Abstract

The precise annotation and accurate identification of neural structures are prerequisites for studying mammalian brain function. The orientation of neurons and neural circuits is usually determined by mapping brain images to coarse axial-sampling planar reference atlases. However, individual differences at the cellular level likely lead to position errors and an inability to orient neural projections at single-cell resolution. Here, we present a high-throughput precision imaging method that can acquire a co-localized brain-wide data set of both fluorescent-labelled neurons and counterstained cell bodies at a voxel size of 0.32 × 0.32 × 2.0 μm in 3 days for a single mouse brain. We acquire mouse whole-brain imaging data sets of multiple types of neurons and projections with anatomical annotation at single-neuron resolution. The results show that the simultaneous acquisition of labelled neural structures and cytoarchitecture reference in the same brain greatly facilitates precise tracing of long-range projections and accurate locating of nuclei.

Introduction

At the cellular level, the connectome precisely annotates a comprehensive map of neural connections in the brain and significantly increases current understanding of how functional brain states emerge from underlying structural substrates. Mapping a cellular mouse connectome requires centimetre-scale imaging with axon resolution. Combined with physical sectioning, Li et al., Ragan et al. and Gong et al. pioneered whole-brain optical microscopic imaging. Neurons with different functions have different sizes, shapes and locations, and even neighbouring neurons of the same cell type differ in their morphologies and projection patterns. Thus, orienting neural projections at single-cell resolution is required for both the precise annotation and accurate identification of the spatial organization of neural structures. To locate neurons and neural circuits, researchers usually map brain images to coarse axial-sampling planar reference atlases. However, individual differences at the cellular level likely lead to position errors, making the precise annotation and accurate identification of projections at single-cell resolution difficult.

Here, we develop an automatic microscopy method called the brain-wide positioning system (BPS) for dissecting and locating neural structures with cytoarchitectonic landmarks at a single-cell resolution. BPS involves a whole-brain real-time counterstain protocol for simultaneously staining cytoarchitectonic landmarks during imaging and a high-throughput multi-channel brain-wide imaging system, termed wide-field large-volume tomography (WVT), for accelerating imaging acquisition at single-cell resolution. We obtain five full-volume, dual-colour mouse brain data sets of cell-type-specific fluorescent protein-expressing neurons and neuronal projections and subsequently reconstruct the three-dimensional (3D) fine structure of these neurons at the axon level using cellular landmarks. This standardized precision imaging approach facilitates quantitative brain-wide studies of 3D neuronal morphology, dendritic arbor and bouton distributions, axonal projection densities and distances and neuronal cell types, based on cytoarchitectonic landmarks at the cellular level. We propose that this method represents a routine tool for highly accurate and precise analysis of the cellular connectome.

Results

As a proof of principle, we revealed the spatial distribution with anatomical annotation of an 8-week-old Thy1-GFP M-line transgenic mouse using BPS (Fig. 1 and Supplementary Fig. 1). The resin-embedded whole-brain sample was then fixed in the anterior-posterior (A-P) direction on a 3D translation stage. The WVT system employed wide-field two-channel fast structured illumination microscopy (SIM) to accelerate imaging acquisition. SIM, which was first demonstrated by Neil et al., has optical sectioning capability comparable to confocal microscopy. The system automatically performed brain-wide data acquisition. The 3D translation stage moved the sample to the microtome, and subsequently the microtome removed the superficial layer, revealing a smooth surface. The sample was then moved to fast SIM, and images were acquired through mosaic scanning in each layer. To avoid the compromised effect of sectioning marks on the machined surface, we set the imaging plane slightly below the surface, generally at 1–2 μm. The sectioning-imaging cycles were repeated until entire brain acquisition was completed (Fig. 1a). To achieve whole-brain counterstaining, we immersed the sample in a fluorescent nuclear staining solution during whole-brain imaging. After sectioning, the dye immediately penetrated the fresh surface, combining with nucleic acid inside cell bodies and staining the soma, proximal dendrites and axon hillocks (Fig. 1b and Supplementary Figs 2, 3). To improve the detection of weak fluorescence signals in thick tissue through SIM, we modified the sample processing protocol. We added a light absorber, Sudan black B (SBB), to the embedded resin to inhibit background fluorescence and enhance the signal-to-noise ratio of the SIM imaging (Supplementary Fig. 4). We also used chemical reactivation of the quenched fluorescent protein on the surface of the sample to enhance the in-focus GFP signal.

