Whole Mouse Brain Reconstruction and Registration to a Reference Atlas with Standard Histochemical Processing of Coronal Sections

Abstract

Advances in molecular neuroanatomical tools have expanded the ability to map in detail connections of specific neuron subtypes in the context of behaviorally driven patterns of neuronal activity. Analysis of such data across the whole mouse brain, registered to a reference atlas, aids in understanding the functional organization of brain circuits related to behavior. A process is described to image mouse brain sections labeled with standard histochemical techniques, reconstruct those images into a whole brain image volume and register those images to the Allen Mouse Brain Common Coordinate Framework. Image analysis tools automate detection of cell bodies and quantification of axon density labeling in the structures in the annotated reference atlas. Examples of analysis are provided for mapping the axonal projections of layer specific cortical neurons using AAV-Cre dependent vectors and for mapping inputs to such neurons using retrograde trans-synaptic tracing with modified rabies viral vectors.

Graphical Abstract

Method is described for processing mouse brains with standard immunohistochemical techniques to reconstruct a whole brain rendering from coronal sections, which is aligned to the Allen mouse brain Common Coordinate Framework for quantitative analysis of labeled neurons and processes.

Introduction

Full understanding of the functional organization of neuronal circuits responsible for the generation of behavior requires mapping the connections of specific neuronal subtypes throughout the entire brain. Decades of neuroanatomical studies using axonal tract tracing techniques, histochemical localization of neurochemical and in situ hybridization localization of mRNA expression provide details of the organization of the circuits of the major brain systems. In the past 15 years several major advances have accelerated the ability to relate neuronal circuits to behavior (Luo et al., 2008, 2018). These included the generation of Cre recombinase driver lines for hundreds of neuron subtypes (Gong et al., 2007; Gerfen et al., 2013; Taniguchi et al., 2011; Harris et al., 2014; Daigle et al, 2018) combined with the development of optogenetic techniques to measure and manipulate the activity of specific neuron subtypes. Another major advance was the development of techniques to map neuroanatomical data in the whole mouse brain and to register that data to a common reference atlas (Oh et al., 2014; Zingg et al., 2014). Registering neuroanatomical data to a common reference atlas allows for data from multiple experimental cases to be used for computational analysis of the organization of brain circuits (Hooks et al, 2018). For example, the Allen Mouse Connectivity Project (Oh et al., 2014) determined the organization of projections from defined cortical areas to the thalamus using cluster analysis of the patterns of overlapping projections of several hundred injection cases in the cerebral cortex. Using a similar approach Hintiryan et al. (2016) identified distinct regions of the striatum based on their receiving convergent projections from distinct areas of the cerebral cortex. These studies confirmed prior concepts of the organization of cortico-thalamic and cortico-striatal systems based on many studies, validating the power of the computational analysis provided by the ability to register whole brain imaging of multiple injection cases into a common reference atlas. While these studies involved considerable resources, similar types of studies may be performed in smaller laboratories to address significant experimental questions. Here techniques are described that implement such analyses for quantifying the connections of defined cortical neuronal subtypes projecting to neuronal subtypes in the cerebral cortex as revealed by trans-synaptic labeling.

Methods

Histology and Imaging

For the purpose of describing whole brain imaging, the processing of a brain in which the modified rabies trans-synaptic tracing technique is used for labeling neurons that provide inputs to pyramidal tract neurons in the primary motor cortex is described (Wall et al., 2011). Mice expressing Cre-recombinase in cortical layer 5 pyramidal tract neurons (Cre driver line Sim1_KJ18, Gerfen et al, 2013) are first injected in the primary motor cortex with a “helper” virus construct that expresses the avian TVA receptor in Cre-expressing neurons, marked by expression of GFP (Wickersham et al., 2007). After 2 weeks, a second injection of a modified rabies construct, which infects neurons expressing the avian TVA receptor labeled with tdTomato (Reardon et al., 2016), is placed in the same location in the primary motor cortex. Animals are euthanized 2 weeks later, perfused trans-cardially with saline followed by 4% formaldehyde, post-fixed overnight and transferred to 20% sucrose in phosphate buffered saline

