Background

The central nervous system has a highly ordered structure: external information is transmitted to a higher center through an ascending pathway, and decision-making information is delivered via a descending pathway to determine an individual’s physiological changes and behavior [1]. The upper brainstem (midbrain and pons), located between the diencephalon and medulla, has dense interconnections with different regions of the nervous system, from the spinal cord to the cortex [2, 3]. Previous studies have indicated that specialized populations of these neurons preferentially innervate the ascending and descending circuits. Dopamine (DA) neurons in the ventral tegmental area (VTA) [4] and different types of neurons in the dorsal raphe (DR) [5,6,7] send most of their axons to ascending circuits and a few to descending circuits. Two separate groups of glutamatergic neurons in the mesencephalic locomotor region (MLR) participate in different functions by projecting to the ascending and descending circuits, respectively [8]. Furthermore, cholinergic neurons in the pontine-tegmental cholinergic system, including the pedunculopontine nucleus (PPN) and laterodorsal tegmental nucleus (LDT), send abundant axons in both the ascending and descending circuits [9]; however, their projection patterns at the single-cell level have not been systematically characterized until now.

As the main cholinergic system in the rostral hindbrain [10], pontine-tegmental cholinergic neurons (PTCNs) send abundant axons along three major trajectories and are involved in various functions [11]. In the ascending dorsal circuits, PTCNs mediate prefrontal serotonin release from the DR [12] and participate in multiple functions, including auditory sensation [13], sensorimotor function [14], and spatial memory [15], by targeting different thalamic nuclei. Additionally, PTCNs innervate different neurons in the striatum (STR) and contribute to exploratory motor behaviors and action strategies [16]. In the ascending ventral circuits, previous studies have mainly focused on cholinergic modulation of the VTA and substantia nigra (SN), which are critically involved in mechanisms of reward [17], addiction [18], locomotion [23].

There are approximately 2000 PTCNs in one hemisphere of the mouse brain [24]. How does the pontine-tegmental cholinergic system innervate many brain regions and contribute to various functions with a limited number of neurons? Notably, the activation of PPN cholinergic neurons plays an opposite role in locomotion in the ascending and descending circuits [11]; however, the relationship between these two circuits is unclear. Although previous studies have acquired whole-brain projections of PTCNs [9], it is still unclear whether the ascending and descending axons belong to different groups or to the same group of neurons, which is urgently needed at the single-neuron level. Previous studies [25] have reconstructed partial axons in serial slices and provided preliminary evidence that individual PPN neurons have complex axonal projections. However, we lack the unabridged single-cell connection of the pontine-tegmental cholinergic system due to limited techniques for tracing and imaging.

In this study, we combined sparsely labeled fluorescence micro-optical sectioning tomography (fMOST) serial technologies [26, 27], and acquired a morphological atlas that may have uncovered the projection logic of PTCNs at the single-cell level.

Results

Sparse labeling and single-cell reconstruction

To obtain the fine morphology of individual cholinergic neurons, we employed a Cre-dependent virus for sparse labeling [28], the fMOST system for whole-brain imaging [26, 27], and GTree software for single-cell reconstruction [4], which mainly project to ascending circuits with a few axons projecting to descending circuits. However, a few of them have synchronous collaterals in the ascending and descending circuits [7]. Glutamatergic neurons in the PPN [38] and VTA [39] also have rich axons in both the ascending and descending circuits, but their single-cell projection patterns remain unknown. Moreover, the activation of PPN cholinergic neurons plays an opposite role in locomotion via the ascending and descending circuits [11], which means that these two circuits may belong to different groups. However, we found that all the reconstructed VTA/SN-projecting PPN cholinergic neurons had collaterals in the PRN. These results show that the opposite functions of the two cholinergic circuits may originate from different neurons or different responses of downstream neurons to the same neurons. Hence, PTCNs, which modulate many ascending and descending circuits, mainly originate from the same group of neurons.

On the other hand, we found that some neurons had richer collaterals in contralateral regions, which may have delivered more information to the contralateral hemisphere. The axial selection of ipsilateral and/or contralateral targets is crucial for integrating bilateral information and coordinated movements. PTCNs decrease significantly in patients with Parkinson’s disease (PD); characterized by abnormalities of movement, including tremors and difficulties with gait and balance [40]. The PPN is an important clinical target for deep brain stimulation in patients with PD [41], emphasizing the role that PTCNs may play in somatic homeostatic regulation by projecting abundant axons to bilateral hemispheres.

