Abstract
Hyperbilirubinemia (HB) is a key risk factor for hearing loss in neonates, particularly premature infants. Here, we report that bilirubin (BIL)-dependent cell death in the auditory brainstem of neonatal mice of both sexes is significantly attenuated by ZD7288, a blocker for hyperpolarization-activated cyclic nucleotide-gated (HCN) channel-mediated current (Ih), or by genetic deletion of HCN1. GABAergic inhibitory interneurons predominantly express HCN1, on which BIL selectively acts to increase their intrinsic excitability and mortality by enhancing HCN1 activity and Ca2+-dependent membrane targeting. Chronic BIL elevation in neonatal mice in vivo increases the fraction of spontaneously active interneurons and their firing frequency, Ih, and death, compromising audition at the young adult stage in HCN1+/+, but not in HCN1−/− genotype. We conclude that HB preferentially targets HCN1 to injure inhibitory interneurons, fueling a feedforward loop in which lessening inhibition cascades hyperexcitability, Ca2+ overload, neuronal death, and auditory impairments. These findings rationalize HCN1 as a potential target for managing HB encephalopathy.
- auditory impairment
- bilirubin
- cochlear nucleus neuron
- GABAergic interneurons
- hyperpolarization-activated cyclic nucleotide-gated channels
- neurotoxicity
Significance Statement
This study demonstrated that bilirubin (BIL) preferentially targets GABAergic interneurons where it facilitates not only gating of HCN1 channels but also targeting of intracellular HCN1 to the plasma membrane in a calcium-dependent manner, resulting in neuronal hyperexcitability, injury, and sensory dysfunction. These findings implicate the HCN1 channel not only as a potential driver for auditory abnormalities in neonatal patients with bilirubin encephalopathy but also as a potential intervention target for clinical management of neurological impairments associated with severe jaundice. Selective vulnerability of interneurons to neurotoxicity may be of general significance for understanding other forms of brain injury.
Introduction
Neonatal hearing impairment is one of the most common disorders detected by newborn screening programs. Among many risk factors, hyperbilirubinemia (HB) scores at the top in causing congenital permanent hearing loss during childhood (Psarommatis et al., 2011; Kanji and Khoza-Shangase, 2012; Huang et al., 2017; Korver et al., 2017; Wroblewska-Seniuk et al., 2017; Xu et al., 2019; Zhai et al., 2021). At present, severe HB occurs in 10–17% of full-term and near-term infants and 36–40% of preterm infants per year in China and ranks first in the spectrum of premature infant diseases (Ma et al., 2009; Z. Yu et al., 2014). An estimated 481,000 term infants worldwide suffer from severe HB annually, including 114,000 neonatal deaths and 63,000 survivors living with moderate and severe disability (Bhutani et al., 2013), presenting a tremendous burden on economy and society in developing and underdeveloped countries (Shapiro, 2010; Muchowski, 2014; Olusanya, 2015; Rose and Vassar, 2015).
Excessive bilirubin (BIL) deposits in the brainstem nuclei [cochlear nucleus (CN), oculomotor nucleus, vestibular nucleus] and other locations (Sugama et al., 2001; Wang et al., 2008; Manchanda et al., 2013), leading to a variety of sensory and cognitive deficits. Animal studies have shown that the blood–brain barrier in the brainstem of newborn rats, more specifically CN, is the weakest, and its permeability is approximately four times more than other contemporaneous sensory systems, even higher than that of globus pallidus and inferior colliculus. The blood–brain barrier in CN is highly permeable to BIL (Roger et al., 1996). CN is the first level of auditory information processing center in the brain where neuronal injury can cascade to downstream auditory nuclei, resulting in dysregulation of audition and even deafness. CN can be subdivided into anteroventral cochlear nucleus, posteroventral cochlear nucleus, and dorsal cochlear nucleus. Diverse cochlear nucleus neurons receive auditory nerve fiber inputs and process various acoustic information (Leão, 2019). Thus, CN neurons are one of the first subset of central neurons affected in HB, which can be detected by changes in the auditory brainstem response (ABR), an important means for early detection of neurological impairments (Shapiro and Nakamura, 2001; Shapiro and Popelka, 2011; Watchko and Tiribelli, 2013; Olds and Oghalai, 2015; Okumura et al., 2021).
The mechanisms of BIL neurotoxicity have been extensively studied for decades and involve the activation of complex signaling cascades associated with excitotoxicity, oxidative stress, calcium overload, inflammation, apoptosis, and so on (Brites et al., 2009; Fernandes et al., 2009; Brites, 2012; Feng et al., 2018). Excitotoxicity is the main cause of neuronal injury due to excessive neural activity driven by elevated presynaptic glutamate release to trigger postsynaptic spiking and/or by upregulated intrinsic excitability to boost spontaneous firings. Although the impact of BIL on excitation is well established (Fernandes et al., 2009; Li et al., 2011; Han et al., 2015; Liang et al., 2017b), much less is known about its effects on inhibition at neonatal stage. Before the opening of the external ear canal, intrinsic discharges induced by hyperpolarization-activated cation channels (HCN) play important roles in facilitating CN development (Yin et al., 2018). Among the four HCN isoforms (HCN1-4), HCN1 and HCN2 channels show a particularly high expression in the CN (Monteggia et al., 2000; Santoro et al., 2000; Notomi and Shigemoto, 2004; Leao et al., 2006). Whether and how excessive BIL targets HCN channels of CN to cause neuronal hyperexcitability remains unknown.
In the present work, we discovered that inhibitory interneurons with predominant expression of HCN1 in CN are particularly susceptible to BIL toxicity. Bilirubin promotes not only the gating of HCN1 on the membrane but also their membrane targeting from an intracellular pool in a Ca2+-dependent manner in these fast-spiking interneurons, resulting in elevated firing rate, Ca2+ overload, and cell death. Our results implicated the HCN1 channel in GABAergic neurons as a key mediator of HB-induced neuronal injury, rationalizing these channels as potentially therapeutic targets to mitigate neonatal brain injury in severe jaundice.
Materials and Methods
Experimental design
The main objective of this study was to explore the vulnerability of auditory neurons to BIL neurotoxicity. For electrophysiology, immunohistochemistry, and animal behavior, sample sizes were chosen based on previous experience, with the minimum number of independent experiments required to reach statistical significance at the 5% level. Sample sizes and statistical tests used are described in detail in the Results section. All behavioral testing was conducted during the day in a quiet room by an experimenter blinded to the identity of the drug treatment group and/or mouse genotype.
Ethical approval
Male and female mice aged P7–P10 and P30–P45 with the same genetic background (C57BL/6J) were used for this study. HCN1−/− (C57BL/6J-HCN1em1Smoc) lacking eight bases (AAATTTGG) were generated and supplied by the Shanghai Nanfang Research Center for Model Organisms. The CRISPR-Cas9 system was used to construct HCN1 knock-out mice: F0 generation mice were determined to be positive by PCR and sequencing and further backcrossed to wild-type C57BL/6J mice. The resulting heterozygous (F1 generation, HCN1+/−) mice were bred to obtain homozygous HCN1−/−.
The homozygous offspring were identified and validated by PCR and sequencing of tail/toe DNA. Inhibitory neuron-specific VGAT-cre-tdTomato reporter mice were obtained by crossing VGAT-ires-cre mice (JAX016962) with Ai14-D mice (Rosa26-tdTomato; JAX 007914). All experimental procedures strictly complied with the relevant ethical regulations and were approved by the Ethics Committee for Animal Experimentation of the Sixth People's Hospital of Shanghai and Shanghai Jiao Tong University.
Preparation of auditory brainstem slices
P7–P10 mice were anesthetized with 1% pentobarbital sodium (0.04 μl/10 g, i.p.) before decapitation. The brain was quickly isolated and transferred to the sectioning chamber and cut into 300 μm slices. All procedures were performed in ice-cold oxygenated (95% O2 and 5% CO2) artificial cerebrospinal fluid (ACSF). The slices containing CN were recovered for 30 min in fully oxygenated ACSF at 37°C and maintained at room temperature (23°C) until use.
