Abstract
The inferior colliculus (IC) represents a crucial relay station in the auditory pathway, located in the midbrain's tectum and primarily projecting to the thalamus. Despite the identification of distinct cell classes based on various biomarkers in the IC, their specific contributions to the organization of auditory tectothalamic pathways have remained poorly understood. In this study, we demonstrate that IC neurons expressing parvalbumin (ICPV+) or somatostatin (ICSOM+) represent two minimally overlapping cell classes throughout the three IC subdivisions in mice of both sexes. Strikingly, regardless of their location within the IC, these neurons predominantly project to the primary and secondary auditory thalamic nuclei, respectively. Cell class-specific input tracing suggested that ICPV+ neurons primarily receive auditory inputs, whereas ICSOM+ neurons receive significantly more inputs from the periaqueductal gray and the superior colliculus (SC), which are sensorimotor regions critically involved in innate behaviors. Furthermore, ICPV+ neurons exhibit significant heterogeneity in both intrinsic electrophysiological properties and presynaptic terminal size compared with ICSOM+ neurons. Notably, approximately one-quarter of ICPV+ neurons are inhibitory neurons, whereas all ICSOM+ neurons are excitatory neurons. Collectively, our findings suggest that parvalbumin and somatostatin expression in the IC can serve as biomarkers for two functionally distinct, parallel tectothalamic pathways. This discovery suggests an alternative way to define tectothalamic pathways and highlights the potential usefulness of Cre mice in understanding the multifaceted roles of the IC at the circuit level.
Significance Statement
The inferior colliculus (IC) is a crucial relay station in the ascending auditory pathway. Our findings revealed that parvalbumin-expressing (PV+) and somatostatin-expressing (SOM+) IC neurons, which are present in all IC subdivisions, represent two minimally overlapping cell classes that have distinct input–output architectures and intrinsic electrophysiological properties, suggesting their distinct roles in auditory processing. Our results preliminarily uncovered how IC neurons expressing different biomarkers could contribute to the organization of the ascending auditory pathway, offered an alternative way to investigate the working mechanisms of the central auditory system, and suggested that IC neurons expressing other biomarkers should be examined as well.
Introduction
In mammalian brains, ascending sensory pathways play critical roles in coding and relaying sensory information, changing internal states, and modulating behaviors (Park and Friston, 2013; Choi et al., 2020; Pardi et al., 2020). In the auditory system, nearly all auditory information ascending to the forebrain passes through the auditory midbrain, also known as the inferior colliculus (IC), making it a major station in the auditory pathway (Batra and Fitzpatrick, 2002; Malmierca et al., 2005; Saldaña et al., 2009). The IC consists of three subdivisions and diverse cell types and plays a critical role in various auditory brain functions, such as vocal communication (Lohse et al., 2020), sound localization (Grothe et al., 2010), and sound-induced innate behaviors (Xiong et al., 2015), that are important for animal's survival and thrival. Given that the thalamus is the major higher brain region targeted by the IC outputs, determining how the auditory tectothalamic pathways are organized would definitely facilitate the interrogation of neural circuit mechanisms underlying the above auditory brain functions.
Our current understanding of the auditory tectothalamic pathways has been mainly based on three traditionally defined IC subdivisions, including the central nucleus (ICC), dorsal cortex (ICD), and external cortex (ICE), which were defined by Golgi method (Cajal, 1995), and based on the tectothalamic connectivity patterns revealed by traditional tracers (Oliver and Hall, 1978; Calford and Aitkin, 1983; Mellott et al., 2014; Clarke and Lee, 2018). The ICC provides driver inputs to the ventral division of the medial geniculate body (MGBv; Lee and Sherman, 2010). This pathway is regarded as the primary auditory tectothalamic pathway and required for relaying sound feature-related information to the forebrain (Grothe et al., 2010). The ICD and ICE provide modulatory inputs to the dorsal and medial division of the MGB (MGBd and MGBm), respectively, serving as the secondary auditory tectothalamic pathways (Lee and Sherman, 2010), which play important roles in mediating salient sound-induced arousal (Wang et al., 2023) and encoding sound detection behaviors (Lee et al., 2023). However, the above three auditory tectothalamic pathways defined by traditional tracers may be oversimplified. When anterograde tracers are injected into each individual subdivision of the IC, axonal fibers from the IC can be observed in all subdivisions of the MGB, and the density of fibers in each thalamic subdivision is not negligible (Oliver and Hall, 1978; Ledoux et al., 1987). In addition, abundant axonal fibers can also be observed in thalamic nuclei surrounding the MGB, such as the posterior intralaminar nucleus (PIN; Cai et al., 2019) and posterior limiting nucleus (POL; Ledoux et al., 1987). These data call for alternative ways to define tectothalamic pathways.
Defining neural pathways based on the expression of specific biomarkers has been increasingly used in the neuroscience field (Zeng, 2022). Neurons expressing different biomarkers in the superior colliculus (SC) project to largely different downstream targets (Shang et al., 2015; Sans-Dublanc et al., 2021; Liu et al., 2022), for example, parvalbumin-expressing (PV+) neurons and neurotensin receptor-expressing (NTSR+) neurons project to the parabigeminal nucleus (Shang et al., 2015) and the paralaminar nuclei of the thalamus (Sans-Dublanc et al., 2021), respectively. In fact, neurons expressing different biomarkers such as parvalbumin (PV; Fujimoto et al., 2017), somatostatin (SOM; Wynne et al., 1995), cholecystokinin (Kreeger et al., 2021), vasoactive intestinal peptide (VIP; Goyer et al., 2019; Beebe et al., 2022; Drotos et al., 2023), and neuropeptide Y (NPY; Silveira et al., 2020, 2023) have also been reported in the IC, and they are distributed throughout the IC rather than localized in an individual subdivision. Whether and how the neurons in the IC would demonstrate a preference for downstream targets depending on their biomarker remained unknown.
Here, by using transgenic mice, cutting-edge viral tools, immunostaining, our newly developed rapid and deformation-free on-slide tissue-clearing method and slice recording, we identified PV+ (ICPV+, 14% of IC cells) and SOM-expressing (ICSOM+, 25% of IC cells) neurons as two cell classes in the IC that are minimally overlapping with each other and present in all three IC subdivisions. Interestingly, no matter in which IC subdivision the PV+ neurons are located, their axon terminals were mainly distributed in the MGBv, the primary auditory thalamus module that only projects to the primary auditory cortex. In contrast, ICSOM+ axon terminals are predominantly distributed in the secondary auditory thalamus, the POL in particular, which mainly projects to the posterior tail of the striatum (TS) followed by the secondary auditory cortex. We also examined the inputs of these two cell classes and found that ICPV+ neurons primarily receive auditory inputs, whereas ICSOM+ neurons integrate polymodal inputs that likely hold behavioral significance. Furthermore, ICPV+ neurons exhibit significant heterogeneity in both intrinsic electrophysiological properties and presynaptic terminal size compared with ICSOM+ neurons. Notably, approximately one-quarter of ICPV+ neurons are inhibitory neurons, whereas all ICSOM+ neurons are excitatory neurons. Collectively, our findings indicate that PV and SOM in the IC can serve as biomarkers for two parallel tectothalamic pathways, which have distinct anatomical organization patterns and functional implications.
Materials and Methods
Experimental details
Key resources
Detailed information regarding the essential resources utilized in this study is available in Table 1.
Abbreviations for brain regions
The majority of abbreviations for brain regions used in this study were from the Allen Mouse (https://scalablebrainatlas.incf.org/main/coronal3d.php?template=ABA_v3#downloads) and the Mouse Brain in Stereotaxic Coordinates, second edition by Franklin, K. B. J. and Paxinos, G ⇓(Table 2).
Key resources
Abbreviations for brain regions
Animals
All animal care procedures and experiments described in this study were conducted in strict accordance with the ethical guidelines and regulations set forth by the Institutional Animal Care and Use Committee (IACUC) at Tsinghua University, Beijing, China. Prior to and following surgical procedures, all mice were maintained under standard environmental conditions of temperature, humidity, and light/dark cycles at the Laboratory Animal Resources Center, Tsinghua University. Appropriate measures were taken to ensure animal welfare and minimize potential discomfort or harm. Adult mice of both sexes (∼2 months, background strain C57BL/6J) were used for tracing and immunofluorescence staining studies, while mice (5–12 weeks, background strain C57BL/6J) of both sexes were used for whole-cell patch experiments. C57BL/6J [wild-type (WT)] mice were procured from Wei Tong Li Hua Experimental Animal Co., Ltd (Beijing, China), and all transgenic mice were obtained from The Jackson Laboratory. The PV-IRES-Cre mice (B6.129P2-Pavlbtm1(cre)Arbr/J, The Jackson Laboratory, Stock#: 017320) and SOM-IRES-Cre mice (Ssttm2.1(cre)Zjh/J, The Jackson Laboratory, Stock#: 013044; Taniguchi et al., 2011) were utilized for labeling PV+ and SOM+ neurons in the IC, respectively. The Vgat-Cre mice (Slc32a1tm2(cre)Lowl/J, The Jackson Laboratory, Stock#: 016962; Vong et al., 2011) were used for labeling inhibitory neurons in the IC. In addition, the Ai14 mice [B6. Cg-Gt (ROSA)26Sortm14(CAG-tdTomato) Hze/J, The Jackson Laboratory, Stock#: 007914; Madisen et al., 2010], also known as Cre-reporter mouse line, were crossed with SOM-Cre (SOM × Ai14), PV-Cre (PV × Ai14), and Vgat-Cre (Vgat × Ai14) mice to achieve cell type–specific expression of red fluorescent protein in neurons within the IC in this study.