The brain is then processed for immunohistochemical localization of GFP to mark the starter cells in the primary motor cortex and tdTomato to label neurons that are trans-synaptically labeled from the starter cell population. The method of histologic processing of the brain sections is critical to whole brain reconstruction. The brain is sectioned coronally at 50μm using a sledge freezing microtome and sections are placed in a 10 well plate with a mesh bottom such that the first section is placed in the first well, the second in the second well, and then the 11^th section is placed in the first well, and so on. This results in the placement of a 1 in 10 series of sections from the brain in each well. The sections in the mesh bottom wells are then processed for immunohistochemical labeling of GFP (goat anti GFP primary, secondary) and tdTomato (rabbit anti RFP, secondary). Following the immunohistochemical procedure the sections from each well are mounted onto gelatin coated slides such that there are 10 slides each with a 1 in 10 series from the brain. The advantage of this process is that it is trivial to place the sections from each well in order from rostral to caudal to assure that the brain sections are organized sequentially. Once mounted onto slides, the sections are stained with a fluorescent Nissl counterstain (Neurotrace Blue). The process for organizing sections is shown in Figure 1.

Figure 1. Reconstruction of whole brain images from coronal sections.

(a) Coronal sections from the brain are collected in a plate of 10 wells, with successive sections being placed in sequence in well 1 through 10 and then repeated such that each well contains a 1 in 10 series through the brain. Sections are processed for immunohistochemical labeling and then mounted onto slides, with each slide containing a 1 in 10 series, which are imaged. (b) The BrainMaker software (MBF Bioscience) first outlines the individual section on each slide and then renumbers them to put them in sequence for the whole brain. Beginning with section 1 adjacent sections in the whole brain series are then overlain and aligned automatically and adjusted manually. (c) The aligned whole brain image is reconstructed into a stack of images, which may be viewed in coronal (XY), topdown (XZ), sagittal (YZ) or 3D views

The slides are imaged using a Zeiss microscope equipped with a Z-axis drive, a Hammamatsu Orca Flash 4 CMOS camera, an LED fluorescent light system to image fluorophores for GFP, RFP and Neurotrace Blue, and a Ludl motorized stage, all of which are controlled with Neurolucida software (MBF Bioscience, Williston, VT). The imaging software provides the ability to image the entire area of the slide that has sections by first establishing a focus matrix with the surface position of each tile in focus and then sequentially imaging each field of view with multiple fluorophores in 8 μm steps in the z-axis through the 50 μm thickness of the section. Labeling in brain sections may have regions with very intense labeling, typically near an injection site, and areas with very fine processes that are not so intensely labeled. To image the range of intensities the selection of the exposure time is critical and may require use of software enabling high dynamic range (HDR) of captured images, which is a feature of Neurolucida acquisition software. Typically, the image of the entire slide is made up of 700–800 tiles of 5–6 z-axis planes which are then stitched together and collapsed into a single plane using either a maximum-projection or Deep Focus algorithm such that the final single plane image of the section is optimally focused. This imaging protocol provides images of individual sections with excellent resolution of labeled neurons and axonal projections, which is essential for quantitative analysis. Other imaging platforms may be used, such as slide scanners that provide similar high-resolution imaging.

Reconstruction of coronal sections into whole brain image volume

Once the brain sections have been imaged, they are registered and reconstructed into a whole brain image volume using the BrainMaker software (MBF Bioscience, Williston, VT). The first step is to separate the multiple sections on each slide in the collection into individual images using image contrast and edge strength to automatically identify and outline sections. Where multiple sections overlap or touch, the software automatically splits contours that include multiple sections and provides an interface for manually splitting and joining contours if errors occur. Next, section order is established for all sections in the collection based on user provided number of slides and section interval (e.g., specifying that each slide contains a 1 in 10 series as described above or a contiguous series of sections.). The ordered sections are extracted from the whole slide images at a low resolution and automatically aligned pair-wise using a multiple stage image registration process that incorporates section shape and image intensities. Multiple registration hypotheses are formed to test whether a section has been flipped during processing—a common occurrence with float mounting. This automated registration process generally provides a reasonably accurate alignment of adjacent sections, however irregularities in sections can cause minor mis-alignments, which are corrected with manual adjustments of rotation and translation in the x and y axes. Following inspection and manual intervention of the alignment results, updated registration transforms from all pairs of images are composed into the same coordinate system which allows all sections to be compiled into a full-resolution 3D volume and saved as a 3D image or a series of 2D images. Typical image formats include jpeg2000 stack (JPX) or image series (JP2) or, tiff image series.