Solutions
The ACSF used for recordings from CN neurons in slices contained the following (in mM): 124 NaCl, 5 KCl, 1.2 KH2PO4, 1.3 MgCl2, 2.4 CaCl2, 24 NaHCO3, and 10 glucose. For cell-attached loose-patch recordings, the pipette solution contained the following (in mM): 150 NaCl, 5 KCl, 1 MgCl2, 2 CaCl2, 10 glucose, and 10 HEPES. For whole-cell recording, the pipette internal solution contained the following (in mM): 130 K-gluconate, 10 HEPES, 0.6 EGTA, 5 KCl, 3 Na2-ATP, 0.3 Na3-GTP, 4 MgCl2, and 10 Na2 phosphocreatine. For the patch-clamp experiment in the HEK293-T cell line, the pipette internal solution contained the following (in mM): 130 KCl, 10 NaCl, 1 EGTA, 0.5 MgCl2, 2 ATP, and 5 HEPES-KOH. The extracellular solution contained the following (in mM): 110 NaCl, 30 KCl, 1.8 CaCl2, 0.5 MgCl2, and 5 HEPES-KOH. We routinely included ATP (Na2-ATP 3 mM) and cAMP (20 µM) in the pipette solution to minimize the rundown of Ih, which is particularly evident in CHO cells. In our hands, stable Ih can be recorded from slices for over 30 min without obvious rundown. All our experiments were designed to be completed within this timeframe (Extended Data Fig. 2-1): for example, Ih was recorded twice in the presence of synaptic blockers, with an interval of 3 min, and this was followed by multiple measurements in BIL solution for another 5 min and during washout for another 5–10 min.
All solutions were adjusted to pH 7.2–7.4 and an osmotic pressure of 300 mOsm. BIL was dissolved in dimethyl sulfoxide to prepare a stock solution and diluted in the ACSF perfusion solution to a working concentration of 9 μM just before experiments while the reservoir was shielded from light exposure. Other reagents used included cAMP (20 μM), ZD7288 (40 μM), d-2-amino-5-phosphonopentanoic acid (d-AP-5, 50 μM), 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo quinoxaline-7-sulfonamide (NBQX, 2 μM), bicuculine (10 μM), strychnine (1 μM), exocytosis inhibitor (TAT-NSF 700, 5 μM), endocytosis inhibitor (dynasore, 40 μM), and calcium chelating agent (BAPTA-AM, 40 μM). All reagents were purchased from Sigma.
Electrophysiology
CN neurons in brain sections were visually identified by a near-infrared differential interference contrast microscope (Zeiss Examiner, A1). All recordings were carried out at room temperature using a MultiClamp 700B amplifier (Molecular Devices, MultiClamp 700B). The patch electrodes were pulled from borosilicate capillary glass (1.5 mm outer diameter; 0.86 mm inner diameter; World Precision Instruments) on a horizontal pipette puller (Sutter Instrument, P-1000). When filled with the internal solution, electrodes had a resistance of 2–3 MΩ for cell-attached recording and 3–5 MΩ for recording in the whole-cell configuration. Spontaneous firings (SFs) or intrinsic firings (IFs, in the presence of synaptic blockers) were recorded from the CN neurons by cell-attach configuration. The electrode was placed onto cell soma where a loose-patch was formed by gentle suction with low resistance seal (20–50 MΩ). Data were acquired in current-clamp I0 mode. The stable discharge pattern of neurons in the near-physiological state was first recorded as baseline before the exposed to different solutions illustrated in the main text and figures.
The HCN current and membrane excitability of CN neurons were measured using whole-cell voltage- and current-clamp configurations as described previously (Rodrigues and Oertel, 2006; Kopp-Scheinpflug et al., 2015). Series resistance varied from 4 to 15 MΩ and was compensated by 70–90%. In HEK-293T cells expressing HCN1 or HCN2 channels, whole-cell voltage-clamp recordings were made using an EPC-10 patch-clamp amplifier (HEKA Elektronik). Data were acquired at 10–20 kHz and filtered at 1–3 kHz using a computer equipped with the Pulse 6.0 software (HEKA Elektronik). Patch pipettes were positioned on the cells using a motorized micromanipulator (MP-225; Sutter Instrument). Cells were visualized using an inverted microscope under phase contrast (TE-2000U; Nikon). Only one recording was conducted per slice or cell culture dish before and after BIL application. The HCN currents were recorded using the same paradigm as that used for neurons and cultures.
HCN currents were recorded by voltage step commands over a voltage range from −160 to −60 mV in −10mV increments (Cuttle et al., 2001). Current amplitude was quantified by subtracting the instantaneous current from the steady-state current. The activation curves were constructed by plotting the Ih amplitudes against the command voltages and fitting them with a Boltzmann function:
Transfection of HEK-293T cell lines in culture
The plasmids (mHCN1-pEGFP and mHCN2-pmCherry) contain a murine HCN1 cDNA (NC 000079.7) or a murine HCN2 cDNA (NC 000076.7) with the kanamycin resistance gene and eukaryotic cytomegalovirus promoter. Recombinant plasmids were constructed by Changsha You Bao Biotechnology. HEK-293T cells were cultured in DMEM supplemented with 10% fetal bovine serum and 1% streptomycin/penicillin at 37°C in the atmosphere of 95% and 5% CO2. Cells were transfected using lipofectamine 3000 and OPTI-MEM with 2 μg plasmids per plate at approximately 40% confluency. The culture medium was replaced 6 h after transfection. Patch-clamp recordings were performed at 18−24 h post-transfection. HCN1-pEGFP–positive cells were identified by the green fluorescence of EGFP, and HCN2-mCherry–positive cells were identified by the red fluorescence. Adherent fluorescent cells from individual colonies were selected and recorded. The experimental solutions are detailed in Solutions.
Membrane and cytosol protein extraction and immunoblotting
Fresh and intact brain tissues from three litter-matched P7 HCN1+/+ mice were obtained and meninges were removed (per group). Tissues were treated with different experimental solutions (i.e., ACSF, CTRL, BIL) for 30 min in darkness. The membrane and cytoplasmic proteins were extracted using a Membrane and Cytosol Protein Extraction Kit (Beyotime), and the samples underwent Western blot assays according to the manufacturer's instructions. After sodium dodecyl sulfate–polyacrylamide gel electrophoresis, proteins were transferred to the nitrocellulose membrane and blocked in 5% nonfat milk/TBST buffer for 1 h at room temperature. Then, the membranes were incubated overnight with primary antibodies at 4°C and secondary antibodies at RT for 1 h while avoiding light, with mild shaking. The following primary antibodies were used: anti-HCN1(ab84816, Abcam, mouse, 1:500, 99 kD), anti-HCN2 (ab84817, Abcam, mouse, 1:1,000, 95 kD), and β-actin (AC026, ABclonal, rabbit, 1:100,000, 42 kD). Anti-mouse 800 (IRDye 800CW, Li-Cor, goat, 1:5,000), anti-rabbit 680 (IRDye 680RD, Li-Cor, goat, 1:100,000) fluorescent secondary antibodies were detected with the Odyssey Imaging System (Li-Cor). Fluorescence values were analyzed using Odyssey software. The relative expression of the target protein was evaluated by comparing the gray value of the target protein band to that of the internal reference protein band. The ratio in the control group was set to 1 to normalize those of other groups.
Calcein AM/PI costaining
Brain slices containing CN (100 μm) were obtained using a vibrating blade microtome (Leica VT1200S). The slices were placed into oxygenated ACSF to recover for 20 min at 37°C and then transferred into different testing solutions for 30 min in darkness. Calcein AM (1 μM) and propidium iodide (PI, 1 μM) were added and incubated for 15 min to stain viable and dead cells, respectively (Karászi et al., 2001; Lecoeur, 2002). Thereafter, slices were washed with ACSF three times and fixed with 4% paraformaldehyde for 30 min. Finally, the sections were mounted on glass slides and observed with confocal fluorescence microscopy (Zeiss, LSM-710). The number of stained cells was measured using ImageJ software (one to two fields of view per slice and one to two slices per animal, and each experiment was performed with at least three animals).
Analysis of HCN1 and HCN2 expression at the single-cell level
Single-cell transcriptomic sequencing data from the mouse cortex and basal ganglia (GSE207848) were utilized to analyze HCN1 and HCN2 expression levels across different neurons at the single-cell level. Transcriptomic data analysis and cell clustering were performed in R using the Seurat package. Cell annotations were derived from the Single-R analysis results and literature. The scores for the GABAergic neuron signatures (Gad1 and Gad2) and glutamatergic neuron signatures (Vglut1 and Vglut2) were computed using the Add Module Score function. Pearson's correlation analysis was performed to calculate the correlation between expression levels of the indicated genes and gene signatures.
