The image datasets in this study were obtained from five 8-week-old, male C57BL/6 mice with Nissl staining (Wu et al., 2014). First, the mice were anesthetized using a 1% solution of sodium pentobarbital (90 mg/kg) and perfused intracardially with 0.01M phosphate-buffered saline (PBS) and 4% paraformaldehyde at a pressure of 38 mmHg. After perfusion, the brains were removed from the skull and post-fixed. Subsequently, the brains were washed with PBS and stained with thionine solution. Then a graded series of ethanol and acetone solutions were used to dehydrate the brains. Finally, the brains were immersed in a graded series of Spurr resin solutions. The total sample preparation time was nearly 25 days. Detailed information about sample preparation can be found in our previous work (Wu et al., 2014). All the animal experiments were performed according to the procedures approved by the Institutional Animal Ethics Committee of Huazhong University of Science and Technology.

The embedded whole-brain specimens were imaged with MOST (Li et al., 2010), which performed simultaneous thin sectioning and imaging while simultaneously recording the image coordinates for automatic aligning of the images. The high-resolution images had a voxel size of 0.35 × 0.35 × 1 μm and were all saved at a depth of 8 bits. For each specimen, uninterrupted imaging was performed for 7 days, and the size of the uncompressed image dataset exceeded 2.4 terabytes with more than 11,000 coronal sections.

Due to the uneven staining and illumination, the original images were preprocessed to correct for brightness and to remove noise using a customized preprocessing program (Ding et al., 2013). Then, the spatial orientation was corrected. The preprocessed raw data is shown in Supplementary Figure 1.

The transform pattern between and the original orientation and the standard orientation was obtained by 3D transformation using Amira software (Visage Imaging, San Diego, California). First, the whole mouse brain with a voxel size of 5 × 5 × 5 μm³ was visualized, and then the 3D mouse brain was manually transformed to the standard orientation such that the dorsal part was horizontal. Ultimately, we obtained a 4 × 4 transformation matrix T (Gong et al., 2013).

A is the original image stack and B is the transformed image stack in 3D. S is the rotation factor, and t is the translation factor in 3 directions. And the gray value of all points in B were acquired with the corresponding points in A through an inverse matrix of T with a customized MATLAB (MathWorks, Natick, MA) program.

Cytoarchitecture with Nissl staining is the basic characteristic that has been used to label brain regions and the nucleus in both the Allen Reference Atlas (Dong, 2011) and the Franklin and Paxinos Mouse Atlas (Paxinos and Franklin, 2012). With reference to the Allen Reference Atlas, the Franklin and Paxinos Mouse Atlas, and the MICe mouse atlas (Dorr et al., 2008), three skilled persons segmented images (voxel size of 5 × 5 × 5 μm³) slice by slice with the filament editor in Amira software and then performed back-to-back amendment and validation manually in any of the coronal, sagittal or horizontal sections. The typical segmented brain region contour in 2D sections and 3D space were shown in Figure 2.

The same with traditional annotating rules, the main annotating rule used here for the mouse cerebral vascular system was “orientation-brain region-artery/vein,” as indicated in the names in black (Supplementary Figure 5). If one position was not sufficient to distinguish the vessels, two or three were used: orientation-orientation-brain region-artery/vein, as indicated in the names in blue. If some branches that connect to the same vessel at the same level were distinguishable with orientation, such as some branches in the pial surface and some third level branches inside the brain, we annotated them “orientation-branch of-name of connected vessel,” as indicated in the names in purple.

To obtain the microvessels in the brain regions, the datasets (voxel size of 0.35 × 0.35 × 1 μm³), with a unified size of 400 × 400 × 400 μm³ were binarized based on the Otsu method (Otsu, 1975; Wu et al., 2014). The binarized vessels were then traced with the skeletonization method in Amira software. The center lines of the vascular network were acquired via Amira, and a radius-adjustable sphere (Peng et al., 2010) was used to estimate the diameters and the results were saved as .swc format.

We assessed the tracing method both qualitatively (Supplementary Figures 6a–c) and quantitatively (Supplementary Figures 6e–i). Qualitative results in 2D and 3D both showed the diameter and centerline matched well with the raw data. For quantitative assessment, recall (R), precision (P), and diameter ratio (D) were used to estimate the traced results with the ground truth labeled manually, according to a reported method (Wu et al., 2014). Images were extracted every 70 μm, and the traced result and ground truth (segmented manually) were put together. We defined B as the overlapped regions, B1 as the total number of ground truth regions, B2 as the total number of traced regions, A1 and A2 as the areas of the traced and manually labeled regions, So R = B/B1, P = B/B2, D = A1/A2 (Wu et al., 2014). The recall reached 95.7%, while the precision was 94.4%. The diameter ratio was 96.7%. The detailed estimated results was shown in Supplementary Figure 6.

