In this study, we develop a comprehensive and precise whole-brain atlas of monosynaptic long-range inputs to histaminergic neurons by integrating a cutting-edge fMOST system with HDC-CreERT2 mice and a retrograde RV tracing system. Importantly, we further select the LS and paraventricular nucleus of the thalamus (PVT) as representative inhibitory and excitatory upstream brain regions, to investigate their differential functional dynamics and causal roles in the sleep-wake cycle. These results provide a foundation for future functional studies of neural circuits projecting to the histaminergic nervous system.
Whole-brain distribution of monosynaptic long-range inputs to histaminergic neurons
To genetically label the monosynaptic long-range input neurons projecting to histaminergic neurons, we first injected a mixture of rAAV-EF1α-DIO-mCherry-F2A-TVA-WPRE-hGH polyA and rAAV-EF1α-DIO-oRVG-WPRE-hGH polyA (TVA:RVG mixing ratio = 1:2) into the TMN of HDC-CreERT2 mice (Fig. 1a). These two helper viruses enabled TVA and RVG protein expression in Cre-positive neurons, where TVA mediated RV-ENVA-ΔG-EGFP entry into the cell body, and RVG facilitated monosynaptic retrograde tracing by RV-ENVA-ΔG-EGFP. Tamoxifen (100 mg/kg, i.p.) was administered for 5 consecutive days starting one day after viral injection to induce Cre recombinase activity (all CreERT2 mice received tamoxifen administration). Three weeks later, RV-ENVA-ΔG-EGFP was injected into the same region (Fig. 1a). Cells co-labeled with TVA-mcherry and RV-EGFP were defined as starter cells (Fig. 1b, d). RV-labeled upstream neurons were distributed across multiple brain regions at whole-brain scale, exhibiting distinct distribution patterns (Fig. 1c). When RV retrograde tracing system was injected unilaterally into the TMN (Supplementary Fig. 1a), RV-labeled upstream neurons were predominantly located in the ipsilateral hemisphere (Supplementary Fig. 1b).
To verify the specificity and reliability of RV monosynaptic retrograde tracing system, we performed a series of control experiments. First, to assess viral expression specificity, we injected rAAV-EF1α-DIO-mCherry-F2A-TVA-WPRE-hGH polyA bilaterally into the TMN of HDC-CreERT2 mice (Supplementary Fig. 1c). Immunohistochemical staining for HDC revealed that the majority of TVA-labeled neurons were co-labeled with HDC (totally 87.27 ± 1.298 %, n = 3), indicating the convincing specificity of Cre-dependent expression manner of rAAV-EF1α-DIO-mCherry-F2A-TVA-WPRE-hGH polyA (Supplementary Fig. 1d, e). Furthermore, the expression specificity of starter cells (totally 86.50 ± 0.4513%, n = 3) was verified by immunohistochemistry in tested mice expressing the complete RV system (Supplementary Fig. 1a, f-h). To further evaluate potential leakage of RV expression, we conducted additional control experiments. Injection of RV-ENVA-ΔG-EGFP alone into the TMN of HDC-CreERT2 mice (Supplementary Fig. 1i) resulted in no labeled neurons throughout the brain (Supplementary Fig. 1f). When co-injecting rAAV-EF1α-DIO-mCherry-F2A-TVA-WPRE-hGH polyA and RV-ENVA-ΔG-EGFP into the TMN of HDC-CreERT2 mice without RVG component (Supplementary Fig. 1k), no labeled neurons were detected outside the TMN (Supplementary Fig. 1l). In wild-type (WT) mice receiving all three components (Supplementary Fig. 1m), only sparse labeling was observed in limited brain regions including medial prefrontal cortex (mPFC), anterior hypothalamus (AHY), and posterior amygdala nucleus (PA) (Supplementary Fig. 1n). This labeling was negligible compared to the extensive labeling observed in HDC-CreERT2 mice (Fig. 1c). These results convincingly demonstrate the specificity and reliability of the monosynaptic retrograde RV tracing methodology in HDC-CreERT2 mice.