Construction of HB mouse models
The mouse model of HB was established by an intraperitoneal injection of BIL as previously described (Lai et al., 2020). BIL was prepared in 0.1 M NaOH, and pH was adjusted to 8−8.2 with 1 M HCl. Mice were injected intraperitoneally (i.p.) once daily with BIL (100 μg/g) for 7 d, beginning from postnatal day 3. Saline (100 μg/g) was administered to the control group. Electrophysiological recordings, immunofluorescence staining, and calcein AM/PI costaining experiments were carried out 12–24 h after the final injection. Mouse ABR texts were conducted at P30 ± 5 d.
Immunofluorescence staining
Mice (P10) were anesthetized with 1% pentobarbital sodium (0.04 μl/10 g, i.p.) and decapitated. The intact brain was quickly stripped off and fixed with 4% paraformaldehyde (PFA, SolarBio) for 24 h. Then, brain tissue was dehydrated, embedded in paraffin, and sectioned at 5 μm thickness. The paraffin-embedded sections were dewaxed and subjected to sodium citrate antigen retrieval. Brain sections were treated with an autofluorescence quencher (Servicebio) and blocked for 1 h with a blocking solution (10% goat serum, 1% bovine serum albumin, and 0.5% Triton X-100 in phosphate-buffered saline). The primary and secondary antibodies were diluted to the desired concentration with an antibody signal enhancer solution before using. They include anti-HCN1 (#ab84816, Abcam, mouse, 1:200), anti-HCN2 (#ab84817, Abcam, mouse, 1:200), anti-tdTomato (#ARG55724, Arigo, goat, 1:200), goat anti-mouse IgG (H + L) secondary antibody, Alexa Fluor 488 (Invitrogen, 1:500), donkey anti-goat IgG (H + L) secondary antibody, Alexa Fluor 568 (Invitrogen, 1:500; see Extended Data Table 1-1 for details of antibodies). The sections were sequentially incubated with anti-HCN1/HCN2 primary antibodies and anti-tdTomato antibodies overnight at 4°C and then at room temperature for 2 h with the respective secondary antibodies in order. Finally, the slides were sealed with an antifluorescence quencher and visualized under a confocal fluorescence microscope. Representative images were taken at ×20 objective for counting (Zeiss, LSM-710) and ×100 oil objective and 3D reconstruction for detailed images of HCN1 and HCN2 (Leica X).
DID membrane/HCN costaining
In a subset of experiments, we stained the membrane by incubating tissue sections after secondary antibody labeling with organic dye DID (C1039, Biyuntian, 5 μM) overnight in the dark at 4°C. Profile analyses of membrane and antibody costaining were used to identify changes in HCN channel distribution in the cell membrane and cytoplasm by confocal fluorescence microscope (Zeiss, LSM-710). After choosing the target neuron on the confocal image and marking the line segment, we quantified the fluorescence intensity profile of DID (purple) and HCN (green) signals by line scan. The two peaks identified by the fluorescence intensity spectrum of DID represent the location of the cell membrane while the peak areas of fluorescence intensity of the HCN1 or HCN2 under the curve were integrated using Graph Pad Prism 9.0 software to represent the level of the target protein on the cell membrane (see Fig. 4 for details).
ABR test
Mice (P30–35) were anesthetized with an intraperitoneal injection of ketamine (40 mg/kg) and xylazine (4 mg/kg) and placed on a heated pad to maintain the core body temperature at 37°C. ABR signals were obtained using three needle electrodes: the recording electrode was inserted subcutaneously at the top of the skull vertex, and the reference and ground electrodes were placed under the skin behind the neck on both sides, respectively. These electrodes were connected to a multichannel preamplifier (TDT, RA16PA), recorded with a real-time digital signal processing module (TDT, RZ5), and processed using BioSigRP software (TDT). The ABRs were averaged over 1,000 presentations, and the ABR threshold was defined as the lowest intensity at which noticeable wave Ⅲ could be evoked. ABR responses were evaluated at 1, 2, 4, 8, 16, 20, 24, 32, and 40 kHz. The test was performed with pure tones (10 ms duration, 0.5 ms rise/fall times, 21.1/s rate) and stimulation with a descending intensity starting from 90 dB SPL in 5 dB steps. The amplitudes at different SPL and frequencies and latency of Wave I, Wave II, and Wave III were quantitatively analyzed for statistical comparison. At least three mice were used for each group.
Statistical analysis
All statistical analyses were performed using SPSS software (version 24.0, International Business Machines). Electrophysiological data were acquired and analyzed using Clampfit 10.7, PatchMaster, and Mini Analysis software. Confocal and immunoblotting imaging data were analyzed using Zen Blue Edition (Carl Zeiss) and ImageJ software. All images and illustrations were prepared and assembled using Prism9 (GraphPad), Adobe Illustrator CS6, and Adobe Photoshop. All data were presented as mean ± SD. Independent-samples t test, paired Student’ s t test, one-way ANOVA with LSD post hoc test, mixed-effects analysis, and Tukey's multiple-comparisons test were performed for statistical significance.
Results
Bilirubin causes inhibitory neuronal death by acting on HCN1 channels
Previous studies showed that BIL amplifies the excitability of CN neurons through the modulation of multiple channels, including voltage- and ligand-gated ion channels (Li et al., 2011; Han et al., 2015; Liang et al., 2017a; Shi et al., 2019; Liu et al., 2023), ultimately contributing to neuronal excitotoxicity. However, whether and how the HCN channels, the main pace-making conductance of CN neurons (Pál et al., 2003; Yin et al., 2018), are involved is unknown, as does the cell-type particularly susceptible to cell death. To address these questions, we first performed cell death assays on CN neurons in brainstem slices (100 μm) from HCN1+/+ and HCN1−/− neonatal mice (P5–P10) under the three conditions using (1) ACSF with synaptic blockers as control; (2) ACSF with synaptic blockers plus BIL (9 μM); and (3) ACSF with synaptic blockers plus BIL (9 μM) and ZD7288 (40 μM), a nonspecific inhibitor of HCN channel. By calcein AM and propidium iodide (PI) double staining, we counted live and dead cells, respectively. As shown by the example confocal images in Figure 1A (left), we found that BIL treatment led to the significant accumulation of PI-labeled cells in the HCN1+/+ groups, but the addition of ZD7288 reduced PI-positive cells. The percentage of cell death measured post hoc by ImageJ (Fig. 1B) indicated that the increased mortality induced by BIL in slices from HCN1+/+ mice was inhibited by ZD7288 [CTRL (n = 8), 20.87 ± 3.08%; BIL (n = 7), 50.21 ± 3.06%; ZD (n = 6), 23.32 ± 3.18%; F(2, 18) = 195.0, p < 0.0001; ANOVA; Tukey's multiple comparisons, PCTRL-BIL < 0.0001, PBIL-ZD7288 < 0.0001, PCTRL-ZD7288 = 0.331]. In contrast, the PI-labeled cells and the mortality had no marked difference between CTRL and BIL-treated slices from HCN1−/− mice [Fig. 1A, right, B; CTRL (n = 9), 14.33 ± 1.91%; BIL (n = 6), 17.80 ± 5.12%; ZD (n = 11), 11.95 ± 1.83%; F(2,23) = 7.93, p = 0.002; Tukey's multiple comparisons, PCTRL-BIL = 0.082, PBIL-ZD7288 = 0.002, PCTRL-ZD7288 = 0.182). These results suggested that HCN1 channels are necessary for mediating BIL-induced increase in neuronal death.