Stereotaxic surgeries
The mice were first anesthetized with 2.5% avertin (300 mg/kg, i.p.) and placed in a stereotaxic apparatus (RWD Life Science). To ensure the comfort of the mice, we administered an erythromycin eye ointment to prevent eye dryness and used an electric heating pad to maintain their body temperature during the surgical process. The skin over the skull midline was then carefully incised with sterilized scissors, and forceps were used to expose the bregma, lambda, and skull surface. Next, a small hole was performed above the IC or auditory thalamus, and a microsyringe pump with a glass pipette (World Precision Instruments) was used to slowly inject viruses (35 nl/min) into the IC or auditory thalamus. The following coordinates (in mm) were used: −3.00 anteroposterior (AP), −2.10 mediolateral (ML), and −3.00 dorsoventral (DV) for the MGBv; −3.00 AP, −1.75 ML, and −2.80 DV for the POL; −1.20 AP1, −0.60 ML, and −0.70 DV for the ICD; −1.20 AP1, −1.00 ML, and −0.8 DV for the ICC; −1.20 AP1, −1.60 ML, and −0.70 DV for the ICE (AP is relative to bregma for auditory thalamus nuclei, AP1 is relative to lambda for the IC; ML is relative to the midline; DV is relative to the brain tissue surface above the targeted brain areas). The glass pipette remained in place for an additional 5 min after injection before being slowly withdrawn. The skin was sutured after retracting the pipette. After surgery, the mice were allowed to recover on a heated pad until they were ambulatory, and then they were returned to their home cage. The mice underwent a postsurgery recovery period of approximately 2–3 weeks to allow for adeno-associated virus (AAV) virus infection and gene expression.
Virus injection and tracing
For cell type–specific anterograde tracing, we employed the injection of 150 nl AAV2/9-hSyn-Flex-synaptophysin-mRuby (Cat# WY3330, 3.6E + 12 V.G/ml, Taitool Bioscience) into the IC of PV-IRES-Cre or SOM-IRES-Cre mice to trace the projection of PV+ or SOM+ neurons in all subdivisions of the IC. Furthermore, we applied the same virus (30–50 nl) injection as local as possible in individual IC subdivision to investigate how PV+ and SOM+ neurons in each subdivision of the IC participate in the tectothalamic pathways. In addition, we conducted control experiments in WT mice by performing AAV2/9-hSyn-Flex-synaptophysin-mRuby injection mixed with CTB-647 (C34778, Thermo Fisher Scientific) into the IC, aiming to confirm the crucial reliance of mRuby expression on Cre (Fig. 2P).
For trans-synaptic anterograde tracing, we injected 80–100 nl scAAV-hSyn-Cre (Cat# S0292-1, 1.25E + 13 V.G/ml, Taitool Bioscience) into the IC of the WT mice. At the same time, we performed localized injection of AAV2/5-Flex-tdTomoto-T2A-synaptophysin-EGFP (Cat# S0161-5, 5.3 + 12E V.G/ml, Taitool Bioscience) into the MGBv or AAV2/9-Flex-synaptophysin-mRuby into the POL, to determine the downstream target of IC-MGBv or IC-POL pathways.
To identify the input of PV+ or SOM+ neurons in the IC, we adopted a widely used rabies virus (RV)–mediated retrograde monosynaptic tracing strategy. Briefly, mixed Cre-dependent AAV helper viruses AAV2/5-hEF1a-DIO-H2B-eGFP-T2A-TVA-WPRE-pA (Cat#S0320-5, 7.3 + 12E V.G/ml, Taitool Bioscience) and AAV2/5-hEF1a-DIO-RVG-WPRE-pA (Cat#S0325-5, 5.5 + 12E V.G/ml, Taitool Bioscience) were injected into the IC of PV-IRES-Cre or SOM-IRES-Cre mice to selectively express TVA and RG in PV+ or SOM+ neurons (Fig. 4A, Day 1). Three weeks later, 100 nl of pseudotyped RV equipped with the TVA selective avian ASLV type A (EnvA; RV-EnvA-△G-dsRed; Cat#R01002, 2.00 + 08E IFU/ml, BrainVTA) was injected into the same injection sites (Fig. 4A, Day 21) to enable dsRed expression in the input neurons of PV+ or SOM+ starter neurons (Fig. 4B, PV+; Fig. 4C, SOM+; AAV-Helper, green; input neurons, red; starter neurons, yellow). For control experiments, we injected the same set of viruses with the same procedures into the IC of SOM-IRES-Cre mice without including either AAV-DIO-RG (Fig. 4G, N = 4 mice, left panel) or AAV-DIO-EGFP-TVA (Fig. 4G, N = 4 mice, right panel), and no input neurons were observed in the control mice, validating the dependency of retrograde tracing on RG or TVA. The mice were killed after RV expression and retrograde spreading for another 7 d (Fig. 4A, Day 28).
To visualize the morphology of ICSOM+ and ICPV+ neurons, we performed injections of AAV-sparse-CSSP-RFP-8E3 (Cat#BC-SL005, 5 + 12E V.G/ml, Brain Case) into the IC of SOM-IRES-Cre and PV-IRES-Cre mice, respectively. We also used a combination of AAV2/1-CAG-FLEX-Flpo-WPRE-pA (Cat#S0273-1, 2.73 + 10E V.G/ml, Taitool Bioscience) and AAV2/9-hSyn-fDIO-mGFP-T2A-synaptophysin-mRuby (Cat#S0920-9, 6.8 + 12E V.G/ml, Taitool Bioscience) for injections into the IC of SOM-IRES-Cre or PV-IRES-Cre mice to visualize the morphology of ICSOM+→POL or ICPV+→MGBv axon terminals (Fig. 5G,H, right panels).
Histology and immunochemistry
Animals were administered an overdose of 2.5% avertin (350 mg/kg, i.p.) and underwent transcardially perfused with 0.9% saline, followed by 4% paraformaldehyde (PFA, P804536-2, Aladdin) in 0.01 M phosphate-buffered saline (1× PBS, Solarbio Life Science). The brains were then extracted, postfixed in 4% PFA at 4°C overnight, and subsequently immersed in 30% sucrose for dehydration and cryoprotection until they sank. The brains were frozen in optimal cutting temperature compound (OCT, Sakura), and coronal slices were obtained using a freezing microtome (CM1950, Leica Biosystems) at a thickness of 50 μm, covering the whole brain. For brain-wide cell counting, every other slice was collected and mounted on gelatin-coated slides. For antibody immunofluorescent staining, coronal brain sections of 40–50 μm were collected into multiple well plates (Corning® Costar®) coated with an ultralow attachment material and filled with 1× PBS. The collected brain sections were rinsed three times for 5 min (3 × 5 min) each with 1× PBS buffer on a shaker. Next, brain sections were incubated for 3 × 15 min with 0.3% Triton X100 in 1× PBS for permeabilization, followed by a 60 min incubation with 5% goat serum (SL038, Solarbio) to block unspecific binding of the antibodies. The brain sections were then incubated with primary antibodies overnight at 4°C, followed by 3 × 15 min washes with 1× PBS, and incubated with secondary antibodies for 2 h. The sections were then thoroughly washed with 1× PBS and coverslipped with a 50% glycerol mounting medium. We used the primary antibody of anti-PV mouse monoclonal antibody (1:1,000, MAB1572, Merck Millipore) combined with the secondary antibody of anti-mouse polyclonal goat antibody conjugated with Alexa Fluor 647 (1:1,000, ab150115, Abcam) to label PV+ neurons in the IC. We also employed the primary antibody of GAD65 + GAD67 rabbit monoclonal antibody (1:500, ab183999, Abcam) with anti-rabbit polyclonal goat secondary antibody conjugated with Alexa Fluor 488 (1:1,000, ab150077, Abcam) to label glutamic acid decarboxylase (GAD)+ neurons in the IC. We used an anti-NeuN rabbit monoclonal antibody conjugated with Alexa Fluor488 (1:1,000, ab19095, Abcam), which was used without secondary antibody incubation, to label NeuN+ neurons in the IC. To validate the specificity of each primary antibody, we performed control experiments following the same procedures as described previously, except that 1× PBS was substituted for the primary antibodies.