To establish a precise and comprehensive whole-brain atlas of monosynaptic long-range inputs to histaminergic neurons, we employed high-resolution whole-brain sectioning, imaging, and reconstruction through fMOST system (Fig. 1e). Briefly, mouse brain samples were sectioned coronally at 1 μm thickness and automatically imaged with a resolution of 0.32 × 0.32 × 1 μm. The raw data, comprising more than 10,000 coronal slices per brain, were registered to the Allen Reference Atlas Common Coordinate Framework (CCF v3.0) for spatial standardization (through careful delineation of major brain structures, manual verification, and automated computation and registration), consistent with our previous methodology. Standardized whole-brain 3D reconstructions of monosynaptic long-range inputs to histaminergic neurons were visualized through coronal, sagittal, horizontal, and full-view projections (Fig. 1f), with representative coronal sections displayed at multiple rostrocaudal levels (Supplementary Fig. 2a). Quantitative distribution analysis of input neurons across all major brain regions was performed using NeuroGPS system followed by manual correction. To avoid potential interference from injection areas, we excluded brain regions containing histaminergic neuron somata (Supplementary Table 3), as implemented in our previous work. In general, we found that the upstream cells targeting histaminergic neurons showed three key distribution patterns: (1) medial preference over lateral regions, (2) ventral preference over dorsal areas, and (3) concentration along part of the anterior-posterior axis (0 mm ~ -2.5 mm from bregma) (Supplementary Fig. 2b-d). At the regional level, hypothalamic nuclei contained the most upstream cells even after excluding the injection site (Fig. 1g), whereas the cerebellum (CB) showed the lowest density (Fig. 1g).
Quantitative analysis revealed the anterior hypothalamic nucleus (AHN, 8.20 ± 1.37%, n = 5) as the predominant source of upstream inputs, accounting for a significantly greater proportion than any other brain region (Fig. 1g and Supplementary Fig. 3b). Substantial inputs were also observed in the striatum, pallidum, hippocampal formation, and cortical subplate, with key nuclei including: the medial amygdalar nucleus (MEA, 3.83 ± 0.27%, n = 5), lateral septal nucleus, rostral part (LSr, 2.44 ± 0.08%, n = 5), and lateral septal nucleus, ventral part (LSv, 2.25 ± 0.36%, n = 5) in the striatum, the bed nuclei of the stria terminalis (BST, 4.74 ± 0.53%, n = 5) in the pallidum, field CA1 (CA1, 3.09 ± 0.44%, n = 5) in the hippocampal formation, and the posterior amygdalar nucleus (PA, 3.08 ± 0.28%, n = 5) in the cortical subplate (Fig. 1g and Supplementary Fig. 3c-f, h). Although the midbrain and thalamus only contained moderate input densities overall, specific nuclei, including the periaqueductal gray (PAG, 2.59 ± 0.41%, n = 5) and paraventricular nucleus of the thalamus (PVT, 2.00 ± 0.35%, n = 5) exhibited a relatively large number of upstream cells (Fig. 1g and Supplementary Fig. 3b, h). In contrast, fewer upstream cells were distributed in olfactory areas (OLF), hindbrain (HB), and the cortex. Spatially, upstream cells were diffusely distributed in OLF and HB (Fig. 1g and Supplementary Fig. 3a, g), whereas cortical inputs displayed a layer V-predominant distribution pattern (Fig. 1g and Supplementary Fig. 3d, f), which warrants further analysis. 3D visualization of major input brain regions (Supplementary Fig. 4) revealed several notable spatial preferences: (1) LS and CA1 showed ventral-dominant distributions (Supplementary Fig. 4a, i); (2) PVT inputs were exclusively localized to its anterior region (Supplementary Fig. 4g). Complete distributions across all sample mice are provided in Supplementary Tables 1 and 2, with major input areas summarized in Fig. 1g and Supplementary Fig. 3a-h.