Projection logic of cholinergic axons ascending to the thalamus

The thalamus is an important relay station for sensorimotor information in mammals and acts as the main pathway for the pontine-tegmental cholinergic system to deliver information in ascending circuits. As the main source of acetylcholine for the thalamus, PTCNs project to the thalamus and form synaptic connections [9, 31]. In this study, we uncovered the logic of projection of the PTCNs to the thalamus at the single-cell level. Consistent with previous studies [42], most PTCNs directly innervate the thalamic nuclei. Individual PTCNs synchronously target multiple thalamic nuclei. We found that more than 80% of the reconstructed PTCNs innervated at least five thalamic nuclei, and some of them simultaneously projected to 18 nuclei in the bilateral thalamus. This indicates that the regulation of PTCNs in the thalamus is mainly diffuse, rather than concentrated in a few regions. The thalamic nuclei receive information from the whole brain and deliver it to diverse cortical areas [33, 34]. We speculate that individual PTCNs send complex information to various thalamic nuclei, which sort the information and then deliver it to specific cortical fields. The diverse functions of thalamic circuits projected from the PTCNs [13,14,15] may originate from the different responses of different thalamic nuclei.

Moreover, we decoded three projection patterns from PTCNs with different preferences in the ipsilateral and/or contralateral thalamus. There may be three different modes of information transmission from the pontine-tegmental cholinergic system to the thalamus for further information progress. Specifically, a large number of PTCNs targeted the bilateral thalamus and some preferred contralateral areas. These may play important roles in coordinated movement and contribute to some movement disorders diseases, such as PD [40]; the cholinergic circuits in the thalamus of these patients exhibit abnormal activity [14].

Morphological differences between the PPN and LDT cholinergic neurons

PPN and LDT cholinergic neurons are two components of the pontine-tegmental cholinergic system that have diverse characteristics with regard to connections and functions [9, 37]. They connect with similar regions with different preferences [9, 11, 31]. In this study, we demonstrated that the PPN and LDT cholinergic neurons have similar projection patterns with different preferences at the single-cell level. Comparatively, the PPN projected more divergent cholinergic axons to the ascending and descending circuits, whereas LDT neurons were more convergent. Various studies have suggested that the PPN and LDT cholinergic neurons target the same regions with different functions, such as the VTA [37] and SN [19]. The functional difference is related to the connectional difference between the two cholinergic nuclei: PPN neurons prefer motor nuclei, whereas LDT neurons prefer limbic nuclei [9, 11, 31]. Furthermore, single PTCNs have abundant and diverse co-projection axons that may contribute to functional differences. The thalamus is one of the main targets of PTCNs, and both individual PPN and LDT cholinergic neurons have rich collaterals in different thalamic nuclei with similar projection logic; however, they also show some differences [9, 31]. On the one hand, the PPN neurons had richer thalamic collaterals on average, and we found that some projected more than 50% of their axons to the thalamus. In contrast, the two groups of cholinergic neurons had different preferences for thalamic nuclei: PPN neurons preferred the thalamic motor nuclei, while LDT neurons preferred the limbic nuclei. In addition, the dendrites of the LDT cholinergic neurons had richer branches and longer dendrites than those of the PPN. As neuronal portals receive information, dendrites influence the number and identity of presynaptic inputs [43], which implies that LDT neurons may receive more afferent information.

Conclusions

Our study elucidated the single-cell morphology of PTCNs on a brain-wide scale and decoded the projection logic of the whole brain and specific circuits. We revealed that individual PTCNs send abundant axons in multiple nuclei in the ascending and descending circuits synchronously and can be grouped into four types with different projection patterns. Furthermore, we uncovered three different projection patterns from the PTCNs to the bilateral thalamus, and two separate pathways to the cortical areas. Our study mapped the axonal projections of the pontine-tegmental cholinergic system; thus providing researchers with a better understanding of its projection logic.

Methods

Animals

Adult (2–4 months) ChAT-ires-Cre mice were used in this study. ChAT-Cre transgenic mice (stock No: 018957) were purchased from Jackson Laboratory. The mice were kept under a condition of a 12-h light/dark cycle with food and water ad libitum. All animal experiments were approved by the Animal Care and Use Committee of Huazhong University of Science and Technology.

Tracer information

For sparse labeling, we employed the CSSP-YFP (5.2 × 1012 VG/mL) packed by Brain Case (Brain Case Co., Ltd., Shenzhen, China). CSSP-YFP virus is produced by co-packaging the rAAV-EF1α-DIO-Flp plasmid and rAAV-FDIO-EYFP plasmid with a ratio of 1:20,000 in the single rAAV production step [28].

Surgery and viral injection

Before virus injection, we anesthetized the mice with mixed anesthetics (2% chloral hydrate and 10% ethyl urethane dissolved in 0.9% NaCl saline) according to their weight (0.1 ml/10 g). The brain of anesthetized mice was fixed with a stereotaxic holder to adjust the position of the skulls. Then a cranial drill (~ 0.5 mm diameter) was employed to uncover the skulls above the targets. For sparse labeling, we injected 100 nl CSSP-YFP into the unilateral PPN or LDT.