Further image processing is used to normalize the signal to account for minor differences in the intensity of label between sections or within sections. For example, to account for within-section variation common observed when visualizing an entire brain section with Nissl or DAPI fluorescent stains, a FIJI macro is used that incorporates a Gaussian Blur to model the intensity variation across the section and normalizes the intensity by inverting this model and adding it to the original image.. (Figure 2) Other image processing algorithms may be useful for discriminating the particular structures to be analyzed.

Figure 2. Normalization of Nissl staining.

Imaging of the Neurotrace blue Nissl image may produce uneven labeling intensity across the section (a). To normalize this labeling a FIJI macro program is used that utilizes a Gaussian blur function to generate a heat map of the signal intensity across the section, which is inverted (b) and added to the original image to produce an image with the signal intensity normalized across the section (c).

Detection of labeled neurons

The centers of labeled cells are identified using a scale-space analysis of the response to Laplacian of Gaussian (LoG) image filters using NeuroInfo software (MBF Bioscience, Williston, VT). Mathematically, applying an LoG filter to a 2D image, I, is described by the following equation:

$$L(x,y;\sigma )=-{\sigma }^2{\nabla }^{2}G(x,y;\sigma )\ast I(x,y),$$

where ∇² is the second order differential Laplacian operator, G is the Gaussian function with standard deviation σ, * and denotes convolution (Mikolajczyk and Schmid, 2001; Lowe, 2004). The LoG kernel, ∇²G is characterized by a positive center with a negative surround, and the result of convolution, L, is an image in which bright circular features are emphasized independent of the background signal (omit the minus sign to emphasize dark circular features). The standard deviation, σ, is a scale parameter that determines the size of the features that are emphasized.

We construct a series of LoG filters with scales determined by the expected range of labeled cell body diameters. Applying the series of filters to an image produces an LoG scale-space—in which scale is an extra dimension in addition to the image’s spatial dimensions (Lindeberg, 1994). The centers of candidate cells are determined by finding local maxima within the LoG scale-space. This scale-space analysis enables identifying candidate cells based on their expected size, excluding smaller and larger circularly-shaped features such as camera noise and cell clusters. Candidate cells are further filtered by the strength of the LoG response to separate labeled cells from spurious objects such as dendrites, which may extend from the cell body. The LoG response of candidate cells is visualized as a histogram that enables interactively setting the LoG strength threshold. The detection result is more sensitive to parameterization in regions where cells are densely packed. In our experience, using a scale range in which the maximum diameter is no more than twice the minimum diameter provides the best ability to detect individual cells within dense regions and avoids erroneously detecting clusters as single objects. Because the number of cells detected within a whole brain is too large for complete manual validation, parameters are tuned in selected regions of interest before being applied to detection across the entire 3D reconstruction.

Registration to Allen mouse brain reference atlas

The aligned whole brain mouse image volume is registered to the Allen Mouse Brain Common Coordinate Framework (CCF) using a two-step registration process. A specialized Nissl average reference image volume is built by co-registering the fluorescent Nissl counterstain channel from serial section 3D reconstruction image volumes from 78 brains to a common template image. Each image was registered to the template image using a multistage optimization of rigid, affine, and nonlinear transforms. The average brain image is constructed by resampling all images in the common template space using isotropic 16 μm voxel spacing and performing a pixel-wise average of all co-registered brains. To limit the effect of damaged tissue on the average brain, pixels that received contributions from fewer than five brains were excluded from the average.