Acute brain electrophysiology and recordings
SOM × Ai14 and PV × Ai14 mice (5–12 weeks old, both sexes) were deeply anesthetized by intraperitoneal administration of 2.5% avertin (300 mg/kg) and were then perfused through the heart using NMDG-HEPES (Ting et al., 2018) artificial cerebrospinal fluid (ACSF) consisting of the following (in mM): 92 NMDG, 2.5 KCl, 1.25 NaH2PO4, 30 NaHCO3, 20 HEPES, 25 glucose, 2 thiourea, 5 Na-ascorbate, 3 Na-pyruvate, 0.5 CaCl2·2H2O, and 10 MgSO4·7H2O. The pH of the solution was adjusted to 7.3–7.4 using 5 M hydrochloric acid, and the osmolality was 300–310 mOsmoles/kg. The solution was prechilled at 4°C and bubbled with carbogen (95% O2 and 5% CO2) beforehand. We obtained approximately four coronal slices, each with a thickness of 300 µm, using a vibratome (Leica VT1200S). Subsequently, the slices were transferred to a prewarmed (34°C) beaker filled with NMDG-HEPES ACSF for a protective recovery spike-in procedure. Afterward, these slices were transferred into HEPES-holding ACSF consisting of the following (in mM): 92 NaCl, 2.5 KCl, 1.25 NaH2PO4, 30 NaHCO3, 20 HEPES, 25 glucose, 2 thiourea, 5 Na-ascorbate, 3 Na-pyruvate, 2 CaCl2·2H2O, and 2 MgSO4·7H2O, with the pH titrated to 7.3–7.4 using 10 M NaOH. The holding chamber was equilibrated at room temperature for a minimum of 1 h prior to the recording session and was maintained at this temperature throughout the 1 d experiment. Individual slices were transferred to the recording chamber as required. Slices were transferred to a submerged recording chamber on the stage of an upright microscope (Olympus BX51WI). The chamber was continually perfused (2–3 ml/min) with recording ACSF solution under constant oxygen and containing the following (in mM): 124 NaCl, 2.5 KCl, 1.25 NaH2PO4, 24 NaHCO3, 12.5 glucose, 5 HEPES, 2 CaCl2·2H2O, and 2 MgSO4·7H2O. Individual SOM+ or PV+ neurons were identified under differential interference contrast or fluorescence and a 40× water immersion objective with an infrared CCD camera (DAGE MTI IR-1000E). Whole-cell recordings were obtained using 8–12 MΩ pipettes pulled from borosilicate glass by a puller (P97, Sutter Instrument), with an in-pipette solution containing 122.5 mM K-glucose, 12.5 mM KCl, 10 mM HEPES, 2 mM Na2ATP, 0.3 mM Na2GTP, 2 mM MgCl2, and 8 mM NaCl. Electrophysiological signals were amplified using a MultiClamp 700B amplifier (Molecular Devices), digitized (Axon Digidata 1440A) at 20–50 kHz, and filtered at 2–10 kHz.
Rapid and nonscaling on-slide tissue clearing
In this study, we developed an innovative and efficient on-slide tissue-clearing methodology that enables rapid and nonscaling processing for visualization of tectothalamic terminal structures. To achieve high-throughput and high-precision imaging, we combined this technique with high-speed spinning-disk confocal imaging. Following prefusion using 4% PFA, the mouse brain was placed in a balanced cryopreserved solution containing 10% dimethyl sulfoxide, 3% sucrose, and 1× PBS solution for 2 d. The OCT-embedded brains were then cryosectioned at approximately −25°C. The resulting brain slices were placed onto gelatin-coated slides and incubated in a high-humidity environment at 37°C for 2 h. After thorough rinsing with 1× PBS solution was conducted for 10 min, the brain slices were subjected to tissue clearing by treating them with a 55% antipyrine (A800873) and 0.1% DAPI (D9542, Sigma-Aldrich) on-slide clearing mounting solution dissolved in deionized water for 2–3 min. The same clearing solution was used for coverslipping the brain slices. Prior to imaging, the slides were stored in a dark environment at room temperature. This optimized protocol effectively preserved tissue morphology and ensured the acquisition of reliable imaging data for our analysis.
Imaging and fluorescent signal recognition
To enable brain-wide cell counting and analysis of neuron distribution, we imaged the slides using the Axio Scan.Z1 (Zeiss Axio Scan) equipped with a 10× objective, allowing for efficient scanning of large areas and capturing an overview of the samples. For higher-resolution imaging and precise visualization of microscale structures and immunofluorescence staining in brain slices, we utilized the high-speed confocal microscopy system (Andor, Oxford Instruments) with a 20× objective. This enabled us to obtain detailed and accurate information about cellular colocalization and fine structural features. To further investigate tectothalamic terminal structures, we employed a 40× objective with a 0.5 μm z-step, enabling us to capture high-resolution images of synapses. During confocal imaging, we ensured a single-field resolution of 2,048 × 2,048, acquired images with an average of two scans, and used 16-bit imaging to preserve the full dynamic range of the data. Finally, to generate complete three-dimensional images, automatic stitching was employed to seamlessly merge multiple fields of view, resulting in a comprehensive and detailed visualization of the samples.
By employing a combination of automated identification and manual assessment, we identified fluorescence signals corresponding to neurons and synaptic structures. To achieve this, we utilized the Spots tool in Imaris9.0 software (Oxford Instruments, https://imaris.oxinst.cn/). Size thresholds were set for the two distinct types of fluorescent signals, with a range of 0.5–2 μm for synaptic structures and 10–12 μm for neurons. Manual intervention was applied to supplement or remove misidentified or missed fluorescent signals. In particular, manual adjustments were necessary when the fluorescent intensity of proximal dendrites or axons was exceptionally high, making it challenging for the software to accurately identify the neuron soma or distinguish neighboring neurons. Subsequently, the positions of the fluorescent markers on the corresponding images, as well as the images themselves, were imported into Rhinoceros 7 software (Rhinoceros, https://www.rhino3d.com/) and aligned with a widely utilized standard brain atlas obtained from the Allen Reference Atlas (AllenInstitute, https://scalablebrainatlas.incf.org/main/coronal3d.php?template=ABA_v3#). Statistical analysis was performed using custom MATLAB scripts (Yuan Lab, MathWorks) for cell counting and localization, as well as custom Grasshopper scripts (Yuan Lab, Rhinoceros) for the quantification of auditory thalamic terminals, to conduct further analysis.
Data analysis
Kernel density estimation
The density distribution of IC neurons was analyzed using kernel density estimation (KDE) with an adjusted quad kernel densitometry method (Silverman, 1986). This approach allowed us to estimate the density distribution of dot signals at any given location (x, y) within the identified neurons. The calculation was derived from the following formula:
The bandwidth (h) in the KDE was determined using the following rule, where Dm represents the median value of the distance between the dots and their average center:
Quantification estimation of tectothalamic synaptic terminals
To quantitatively analyze the tectothalamic pathways, we focused on the brain slice at the AP = −2.97 mm coordinate, which represents a typical section of the auditory thalamus. By using the “fluorescent signal recognition” method described above, we identified the axon terminals in different subdivisions of the auditory thalamus and determined the total number of terminals originating from ICSOM+ or ICPV+ projection neurons. We then calculated the percentage (P) and weighted density (D) of synaptic terminals in each auditory thalamus subdivision to assess the strength of the connection between IC neurons and each auditory thalamus nucleus. Specifically, P represents the percentage of terminals in a given auditory thalamus subdivision relative to the total number of terminals across all subdivisions. We only presented the P values for ICSOM+ and ICPV+ neuron projections in the MGBv, MGBd, POL, and PIN, as their projections to other subdivisions of the auditory thalamus are minimal (Fig. 2J–O, third column, bar charts). Additionally, we included the weighted density (D) due to the significant differences in the size (A) of these four auditory thalamus subdivisions (Fig. 2J–O, third column, line graphs).
To visualize the distribution of ICSOM+ and ICPV+ terminal projections to the auditory thalamus, we conducted a random and even sampling of 3,000 terminals from each mouse brain slice at the −2.97 mm coordinate. These selected terminals were used as representative samples to illustrate the average distribution pattern across the entire sample (Fig. 2J–L,M,O, second column). We observed a low number of terminals originating from SOM+ neurons in the ICC projecting to the thalamus, and this dataset illustrated the distribution of 300 terminals representing the projection of ICCSOM+ neurons to the auditory thalamus (Fig. 2N, second column).
Data analysis for retrograde tracing
To evaluate the similarity in the number of RV-mediated retrograde monosynaptic tracing labeled neurons among input sources, we used Pearson's correlation coefficient of normalized inputs, calculated as the total number of input neurons from a specific brain region divided by the total number of input neurons.
Electrophysiology data analysis
All electrophysiological data were analyzed using our custom-written MATLAB program. Traces obtained from PV+ or SOM+ neurons, depicting their responses to current injections, were utilized to extract several electrophysiological parameters. The resting membrane potential was determined as the mean value from trials without any injected current (Fig. 6C). The rheobase was defined as the minimum current required to elicit the first single action potential in a neuron (Fig. 6D). The time constant was derived from a single exponential decay function fitted between 10 and 95% of the response to a −80 pA current injection (Fig. 6F). Sag was quantified by measuring the voltage response during a hyperpolarizing current injection of −80 pA (Fig. 6E):
Terminal size analysis
We utilized synaptophysin-mRuby labeling to identify tectothalamic axon terminals in the auditory thalamus. By combining the turntable confocal microscope with our previously described “rapid and nonscaling on-slide tissue-clearing” method (see detailed methods above), we successfully achieved high-throughput three-dimensional fluorescence imaging at the single-synapse scale in large-scale brain regions. Our data processing workflow involved several steps. First, we performed deconvolution based on the point spread function of the microscope system. Next, background subtraction was applied, and the volume of fluorescent signals was reconstructed using a fluorescence intensity threshold. If necessary, point recognition algorithms were used to separate closely adhered signals and generate models for each fluorescent dot signal. Finally, we extracted the fluorescence intensity and volume information from each three-dimensional reconstructed object.