BIL exacerbated inhibitory neuronal death by acting on HCN1 channels. A, Representative confocal images of calcein AM/PI staining of CN neurons in HCN1+/+ (left panels) and HCN1−/− brain slices (right panels). Live cell (calcein AM, green), dead cell (PI, purple). Note that BIL led to the accumulation of PI-labeled cells in HCN1+/+ slices, but this change was attenuated by ZD7288 or deletion of HCN1. B, The bar chart summarizes the mortality rates induced by BIL in slices from HCN1+/+ and HCN1−/− mice under different conditions. HCN1+/+: CTRL (n = 8), 20.87 ± 3.08%; BIL (n = 7), 50.21 ± 3.06%; ZD (n = 6), 23.32 ± 3.18%; F(2, 18) = 195.0, p < 0.0001, ANOVA; Tukey's multiple-comparisons, PCTRL-BIL < 0.0001, PBIL-ZD7288 < 0.0001, PCTRL-ZD7288 = 0.331. HCN1−/−: CTRL (n = 9), 14.33 ± 1.91%; BIL (n = 6), 17.80 ± 5.12%; ZD (n = 11), 11.95 ± 1.83%, F(2,23) = 7.93, p = 0.002; Tukey's multiple comparisons, PCTRL-BIL = 0.082; PBIL-ZD7288 = 0.002, PCTRL-ZD7288 = 0.182. C, Representative confocal images of inhibitory CN neurons costained with calcein AM in VGAT-cre-tdTomato mice with (right) or without bilirubin exposure (left panels). See Extended Data Figure 1-1 for more details. D, Statistical plot of the percentage of tdTomato and calcein AM cotarget neurons among tdTomato-positive neurons from (C). BIL decreased the costaining of calcein AM and VGAT-cre-tdTomato neurons in VGAT-cre-tdTomato mice [CTRL (n = 11), 30.17 ± 9.15%; BIL (n = 11), 14.42 ± 5.78%, t(20) = 4.825, PCTRL-BIL = 0.001, independent-samples t test]. E, Pie chart depicting the percentage of tdTomato-positive cells (25/28) with electrophysiological characteristics consistent with stellate cells in VGAT-cre-tdTomato mice. F, Example traces of membrane potential responses (injected with −300 to +500 pA) and Ih of tdTomato-positive (up, purple) and tdTomato-negative cells (down, black). G, Representative immunofluorescence staining of HCN1 (anti-HCN1) or HCN2 channels (anti-HCN2) of inhibitory neurons in VGAT-cre-tdTomato mice. H, Confocal images of cochlear nucleus neurons in VGTA mice colabeled by HCN1 and HCN2 antibodies. Note distinct patterns of HCN1 and HCN2 labeling of neurons, some of which express only HCN1 (indicated by the white arrow, upper row). Even in neurons labeled by both antibodies, high-power mirror scanning reveals distinct sublocalization of HCN1 and HCN2 distribution within the same neuron (bottom). Error bars represent means ± SD; one-way ANOVA with LSD post hoc test, independent-samples t test. See Extended Data Table 1-1 for more details.
Table 1-1
Antibody and dye information. Download Table 1-1, DOCX file.
Figure 1-1
Single-Cell Level Expression Analysis of HCN1 and HCN2. (A) Transcriptomic data analysis and cell clustering were performed in R using the Seurat package. Neurons were divided into GABAergic and glutamatergic. (B-C) Scores for GABAergic neuron signatures (GAD1, GAD2) and glutamatergic neuron signatures (vglut1, vglut2) were computed using the Add module score function. (D) Pearson correlation analysis was performed to calculate the correlation between the expression of indicated genes and signatures. (E-F) Plots showing the relative expression of GABAergic neuron signatures and glutamatergic neuron signatures in HCN1 [t (23528) = 25.69, P < 0.0001] or HCN2 [t (11115) = 25.29, P < 0.0001] dominated neurons. Download Figure 1-1, TIF file.
Figure 2-1
Summary of electrophysiological properties of other CN neurons. (A) Time courses of firing frequency of cochlear nucleus neurons under different conditions as labeled. The firing frequency significantly increased with 9 µM bilirubin and remained stable after 5 minutes of perfusion. (B) No sex difference for bilirubin effects in C57 and VGAT cre-td Tomato mice was observed. (C) Bilirubin had the same effect on both C57 and VGAT cre-td Tomato mice. (D) Other types of neurons were not sensitive to bilirubin. Typical raw data of membrane potential response evoked by current injection (left) and Ih amplitude before and after bilirubin treatment (right). (E) Summary plot showing a lack of changes of Ih amplitude by bilirubin among different types of CN neurons. Download Figure 2-1, TIF file.
To distinguish whether bilirubin-induced cell death varies between excitatory and inhibitory neurons, we employed VGAT-cre-tdTomato reporter mice in which all GABAergic neurons are fluorescence tagged. We performed calcein AM staining in CN slices from VGAT-cre-tdTomato mice slices before and after BIL treatment (Fig. 1C) and found that BIL significantly decreased calcein AM costained tdTomato-positive cells. The percentage of colabeled neurons measured in ImageJ showed that BIL reduced their viability by half [CTRL (n = 11): 30.17 ± 9.15%, BIL (n = 11): 14.42 ± 5.78%, t(20) = 4.825, PCTRL-BIL = 0.001, independent-samples t test, Fig. 1D]. These results prompted us to reanalyze single-cell transcriptomic sequencing data (Allen et al., 2023). We found that the expression of the HCN1 gene was positively correlated with signatures of inhibitory neurons and negatively with excitatory neurons, whereas HCN2 showed a weak correlation (Extended Data Fig. 1-1). This implies that the sensitivity of inhibitory neurons to BIL may be linked to the expression of the HCN1 channel. To test this, we performed patch-clamp recordings from these tdTomato-positive cells, and most of the cells [82%(25/28), Fig. 1E] were consistent with the electrophysiological characteristics of stellate cells (SCs; Schwarz and Puil, 1997; Cao et al., 2008; Cao and Oertel, 2017; Xie and Manis, 2017; Müller et al., 2019), the main inhibitory interneurons in CN (Fig. 1F), including high-frequency repetitive firing in response to depolarizing current steps and large HCN-mediated sags and rebound potentials to hyperpolarizing current injections in current-clamp configuration (−300 to +500 pA, 100 pA increment, from a holding membrane potential of ∼ −65 mV). Voltage-clamp recordings further showed that hyperpolarizing voltage steps (holding potential −60 to −150 mV in −10 mV increment, 4.5 s) evoked Ih currents with fast activation time course and the voltage dependence, reminiscent of those of HCN1 channels (Biel et al., 2009). The tdTomato-negative cells exhibited completely different electrophysiological characteristics from the positive ones in the same slices, as shown in Figure 1F. To validate this, we employed VGAT-cre-tdTomato reporter mice to first localize HCN1 and HCN2 in CN. Immunofluorescence staining of brain slices was performed using HCN1/HCN2 antibodies and tdTomato fluorescent protein antibodies. Images (Fig. 1G) showed neurons stained for HCN1/HCN2 protein in green, while the tdTomato labeling (magenta) marked the VGAT-positive neurons in VGAT-cre-tdTomato reporter mouse. We found a high overlap of HCN1 protein and tdTomato fluorescent protein in CN neurons, where the colabeling of HCN2 protein was low (Fig. 1G). These results suggested that HCN1 is indeed more preferentially expressed in inhibitory neurons than HCN2. Furthermore, confocal imaging and analyses revealed that some neurons expressed both HCN1 and HCN2 proteins, while others only expressed HCN1 (indicated by the white arrow in Fig. 1H). High-resolution scans and 3D reconstructions of neurons with apparent coexpression of HCN1 and HCN2 at 100× magnification showed different sublocalizations of these two proteins (Fig. 1H), suggesting HCN1 and HCN2 unlikely form heteromeric channels in CN neurons. These results lead us to postulate that BIL preferentially exacerbates the death of inhibitory neurons, largely SCs with dominant expression of HCN1 channels.
BIL elevates the intrinsic excitability and Ih amplitude of SCs by targeting HCN1 channels
Given that SCs represented >80% of inhibitory neurons that appeared to be particularly vulnerable to BIL neurotoxicity, we used a K+-based intracellular solution to first identify SCs from CN slices in current mode and examine how BIL affects the magnitude and kinetics of Ih mediated by native HCN channels in this subset of inhibitory neurons in slices from HCN1+/+ and HCN1−/− mice. Whole-cell voltage-clamp recordings of Ih from CN cells were made in the presence of synaptic blockers. Following establishing whole-cell configuration, we evoked Ih by a series of hyperpolarized voltage steps using the same paradigm (as in Fig. 1H). As exemplified in Figure 2A,B, BIL significantly enhanced Ih in SCs from HCN1+/+ mice (−508.35 ± 268.51 pA vs −644.54 ± 294.46 pA, N = 13, p = 0.001, Fig. 2C), and the effect cannot be easily washed out (−714.17 ± 242.48 pA, N = 6, Tukey's multiple-comparisons test). In contrast, BIL induced no changes in HCN1−/− SCs (−260.37 ± 126.56 pA vs −261.74 ± 124.15 pA vs −354.00 ± 124.81 pA, N = 14, 14, 6; Fig. 2E). No statistical differences in time constant (τ) from exponential fits to activation time courses of evoked Ih (Fig. 2D,F) were identified between before and after BIL solutions perfusion in all groups (p > 0.05, Tukey's multiple-comparisons test). Note that activation time constants of Ih appeared to be much slower (by ∼10-fold) in SCs from HCN1−/− than those from HCN1+/+ mice, implicating a partial compensation by HCN2 which is known for slower gating kinetics than HCN1. As a result, the intrinsic excitability of SCs was substantially reduced in HCN1−/− slices as shown in Figure 2A,B. BIL increased the number of spikes induced by depolarization injection current (+100 pA) in SCs from HCN1+/+ mice [CTRL, 11.08 ± 6.71 (N = 13); BIL, 14.77 ± 5.8 (N = 13); wash, 11.83 ± 4.75 (N = 6)], but had no effect on SCs from HCN1−/− mice [CTRL, 8.3 ± 6.85 (N = 10); BIL, 8.7 ± 6.82 (N = 10); wash, 7 ± 2 (N = 3)]. The sex of mice had no effect on the experimental results (Extended Data Fig. 2-1). We have also examined the effects of BIL on other cell types (e.g., bushy, fusiform, cartwheel, and tuberculoventral cells in CN on the basis of their electrophysiological signatures) and found that BIL exerted no significant impact on Ih (Extended Data Fig. 2-1). In summary, BIL enhanced the intrinsic excitability by specifically enlarging Ih currents mediated by HCN1 in SCs, and this effect was ablated in SCs from HCN1−/− mice. These results demonstrated that the HCN1 channel is necessary for mediating hyperexcitability of the vast majority of inhibitory neurons by BIL in neonatal CN.