The Nissl average image is registered to the CCF two-photon autofluorescence reference image using manual identification of 300 corresponding landmarks in the two images. The landmarks are used to construct an affine and nonlinear transform that maps points from the Nissl average image onto the CCF. The nonlinear transform used B-splines to describe deformation of a regular grid using 12³ control points. The resulting transform mapped the 300 Nissl average landmarks to their corresponding CCF landmarks with a mean error of 6.2 μm, less than a single voxel of the CCF.

Individual serial section 3D reconstruction images are registered to the Nissl average image using a multistage registration optimization of rigid, affine, and nonlinear transforms. Cell locations are then transformed into the CCF using the composition of the individual-to-Nissl-average transform and the Nissl-average-to-Allen transforms. Note that this two-step registration is determined to produce more accurate anatomical registration than direct multimodal registration of an individual serial section 3D reconstruction to the CCF. This is at least in part due to distortions caused by the physical process of cutting and mounting histological sections in this study compared to the en-bloc imaging used to create the CCF. In other words, the individual brains in this study were more similar to each other than to the CCF reference image. The image-based registration of an individual brain to the Nissl average encompassed the variation among brains while the landmark-based registration to the CCF encompassed the variation due to histological processing and imaging protocols.

The anatomical region for each cell is identified using nearest-neighbor interpolation of the CCF annotation volume. Cell counts are tabulated for each region of the CCF. The position of the mapped point relative to the midsagittal plane of the CCF is used to distinguish between left and right brain hemispheres. Detected cells registered to the CCF are displayed throughout the whole brain reconstruction for each section viewed against either the 2 photon CCF images or the CCF atlas annotation (Figure 3).

Figure 3. Detection of labeled neurons and registration to the CCF.

A section from the whole brain image stack is selected (a) and a subregion containing the labeled neurons is selected (b). Using NeuroInfo software (MBF Bioscience) the “2D cell detection tool” is used to determine the optimal LoG cell detection parameters. (c) In the first step maximum and minimum cell size parameters are chosen and the 2D cell detection tool is run on the selected are of the image for the particular channel with the cells to be detected (red in this case for labeled neurons presynaptic to the starter cells). Selected cells are marked with white dots. A histogram is generated displaying the number of detected objects dependent on the value of the LoG Strength value. The LoG algorithm detects many objects in the image with low Strength values that are not cells. (d) In the second step the “Strength Value” is adjusted such that only neurons are detected. Different areas may be sampled and tested with the adjusted “Strength Value” threshold set to assure that the parameters selected optimally detect neuron cell bodies. Once these values have been selected, the 2D cell detection function is run on the whole brain image stack. The whole brain image stack is registered to the CCF 2 photon background image. (e) A selected section displays detected neurons in the registered whole brain image stack to the CCF 2 photon background image stack. Neurons displayed in green are the starter neurons and those displayed in red are those that project to the starter cells. (f) Detected cells are displayed for each section in the CCF annotated atlas.

Determination of registration accuracy

Accuracy of the registration is verified in two ways (Figure 4). First, eight well-defined anatomical landmarks visible in individual 10 μm isotropic voxel aligned brains (“10 μm brains”) are compared to each other in standard coordinates. These landmarks include the most anterior midline point on the dorsal surface of the corpus callosum (~1.1 mm anterior to bregma), the dorsal most points in the white matter elbow under vibrissal motor cortex (same plane as corpus callosum), the most anterior midline point on the dorsal surface of the anterior commissure (~0.1 mm anterior to bregma), and the posterior locations where the fasciculus retroflexus separates from the 3^rd ventricle (~2.1 mm posterior to bregma). For each anatomical landmark, the coordinates of the anatomical landmark in ten example brains are averaged and the distance of each individual brain’s landmark from the average location is determined. This error ranges from 51.0–69.6 μm (grand mean±SD, 62.87±5.34 μm). If only the displacement in the xy plane is considered (since 80 μm planes required interpolation to achieve 10 μm brain sampling), variation in the planar offset is considerably less (grand mean±SD, 21.28 + 3.67 μm). This data suggests that anatomical locations across individual brains are well-aligned at 50 μm.