To accurately quantify the size of axon terminals using a fluorescence microscope, we employed fluorescent microbeads with known sizes (e.g., 0.5 μm, 1 μm, 2 m, and 3 μm) that closely resembled the size range of real synaptic terminals (De Banne et al., 2011). By imaging these microbead samples, which exhibited varying fluorescence intensities achieved through gradient fluorescence quenching, we established a relationship function between the parameters of fluorescent signals obtained from the fluorescence microscope and the corresponding real sizes of the microbeads. The microbeads were injected into either agarose gel or the mouse brain, and their reconstructed volumes were determined, enabling us to establish a robust correlation between the fluorescence signal parameters and the actual sizes of synaptic terminals. Subsequently, we performed a linear fitting analysis by taking the negative reciprocal of the mean fluorescence intensity of the microbeads and their corresponding reconstructed volumes. The results revealed a strong positive linear correlation between these two parameters. Based on this correlation, we constructed a surface representation in three-dimensional space by lofting a fitting line, which accurately described the relationship between the real size of the targeted microbeads and their reconstructed volumes. This surface provided a comprehensive depiction of the distribution of both the negative reciprocal of the fluorescence intensity and the reconstructed size, allowing for precise estimation of the actual sizes of synaptic terminals based on their fluorescence signal characteristics. The established model was then applied to reconstruct the volumes of all fluorescent microbeads of different sizes (0.5 μm, 1 μm, 2 μm, and 3 μm). As anticipated, we observed three distinct peaks in the diameter distribution function after identifying the peak values using the Fourier transform and fitting Gaussian distributions to the reference peaks. Notably, the distribution of reconstructed volumes exhibited a clear superposition of three normal distributions, with each mean value corresponding to the true size of the microbeads. We further calculated the mean values and areas under the curve of the fitting normal distributions. The results demonstrated that the areas under the fitting distribution curves (or peaks) accurately reflected the real proportions of the three given microbead sizes. This confirms that our methodology effectively represents the true size and number of fluorescent microbeads. In conclusion, our approach not only accurately estimates the real size of fluorescent signals ranging from 0.5 to 2 μm but also accurately captures the distribution of fluorescent signals with mixed sizes.
Quantification and statistical analysis
Statistical analysis and tests were conducted using GraphPad Prism (GraphPad Software) or MATLAB (MathWorks) with a two-tailed test. The normality of data was assessed using the Shapiro–Wilk test. For normally distributed data, a t test was employed, while a nonparametric Mann–Whitney U test was used for nonnormally distributed data. In cases of multiple comparisons, post hoc tests such as Bonferroni’s and Dunn's tests were utilized for one-way ANOVA and the Kruskal–Wallis H test, respectively. The data are presented as means ± SEMs in both figures and text. Paired tests for two groups were performed using the paired t test for normally distributed data and the Wilcoxon matched-pairs signed rank test for nonnormally distributed data. Multiple paired tests were analyzed using two-way ANOVA with Geisser–Greenhouse correction. The statistical details, including the specific tests used, statistics, significance levels, sample sizes, and animal numbers, are provided in the figure legends. Significance levels are denoted as *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, and n.s. for nonsignificant (p > 0.05).
Results
PV+ and SOM+ neurons are the two cell classes present in all IC subdivisions
PV+ and SOM+ neurons have been widely observed and extensively studied in many cortical and subcortical regions, and they usually play distinct roles in sensory processing and animal behaviors (Shang et al., 2015; Tremblay et al., 2016; Anderson et al., 2018; Antonoudiou et al., 2020; Shen et al., 2022). As mentioned earlier, these two classes of neurons are present in the IC as well, but their connectivity pattern remains unknown. Therefore, we decided to focus on these two cell classes in the present study.
We first injected AAV-sparse-CSSP-RFP-8E3 into the IC of SOM-IRES-Cre or PV-IRES-Cre mice to characterize the morphology of sparsely labeled PV+ and SOM+ neurons. We found that both classes of neurons were stellate neurons, which typically extended their dendrites beyond the single fibro-dendritic lamina (Oliver, 2005; Fig. 1A; SOM+, n = 34 neurons from N = 3 mice; PV+, n = 25 neurons from N = 3 mice). We next determined the distribution of PV+ and SOM+ neurons in the IC. To this end, we performed anti-NeuN immunostaining using brain slices from SOM-IRES-Cre × Ai14 (SOM-Ai14) mice to observe ICSOM+ neurons (Fig. 1B, first row, left panel). Although the SOM-IRES-Cre mouse line has been widely used (Taniguchi et al., 2011; Muñoz-Castañeda et al., 2021), we were unable to validate ICSOM+ neurons using the anti-SOM immunostaining approach because antibodies tested did not work for an unknown reason. Regarding PV+ neurons, we took an anti-PV immunostaining approach instead using WT mice (Fig. 1B, second row, left panel) because very dense PV+ neurites are present in the ICC (Fig. 1C), hindering the observation of the PV+ cell body. We performed NeuN immunostaining using both SOM-IRES-Cre × Ai14 and WT mice to count neurons in the IC (Fig. 1B, middle panel).
PV+ and SOM+ neurons are two cell classes present in all IC subdivisions. A, Confocal 10× images showing the morphology of ICSOM+ and ICPV+ neurons by sparse labeling. Scale bar, 50 μm. B, Confocal 20× showing SOM+ neurons (first row, left, red), NeuN staining (first row, middle, green), an overlay (first row, right) in the IC and PV+ neurons (second row, left, red), NeuN staining (second row, middle, green), and an overlay (second row, right) in the IC. Scale bar, 200 μm. C, Example images showing dense PV+ neurites in the ICC in PV-IRES-Cre × Ai14 mice slices. Scale bar, 200 μm. D, Confocal 20× showing PV+ neurons (top, green), SOM+ neurons (middle, red), and an overlay (bottom) in the IC. Scale bar, 200 μm. Insets, enlarged view of the dotted box, and the white arrows mark IC neurons expressing SOM or PV, while the yellow arrows mark IC neurons expressing both SOM and PV. Neurons expressing PV and SOM were minimally overlapping in the IC (bottom). Scale bar, 10 μm. E, Illustrative diagrams showing the distribution of PV+ neurons (green dots) and SOM+ neurons (red dots) in different directions within the IC. AP, anteroposterior; ML, mediolateral; DV, dorsoventral. F, Percentage of PV+ neurons (left bar, green) and SOM+ neurons (right bar, red) in the IC. n = 8 slices from N = 4 mice. G, The percentages of PV+ neurons in different subdivisions of the IC (ICE, ICC and ICD). n = 8 slices from N = 4 mice; one-way ANOVA with post hoc, **p < 0.01, n.s.p > 0.05. H, The percentages of SOM+ neurons in different subdivisions of the IC (ICE, ICC, and ICD). n = 8 slices from N = 4 mice; one-way ANOVA with post hoc, **p < 0.01, ****p < 0.0001. Data are means ± SEM. I, Percentage of PV+ neurons of NeuN neurons in the rostro-caudal axis of the IC. n = 8 slices from N = 4 mice, unpaired t test, p > 0.05. J, Percentage of SOM+ neurons of NeuN neurons in the rostro-caudal axis of the IC. n = 8 slices from N = 4 mice, unpaired t test, p > 0.05. K, Percentage of PV+ and SOM+ coexpression neurons (merge) among PV+ or SOM+ counted neurons; n = 6 slices from N = 2 mice. Data are means ± SEM. L, KDE of neurons in the IC expressing NeuN (first row, blue), PV (second row, green), and SOM (third row, red) and an overlay of SOM and PV (fourth row) from rostral (AP = −4.57) to caudal (AP = −5.37) coronal slices with a spacing of 200 μm; n = 20 slices from N = 4 mice per group; PV, SOM, central nucleus, dorsal cortex, and external cortex (ICC, ICD, and ICE, respectively) of the IC.
By performing anti-PV immunostaining using SOM-Ai14 slices, we found that neurons expressing these two biomarkers were mutually exclusive (Fig. 1D,K; n = 6 slices from N = 2 mice; fraction of coexpression, 2.3 ± 0.2% of SOM+ neurons, 4.4 ± 0.3% of PV+ neurons). Therefore, PV+ and SOM+ neurons are two minimally overlapping classes of neurons in the IC in terms of the biomarker expressed.
We found that both PV+ and SOM+ neurons were distributed across all three subdivisions of the IC and in all directions (Fig. 1E; PV+, green; SOM+, red), and they represented 14.0 ± 1.0% and 26.0 ± 1.9% of neurons counted, respectively (Fig. 1F; n = 8 slices from N = 4 mice). In terms of specific subdivisions, PV+ neurons exhibited a significantly higher percentage in the ICC (∼19%) compared with the ICE (∼11%) and ICD (∼12%; Fig. 1G; n = 8 slices from N = 4 mice; one-way ANOVA with post hoc, **p < 0.01), while SOM+ neurons were significantly more populated in the ICE (∼34%) and ICD (∼27%) than in the ICC (∼15%; Fig. 1H; n = 8 slices from N = 4 mice; one-way ANOVA with post hoc, **p < 0.01, ****p < 0.0001). Furthermore, the distribution of these two classes of neurons did not vary along the rostro-caudal axis of the IC (Fig. 1I,J; n = 8 slices from N = 4 mice per group; unpaired t test, p > 0.05).
To more precisely determine the distribution pattern of PV+ or SOM+ neurons in the IC, we performed KDE (see Materials and Methods for more details) using five coronal slices with a spacing of 200 μm (n = 20 slices from N = 4 mice per group). Although NeuN was evenly distributed throughout the IC (Fig. 1L, first row), the density of PV+ neurons decreased gradually from the ICC to ICD or ICE, while that of SOM+ neurons demonstrated an opposite decreasing direction (Fig. 1L; PV+, second row; SOM+, third row). In the ICD, we observed a higher density of SOM+ in the superficial layers and PV+ neurons in the deeper layers. Collectively, the density of these two classes of neurons showed a largely complementary distribution pattern, but they were comparable around the borders between ICC and ICD or ICE (Fig. 1L, fourth row).
