. For each animal, three sets of raw speckle images were acquired in <15 s (250 frames in each set; image width, 752 pixels; image height, 580 pixels; exposure time, 20 ms). A speckle contrast image was calculated from each raw speckle image using a sliding grid of 2.5 mm 3 2.5 mm. A mean speckle contrast image was calculated for each set and used to estimate the relative cerebral blood flow (rCBF). The rCBF in the ipsilateral (ischemic) hemisphere was normalized by the mean rCBF in the contralateral (non-ischemic) hemisphere. Speckle images were obtained and processed by the software mFLPI2MeasV2.0, rCBF data from all pooled hemispheres were obtained by the software moorFLPIReviewV50. All analyses were randomized.
Infarct volume measurement 24 hours or 7 days after tMCAO, animals were anesthetized using a 2% isoflurane-oxygen mixture. Brains were extracted and coronally sectioned into 1 mm slices, which were then stained with 2% 2,3,5-triphenyltetrazolium chloride (TTC) for 20 min at 37 C. The infarct volume was analyzed using ImageJ and the infarct volumes were calculated according to the following formula: Corrected infarct volume (%) = [contralateral hemisphere volume À (ipsilateral hemisphere volume À infarct volume)] / contralateral hemisphere volume 3 100%.
Total bilirubin (TB) and unconjugated bilirubin (UCB) measurement Trpm2 +/+ , Trpm2 -/and D1066A C57BL/6 mice were anesthetized by intraperitoneal injection of pentobarbital (55 mg/kg) 24 hours before tMCAO operation, and then fixed in the stereotactic setup (RWD Life Science). The skin above the skull from the base of the neck up to in between the eyes was cut open. A hole directly above the right ventricle was made with a grinding drill to allow insertion of the micro syringe into the lateral ventricle (coordinates: 1.1 mm laterally to the right and 0.5 mm posterior of the bregma, 2.5 mm deep). CSF was collected by using the micro injection pump (speed: 0.2 ml/min), after which the syringe was withdrawn, and the skin was sutured. The mice were removed from the stereotactic setup and 50 ml blood was taken through the orbital venous plexus. Mice were then placed on 37 C heating blanket till they regained consciousness and returned to the cage. 1 or 24 hours after tMCAO, animals were anesthetized using a 2% isoflurane-oxygen mixture. CSF and blood were collected in the same way before sacrifice. All samples were centrifuged at 3000 r/min. Total bilirubin concentration of supernatants was measured with the Bilirubin Reagent Kit (Sigma). Samples were transferred to a 96-plates and mixed with reaction solution before measurements of their absorbance were made at 530 nm 10 minutes later. UCB concentration was measured with UnaG, a bilirubin-inducible fluorescent protein from Japanese eel muscle. For standard calibration curve of fluorescence intensity, a 100 ml reaction mixture containing 50 ml UnaG solution (1 mM) and 50 ml artificial bilirubin solution with concentration gradient (1.4278 mM, 0.7139 mM, 0.3573 mM, 0.1785 mM, 0.0893 mM and 0 mM) was prepared. After 10 minutes of reaction, the fluorescence intensity was detected by microplate reader (Synergy H1M, Bio Tek) with fluorescence filters for excitation and emission wavelengths of 485 and 528 nm, respectively. Serum and CSF samples were diluted 20-fold with PBS and fluorescence intensity was measured. The UCB levels were extrapolated from the standard curve.
Oxygen-glucose deprivation model in brain slices Mice were anesthetized with isoflurane, brain slices containing cortical neurons were sectioned at a thickness of 300 mM and incubated (37 C) for 30 min. 82 The slices were first washed with glucose-free artificial cerebrospinal fluid (ACSF) solution for 10 min prior to OGD to deplete the remaining glucose from extracellular space. Finally, the brain slices were transferred to a 24-well plate containing glucose-free ACSF, and the OGD experiment was performed in a hypoxia incubator (95% N 2 , 5% CO 2 , 37 C). During the hypoxia process, 5 ml of glucose-free ACSF was collected at different time points for the measurements of UCB concentration.

Immunofluorescence
All mice were anesthetized with 1% pentobarbital (0.04 ml/10 g) before cutting the sternum to expose heart. A needle was inserted into the left ventricle to perfuse the pre-oxygenated cold ACSF with low molecular weight heparin. The ACSF was perfused until the