See the figures on nature.com here.

Whole-brain neuron-mapping with cytoarchitecture

The whole-brain data set from a Thy1-GFP M-line transgenic mouse, including GFP-labelled neurons and propidium iodide (PI)-stained cytoarchitecture, was acquired at a voxel resolution of 0.32 × 0.32 × 1 μm in a coronal plane. The images of the typical structure of the hippocampus, cortex and cellular layers of the cortex (Fig. 2a,b) demonstrate the method used in the present study to identify classical regions through PI-stained landmarks. Neuron arbors, axon boutons and dendritic spines were distinguished in the reconstruction of neurons (Fig. 2c–e). The overlapping of soma with GFP-labelled neurons and PI-stained cellular nuclei (Fig. 2f–h) illustrated single-cell co-location accuracy in the same brain using this BPS method. The sagittal reconstruction of continuous projections demonstrated that the high resolution and self-registration of BPS guarantee the data integrity of the whole-brain data set (Supplementary Fig. 5). Precision imaging through BPS not only reveals the complexity of fibre orientations, even in a small tissue volume (400 × 400 × 400 μm), but also distinguishes individual axons in the dense fibre bundle (Fig. 3 and Supplementary Movie 1).

Reconstruction and localization of single pyramidal neurons

Furthermore, to illustrate the accurate co-localization of the same neuron types with cellular landmarks in the same region, we examined classical function columns and determined the complete 3D morphology of barrel cortex layer V/VI pyramidal neurons.

Read whole article on nature.com here.

Quantitative morphology analysis of pyramidal neurons

We also orientated and traced long-range projection neurons in the barrel cortex in the first Thy-1 GFP data set. Read more.

Mapping long-range projections with nuclei locations

The co-localization of axonal long-range projections with cellular landmarks is a prerequisite for the precise analysis of the cellular connectome. To illustrate the accurate identification of the nuclei that the axons pass through and precise tracing of single axons, we performed two-channel whole-brain imaging at a voxel size of 0.32 × 0.32 × 2 μm on an 8-week-old C57BL/6J mouse injected with adeno-associated virus vector plasmid ... Read more.

Discussion

Here, we report a high-throughput method for the precise reconstruction and accurate localization of specific labelled neurons and projections in the whole brain. We observed the integrated cytoarchitecture of barrel columns in the barrel cortex using real-time counterstaining during whole-brain imaging and traced layer V/VI pyramidal neurons. In addition, we precisely traced a representative single-neuron projection on these PI-counterstained landmarks and identified the nuclei through which this projection passed. The diverse and complex spatial relationships between these neurons and nuclei suggest the necessity of reconstructing and analysing neuronal morphology and neural connections with accurate anatomical annotation. In this regard, BPS provides two major advantages.

First, the acquisition of cytoarchitecture landmarks in whole-brain imaging is simple and feasible. In traditional neurobiological experiments, the registration of neural images to a planar brain reference atlas is typically required. Individual differences among subjects, unavoidable deformation during tissue preparation, and interval-sampling planar reference atlases inevitably generate location errors in these forced registrations. Thus, staining the cytoarchitecture landmarks and labelling the neural structures in the same brain are necessary to avoid these errors. However, low-permeability nucleic acid dyes, such as PI and DAPI, are traditionally used in slide staining to label cytoarchitecture landmarks, and these stains are difficult to use when staining the whole mouse brain. Here, we propose a new concept of real-time counterstaining during whole-brain imaging, rather than whole-brain counterstaining. The low permeability of nucleic acid dyes results in superficial staining during whole-brain imaging without background interference from deep layers in the same channel. Moreover, the sectioning-staining-imaging cycle guarantees consistent PI staining in all coronal planes. This real-time staining approach avoids additional sample preparation. The stained cytoarchitecture and labelled neurons in the same field of view (FOV) are simultaneously imaged. Thus, neither additional data acquisition time for anatomical landmarks nor extra two-channel registration is needed. This concept can be applied to other mechanical section-based whole-brain imaging techniques, such as serial two-photon (STP) tomography and fluorescence micro-optical sectioning tomography (fMOST), for acquiring anatomical references in the same brain.