BIL elevates the Ih amplitude of CN neurons by targeting the HCN1 channel. A, B, Typical raw traces of membrane potential responses (left) and Ih (right) in stellate cells of HCN1+/+ and HCN1−/− mice before (black) and after (orange) BIL treatment and wash (pale yellow). C, E, Summary data showing that BIL elevated the Ih amplitude of Stellate neurons in HCN1+/+mice (−508.35 ± 268.51 pA vs −644.54 ± 294.46 pA, N = 13, p = 0.001) and the effect cannot be washed (−714.17 ± 242.48 pA, N = 6, Tukey's multiple-comparisons test), but had no significant effect on HCN1−/− mice (−260.37 ± 126.56 pA vs −261.74 ± 124.15 pA vs 354.00 ± 124.81 pA, N = 14,14,6). The diamonds show female data, while the circles show male data. D, F, Summary data showing that BIL did not affect the τ of Ih activation time courses in either HCN1+/+ (CTRL vs BIL vs wash, τ −120: 171.83 ± 109.7 ms vs 190.53 ± 128.22 ms vs 226.15 ± 130.89 ms, p = 0.304; τ −130: 135.51 ± 69.75 ms vs 137.41 ± 63.01 ms vs 143.27 ± 64.25 ms, p = 0.188; τ −140: 99.8 ± 23.93 ms vs 108.86 ± 36.43 ms vs 120.13 ± 42.09 ms, p = 0.212, N = 8, 8, 4, mixed-effects analysis) or HCN1−/− mice [τ −120: 864.74 ± 360.41 ms vs 861.71 ± 425.66 ms vs 1014.2 ± 334.07 ms, p = 0.269; τ −130: 534.46 ± 213.87 ms vs 598.62 ± 267.12 ms vs 785.49 ± 304.82, p = 0.048; τ −140: 412.94 ± 133.8 ms vs 403.88 ± 139.82 ms vs 469.94 ± 112.85 ms, p = 0.358, N = 13, 13, 6, mixed-effects analysis]. Raw traces (G, I) and summary data (H, J) showed the changes of spike number induced by +100 pA depolarization step in different perfusion fluids of HCN1+/+-mice [CTRL, 11.08 ± 6.71 (N = 13); BIL, 14.77 ± 5.8 (N = 13); wash, 11.83 ± 4.75 (N = 6)] and HCN1−/− mice [CTRL, 8.3 ± 6.85 (N = 10); BIL, 8.7 ± 6.82 (N = 10); wash, 7 ± 2 (N = 3)]. Error bars represent means ± SD; paired Student’s t test; mixed-effects analysis, Tukey's multiple-comparisons test. See Extended Data Figure 2-1 for more details.
BIL increases Ih mediated by the HCN1 but not the HCN2 channel
Given faster activation kinetics of Ih and stronger sensitivity to BIL in SCs than other types in HCN1 genotype-specific manner, we postulated that HCN1 channels may be the dominant isoform on which BIL acts to upregulate HCN1 activity and/or level and boost the neuronal excitability. In contrast, HCN2 is more likely associated with other cell types and insensitive to BIL, as supported by IHC results (Fig. 1D). To directly test HCN isoform-specific sensitivity, we employed the HEK-293T cell line transiently transfected with human HCN1 or HCN2 channel as a reduced system to study the effects of BIL independent other confounding factors in neurons. We performed patch-clamp recordings from these cells as illustrated by example Ih current traces at −150 mV after HCN1 or HCN2 overexpression (Fig. 3A,D). We found opposite effects of BIL on Ih mediated by two isoforms: the HCN1 Ih amplitude was increased (CTRL, −1,373.9 ± 825.30 pA; BIL, −1,662.71 ± 992.19 pA; N = 12, t(11) = 3.812, p = 0.003, paired t test, Fig. 3B,C), but the HCN2 Ih amplitude was decreased (CTRL, −1,537.10 ± 1,128.25 pA; BIL, −1,308.84 ± 951.88 pA; N = 11, t(10) = 3.441, p = 0.006, paired t test, Fig. 3E,F). This was also supported by opposite changes in the direction of the maximal Ih of HCN1 verse HCN2 in their activation curves before and after BIL (Fig. 3B,E). Notably, Ih by HCN1 channels treated with BIL displayed a right shift toward more depolarized potentials as indicated by V50 values (CTRL, −103.43 ± 10.15 mV; BIL, −99.96 ± 10.37 mV; N = 10, t(9) = 4.406, p = 0.002, paired t test, Fig. 3B), whereas Ih by HCN2 channels shifted in the hyperpolarized direction (CTRL, −119.03 ± 4.65 mV; BIL, −126.61 ± 7.18 mV; N = 11, t(10) = 5.336, p < 0.0001, paired t test, Fig. 3E). The τ values of Ih activation by HCN1 under different step voltages (e.g., −130 mV, −140 mV, −150 mV) were shortened, and those of Ih by HCN2 were lengthened after BIL treatment (p < 0.05, Fig. 3G,H). Again, these values differed by an order of magnitude, in line with those measured from native SCs in HCN1+/+ and HCN1−/−. These observations indicated that BIL increases Ih amplitude of HCN1 channels and accelerates their activation but does the opposite to Ih mediated by HCN2 channels in HEK-293T cell lines.
BIL increased the Ih of HCN1 but not the HCN2 channel in HEK-293T cell line. A, D, Typical traces of HCN1/HCN2 currents recorded at a membrane potential of −150 mV from HEK-293T cell transient transfected with HCN1/HCN2 before and after BIL application were superimposed to show changes in amplitude and kinetics of Ih. B, E, The activation curves of the HCN1 channel were shifted to more depolarized potentials (CTRL, −103.43 ± 10.15 mV; BIL, −99.96 ± 10.37 mV, N = 10, t(9) = 4.406, p = 0.002), but the activation curves of HCN2 channel were shifted to more hyperpolarized potentials (CTRL, −119.03 ± 4.65 mV; BIL, −126.61 ± 7.18 mV, N = 11, t(10) = 5.336, p < 0.0001). C, F, The summary data showing changes in the Ih amplitude (at −150 mV) before and after perfusing BIL. BIL increased the HCN1 current (CTRL, −1,373.9 ± 825.30 pA; BIL, −1,662.71 ± 992.19 pA, N = 12, t(11) = 3.812, p = 0.003) but decreased the HCN2 current amplitude (Ih, −1,537.10 ± 1,128.25 pA vs −1,308.84 ± 951.88 pA, N = 11, t(10) = 3.441, p = 0.006). G, H, The tau values (ms) of HCN1 channel were shortened after perfusion BIL (τ −130: 223.60 ± 64.61 ms vs 187.29 ± 49.83 ms, N = 9, t(8) = 3.966, p = 0.004; τ −140: 213.25 ± 68.50 ms vs 172.36 ± 49.31 ms, N = 9, t(8) = 4.763, p = 0.001; τ −150: 179.63 ± 56.88 ms vs 153.32 ± 46.86 ms, N = 9, t(8) = 5.323, p = 0.001], but the tau of HCN2 channel was increased by BIL [τ −130: 1,702.17 ± 328.02 ms vs 2,422.59 ± 990.60 ms, N = 8, t(7) = 2.434, p = 0.045; τ−140: 1,326.66 ± 294.71 ms vs 1,968.08 ± 850.18 ms, N = 10, t(9) = 3.460, p = 0.007; τ −150: 1,059.55 ± 228.58 ms vs 1,378.18 ± 407.50 ms, N = 9, t(8) = 3.724, p = 0.006). Error bars represent means ± SD; paired Student’s t test.