Building upon our previous findings regarding the projection heterogeneity of histaminergic neurons, we sought to examine whether subpopulations projecting to different downstream areas also exhibit differential input patterns. Firstly, we injected a 1:2 mixture of rAAV-EF1α-DIO-mCherry-F2A-TVA-WPRE-hGH polyA and rAAV-EF1α-DIO-oRVG-WPRE-hGH polyA into the bilateral TMN of HDC-CreERT2 mice. After three weeks, RV-ENVA-ΔG-EGFP was injected into either the MS or SC (Supplementary Fig. 7f). Notably, both MS- and SC-projecting histaminergic neuron subpopulations received dense inputs from the mPFC, LS, PVT, MEA, PA, and vCA1 (Supplementary Fig. 7g-l). However, comparative analysis revealed significant differences in input density (ratio of input cells/starter cells) from several upstream regions -- including the mPFC, LS, PA, and vCA1 -- between MS- and SC-projection-defined subpopulations and the general TMN histaminergic neurons (Supplementary Fig. 7m-r). These findings support the hypothesis that histaminergic neurons with distinct efferent projections differentially integrate afferent inputs from upstream brain regions.
Specific layer distribution and co-projection characteristics of cortical inputs to histaminergic neurons
The cortex is well-established to play crucial roles in emotional and sensory processes. Our analysis of fMOST raw data (Fig. 2c) and the quantified whole-brain input atlas (Fig. 1g) revealed a predominant layer V distribution of cortical inputs to histaminergic neurons, suggesting potential functional specialization. To characterize detailed cortical input organization, we first performed 3D visualization of all cortical input neurons (Fig. 2a). Notably, input neurons were widely distributed across most cortical regions, albeit with varying densities (Fig. 2b). Quantitative analysis demonstrated that most of cortical inputs originated from layer V (61.77 ± 5.54%, n = 5), followed by layer VIa (23.19 ± 4.35%, n = 5), while other cortical layers contained sparse inputs (Fig. 2d). At the regional level, the layer V preference was conserved across most cortical areas, with the notable exception of the infralimbic area (ILA), a subregion of mPFC that exhibited greater input density in layer VIa than in layer V (Fig. 2e).
To further elucidate the co-projection patterns of different cortical input neurons and their potential regulatory mechanisms at the whole-brain circuitry level, we performed single-neuron morphological reconstruction using the fMOST raw data. Following semiautomatic tracing and manual verification, 150 well reconstructed neurons were obtained from various cortical areas. These neurons were then registered to CCF v3.0 for 3D visualization and projection target analysis (Fig. 2f-h and Supplementary Fig. 5). At the whole-cortex level, our analysis revealed that except for hypothalamus and cortex, the majority of input neurons co-projected to multiple subcortical regions, primarily the striatum, midbrain, pallidum, and thalamus (Fig. 2h). Notably, all reconstructed neurons innervated exclusively the ipsilateral hemisphere (Fig. 2f). Region-specific projection patterns were identified in certain cortical areas. For example, the entorhinal area was the sole cortical region exhibiting co-projections to HPF. Moreover, mPFC, orbital area (ORB), and agranular insular area (AI) displayed selective connectivity with OLF. Given the distinct somatic distribution patterns between two mPFC subregions, the ILA and PL, we further compared their co-projection patterns. Strikingly, ILA neurons preferentially co-projected to anterior regions (e.g., OLF and STR), whereas PL neurons preferentially co-projected to posterior areas (e.g., PAL, TH, and MB) (Supplementary Fig. 5b, c), suggesting distinct functional circuitry for these subregions.