ICPV+ or ICSOM+ axon terminals predominantly innervate the primary or secondary auditory thalamus
We then injected AAV-Flex-synaptophysin-mRuby into the IC of PV-IRES-Cre or SOM-IRES-Cre mice to label the synaptic terminals of PV+ or SOM+ neurons in a Cre-dependent manner (Fig. 2A). Interestingly, in PV-IRES-Cre mice, the vast majority of PV+ synaptic terminals were localized in the ipsilateral MGBv (Fig. 2B, second row; Fig. 2D), which mainly comprises of bushy cells (Fig. 2E), even when the virus was injected in a subdivision-independent manner (Fig. 2B, first row), and only very few terminals were observed in the contralateral MGBv, the medial part of the superior olivary complex (SOCm) and the SC (Fig. 2F). In contrast, SOM+ synaptic terminals were mainly distributed in the ipsilateral POL (Fig. 2C, second row; Fig. 2G), which mainly comprises of bipolar cells (Fig. 2H), when virus transduction was observed in all three subdivisions (Fig. 2C, first row), and were very sparsely located in the lateral part of ipsilateral SOC (SOCl) and in the SC (Fig. 2I). No mRuby-labeled synaptic terminals was observed in the thalamus (Fig. 2P, right panel; n = 4 sides from N = 2 mice) when the same virus, mixed with CTB-647 to demonstrate injection sites, was injected into the IC of WT mice (Fig. 2P, left panel), validating the essential dependency of mRuby expression on Cre.
ICPV+ or ICSOM+ axon terminals predominantly innervate the primary or secondary auditory thalamus. A, Experimental procedures for cell type–specific anterograde tracing. Top panel: injection of AAV-Flex-synaptophysin-mRuby into the IC of PV-IRES-Cre or SOM-IRES-Cre mice. Bottom panel: anterograde projections from the IC to the major subdivisions of the auditory thalamus. B, C, Example images showing when the virus was injected in a subdivision-independent manner into the IC of PV-IRES-Cre (top green box) or SOM-IRES-Cre (top red box), scale bar = 200 μm, the projections to the auditory thalamus (PV+, bottom green box; SOM+, bottom red box) in three AP distance (AP = −2.77, −2.97, −3.17), scale bar = 300 μm. D, Example images showing when the virus was injected in a subdivision-independent manner into the IC of PV-IRES-Cre, the projections to the ipsilateral auditory thalamus from anterior to posterior coordinate. Scale bar, 300 μm. E, Confocal 10× image showing the morphology of MGBv neuron by sparse labeling. Scale bar, 50 μm. F, Example images showing when the virus was injected in a subdivision-independent manner into the IC of PV-IRES-Cre, the projections to brain regions other than the ipsilateral auditory thalamus. Distribution of axonal terminals in the contralateral MGBv (con-MGBv), SOCm, and SC. Scale bar, 300 μm. G, Example images showing when the virus was injected in a subdivision-independent manner into the IC of SOM-IRES-Cre, the projections to the ipsilateral auditory thalamus from anterior to posterior coordinate. Scale bar, 300 μm. H, Confocal 10× image showing the morphology of POL neuron by sparse labeling. Scale bar, 50 μm. I, Example images showing when the virus was injected in a subdivision-independent manner into the IC of SOM-IRES-Cre, the projections to brain regions other than the ipsilateral auditory thalamus. Distribution of axonal terminals in the SOCl and SC. Scale bar, 300 μm. J–L, Left panel: example images showing when the virus was injected into the ICE (J), ICC (K), or ICD (L) of the PV-IRES-Cre mice, scale bar = 200 μm. Middle panel: distribution of ICPV+ terminals to the auditory thalamus sampling of 3,000 terminals from each mouse brain slice (N = 3) at the −2.97 mm coordinate. Right panel: percentage (bar chart) and weighted density (line chart) of ICPV+ neurons projecting to different subdivisions in the auditory thalamus, N = 3 per group. M–O, Left panel: example images showing when the virus was injected into the ICE (M), ICC (N), and ICD (O) of the SOM-IRES-Cre mice, scale bar = 200 μm. Middle panel: distribution of ICSOM+ terminals to the auditory thalamus sampling of 300 (N) or 3,000 (M,O) terminals from each mouse brain slice (N = 3) at the −2.97 mm coordinate. Right panel: percentage (bar chart) and weighted density (line chart) of ICSOM+ neurons projecting to different subdivisions in the auditory thalamus, N = 3 per group. P, Validation of the dependency of mRuby expression on Cre. Left panel: example image showing injection AAV-Flex-synaptophysin-mRuby with CTB-647 into the IC of WT mice. Scale bar, 200 μm. Right panel: the projections to the auditory thalamus. Scale bar, 300 μm. Q, Schematic diagram illustration of the primary projection of PV+ or SOM+ neurons in individual subdivisions in the IC to the auditory thalamus. PV+ neurons (left panel) in the ICE (purple), ICC (ochre), or ICD (cyan) predominately project to the MGBv. SOM+ neurons (right panel) in the ICE, ICC, or ICD predominately project to the POL.
To characterize the distribution of PV+ or SOM+ synaptic terminals in the thalamus originating from individual IC subdivisions, we injected AAV-Flex-synaptophysin-mRuby into each individual IC subdivisions of PV-IRES-Cre or SOM-IRES-Cre mice as localized as possible (N = 3 mice per group). We found that, regardless of the subdivision injected, the majority of PV+ terminals were observed in the MGBv (Fig. 2K, ICC, 94%; Fig. 2L, ICD, 97%; Fig. 2J, ICE, 87%; N = 3 per group), while SOM+ terminals following localized virus injection in IC subdivisions were predominantly located in the secondary auditory thalamic nuclei surrounding the MGBv, in which the POL had the most axon terminals (Fig. 2M, ICE, 70%; Fig. 2O, ICD, 67%; Fig. 2N, ICC, 50%; N = 3 per group), followed by the MGBd and PIN. These results were consistent with the phenomenon observed following subdivision-independent virus injections (Fig. 2B,C) and demonstrated a strong preference for ICPV+ and ICSOM+ projections for the primary and secondary auditory thalamus, respectively (Fig. 2Q).
Since the connectivity of the POL remained poorly understood as a subdivision of the auditory thalamus (Márquez-Legorreta et al., 2016), we intended to determine the downstream target regions of POL neurons receiving ICSOM+ inputs. Given that we had no access to the viral tool that can enable cell type–specific trans-synaptic anterograde labeling, we estimated the downstream regions by injecting scAAV2/1-Cre-PA virus into the IC of WT mice to trans-synaptically express Cre in thalamic neurons receiving IC inputs, followed by targeted injections of AAV2/9-Flex-synaptophysin-mRuby into the POL to enable Cre-dependent expression of mRuby in POL terminals (Fig. 3A). Only the cases exhibiting local virus transduction in the POL (Fig. 3B, injection site; N = 4 mice) were used for further analysis. In addition, we revealed the downstream target region of the MGBv by using a similar tracing strategy except that AAV2/5-Flex-tdTomoto-T2A-synaptophysin-EGFP was locally injected into the MGBv (Fig. 3E,F, injection site; N = 4 mice).
The POL or MGBv axon terminals predominantly innervate the TS or AUDp. A, Experimental procedures for anterograde tracing of IC→POL pathway. Left panel: scAAV2/1-Cre-PA virus was injected into the IC of WT mice. Right panel: targeted injections of AAV2/9-Flex-synaptophysin-mRuby into the POL. D, dorsal; M, medial. B, Example images showing expression of mRuby localized in the POL neurons. Scale bar = 300 μm. C, Example images showing several target brain regions of the synaptic terminals of POL neurons receiving IC inputs. Scale bar = 300 μm. D, Normalized outputs of the synaptic terminals of POL neurons receiving IC inputs in several brain regions. N = 4 mice. Data are means ± SEM. E, Experimental procedures for anterograde tracing of IC→MGBv pathway. Left panel: scAAV2/1-Cre-PA virus was injected into the IC of WT mice. Right panel: targeted injections of AAV2/5-Flex-tdTomoto-T2A-synaptophysin-EGFP into the MGBv. F, Example images showing expression of mRuby localized in the MGBv neurons. Scale bar, 300 μm. G, Example images showing major target brain region of the synaptic terminals of MGBv neurons receiving IC inputs. Scale bar, 300 μm.
We found that ∼70% of the synaptic terminals of POL neurons receiving IC inputs (ICPOL) were concentrated in the ipsilateral TS, followed by the AUDv (∼12%) and secondary supplemental somatosensory area (SSs; ∼12%; Fig. 3C,D). Only sparse terminals were observed in the lateral dorsal nucleus of thalamus (LD), subparafascicular nucleus (SPF), zona incerta (ZI), SC, and the periaqueductal gray (PAG; Fig. 3C). In contrast, the terminals of MGBv neurons receiving IC inputs (ICMGBv) were exclusively distributed in the layer 4 of AUDp (Fig. 3G), which was consistent with the classic IC→MGBv→A1 circuit model established by traditional tracing method (Ehret and Romand, 1997). These data suggested that the major downstream target of POL neurons receiving ICSOM+ inputs is the TS and that the ICSOM+→POL circuit belongs to the secondary auditory pathway because ICPOL neurons only project to the secondary auditory and somatosensory cortices.