Acute exposure to BIL enhances the trafficking of HCN1 from the cytosol to the membrane
Because BIL appeared to affect both the amplitude of Ih mediated by HCN1 channels and their voltage dependence, we postulated that the membrane targeting of the HCN1 channel be a potential pathway for BIL to upregulate Ih. To this end, we first tested reagents that are known to block the insertion and internalization of membrane proteins. We pretreated brainstem slices with Dynasore, a dynamin inhibitor that blocks the membrane proteins internalization, or TAT-NSF 700, a fusion polypeptide of N-ethyl-maleimide sensitive factor (NSF) inhibitor that blocks insertion of membrane proteins from the cytosolic pool, for 30 min and then recorded Ih from SCs before and after BIL (5 min). Interestingly, TAT-NSF 700 prevented the increase of Ih induced by BIL (CTRL, −309.34 ± 253.75 pA; BIL, −336.47 ± 263.39 pA; N = 11, t(10) = 1.623, p = 0.136, paired t test, Fig. 4A,B), while Dynasore failed to block the effect (CTRL, −707.26 ± 552.67 pA; BIL, −808.38 ± 616.48 pA; N = 12, t(11) = 4.411, p = 0.001, paired t test, Fig. 4A,B). It is well known that BIL can increase intracellular Ca2+ release and augment Ca2+ influx, both of which participate in its neurotoxicity (Liang et al., 2017a, b). To address whether Ca2+ participates in the membrane targeting of HCN1 channels by BIL, we pretreated the slices with BAPTA-AM to chelate intracellular Ca2+ before BIL application and found upregulation of Ih in SCs was blocked (CTRL, −598.13 ± 266.14 pA; BIL, −577.75 ± 281.59 pA; N = 12, t(11) = 0.882, p = 0.397, paired t test, Fig. 4A,B). These results suggested that BIL acutely promotes HCN1 channel membrane targeting from its cytosol pool in a calcium-dependent manner.
BIL increased the Ih amplitude by facilitating HCN1 protein trafficking to the plasma membrane in a calcium-dependent manner. A, Example traces of Ih (−150 mV) of CN neurons before and after BIL in slices pretreated with TAT-NSF700, Dynasore, or BAPTA-AM. B, The summary data of three sets of experiments in A. TAT-NSF 700 pretreated group: CTRL, −309.34 ± 253.75 pA; BIL, −336.47 ± 263.39 pA, N = 11, t(10) = 1.623, p = 0.136. Dynasore pretreated group: CTRL, −707.26 ± 552.67 pA; BIL, −808.38 ± 616.48 pA, N = 12, t(11) = 4.411, p = 0.001. BAPTA pretreated group: CTRL, −598.13 ± 266.14 pA; BIL, −577.75 ± 281.59 pA, N = 12, t(11) = 0.882, p = 0.397. Paired Student’s t test. C, Western blots of HCN1 and HCN2 proteins from the same tissue extracted by membrane protein extraction kit after different treatments. D, E, Summary plots from densitometric analyses for HCN1 and HCN2 channel expression in membranes and cytoplasm in (C). F–I, Typical confocal image showing colabeling of HCN channels [HCN1 (F) or HCN2 (H)] and cell membrane fuel DID. G, I, Fluorescence intensity profiles of the DID and HCN signals along the white line between the groups [with (middle) or without (left) BIL] in F and H. The two peaks on the purple fluorescence intensity spectrum mark the location of the cell membrane where the integrated green fluorescence density of HCN1 or HCN2 under the curves are assigned as membrane-targeted HCN1 or HCN2 proteins in the CTRL (G, right) and BIL group (I, right). Error bars represent means ± SD, independent-samples t test, Tukey's multiple-comparisons test.
To further validate that BIL upregulates the HCN channel expression levels at the membrane, we performed western blots to separate membrane and cytosolic HCN1 and HCN2 protein from CN slices using a Membrane and Cytosol Protein Extraction Kit (Y. Yu et al., 2022). As depicted by representative images in Figure 4C, the total level of HCN1 channels in the membrane and cytoplasm was not apparently different in ACSF with or without synaptic blockers, but the membrane fraction of HCN1 protein was increased, and the cytosolic fraction was decreased following the treatment with BIL. On the contrary, HCN2 channel proteins were downregulated in the membrane fraction and upregulated in the cytosolic fraction (Fig. 4C), in line with the results from patch-clamp recordings. Using β-actin as the internal reference, we normalized expression levels of membrane and cytoplasmic proteins to quantitatively evaluate the changes in their subcellular redistribution. We found synaptic blockers did not significantly alter the distribution of HCN1 protein on the membrane and in the cytoplasm [94 ± 18% (p = 0.793) and 93 ± 17% (p = 0.687) of that ACSF solution, n = 4). However, BIL treatment significantly elevated the membrane level (208 ± 65% of that in CTRL, n = 4, p = 0.035) and slightly decreased in the cytoplasm (∼82 ± 9% that in CTRL, n = 4, p = 0.176; Fig. 4D). On the other hand, the membrane level of HCN2 channel protein decreased (76 ± 22% of that in ACSF, n = 4, p = 0.217) while cytoplasmic level increased (137 ± 16% of that in ACSF, n = 4, p = 0.041) by synaptic blockers and further decreased by bilirubin [to 96 ± 16% (p = 0.202) and 112 ± 7% (p = 0.038) of that in CTRL, n = 4, Tukey's multiple-comparisons test; Fig. 4E).
In a separate set of experiments (Fig. 4F–H), we performed membrane dye DID and HCN channel immunofluorescence costaining to consolidate the findings from immunoblotting. Profile analyses were used to identify changes in HCN channel level in the proximity of the cell membrane. After choosing the target neuron on the confocal image, we performed line scan analyses of the fluorescence intensity profile of DID (purple) and HCN (green) signals (Fig. 4G,H). Using the two peaks of the DID fluorescence intensity spectrum (green) to first mark the location of the cell membrane, we then integrated the areas of the HCN1 or HCN2 fluorescence intensity curves in CTRL and BIL groups. Statistical analysis showed that the integrated fluorescence density of HCN1 on the cell membrane increased after BIL treatment (70.53 ± 17.71 vs 130 ± 38.37 μm*fu, n = 15, p < 0.0001, Fig. 4G) but HCN2 decreased (99.95 ± 54.45 vs 65.05 ± 23.33, n = 20, p = 0.009, unpaired t test, Fig. 4I). Despite limited resolution of confocal microscopy, these results supported the interpretation that BIL promotes HCN1, but not HCN2 channel membrane targeting.
Hyperbilirubinemia primes hearing loss in vivo
To extend our findings from acute actions of BIL in in vitro studies, we examined the effects of chronic BIL in vivo by generating a mouse HB model for which repeated injections of BIL preceded brain slice experiments [100 µg/g for 7 d (P3–P9), i.p.]. Control mice received saline injections instead. During the modeling process, yellow skin was observed in most HB mice (first in the abdomen, spread to the subcutaneous neck, and finally in all the limbs), and growth and development were slowed down as seen by a slower rate of weight gain than the saline group. These HB phenotypes were more apparent in HCN1+/+ mice than HCN1−/− mice, and in fact, some HCN1+/+ HB mice died (2/11) before completing the BIL injection regime.