Distinct organization of inhibitory and excitatory inputs to histaminergic neurons
Histaminergic neurons are known to widely express GABA receptors and NMDA/AMPA receptors, mediating substantial inhibitory and excitatory inputs that modulate histaminergic signaling. However, the precise distribution of GABAergic and glutamatergic neurons among most upstream brain areas remains poorly characterized, impeding functional studies of input circuits to histaminergic neurons. To address this, we performed RNAscope in situ hybridization on coronal slices containing major upstream brain regions of RV-labeled mouse brains (Fig. 3a). We selected GAD1 as the molecular marker for GABAergic neurons. For glutamatergic neurons, since vglut2 is widely expressed in subcortical brain areas but not in the cortex and hippocampus, while CaMKIIα serves as an appropriate marker for cortical and hippocampal areas, we accordingly used GAD1-C1 and vglut2-C2/CaMKIIα-C3 probes. To compensate for the potential RV-EGFP quenching during the in situ hybridization process, we performed GFP immunohistochemical staining before confocal imaging. Quantitative analysis revealed that RV-EGFP labeled neurons in the LS and medial amygdalar nucleus (MEA) were predominantly GAD1-positive (LS: 98.82 ± 0.80%, MEA: 97.73 ± 2.27%, n = 4), with minimal vglut2/CaMKIIα signal (LS: 0.61 ± 0.36%, MEA: 0%, n = 4), indicating predominantly inhibitory inputs from these regions (Fig. 3b, i, j). BST also showed most RV-EGFP labeled neurons co-labeled with GAD1 (89.17 ± 1.78%, n = 4), with only a small fraction co-labeled with vglut2 (7.45 ± 1.29%, n = 4, Fig. 3f, j). In contrast, the PVT, CA1, and PA/BMAp contained primarily vglut2/CaMKIIα-positive inputs (PVT: 98.77 ± 0.95%, CA1: 97.09 ± 0.66%, PA/BMAp: 74.38 ± 2.92%, n = 4), with negligible GAD1 co-expression (PVT: 0.23 ± 0.23%, CA1: 0%, PA/BMAp: 2.1 ± 0.81%, n = 4), indicating basically excitatory inputs from them (Fig. 3d, g, h, j). On the other hand, PAG and anterior hypothalamic areas (AHY) exhibited mixed inputs, with substantial GAD1-positive (PAG: 70.14 ± 4.68%, AHY: 61.29 ± 4.82%, n = 4) and vglut2/CaMKIIα-positive populations (PAG: 26.56 ± 4.82%, AHY: 22.63 ± 1.43%, n = 4) (Fig. 3c, e, j). Given the heterogeneous composition of PAG and AHY input neurons, we further reconstructed the distribution of GABAergic and glutamatergic somata in coronal section. Spatial analysis revealed that GABAergic input neurons were widely distributed throughout the AHY, while glutamatergic input neurons were concentrated near the third ventricle (V3) and scattered across other areas. In PAG, GABAergic input neurons preferred ventral and lateral areas, whereas glutamatergic neurons were uniformly distributed (Supplementary Fig. 6a, b). These findings systematically characterize the major inhibitory and excitatory components of inputs to histaminergic neurons, establishing an important foundation for precise functional circuit studies in the future.
Functional characterization of input circuits to histaminergic neurons remains limited compared to advances in histaminergic output circuit research, and functional analysis may in turn, validate the significance of established structural connections. Given the well-established role of the histaminergic system in sleep-wake regulation, we sought to preliminary validate the biological significance of differential modulation by excitatory and inhibitory inputs on histaminergic neurons. Accordingly, we selected two representative upstream brain regions, the