Figure 4. Analysis of the accuracy of registration to the CCF.

For a set of 8 anatomical points in 10 example brains the distance to the corresponding point in the CCF was calculated to determine the accuracy of the registration. Points were selected at 3 levels. (a) At the level of the rostral most crossing of the corpus callosum (1), the dorsal peak of the white matter on the left (2) and right (3), and the point where the lateral ventricle intersects the white matter on the left (4) and right (5). (b) The level at which the anterior commissure crosses between the hemispheres (6). (c) The level when the fasciculus retroflexus begins to descend on the left (7) and right (8). (d) The mean offset between the experimental brain and the CFF for these 8 points in 10 brains varied by ~62.9 μm ± 5.3 μm. Due to sectioning at 80 μm planes, the offset within the plane (ignoring anterior/posterior offset) was even more precise (planar offset column). As a secondary measure of alignment quality, a curve along the dorsal surface of cortex (e) for each of N=10 brains was compared to the mean cortical surface across a range of anterior/posterior positions (f: +1.1 mm anterior to bregma to −2.4 mm posterior). Individual brains in black and mean surface in red for each plane (g). Red traces include +/− sd. Except for one case with significant slicing artifact (top right), curves differed by ~59.0 μm. (Adapted from Hooks et al., 2018)

A second measure of alignment quality tests how well the dorsal surface of the cortex in a sample brain matched the mean cortical surface across the average of 10 total brains. The dorsal edge of the aligned 10 μm brains is detected and plotted as a curve for a given plane from 0.5 mm to 3.0 mm lateral to the midline for the plane 1.1 mm anterior to bregma in standard coordinates. This curve is averaged with the other brains, and the mean offset (±SD) was computed. For this plane, the SD was 57.49 μm. This process is repeated for planes along the rostrocaudal axis in 0.5 mm steps to 2.4 mm posterior to bregma. Error about the mean is similar (grand mean of SD, 59.03 μm), except in two sections where additional tissue in the individual brains on the dorsal surface of cortex near the midline increases the average. The fitted cortical surface of individual brains is plotted in black and mean surface plotted in red for each plane to compare. The significant slicing artifact in one case at 0.1 mm anterior and −0.4 mm posterior can be seen. Collectively, this suggests that the alignment quality is satisfactory for use of 50 μm aligned brains. Thus, aligned brain images were down-sampled to 50 μm isotropic voxels (156×217×248) using custom FIJI software routines. This also made computations more efficient by reducing the data set size.

Analysis of distribution of detected cells in specific brain structures in multiple cases.

In addition to being able to view detected cells registered to the CCF atlas an Excel spreadsheet with the numbers of neurons located in each of the structures in the CCF atlas is generated. Data include the number of neurons in each brain structure for each hemisphere, including for cortical areas the cell counts for each layer as well as the mean LoG strength value for the cells in each structure. Data for individual selected neurons is also available, including both the coordinates and LoG strength value. An advantage of having detected neurons registered to the CCF is that data from different cases can be compared directly. As an example Figure 5 shows the distribution of detected neurons from 2 experimental cases using trans-synaptic labeling with modified rabies virus injections, one with an injection in MOs and the other injected in SSp. Viewing the 3D reconstruction provides an overview of spatial differences in the distributions the neurons projecting to the starter cells from the two cases, while coronal sections display the detected neurons at specific levels within structures in the annotated atlas including cells located in different cortical layers and in specific thalamic nuclei. Additionally, the cell counts for the two cases can be compared from the data provided in the Excel spreadsheet.

Figure 5. Comparison of cases registered to the CCF.