We first characterized the basic electrophysiological properties of CN neurons in HCN1+/+-HB mice (at P10) with noninvasive cell-attached configuration in the presence of synaptic blockers. We found a significant fraction of CN neurons in HCN1+/+-HB mice showed intrinsic firings [i.e., IF (+)], while others were silent [i.e., IF (−)]. However, a higher proportion of IF (+) neurons was found in HCN1+/+ HB mice than saline-treated controls [saline 35.14% (39/111) vs HB 56.52% (73/129), Fig. 5A], but there was no obvious difference between the two groups in HCN1−/− mice [saline 10.5% (8/76) vs HB 15.8% (3/19), Fig. 5B]. The HCN1+/+ HB group showed a shorter interevent interval (IEI) [saline (n = 26) 144.53 ± 1.16 ms vs HB (n = 40) 83.14 ± 9.32 ms, t(64) = 3.513, p < 0.001, independent-samples t test, Fig. 5C] with a higher frequency of IFs [saline 7.61 ± 6.26 spikes/s (n = 24) vs HB 13.06 ± 7.44 spikes/s (n = 37), t(59) = 2.972, p = 0.004, independent-samples t test, Fig. 5E] than the saline group, whereas the HCN1−/− group exhibited the opposite trend [IEI: saline (n = 8) 176.75 ± 75.52 ms vs HB (n = 3) 185.00 ± 22.24 ms, Mann–Whitney U = 6, p = 0.279. IFF: saline (n = 8) 7.42 ± 5.47 spikes/s vs HB (n = 3) 3.00 ± 2.00 spikes/s, Mann–Whitney U = 6, p = 0.279]. Raw traces of IFs (Fig. 5G,H), Ih, and membrane potential responses evoked by current injection (Fig. 5I,J) of the HB and saline groups in HCN1+/+ and HCN1−/− mice were shown. The electrophysiological characteristics of HB and saline model mice are similar to their respective control (HCN1+/+ or HCN1−/−). In HCN1+/+ model mice, also a majority of these IF (+) neurons (∼80%) were SCs, but the HCN1+/+ HB group shows a larger Ih [saline (n = 19) −466.08 ± 341.53 pA vs HB (n = 27) −651.25 ± 501.46 pA, t(44) = 2.436, PIh = 0.019, independent-samples t test, Fig. 5I] and faster sag (Fig. 5I) compared with the HCN1+/+ saline group, whereas the HCN1−/− model groups exhibit smaller currents [saline (n = 19) −338.22 ± 217.55 pA vs HB (n = 19) −195.33 ± 181.38 pA, t(38) = 2.256, PIh = 0.030, independent-samples t test, Fig. 5J] with shallower and slower sag (Fig. 5J). These results indicated that CN neurons in HCN1+/+ HB mice exhibit elevated intrinsic excitability.
In vivo upregulation of neuronal excitability, Ih, and mortality of CN neurons in HB mouse models. A, B, A bar graph summarizing the percentages of IF (+) and IF (−) neurons in the saline and HB groups in HCN1+/+ mice [IF (+): 35.1% (39/111) vs 56.5% (73/129)] and HCN1−/− mice [IF (+): 10.5% (8/76) vs 15.8% (3/19)]. C, D, Representative curves showing the IEI of the HB model and saline groups in HCN1+/+ or HCN1−/− mice. Compared with the saline group, IEI was shortened [saline (n = 26) 144.53 ± 1.16 ms vs HB (n = 40) 83.14 ± 9.32 ms, t(64) = 3.513, p < 0.001] in CN neurons from the HCN1+/+ HB mice and appeared to be prolonged in HCN1−/− HB mice [saline (n = 8) 176.75 ± 75.52 ms vs HB (n = 3) 185.00 ± 22.24 ms, Mann–Whitney U = 6, p = 0.279]. E, F, The frequency of IFs was increased in HCN1+/+ mice [saline 7.61 ± 6.26 spikes/s (n = 24) vs HB 13.06 ± 7.44 spikes/s (n = 37), t(59) = 2.972, p = 0.004], but there was no significant change in HCN1−/− group [saline (n = 8) 7.42 ± 5.47 spikes/s vs HB (n = 3) 3.00 ± 2.00 spikes/s, Mann–Whitney U = 6, p = 0.279]. G, H, Raw traces of IFs of the HB and saline groups in HCN1+/+ and HCN1−/− mice. I, J, The typical traces of Ih and membrane potential response were recorded in four groups. K, L, Summary plots showing the Ih amplitude in HCN1+/+ mice [saline (n = 19) −466.08 ± 341.53 pA vs HB (n = 27) −651.25 ± 501.46 pA, t(44) = 2.436, PIh = 0.019] and HCN1−/− mice [saline (n = 20) −338.22 ± 217.55 pA vs HB (n = 20) −195.33 ± 181.38 pA, t(38) = 2.256, PIh = 0.030]. M, N, Typical confocal images of calcein AM/PI staining of CN neurons in saline and HB mice (E). The PI-positive cells markedly increased in HCN1+/+ HB mice [saline (n = 6) 15.18 ± 2.96%, HB (n = 6) 62.86 ± 4.74%, t(10) = 20.88, p < 0.001], but not in HCN1−/− saline mice or HCN1−/− HB mice [saline (n = 14) 12.19 ± 2.77%, HB (n = 9) 14.41 ± 5.04%, t(21) = 1.367, p = 0.186]. Error bars represent means ± SD, independent-samples t test.
Calcein AM and PI double staining were also used to evaluate the viability of CN neurons in slices from HB mice. The PI-positive cells and mortality were markedly increased in CN sections from HCN1+/+ HB mice [saline (n = 6) 15.18 ± 2.96%, HB (n = 6) 62.86 ± 4.74%, t(10) = 20.88, p < 0.001, independent-samples t test, Fig. 5G,H]. However, such differences were absent between HCN1−/− saline mice and HCN1−/− HB mice [saline (n = 14) 12.19 ± 2.77%, HB (n = 9) 14.41 ± 5.04%, t(21) = 1.367, p = 0.186, Fig. 5M,N). These results showed that BIL robustly exacerbates the CN neurons’ death, particularly SCs with upregulated HCN1 channels in vivo.
Clinically, children with bilirubin encephalopathy are prone to ABR abnormalities and even permanent, irreversible hearing loss; therefore, ABR tests were used to assess the long-term impact of HB on the hearing function of mouse models when they reach >30 d. Representative ABR waveforms were shown in Figure 6A, and the ABR thresholds at 1, 2, 4, 8, 16, 20, 24, 32, and 40 kHz in HCN1+/+ saline and HCN1+/+ HB mice were shown in Figure 6B. HCN1+/+ HB mice endured high-frequency hearing loss as characterized by elevated ABR thresholds in 32 and 40 kHz than HCN1+/+ saline mice (32 kHz, 60.83 ± 8.61 dB SPL vs 77.5 ± 11.58 dB SPL, p = 0.007; 40 kHz, 65 ± 0.00 dB SPL vs 79.38 ± 9.80 dB SPL, p = 0.006; independent-samples t test, Fig. 6B), but there were no significant ABR threshold changes at other frequencies. Analyses of Wave I, Wave II, and Wave III amplitudes at different SPL and frequencies as well as their latency were carried out from typical traces as shown in Figure 6C. The ABR amplitude was diminished while the latency of the ABR waveforms was prolonged in HCN1+/+ HB mice with the boundary of Waves III, IV, and V being less clear, forming waveform fusion partially. Measurements of latency and the amplitudes of different waves of ABRs revealed no significant differences across various sound pressure levels and frequencies (Extended Data Table 6-1, p > 0.05, n = 6). Other than 32 or 40 kHz, ABR thresholds and other parameters at all other frequencies tested were similar between the HCN1−/− saline and HCN1−/− HB mice with subtle differences (Fig. 6B, p > 0.05, n = 3). Taken together, these results indicated that the HCN1 channel is a key mediator of high-frequency hearing abnormalities in hyperbilirubinemia.
HCN1+/+ but not HCN1−/− mice displayed hearing loss after HB priming. A, Typical ABR traces of HCN1+/+ saline and HCN1+/+ HB mice are shown. B, ABR thresholds were selectively elevated in the HCN1+/+ HB mice (solid orange circles) with a 20 dB hearing loss above 32 kHz [saline (6), 60.83 ± 8.61 dB SPL; HB (12), 77.50 ± 11.58 dB SPL, t(16) = 3.103, p = 0.007] and 40 kHz [saline (6), 65.00 ± 65 dB SPL; HB (12), 79.38 ± 9.80 dB SPL, t(11) = 3.226, p = 0.008] compared with HCN1+/+ saline. ABR thresholds at all frequencies tested were similar between the HCN1−/− saline mice and HCN1−/− HB mice (p > 0.05). C, The representative ABR traces at 16 kHz and 105 dB SPL from the HCN1+/+ saline mice (black) and for the HCN1+/+ HB mice (orange) are superimposed to show a diminishment in amplitude and latency of the ABR waveforms in HCN1+/+ HB mice. See Extended Data Table 6-1 for more details.
Table 6-1
Analysis of wave amplitude and latency of ABRs in HCN1+/+ and HCN-/- mice before (A) and after bilirubin priming (B). Download Table 6-1, DOCX file.