LS (predominantly inhibitory neurons) and PVT (predominantly excitatory neurons), for the following study. These regions were chosen based on two key observations: (1) their substantial input neuron populations to histaminergic neurons (this study), and (2) the dense histaminergic fiber innervation revealed in our previous work, suggesting their potential importance in both afferent and efferent histaminergic circuits. To further confirm monosynaptic connectivity between LS/PVT and histaminergic neurons, we employed a Cre-dependent WGA monosynaptic anterograde tracing system. For LS GABAergic neurons, a 1:1 mixture of rAAV-VGAT1-CRE-EGFP-WPRE-hGH polyA and rAAV-CAG-DIO-mWGA-mCherry was injected into the bilateral LS of HDC; AI47 mice (Fig. 4a). Three weeks post-injection, WGA-mCherry/GFP co-labeled neurons in the TMN confirmed the monosynaptic LS→histaminergic neuron connection (Fig. 4b). Similarly, PVT glutamatergic neuron tracing was performed by injecting a 1:1 mixture of rAAV-VGLUT2-CRE-WPRE-hGH polyA and rAAV-CAG-DIO-mWGA-mCherry into the PVT of HDC; AI47 mice (Fig. 4c), and WGA-mCherry/GFP co-labeled neurons were also found in the TMN (Fig. 4d). On the other hand, considering different upstream brain regions may also exhibit input variability to histaminergic neurons, we performed WGA-based anterograde tracing from the PVT and LS in the same mouse. We injected a 1:1 mixture of rAAV-VGLUT2-CRE-WPRE-hGH polyA and rAAV-CAG-DIO-mWGA-mCherry into the PVT, and a 1:1 mixture of rAAV-VGAT1-CRE-EGFP-WPRE-hGH polyA and rAAV-CAG-DIO-WGA-EGFP into the bilateral LS of WT mice simultaneously (Supplementary Fig. 7a-c). Quantitative analysis revealed that while a substantial subset of histaminergic neurons received convergent inputs from both LS (38.54 ± 4.132%, n = 3) and PVT (41.54 ± 4.435%, n = 3), the majority exhibited selective innervation from either region (Supplementary Fig. 7d, e), demonstrating significant input variability of histaminergic neurons from different brain regions.
To functionally characterize the monosynaptic connections between LS/PVT and histaminergic neurons, we performed whole-cell patch-clamp recordings in acute brain slices containing histaminergic neurons receiving red light-evoked opsin ChrimsonR-expressing inputs from LS/PVT. For LS GABAergic inputs, we injected a 1:1 mixture of rAAV-VGAT1-CRE-EGFP-WPRE-hGH polyA mixed and AAV2/9-hSyn-FLEX-ChrimsonR-tdTomato-WPRE-pA into the bilateral LS of HDC; AI47 mice (Fig. 4e). Upon 5 ms red light stimulation, robust inhibitory post synaptic currents (IPSCs) were recorded in the GFP-labeled histaminergic neurons immediately (Fig. 4f, g). Notably, these IPSCs persisted in the presence of TTX and 4-AP but were abolished by the GABA receptor antagonist bicuculline, confirming direct monosynaptic GABAergic transmission from the LS to histaminergic neurons (Fig. 4g). Analogously, for PVT glutamatergic inputs, we injected the same opsin-expressing virus mixture into the PVT of HDC; AI47 mice (Fig. 4h). Upon 5 ms red light stimulation, robust excitatory post synaptic currents (EPSCs) were recorded in the GFP-labeled histaminergic neurons immediately (Fig. 4i, j). Notably, these EPSCs persisted in the presence of TTX and 4-AP but were abolished by the NMDA receptor antagonist APV and AMPA antagonist CNQX, verifying direct monosynaptic glutamatergic input from the PVT to histaminergic neurons (Fig. 4j). Collectively, these results provide both structural and functional evidence for direct LS/PVT→histaminergic neuron connectivity.