(a) 3D rendering of the CCF with neurons detected from 2 cases in which rabies virus had been injected into the secondary motor cortex (MOs) in one case and the primary somatosensory cortex (SSp) in the other. For the MOs case, helper virus infected starter cells are labeled light blue and neurons providing inputs to the starter cells are labeled dark blue. For the SSp case, starter cells are labeled orange and input neurons labeled red. (b) Top down horizontal view of the MOs and SSp case neurons allow comparison of the distribution of afferent neurons. (c-j) Selected coronal sections from the CCF display detected neurons from the two cases allowing comparison of distribution in specific layers of cortical areas and thalamic nuclei. The specific section shown is indicated in the lower left of each coronal image from CCF section 160 to 315.

Discussion

Advances in genetic techniques provide ever increasing details of the connections of genetically specific neuronal subtypes (Luo et al., 2008; 2018; Tervo et al., 2016; Wall et al., 2014). The ability to analyze labeled neurons and their axonal projections in the whole mouse brain registered to a common mouse brain reference atlas enables understanding of the organization of neural circuits underlying behavior. Here a process is described to provide such analysis. Brains are sectioned coronally and processed with standard immunohistochemical techniques. Critical to the reconstruction of whole brain images from these sections is organizing the brain sections in a manner that facilitates the efficient imaging and ordering of those sections into a series that can be reconstructed into a whole brain series (Figure 1a). BrainMaker software (MBF Bioscience) registers the series of brain sections to produce a 3D reconstruction of the whole brain (Figure 1b,c). As necessary, images in the reconstructed series may be processed to normalize labeling within and across sections (Figure 2). The reconstructed whole brain is registered to the CCF with NeuroInfo software and the accuracy of the registration is determined by measuring the offset between fiducial landmarks in the registered brain volume to the AMBCCF landmarks (Figure 3). Using NeuroInfo software (MBFBioscience) labeled neurons are detected through the whole brain series and the location of these neurons are registered to the CCF atlas, producing both images of the detected cells overlain on the CCF atlas and an Excel spreadsheet with the number of detected neurons in brain structures (Figure 4). Detected cells registered to the CCF from different cases can be combined to compare and analyze the distribution of neurons providing inputs to different neuron populations (Figure 5).

Critical to the successful production of whole brain reconstructions are the histologic and imaging processes. The histologic method required to obtain optimal image signal depends on the marker used to label the particular neuroanatomical structure. Even with viral vectors that express a fluorescent tag, immunohistochemical amplification provides enhanced labeling of axonal and dendritic processes that may not be sufficiently labeled with the endogenous fluorophore. Immunohistochemical labeling involves sectioning using either a freezing microtome or vibratome and collection of the free-floating sections for processing through reagents. To collect every section to generate a whole brain image series requires organizing the tissue sections through the process. A simple procedure is to collect brain sections as they are sectioned into wells containing a 1 in 10 series in 10 processing wells, which facilitates mounting the sections onto slides in their proper order (Figure 1). Applying a fluorescent Nissl Stain or nuclear stain is essential for registering the brain sections to a standard brain atlas. The method for imaging the brain sections described here produces high resolution images sufficient to produce whole brain reconstructions. The essential elements of the imaging process involve capture of tiles of images across the brain sections using a 10x objective incorporating high dynamic range acquisition (HDR), which are stitched together in the XY planes followed by collapsing the Z planes into a single plane with labeled structures optimally focused. Using HDR areas of the image with high density of label to be imaged without saturation to be able to resolve fine morphologic structures while at the same time imaging areas with fine labeled structures such as axonal arborizations to also be resolved. To produce an image by stitching imaged tiles together requires that the captured image tiles have uniform illumination across the imaged area. This is accomplished by adjusting the illumination parameters and cropping the captured area to that which has even illumination. For each image tile, capturing multiple levels through the Z axis provides for merging those levels into a single plane such that the optimal focus for each structure is obtained. In addition to the hardware and software described here, there are other platforms that may be used, including slide scanning systems or confocal microscopes.

Producing a whole brain reconstruction of the imaged brain sections involves separating the sections of images from each imaged slide, putting these sections into serial order from the front of the brain to the back and aligning the sections sequentially using both rotation and translation of the images. BrainMaker software (MBF Bioscience) accomplishes this with a stepwise work flow. Alternatively, this process may be performed using FIJI (Paletzki and Gerfen, 2015). Once the whole brain image series is aligned, further image processing may be performed on the images using FIJI, Photoshop or other image processing software (Paletzki and Gerfen, 2015). One such image process is to normalize the uneven illumination of the Nissl labeling. Additionally, image processing techniques can be used to enhance the contrast of labeled structures, decrease background labeling and remove spurious artifacts.