ICSOM+ neurons receive significantly more polymodal inputs compared with ICPV+ neurons
The above data showed that ICPV+ and ICSOM+ neurons project to distinct thalamic nuclei, but would they receive inputs from distinct brain regions as well? To address this question, we adopted a widely used RV-mediated retrograde monosynaptic tracing strategy (Callaway and Luo, 2015; Fig. 4A; see Materials and Methods for more details). The locations of starter cells from all mice were shown in Figure 4F (N = 4 mice per group, AP = −4.97). Since the number of input neurons and starter neurons was linearly correlated (Fig. 4B,C,H), we determined the input strength of a specific region by dividing the number of input neurons in that region by the total number of input neurons (N = 4 mice per group). We categorized the retrogradely labeled brain regions into auditory-related and polymodal-related regions based on their primary functional roles (Fig. 4I–K).
ICSOM+ neurons receive significantly more polymodal inputs compared with ICPV+ neurons. A, Schematic diagram showing RV-mediated monosynaptic retrograde tracing. Left: time course of virus injection and histology. Right: construction of AAV helper viruses (AAV-DIO-EGFP-TVA and AAV-DIO-RG) and pseudotyped RV. B, C, Left panel: representative fluorescent images showing the infection of RV and AAV in the IC of PV-IRES-Cre (B) or SOM-IRES-Cre (C) mice, scale bar = 200 μm. Right panel: enlarged view of the boxed area showing the expression of AAV-Helper, input, and starter, respectively, scale bar = 25 μm. White arrows in the right panel showing starter neurons expressing both EGFP and DsRed. D, Example images showing major inputs of the PV+ neurons in the IC, and the arrangement of the images is from the anterior to the posterior. Scale bar, 200 μm. E, Example images showing major inputs of the SOM+ neurons in the IC, and the arrangement of the images is from the anterior to the posterior. Scale bar, 200 μm. F, The locations of the starter cells from PV-IRES-Cre (left panel) and SOM-IRES-Cre (right panel) in −4.97 AP. The bottom right showed the mouse of each group. G, Left panel: validation of RG-dependence of RV-mediated monosynaptic retrograde tracing. No retrogradely labeled neurons were observed when AAV-DIO-RG was not injected. Scale bar, 200 μm. Right panel: validation of TVA-dependence of RV-mediated monosynaptic retrograde tracing. No retrogradely labeled neurons were observed when AAV-DIO-TVA-EGFP was not injected. Scale bar, 200 μm. H, A linear relationship was detected between the number of starter and input neurons. I, Normalized inputs of ICSOM+ and ICPV+ neurons from auditory-related and polymodal-related brain regions. N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, ***p < 0.001, ****p < 0.0001. Data are means ± SEM. J, Normalized inputs of ICSOM+ and ICPV+ neurons from auditory-related brain regions. N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, **p < 0.01, ***p < 0.001, ****p < 0.0001. K, Normalized inputs of ICSOM+ and ICPV+ neurons from polymodal-related brain regions. N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, **p < 0.01. Data are means ± SE.M.
We found that ICPV+ neurons received significantly more inputs from auditory-related regions (Fig. 4I; N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, ****p < 0.0001), particularly significant inputs from the contralateral IC (∼35%), ipsilateral VNLL (∼32%), and contralateral CN (∼18%; Fig. 4D,J; N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, **p < 0.01, ***p < 0.001, ****p < 0.0001), suggesting that ICPV+ neurons belong to the classic ascending auditory pathway. In contrast, ICSOM+ neurons received significantly more inputs from polymodal sensorimotor regions (Fig. 4I; N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, ***p < 0.001), notably significant inputs from the PAG (∼14%), SC (∼11%), and CUN (∼11%; Fig. 4E,K; N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, p < 0.01). Interestingly, ICSOM+ neurons received very few inputs from contralateral CN (Fig. 4E,J), further indicating that these neurons do not belong to the primary auditory pathway. There was no significant difference between the two neuron classes in receiving inputs from other brain regions (Fig. 4J,K; N = 4 mice per group, two-way ANOVA with Geisser–Greenhouse correction, p > 0.05).
Together with the outputs of ICPV+ and ICSOM+ neurons, these data suggest that ICPV+ neurons may be mainly responsible for relaying sound feature-related information from the periphery to the auditory cortex via the primary auditory thalamus, while ICSOM+ neurons can integrate polymodal, behaviorally meaningful inputs to modulate mice's behavioral states via the POL→TS pathway.
ICPV+ neurons are more heterogenous in synaptic morphology than ICSOM+ neurons
The distinct input–output architectures of ICSOM+ and ICPV+ neurons inspired us to hypothesize that these two classes of neurons may have distinct anatomical features, which could contribute to the potential functional distinctions between them.
To test the above hypothesis, we first performed anti-GAD and anti-PV immunostaining using brain slices containing the IC from SOM- Ai14 and VGAT-Ai14 mice, respectively. GAD (Ono et al., 2005; Ito et al., 2009) and VGAT (Oberle et al., 2023), which are commonly used markers for GABAergic neurons, are coexpressed in IC neurons (Fig. 5A). We found that, in all IC subdivisions, no SOM+ neurons were labeled by anti-GAD immunostaining (Fig. 5B; n = 12 slices from N = 3 mice), indicating that all SOM+ neurons in the IC are excitatory. In contrast, ∼25% of counted PV+ neurons were VGAT+, and this phenomenon was IC subdivision-independent (Fig. 5C,E; n = 12 slices from N = 3 mice; ICE, 26 ± 1.5%; ICC, 21 ± 3%; ICD, 28 ± 3.6%; one-way ANOVA with post hoc, p > 0.05), indicating that ICPV+ neurons can be divided into inhibitory and excitatory groups. These observations were further confirmed by anti-GAD immunofluorescent staining of mRuby-labeled ICSOM+ or ICPV+ terminals in the thalamus, which showed that all ICSOM+→POL terminals were GAD− whereas ICPV+→MGBv terminals were either GAD− or GAD+ (Fig. 5D), and the inhibitory ICPV+→MGBv terminals accounted for ∼12% of the total ICPV+→MGBv terminals (Fig. 5F; n = 6 slices from N = 3 mice).
ICPV+ neurons are more heterogeneous in synaptic morphology than ICSOM+ neurons. A, Confocal 20× images showing VGAT neurons (red, white arrows, left), GAD65/67 staining (green, yellow arrows, middle), and an overlay (right). VGAT and GAD are coexpressed in the IC. Scale bar, 10 μm. B, Confocal 20× images showing ICSOM+ neurons (red, left), GAD65/67 staining (green, middle), and an overlay (right). Scale bar, 200 μm. Insets, enlarged view of the dotted box, and the white arrows mark IC neurons expressing GAD65/67, while the yellow arrows mark IC neurons expressing SOM. There was no overlap between SOM+ neurons and GABAergic neurons (right). Scale bar, 10 μm. C, Confocal 20× images showing ICVGAT+ neurons (red, left), PV staining (green, middle), and an overlay (right). Scale bar, 200 μm. Insets, enlarged view of the dotted box, and the white arrows mark IC neurons expressing PV, while the yellow arrows mark IC neurons expressing VGAT. There were some overlaps between PV+ neurons and VGAT+ neurons (right). Scale bar, 10 μm. D, Confocal 40× images showing ICSOM+→POL terminals (left panel; red channel: synaptophysin-mRuby), ICPV+→MGBv terminals (right panel; red channel: synaptophysin-mRuby), and GAD65/67 staining (cyan channel). There was no overlap between ICSOM+→POL terminals and GAD65/67 (left panel), while there were some overlaps between ICPV+→MGBv terminals and GAD65/67 (right panel, yellow arrows). Scale bar, 5 μm. E, Percentage of PV neurons colabeled with VGAT (PVVGAT) among the total PV+ neurons in individual IC subdivision (ICE, ICC, and ICD). n = 12 slices from N = 3 mice; one-way ANOVA with post hoc, p > 0.5. F, The percentage of inhibitory ICPV+→MGBv terminals, n = 6 slices from N = 3 mice. G, H, Confocal 40× images showing the morphological features of ICSOM+→POL terminals (G) and ICPV+→MGBv terminals (H). Scale bar, 100 μm. The top right panel showing sparer terminals by sparse labeling of ICSOM+ (G) and ICPV+ neurons (H), and the bottom right panel showing a schematic diagram of the morphology of ICSOM+→POL (G) and ICPV+→MGBv terminals (H). Scale bar, 5 μm. I, Average terminal sizes of ICSOM+→POL (N = 6 mice) and ICPV+→MGBv (N = 12 mice). Unpaired t test, p < 0.0001. J, Large terminal (defined as terminals with sizes larger than 1.5 μm) proportion of ICSOM+→POL (N = 6 mice) and ICPV+→MGBv (N = 12 mice). Unpaired t test, p < 0.0001. K, Terminal size variation of ICSOM+→POL (N = 6 mice) and ICPV+→MGBv (N = 6 mice). Unpaired t test, p < 0.001. L, Terminal skewness of ICSOM+→POL (N = 6 mice) and ICPV+→MGBv (N = 12 mice). Unpaired t test, p < 0.001. Data are means ± SEM. M, Terminal size distribution of ICSOM+→POL (N = 6 mice) and ICPV+→MGBv (N = 12 mice). The SEM is shown as light shadows. The bins are 0.1 μm. N, Confocal 100× images showing axonal terminals of ICPV+→MGBv (first column, red), anti-GAD65/67 (first column, green), and merge (second column; pink arrow: excitatory PV+ terminals; cyan arrow: inhibitory PV+ terminals), scale bar = 3 μm. O–R, Average terminal sizes (O), large terminal proportions (P), terminal size variation (Q), and terminal skewness (R) of inhibitory PV+ terminals, excitatory PV+ terminals, and SOM+ terminals, unpaired t test, ***p < 0.001, ****p < 0.0001, n.s.p > 0.05. Data are means ± SEM.