The balanced roles of LS- and PVT-TMN circuits in sleep-wake regulation
Both LS and PVT exhibit high densities of histaminergic nerve fibers and substantial populations of input neurons projecting to histaminergic neurons, suggesting their importance in modulating core histaminergic functions. Given the well-established role of the histaminergic system in sleep-wake regulation, and the reported involvement of PVT glutamatergic and LS GABAergic neurons in this process, we sought to reconcile the apparent paradox that these functionally opposing nuclei appear to exert similar effects on sleep-wake transitions. To specifically examine histaminergic-projecting neurons in LS and PVT, we employed a Cre-dependent monosynaptic retrograde GCaMP system. Briefly, we first injected a 1:2 mixture of rAAV-EF1α-DIO-mCherry-F2A-TVA and rAAV-EF1α-DIO-N2cG into the bilateral TMN of HDC-CreERT2 mice, followed by CVS-EnvA-ΔG-GCaMP6s injection three weeks later, with simultaneous optic fiber implantation in the LS or PVT (Fig. 5a, e). GCaMP6s-labeled neurons were found in the TMN and upstream brain regions (LS and PVT under optic fiber, Fig. 5b, c, f, g). Simultaneous EEG/EMG and fiber photometry recordings revealed that both LS and PVT input neurons exhibited decreased calcium signals during wake-to-non-rapid eye movement (NREM) sleep transitions (Fig. 5d, h, i, m), and increased calcium signals of input neurons during sleep (NREM or rapid eye movement (REM))-to-wake transitions (Fig. 5d, h, j, l, n, p). Notably, we observed divergent responses during NREM-to-REM transitions: PVT neurons exhibited decreased activity, whereas LS neurons showed no such reduction (Fig. 5d, h, k, o and Supplementary Fig. 8). These results demonstrate that histaminergic-projecting neurons in both LS and PVT coordinately modulate sleep-wake transitions, consistent with previous reports, while revealing previously uncharacterized state-specific differences in NREM-to-REM sleep transitions.
The differential functional dynamics during NREM to REM sleep transitions suggest differential regulatory roles of these input circuits. To assess their effects, we further employed chemogenetic activation approaches on LS/PVT input circuits during mouse sleep-wake cycle. Specifically, we injected a 1:1 mixture of rAAV-VGAT1-CRE-EGFP-WPRE-hGH polyA and rAAV-hSyn-hM3D(Gq)-EGFP-WPRE-hGH polyA into the bilateral LS, or a 1:1 mixture of rAAV-VGLUT2-CRE-WPRE-hGH polyA and rAAV-hSyn-hM3D(Gq)-EGFP-WPRE-hGH polyA into the PVT of WT mouse brain, with rAAV-CAG-DIO-mCherry-WPRE-hGH polyA as control (Fig. 6a-d). Following three weeks of expression, we implanted a bilateral cannula into the TMN for CNO administration delivery along with electrodes for EEG and EMG recording. After 7 days, we performed EEG/EMG recording during sleep-wake cycle under chemo-genetic manipulation, and found that consistent with previous reported regulatory roles of whole LS/PVT neurons, both LS-hM3Dq (54.22 ± 1.504%, n = 4) and PVT-hM3Dq (56.63 ± 0.9636%, n = 4) groups showed significantly increased wakefulness compared to the control group (48.31 ± 1.558%, n = 4) (Fig. 6e-i). However, REM sleep parameters revealed striking differences (Fig. 6e-g, i): LS-TMN activation enhanced REM duration (whole-day: 6.705 ± 0.695% vs control 2.650 ± 0.049%; sleep-period: 14.73 ± 1.657% vs control 5.150 ± 0.242%; n = 4), whereas PVT-TMN activation suppressed it (whole-day: 1.063 ± 0.259%; sleep-period: 2.420 ± 0.541%, n = 4) (Fig. 6i, j). Moreover, NREM-to-REM sleep transition frequency also differed markedly (LS: 371.0 ± 20.08; PVT: 82.50 ± 20.04; control: 177.3 ± 10.26; n = 4) (Fig. 6k). To further explore the sleep/wake situations of tested mice during the daytime and nighttime, we performed circadian analysis: (1) mice spend more time awake during the nighttime than daytime in both the mCherry, LS-hM3Dq, and PVT-hM3Dq group (Fig. 6l, p), (2) LS-hM3Dq mice showed elevated REM (day: 7.063 ± 1.138%; night: 6.333 ± 0.458%; n = 4) relative to controls (day: 3.483 ± 0.396%; night: 1.820 ± 0.392%; n = 4), while PVT-hM3Dq mice exhibited reduced REM (day: 1.268 ± 0.429%; night: 0.860 ± 0.141%; n = 4) (Fig. 6l-s). These findings reveal fundamental differences in how LS and PVT input circuits regulate sleep architecture, highlighting the complexity of histaminergic network regulation. Our whole-brain atlas thus provides a critical foundation for decoding these circuit-level mechanisms in the future.