Histology and immunostaining

The histological operations followed previous studies [33]. Shortly, four weeks after AAV injection, anesthetized mice were perfused with 0.01 M PBS (Sigma-Aldrich, USA), followed by 2.5% sucrose and 4% paraformaldehyde (PFA, Sigma-Aldrich, USA) in 0.01 M PBS. Then, the brains were removed and post-fixed in 4% PFA solution overnight.

For immunofluorescent staining, some samples were sectioned in 50 μm coronal slices with the vibrating slicer (Leica 1200S). All sections containing PPN or LDT were selected to characterize the labeled neurons in inject site. These sections were blocked with 0.01 M PBS containing 5% (wt/vol) bovine serum albumin (BSA) and 0.3% Triton X-100 for 1 h at 37 °C. Then the sections were incubated with the primary antibodies (12h at 4 °C): anti-ChAT (1:500, goat, Sigma-Aldrich, AB144P). Then the sections were washed in PBS five times at room temperature. Next, these sections were incubated with the fluorophore-conjugated secondary antibody (1:500, Abcam: Alexa-Fluor 647, donkey anti-goat) for 2 h at room temperature. After rinsing with PBS, DAPI (1 ng/mL) was performed on stained sections for 5 min, and sections were finally mounted after washing. We acquired the stained information of sections with the confocal microscope (LSM 710, Zeiss, Jena, Germany).

Imaging and 3D visualization

For whole-brain imaging, virus-labeled samples were dehydrated with alcohol and embedded with resin. Then the whole brain datasets were acquired with the fMOST system. In short, we fixed the embedded sample on the base and acquired the image of the top surface with two fluorescent channels; the imaged tissue was subsequently removed by diamond wheel. Thus, we obtained the continuous whole-brain dataset layer by layer with high resolution (0.32 μm × 0.32 μm × 1 μm).

For 3D visualization and statistical analysis of whole-brain datasets, we registered the whole-brain datasets to the Allen CCFv3 [30]. The methods of registration have been described previously [44]. Briefly, we employed image preprocessing to correct uneven illumination and then remove background noise. The down-sampling data (the voxel resolution of 10 μm × 10 μm × 10 μm) was uploaded into Amira software (v6.1.1, FEI, Mérignac Cedex, France) to distinguish and extract regional features of anatomical invariants, including the outline of the brain, the ventricles, and the corpus striatum, etc. Next, the gray-level-based registration algorithm (SyN) was employed to register the extracted features. Basic operations including extraction of areas of interest, resampling, and maximum projection performed via Amira software and Fiji (NIH).

Morphological reconstruction of single neurons

83 neurons were reconstructed from six brains. For single-cell morphological analysis, we reconstructed the morphology of sparsely labeled neurons with semi-automatic methods following previous studies [29]. Briefly, we acquired the spatial coordinates of labeled somas in high-resolution data and transformed the data format of GFP-labeled data from TIFF to TDI type via Amira first.

Then the data block containing the given soma was loaded into GTree software and we assigned the soma as the initial point and marked all its axons with unfinished tags. Next, we selected one uncompleted fiber and traced it in the next block with automatic tracing. Then we checked the traced fiber and marked its branches with unfinished tags. We repeated the above procedure until the selected fiber was finished and then we reconstructed the remaining unfinished axons until all the axons were achieved. The reconstructed neurons were checked back-to-back by three persons. The tracing results were saved in SWC format. Meanwhile, we registered the PI-labeled data and the corresponding tracing results to the reference atlas with the methods mentioned above. Considering the distribution of axons and synaptic connection does not always correlate [31], we counted the terminal branches of reconstructed neurons to represent the connection between single cholinergic neurons and their targets.

Statistical information

To distinguish the axons of passage vs. terminal innervations, we take the ended axons as terminal axons while others as passage axons. For terminals quantification, the reconstructed neurons carried spatial information of all nodes that we could calculate the ended nodes based on registered single neurons to obtain the terminals in different regions. In addition, we employed the Amira software to quantify the length of dendrites and axons of individual neurons. Statistical graphs were generated using GraphPad Prism v.8.02 and Microsoft Excel (Office 2020). We employed GraphPad Prism v. 8.02 for the significance test, Neurolucida360 software for the polar histogram, and MATLAB (2017a) for the Sholl Analysis. We conducted one-way ANOVA followed by Tukey’s post hoc tests to compare the difference among four projections patterns of same nuclei while two-tailed t-tests to compare the difference between the PPN and LDT. The confidence level was set to 0.05 (P value) and all results were presented as mean ± SEM.