We next characterized the morphological features of ICSOM+ and ICPV+ terminals in the POL and MGBv, respectively, to estimate their potential influences on their postsynaptic thalamic neurons. Toward this end, we injected AAV2/9-Flex-synaptophysin-mRuby into the IC of SOM-IRES-Cre or PV-IRES-Cre mice to label ICSOM+ or ICPV+ terminals and utilized our newly developed rapid and deformation-free on-slide tissue-clearing method to visualize the terminal morphology and terminal size (see Materials and Methods for more details). We used AAV2/9-Flex-synaptophysin-mRuby-T2A-EGFP to confirm the localization of mRuby signals at axonal swellings suggesting presynaptic terminals, including boutons and varicosities (Fig. 5G,H). Since ICPV+→MGBv terminals are composed of both excitatory and inhibitory ones, we injected AAV-Flex-synaptophysin-mRuby into the IC of PV-IRES-Cre mice to label PV+ terminals (Fig. 5N, red fluorescence) and then stained thalamic slices with GAD65/67 antibodies (Fig. 5N, green fluorescence), which labeled PV+ inhibitory axonal terminals (Fig. 5N, merge).
Compared with the sizes of ICSOM+→POL terminals, those of ICPV+→MGBv terminals were averagely larger (Fig. 5G–I; SOM+ group, N = 6 mice; PV+ group, N = 12 mice; PV+ terminals, 1.4 ± 0.07 μm; SOM+ terminals, 1.1 ± 0.02 μm; unpaired t test, p < 0.0001), partly contributed by a higher proportion of terminals that were larger than 1.5 μm (Fig. 5J; PV+ terminals, 0.35 ± 0.07; SOM+ terminals, 0.12 ± 0.02; unpaired t test, p < 0.0001) and more heterogeneous (Fig. 5K, PV+ terminals, 0.16 ± 0.02; SOM+ terminals, 0.09 ± 0.01; unpaired t test, p < 0.0001), and their distribution was more deviated from the normal distribution (Fig. 5L; PV+ terminals, 0.67 ± 0.15; SOM+ terminals, 0.24 ± 0.06; unpaired t test, p < 0.0001). All of the above phenomena can also be found in the distribution diagram of the two groups’ terminal size (Fig. 5M), which additionally showed that the representative sizes of ICPV+→MGBv terminals spanned from 1.0 to 1.8 μm, but those of ICSOM+→POL terminals were restricted to ∼1.1 μm.
We then reconstructed the terminals and analyzed the morphological parameters of PV+ excitatory or PV+ inhibitory terminals using the same method. First, we found that there was no significant difference between PV+ excitatory and PV+ inhibitory terminals across all morphological parameters (Fig. 5O–R, N = 6 mice; unpaired t test, p > 0.05). Second, PV+ excitatory terminals were significantly larger and more heterogeneous than SOM+ terminals (Fig. 5O–R, SOM+ group, N = 6 mice; PV+ group, N = 6 mice; unpaired t test, ***p < 0.001, ****p < 0.0001), which are excitatory as well. These results indicate that the morphology of PV+ terminals was very likely independent of their excitatory/inhibitory nature.
Taken together, the synaptic morphology of ICPV+ neurons is distinct from and more heterogeneous than that of ICSOM+ neurons.
ICPV+ neurons are electrophysiologically more heterogenous than ICSOM+ neurons
We next conducted whole-cell current-clamp recordings from ICPV+ or ICSOM+ neurons by using slices from PV×Ai14 or SOM×Ai14 mice to characterize their intrinsic electrophysiological properties. We found that the firing of recorded ICSOM+ neurons (n = 22 cells from N = 4 mice) in response to current injection uniformly demonstrated adapting profile (Fig. 6A, 1). With the increase of current amplitude, the firing rate of these neurons increased nearly linearly at first and then reached a plateau (Fig. 6B, red triangle). In contrast, three types of firing profiles were observed in ICPV+ neurons (n = 28 cells from N = 4 mice), including rapid inactivating (PV-RI; Fig. 6A, 3; n = 12 cells), slowly adapting (PV-SA; Fig. 6A, 4; n = 12 cells), and fast spiking (PV-FS; Fig. 6A, 2; n = 4 cells). We found that the posthyperpolarization rebound was often observed in PV+ neurons, which also demonstrated a large current sag when a negative current was injected (Fig. 6A, 2–4). The characteristic feature of PV-RI neurons was they only fired one or two action potentials at the onset of the current injection, and their firing rate barely changed with the increase of current amplitude (Fig. 6B, green squares). The firing rate of PV-FS neurons was considerably higher than that of PV-SA neurons. Additionally, we observed an elevated coefficient of covariation of interspike intervals (ISI) in these PV-SA neurons, which suggests the presence of SK channels. Although the firing rate of both classes of neurons increased largely linearly with current amplitude, that of PV-FS neurons increased much faster compared with PV-SA neurons (Fig. 6B). Thus, the firing mode of ICPV+ neurons was more diverse than that of ICSOM+ neurons.
ICPV+ neurons are electrophysiologically more heterogeneous than ICSOM+ neurons. A, Representative traces from whole-cell current-clamp recordings obtained from SOM+ and PV+ neurons in the IC. (A1) shows adapting profile of ICSOM+ neurons, (A2) FS of ICPV+ neurons, (A3) rapid inactivating of ICPV+ neurons, and (A4) slowly adapting of ICPV+ neurons. Scale bar, 40 mV/300 ms. B, Mean values of step current versus spikes/s (I/F) curves calculated from all recorded ICSOM+ (red triangles, n = 22 cells from N = 4 mice) and ICPV+ neurons showing the most common spiking classes: FS (PV-FS; green triangles, n = 4 cells from N = 4 mice), slowly adapting (PV-SA; green circles, n = 12 cells from N = 4 mice), and rapid inactivating (PV-RI; green squares, n = 12 cells from N = 4 mice). C, Comparison of resting membrane potential for SOM+ neurons (n = 22 cells from N = 4 mice), PV-RI neurons (n = 12 cells from N = 4 mice), and PV-SA neurons (n = 12 cells from N = 4 mice), one-way ANOVA test with multiple comparisons, p < 0.05. D, Comparison of rheobase for SOM+, PV-RI, and PV-SA neurons. Kruskal–Wallis test with multiple comparisons, **p < 0.01, ***p < 0.001. E, Comparison of sag ratio for SOM+, PV-RI, and PV-SA neurons. Kruskal–Wallis test with multiple comparisons, *p < 0.05, **p < 0.01. F, Comparison of time constant for SOM+, PV-RI, and PV-SA neurons. Kruskal–Wallis test with multiple comparisons, p < 0.05, **p < 0.01. Data are means ± SEM. G, Two-dimensional t-SNE representation of electrophysiological properties for two molecular cell classes. Clustering was performed using the electrophysiological parameters we analyzed.
We further compared the membrane properties between ICSOM+ neurons and two subclasses of ICPV+ neurons (PV-RI and PV-SA), and PV-FS neurons were not included in the statistical analysis due to their paucity (4 out of 28 recorded PV+ neurons). We found that PV-RI and PV-SA neurons are significantly different from each other in terms of time constant (Fig. 6F; PV-RI, 8.13 ± 1.65 ms; PV-SA, 13.94 ± 1.37 ms; Kruskal–Wallis test with multiple comparisons, p < 0.05), but no significant difference was observed for the other three intrinsic membrane properties examined, including resting membrane potential, rheobase, and sag ratio (Fig. 6C,E, Kruskal–Wallis test with multiple comparisons, p > 0.05). In contrast, ICSOM+ neurons, which are adapting neurons, are significantly different from at least one ICPV+ subclass in all the four intrinsic membrane properties examined. Specifically, ICSOM+ neurons had significantly lower resting membrane potential (Fig. 6C; SOM+, −64.6 ± 1.9 mV; PV-RI, −55.8 ± 4.2 mV; PV-SA, −55.1 ± 2.5 mV; one-way ANOVA test with multiple comparisons, p < 0.05), lower rheobase (Fig. 6D; SOM+, 33.6 ± 5.7 pA; PV-RI, 126.3 ± 21.8 pA; PV-SA, 83.8 ± 14.9 pA; Kruskal–Wallis test with multiple comparisons, **p < 0.01, ***p < 0.001), and higher sag ratio (Fig. 6E; SOM+, 0.89 ± 0.03; PV-RI, 0.70 ± 0.05; PV-SA, 0.63 ± 0.06; Kruskal–Wallis test with multiple comparisons, *p < 0.05, **p < 0.01), suggesting significant differences in the expression of potassium, sodium, and hyperpolarization-activated channels, respectively, between ICSOM+ and ICPV+ neurons. We also noticed that the rheobase and sag ratio of ICPV+ neurons were distributed in a wide range. In addition, the time constants of ICSOM+ neurons were significantly higher than those of PV-RI neurons (Fig. 6F; SOM+, 16.16 ± 1.95 ms; PV-RI, 8.13 ± 1.65 ms; Kruskal–Wallis test with multiple comparisons, p < 0.01), suggesting that the soma of ICSOM+ neurons was larger.