Discussion
Central neurotoxicity of bilirubin is associated with intricate signaling cascades such as excitotoxicity (Yin et al., 2015; Liang et al., 2017b), calcium overload (Liang et al., 2017a), oxidative stress (Fernandes et al., 2009), inflammation (Brites, 2012), and cell death (Ostrow et al., 2004; Watchko, 2006; Feng et al., 2018). In this study, we discover that GABAergic inhibitory neurons are particularly vulnerable to HB-induced toxicity, resulting in long-term impairments of ABRs. Most of these inhibitory neurons are fast-spiking SCs that express high levels of HCN1 channels and drive rhythmic firings as intrinsic pacemaker conductance. Bilirubin increases Ih potentially by promoting the HCN1 channel gating and its calcium-dependent targeting of the plasma membrane, both of which likely contribute to the hyperexcitability phenotype of interneurons and their early death. A loss of inhibition in the developing brain at the neonatal stage may ultimately exacerbate global neuronal injury by priming long-term neurological impairments as evidenced by auditory impairments in ABRs in vivo when mice reach the young adult stage. Our findings suggest that intrinsically active pace-making conductance HCN1 channel is one of the key molecular substrates for bilirubin to underpin early excitation and inhibition imbalance, fueling a vicious cycle of neuronal hyperexcitability, Ca2+ overload, and neurotoxicity relevant to hearing loss and other forms of brain injury in patients with bilirubin encephalopathy.
The susceptibility of GABAergic interneurons to bilirubin neurotoxicity is of particular significance. The fast-spiking mode of interneurons means that these neurons endure both heavy Ca2+ load and metabolic/oxidative stress. All of these require high levels of endogenous Ca2+-binding proteins (CBPs) to buffer. Indeed, the auditory brainstem represents a region with the highest level of CBPs including calbindin, calretinin, and parvalbumin according to the Human Protein Atlas database, but their developmental upregulation may fall behind the buffer capacity needed to cope with hyperexcitability state–driven Ca2+ overload under severe HB. Interneurons during early brain development are excitatory due to high intracellular chloride concentration, underpinning a delayed reversal of polarity in GABAergic synaptic response from excitation to inhibition (Jovanovic et al., 2017). Bilirubin may have hijacked this excitatory GABAergic mechanism to elevate excitation and further Ca2+ overload, making GABAergic interneurons particularly vulnerable to stressful insults. Our experimental results lead us to suggest that early interneuron injury and death during HB temporally precede that of other excitatory neurons and that a loss of inhibition may underlie severe impairments of the auditory brainstem and potentially other brain regions such as the cerebellum and hippocampus. Our findings necessitate future studies of interneuron-specific susceptibility to other forms of neurotoxicity beyond HB, such as hypoxia, acidosis, and ischemic insults in neonates.
Before the opening of the external auditory canal, supporting cells in the development cochlea release ATP to prime inner hair cells and drive downstream activity of peripheral spiral ganglion neurons that innervate CN neurons and refine the tonotopic maps (Tritsch et al., 2007, Milenkovic et al., 2009, Jovanovic et al., 2017; Jovanovic and Milenkovic, 2020). In this study, we found that HCN1 and HCN2 channel subtypes are expressed widely in neonatal cochlear nucleus neurons, but HCN1 predominantly drives intrinsic firings independent of exogenous afferents. Among developing CN neurons, SCs appeared to be the main subset of cells being active with enriched HCN1 channels. Our previous study demonstrated that intrinsic firings mediated by the HCN1 channel may play an important role in the formation and pruning of neural circuits in the early postnatal period because spontaneously active neurons show accelerated development in excitability, spike waveform, and firing pattern as well as synaptic remodeling toward mature phenotypes compared with those without such activity (Yin et al., 2018). Genetic deletion of HCN1 cannot be fully compensated by HCN2 or other pacemaker conductance, implicating its necessary role in driving the wiring and firing of developing circuits. This may be of general importance. For example, the cerebellum is also a bilirubin-susceptible region (Wang et al., 2008; Manchanda et al., 2013) where cerebellar basket cells, the major molecular layer interneuron, express high levels of HCN1 proteins and GAD67 proteins (Luján et al., 2005; Zhou et al., 2020). The inhibitory neuron damage and even death under long-term BIL exposure cause excitatory/inhibitory imbalance, ultimately compromising hearing to perform complex tasks such as pitch detection and speech recognition in noisy environments (Yizhar et al., 2011) as well as sensorimotor gating deficits including movement and balance control commonly associated HB-induced cerebellar damage.
Previous studies have reported that HCN1 and HCN2 channels are coexpressed in neocortex and hippocampus neurons, but whether they form heteromecic channels remains uncertain (Chen et al., 2001). In mammalian cochlear SGN, HCN1 and HCN2 channel immunofluorescence colabeling did not support the presence of heteromeric assembly (Luque et al., 2021). In this study, we found that SCs do express other HCN channel subtypes which can compensate for a loss of HCN1 in HCN1−/− mice. However, our observations showed that the electrophysiological characteristics of Ih from SCs in HCN1+/+ mice resemble those of HEK-293T-HCN1 cell line and that HCN1, but not HCN2 channels, was colabeled with inhibitory neuron markers, with homomeric HCN1 being the dominant subtype in inhibitory neurons.
BIL enhances the neuronal excitability of other neurons by affecting a variety of voltage- and ligand-gated ion channels to elevate both intrinsic excitability and synaptic drive for neurons to fire action potentials. As a part of the programmed patterning of an auditory circuit, HCN1 channels pivot the excitable state of neonatal neurons, fast-spiking neurons in particular, and their connectivity with other neurons. HCN1 channels are activated during hyperpolarization to regulate the resting membrane potential and facilitate rebound spiking as a result of their slow inactivation (Lee and MacKinnon, 2017), while other channels may synergistically contribute to excessive firings under HB conditions. For example, bilirubin can augment cation currents by Cav2.1 channels and Nav1.1 channels (Liang et al., 2017a; Shi et al., 2019), which do not function on their own unless neurons reach the threshold for discharge to feedforward hyperexcitability and Ca2+ overload. Therefore, HCN1 channels are the key determinant for the all-or-none excitable state of neonatal neurons and thus can be considered as the initiator for bilirubin to propel a vicious cycle of excessive discharge, Ca2+ overload, and cell death.
Subtle hearing phenotypes were previously reported in HCN1−/− mice with the ABR threshold being elevated 20 dB SPL at 32 and 48 kHz relative to that in HCN1+/+ mice (Ison et al., 2017). However, our study revealed no major differences in ABR threshold between HCN1−/− mice and wild-type mice. This may be due to the different genetic backgrounds of the mice, being B6129SF2/J strain and B6129-HCN1tndl/J strain in the literature, while we employed C57BL/6J mice and C57BL/6J-HCN1em1Smoc strain. In addition, the increased expression of HCN1 channels and Ih and apoptosis in SGNs has been associated with age-related hearing loss (AHL; Shen et al., 2018). Although age-dependent susceptibility of GABAergic interneurons remains to be explored, HCN1 being a risk factor in AHL is in line with our results showing that HCN1+/+ HB mice with enhanced HCN1 function experience hearing loss, similar to those of patients with abnormal hearing in bilirubin encephalopathy (Shapiro, 2010). Strikingly, HCN1+/+ HB mice preconditioned with HB displayed high-frequency hearing loss and decreased ABR amplitude, this may be due to HCN1 channels being most prevalent in fast-spiking interneurons, namely, SCs. On the contrary, HCN1−/− mice under the same HB condition showed resilience to all aspects of BIL toxicity, indicating HCN1 is likely a key molecular substrate necessary for BIL to impair auditory and other brain functions in vivo. Our findings thus rationalize HCN1 channels as a potential target for strategizing prevention and treatment of brain injury and encephalopathy in severe jaundice.
Data Availability Statement
Data that support the findings of this study are available in the article or can be requested from the author.
Footnotes
This work was supported by the State Key Program of National Natural Science of China (82330034 to S-K.Y. 82430036 to H-B.S.) and the National Natural Science Foundation of China (82020108008 to H-B.S., 82301308 to L.-N.G., 82301309, 81800903, and 81900935 to H-W.L., M.L., and X.L.Y, respectively) and Canadian Institutes of Health Research (PJT-156034 and PJT-156439), Natural Science and Engineering Research Council (RGPIN-2017-06665), and Canada Research Chair Program (CRC-2018-00195 to L-Y.W.).
↵*L-N.G., H-W.L., and K.L. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Hai-Bo Shi at hbshi{at}sjtu.edu.cn or Lu-Yang Wang at luyang.wang{at}utoronto.ca or Shan-Kai Yin at skyin{at}sjtu.edu.cn.