Registration of the reconstructed whole brain image series to the CCF is performed in NeuroInfo software (MBF Bioscience). The process described above uses a multistage optimization of rigid, affine, and nonlinear transforms to register the image sections in the reconstructed whole brain image series to a Nissl average brain series that is generated from the average of the Nissl images of 78 brains. This registration is performed based on registration based on image intensity values. The whole brain image series registered to the Nissl average template is then registered to the CCF. Registration of the Nissl Average brain to the CCF is based on alignment of 300 fiducial markers in the Nissl average brain with their corresponding locations in the CCF. The accuracy of the registration of the Nissl average brain to the CCF is approximately 6 μm. The accuracy of the registration of the original images to the CCF is determined by selecting fiducial locations in the original section image and determining the offset to the corresponding location in the CCF. This offset is determined to be approximately 80 μm (Figure 3). There are many alternative methods to register brains to a standard atlas. An effective method is to place fiducial markers in the experimental brain and the target atlas and use a program such as 3D Slicer (Federov et al., 2012) to develop transforms to register the experimental brain to the CCF. While this method provides for accurate registration it is somewhat more labor-intensive requiring placement of many fiducial markers in each experimental brain.

The process described for constructing a whole mouse brain image volume registered to the CCF provides the ability to analyze quantitatively the connections of specific neuron subtypes for understanding the organization of neural circuits. In the example described the modified rabies trans-synaptic technique was used to map the neurons providing inputs to layer 5 PT neurons. This technique has been used in prior studies to describe the distribution of inputs to specific subtypes of layer 5 neurons in the visual cortex (Kim et al., 2015) and from somatosensory and prefrontal cortical areas (Denardo et al., 2015). Results from these studies provided detailed information of the laminar organization both locally and of long-range inputs of circuits involved in processing information within the cerebral cortex. While such data provides important information, it would be enhanced if it were to be put in the framework of a standard atlas such as the CCF as this would allow comparison of the data not only between brains in the same study, but between brains from different studies. Additionally, mapping such data from whole brain reconstructions allows for analysis not only of the numbers of neurons in particular structures but also for analysis of patterns of topographic or non-topographic organization of connectivity within such structures.

The process for registering whole brain image volumes to the CCF has been used for analysis of labeled axonal projections from specific neuron subtypes in the cerebral cortex (Hooks et al., 2018). In this study comparison of the projections of layer 5 neurons from somatosensory and motor cortical areas to the striatum from over 80 cases revealed the distinct regions within the striatum related to different cortical areas. Moreover, it was demonstrated that cortical areas connected with each other projected to overlapping areas within the striatum and that the topographic organization of these projection patterns was more precise from primary somatosensory areas than from primary and secondary motor areas. Analyzing data in the framework of registered whole brain reconstructions provides new insights into how cortical information is integrated within basal ganglia circuits. Other experimental paradigms that would benefit from such analysis include mapping the induction of immediate early genes such as c-fos or arc in different brain areas during different behaviors.

In this study a process is described for reconstruction of whole brain image volumes and registration to the standard CCF using coronal brain sections processed with standard histochemical techniques. Recent advances in labeling and imaging whole brains using clearing techniques combined with light sheet imaging provide an alternative method (Renier et al., 2015). This approach offers some advantages in that the brains are imaged intact without having to reconstruct whole brain volumes from coronal sections. Although techniques such as iDISCO (Renier et al., 2015) provide adequate immunohistochemical labeling for some antigens, the sensitivity for labeling of other antigens may be problematic. As this approach is developed it may replace the necessity of using standard brain sectioning and histologic processing. Whole mouse brain image volumes produced with these techniques may be processed and registered to the CCF and analyzed by adapting the methods described for processing coronal brain sections.