Considering that ICSOM+ and ICPV+ neurons demonstrated distinctions in multiple dimensions of intrinsic electrophysiological properties, we performed clustering and dimensionality reduction using the t-SNE algorithm to obtain a two-dimensional embedding of these two cell classes (Fig. 6G). The t-SNE plot showed two well-separated clusters, where one was mainly composed of ICSOM+ and the other one predominantly comprised ICPV+ neurons. The more dispersed distribution of ICPV+ neurons was consistent with their higher degree of heterogeneity in intrinsic electrophysiological properties.
Discussion
In the present study, we showed that PV+ and SOM+ neurons in the IC, regardless of their location, predominantly contribute to the primary and secondary auditory pathways, respectively. Remarkable differences in input sources, intrinsic electrophysiological properties, and axon terminal features collectively suggested that these two classes of neurons are functionally distinct (Fig. 7). Our findings reconciled the discrepancies among previous studies characterizing tectothalamic pathways based on the subdivisions of the IC and provided an anatomical framework for molecular marker-based, functional interrogation of the tectothalamic pathways.
Schematic diagram illustrating the secondary and primary auditory pathways by ICSOM+ and ICPV+ neurons. The schematic diagram contains the distribution, intrinsic electrophysiological property, axon terminal features, and input–output architecture of ICSOM+ (red) and ICPV+ (green) neurons. The darker color represents higher distribution density or stronger projections of ICSOM+ or ICPV+ neurons.
Biomarker-based versus subdivision-based auditory tectothalamic pathways
Traditionally, auditory tectothalamic pathways have been defined based on the three subdivisions of the IC, highlighting topographic innervation of the auditory thalamus by different IC subdivisions. This definition was mainly derived from retrograde tracing studies using traditional tracers such as HRP (Calford and Aitkin, 1983) and retro-beads (Mellott et al., 2014). However, data in these studies has also shown that an individual thalamic subdivision can actually receive inputs from all IC subdivisions despite the preference for certain IC subdivision. Echoing this observation, studies using traditional anterograde tracers show that neurons in an individual IC subdivision can send their axons to several auditory thalamic subdivisions (LeDoux et al., 1985). These results strongly suggest that the auditory tectothalamic pathways defined by IC subdivisions may have been oversimplified. In the present study, by combining transgenic mice with cutting-edge viral tracing tools and our newly developed tissue-clearing method, we showed that PV+ and SOM+ neurons project to distinct auditory thalamic subdivisions. This finding provides an alternative approach to defining auditory tectothalamic pathways and can likely interpret the intricate observations made by using traditional tracers. Nevertheless, our results do not exclude the possibility that IC neurons expressing other biomarkers may also contribute to distinct tectothalamic pathways, although it has been shown that VIP+ and NPY+ neurons do not do so (Silveira et al., 2020; Beebe et al., 2022).
Potential function of ICPV+ and ICSOM+ neuron–mediated auditory pathways
We showed that ICPV+ neurons represent ∼15% of IC neurons and that receive inputs predominantly from the contralateral IC, ipsilateral NLL, and contralateral CN, which are the major stations in the ascending auditory pathway (Oliver et al., 2018), strongly suggesting that ICPV+ neurons mainly receive auditory inputs. We also demonstrated that ICPV+ neurons are highly heterogeneous in terms of their intrinsic electrophysiological properties and axon terminal size. This remarkable heterogeneity can likely enhance the robustness of the representation and transmission of time-varying acoustic features by ICPV+ neurons (Chen et al., 2022). Furthermore, the MGBv, the primary auditory thalamus, is the dominating downstream target of ICPV+ neurons. These data collectively suggest that ICPV+ neurons may play an important role in auditory processing by providing the MGBv with both feedforward excitation and inhibition.
In contrast, ICSOM+ neurons, which represent ∼25% of IC neurons, can likely integrate auditory inputs with polymodal, threat-indicative inputs from multiple sources, such as the PAG (Lefler et al., 2020), SC (Basso et al., 2021), and CUN (Caggiano et al., 2018), suggesting that ICSOM+ activity correlates with animal behaviors. Interestingly, both the electrophysiological and anatomical heterogeneity of ICSOM+ neurons are much lower than those of ICPV+ neurons, suggesting that these neurons are more suitable for sensory detection rather than discrimination and capable of effective information transmission. Furthermore, ICSOM+ neurons predominantly project to the POL, which in turn mainly innervates the TS. It has been shown that TS neurons can mediate alerting stimulus-induced arousal (Wang et al., 2023) and the reinforcement of defensive behavior (Menegas et al., 2018) in freely moving mice. Together, our data suggest that ICSOM+ neurons are responsible for detecting polymodal, alerting stimuli, and effectively triggering the transition of behavioral state to defense.
Molecular biomarkers as an approach for classifying IC neurons
Neuron types in the IC have been identified using many different parameters such as dendritic morphology (Oliver and Morest, 1984), neurotransmitter synthesis (Beebe et al., 2016), potassium currents (Sivaramakrishnan and Oliver, 2001), axonal target (Chen et al., 2018), and synaptic morphology such as glutamatergic axosomatic terminals (Lee and Sherman, 2010). Biomarkers also have been used such as VIP and NPY (Goyer et al., 2019; Silveira et al., 2020). Thus, neuron types in the IC, like elsewhere in the CNS, must be defined according to each of these characteristics since single parameters are inadequate. Some pioneering studies have successfully linked the biomarkers expressed by IC neurons to specific cellular properties. For example, VIP+ neurons, which are excitatory neurons expressing GluN2C/D-containing NMDA receptors (Drotos et al., 2023), are characterized by their sustained firing pattern (Goyer et al., 2019). NPY+ neurons, which are inhibitory neurons (Silveira et al., 2020), are featured by the coexistence of GABAergic and NPY signaling (Silveira et al., 2023), which can likely regulate excitation in local IC circuits.
In the present study, we found that all SOM+ neurons are excitatory neurons with similar firing pattern, intrinsic membrane properties, axonal target, and synaptic morphology, suggesting that SOM+ neurons are highly homogeneous. In contrast, PV+ neurons are quite heterogeneous in terms of neurotransmitter synthesis, firing pattern, intrinsic membrane properties, and synaptic morphology. Despite the heterogeneity, PV+ neurons predominantly project to the MGBv, suggesting that PV can be used as a biomarker for a class of neurons with the same axonal target. Nevertheless, PV+ neurons should be further classified in the future.
Among the identified IC cell classes with distinct biomarkers, SOM+, VIP+, and three-quarters of PV+ neurons are excitatory neurons. Since a majority of IC glutamatergic neurons express NPY Y1 receptor (Silveira et al., 2023), which mediates an inhibitory effect of NPY-binding, it's very likely that the excitability of SOM+, VIP+, and PV+ neurons and their influence on downstream targets can be regulated by NPY+ neurons. For a more thorough understanding of the multifaceted roles of the IC in sound processing, determining the local interactions between different cell classes is extremely meaningful and should be encouraged along with further classification of IC neurons.
Technical limitations
Firstly, the Golgi method applied to the mouse IC reveals disc-shaped/flat cells defining the ICC (Meininger et al., 1986). However, biomarker-based papers consistently describe VIP+, NPY+, SOM+, and PV+ cells as stellate in single brain slices (Goyer et al., 2019; Silveira et al., 2020). Unfortunately, fully assessing dendritic morphology usually involves labeling cells in the intact brain and reconstructing 3D morphology. When viewed orthogonally to the lamina, IC neuron dendritic fields may seem stellate, potentially requiring 3D rotation. It's possible that our method may not have sufficiently explored dendritic morphology, potentially resulting in an oversight of flat dendritic fields.
Secondly, given that an anterograde tracing virus that can enable cell type–specific trans-synaptic labeling was not available to us, we were not able to selectively characterize the downstream targets of POL neurons receiving ICSOM+ inputs. Although the axon terminals of POL neurons are predominantly distributed in the TS, ∼20% of the terminals were also observed in the AuDv and SSs. Therefore, the exact distribution pattern of the axon terminals of POL neurons receiving ICSOM+ input remains unclear. Nevertheless, our data showed that the axon terminals of ICSOM+ neurons cover the entire POL in both AP and DV directions with high density (Fig. 2G), suggesting that most POL neurons receive ICSOM+ input. Thus, it is very likely that the projection pattern of POL neurons revealed by our cell type–independent approach applies to ICSOM+-targeted POL neurons as well.
Footnotes
We thank B.H. for instrumental support, S.S. for the SOM-IRES-Cre mice supplement, and Y.Z. for assistance with using the Andor Oxford Instruments (Imaging Core Facility, Technology Center for Protein Sciences, Tsinghua University). K.Y. receives funding from the STI 2030-Major Projects 2021ZD0200300, National Natural Science Foundation of China (31871057, 32070993, 81527901, T2341003), Beijing Municipal Science & Technology Commission (Z181100001518004, Z181100001518006), and Guoqiang Institute, Tsinghua University. N.W. receives funding from the National Natural Science Foundation of China (81770993) and Capital’s Funds for Health Improvement and Research (2020-1-2032). M.L. receives funding from the Postdoctoral Funding Program (LBH-Z23220).
The authors declare no competing financial interests.
↵*M.L. and Y.G. contributed equally to this work.
↵‡K.Y. is the lead contact.
- Correspondence should be addressed to Kexin Yuan at kexinyuan{at}mail.tsinghua.edu.cn or Ningyu Wang at 2460331882{at}qq.com.