Second, WVT efficiently shortens the time required for data acquisition during full-volume whole-brain imaging at single-neuron resolution. The time required for data acquisition is one of the decisive factors for the use of whole-brain imaging as a routine approach in neuroanatomical studies. Currently, STP is considered the fastest approach; however, this approach sacrifices the continuity of the whole-brain imaging data set. fMOST and 2p-fMOST (ref. 17) distinguish the detailed neural structures in full-volume whole-brain imaging, requiring more than 1 week for completion. As all these techniques employ a point-scanning approach, it is challenging to improve imaging throughput. Taking advantage of the high throughput of the wide-field and the background inhibition of SIM, BPS achieves full-volume imaging of the mouse brain with high resolution within 3 days. The high continuity and self-registration of the whole-brain imaging data set through BPS facilitates the tracing of the detailed morphology of neurons, including axons, dendritic arbors, boutons and spines in the whole brain. With the rapid development of an objective with a larger FOV, a camera with a larger sensor size, and a 3D stage with longer travel, we expect that the throughput of whole-brain imaging will further improve. Compared with point-scanning whole-brain imaging, the BPS approach has significant advantages, with potential for imaging larger tissues, such as whole primate brains.

Recently, the CUBIC technique has been a notable breakthrough in counterstaining mouse brains. However, it remains unknown whether the chemically treated samples are sufficiently hard to facilitate mechanical sectioning, such as the samples used in STP or fMOST. Thus far, CUBIC is employed in light-sheet microscopy to image counterstained brains and has not shown potential for use in detailed whole-brain labelling and counterstaining images at axonal resolution.

Compared with current methods, BPS technology represents the technical advance of providing co-located cytoarchitecture for neural structure in the same whole brain. In addition, this technology also has the advantages of automation, high throughput, high resolution and robustness to accelerate the acquisition of high-resolution whole-brain data sets. BPS enables us to fast acquire both the neural structures and their own anatomic reference simultaneously. It helps to accurately identify and precisely annotate the brain-wide neural structures, which is difficult to be achieved by previous methods. Potentially, BPS can be routinely used to examine the intersubject variability of neurons. Consistent with previous studies, the results of the present study demonstrate that in different brains of mice that are the same strain, even the same types of neurons in the same brain areas differ in terms of morphology and distribution. Revealing the intersubject variability of an axonal projection is a fundamental requirement in neuroanatomy. Axonal projections of individual neurons correlate with sublaminar location, dendritic morphology, intrinsic and synaptic properties and local connectivity patterns of those neurons, respectively. The accurate tracing of axonal projections relies on the precise localization of brain regions or nuclei. Benefitting from the cytoarchitecture of the brain, the brain-wide images obtained using the BPS system can facilitate a comparison of the fine intersubject differences between axonal projections. Neuroscientists have made significant advances in the genetic labelling of specific neural circuits25. The combination of the genetic dissection of neural circuits and BPS technology will accelerate studies of the intersubject variability of neural circuits.

Full-volume imaging at high resolution inevitably generates large data sets, resulting in a crucial challenge to store, transfer, compute, manage and analyse these data. The most fundamental and essential challenge to mine large neural data sets is to recognize and reconstruct the complete morphology of single neurons from the raw data. This challenge is recognized worldwide as an open question and bottleneck. As a breakthrough, the BigNeuron project of the Allen Institute for Brain Science and the Blue Brain Project in Europe have focused on computational efforts. Recently, Markram et al. reported a simulation method of partial cortex circuits based on vast amounts of experimental and model data.

In summary, the BPS method described in the present study facilitate the acquisition of a more detailed morphology of neurons and accurately identify the brain regions or nuclei in which these are located, connected with and passing through. We propose that this method has many potential applications for neuroscience and will promote future cellular connectome studies using a traditional interval-sampling 2D atlas to precisely generate ‘personalized’ 3D maps for each brain at cellular resolution. This method could also contribute to cell type, projectome and connectome studies.