The parameters used for quantitative statistics of the vascular network were fractional vascular volume (Fv) and normalized vascular length (Nl). Firstly, the vessels were separated into 3 groups according to diameter: D > 20 μm; 8 < D ≤ 20 μm; and D ≤ 8 μm. Then, the Fv and Nl were calculated with a MATLAB toolbox, and the results were revised according to the estimated linear shrinkage of 25.9 ± 1.6% (Wu et al., 2014).

To obtain the distribution pattern of arteries and veins within different brain regions, each artery and vein, together with the connected penetrating branches, was first designated with a separate gray value during segmentation, while the 19 brain regions were labeled with different gray values. The segmented results of the vessels and brain regions were saved as separate image stacks with different gray values (voxel size of 5 × 5 × 5 μm³). However, both the vessels and brain regions were acquired from the same dataset; thus, their coordinates were matched, and the distribution in different brain regions of each artery and vein could be calculated with the volume of all vascular branches in 3D. The volume of the vessels was calculated by adding together the voxels in in the 3D binarized image stack. The radii of the pial surficial vessels were missing and thus the artery was set to 150 μm and the veins were set to 250 μm uniformly according to the reported mean value (Müller et al., 2008). The calculations and mean distribution results for 5 mouse brains were performed using a customized MATLAB program.

To construct the precise arterial and venous vascular system in the whole mouse brain, we acquired serial datasets in five 8-week-old male C57BL/6 mice with a voxel resolution of 1 μm using modified Nissl staining and MOST imaging. We traced the pial surface arteries and veins. First, the brain contours of the whole mouse brain were segmented within serial coronal sections (Supplementary Figure 2) based on cytoarchitecture, using the Otsu method (section Materials and Methods), and the 3D brain surface was visualized with surface rendering using Amira (Visage Imaging, San Diego, California) software (Figures 1A,C). This high-quality imaging technology allowed the continuous dents of pial surface vessels to be clearly distinguished on the reconstructed brain surface. Then, based on these dents, we semi-automatically traced the mouse pial surface vessels (see the section Materials and Methods). Finally, arteries, including the middle cerebral artery (Mcer) and pial branches of the anterior cerebral artery (Acer) and posterior cerebral artery (Pcer), and veins, including the superior sagittal sinus (Sss), rostral rhinal vein (Rrhv), caudal rhinal vein (Crhv), and transverse sinuses (Trs) were acquired (Figures 1B,D). The diameter of the smallest terminal branches of the pial surface vessels was as small as 25–30 μm.

To acquire the structures of the cerebral vascular system of the whole mouse brain and to distinguish arteries and veins, we continuously traced the penetrating vessels in large areas using the regional growing method (section Materials and Methods), along the connecting paths with the identified arteries and veins in the pial surface. The fine vascular connecting paths from the pial surface to the deep-brain across the whole brain were acquired, and the diameters of the arterial terminal branches were as small as 8.3 ± 2.6 μm, while the venous terminal branches were as small as 8.7 ± 3.7 μm. To provide a better understanding of the connecting path, Supplementary Figure 3 showed the path between Mcer and the anterior striate artery (Astr) in an individual mouse brain as an example, as a 2D section and a 3D volume rendering. The existing arterial system was fine to about 20 μm (Ghanavati et al., 2014). Here, we reconstructed the 3D fine vascular structures of both the arteries and veins in one and the same mouse brain to make the precise vascular atlas.

We accurately located the cerebral vascular system in the whole mouse brain. Cytoarchitecture is regarded as the gold standard in brain regions labeling (Dong, 2011; Paxinos and Franklin, 2012). Datasets acquired with modified Nissl staining and serial high resolution imaging, provide both vascular structures and cytoarchitecture. Because the cerebral vascular structures and brain regions were imaged at the same time and in the same mouse brain, no more registration was needed. We manually segmented 19 main brain regions (Figure 2, Supplementary Table 1) in each of 5 whole mouse brains (section Materials and Methods) referencing the Allen Reference Atlas (Dong, 2011), the Franklin and Paxinos Mouse Atlas (Paxinos and Franklin, 2012), and the MICe (Mouse Imaging Center) mouse atlas (Dorr et al., 2008). Subsequently, we accurately located the traced vessels in 3D, based on the segmentated brain regions.

We annotated the traced cerebral vascular system of arteries and veins based mainly on the following naming rule (Supplementary Figure 5): “orientation-brain region-artery/vein.” The fine 3D distribution and annotation results of the traced cerebral vascular system in an individual mouse brain were presented (Figures 3, 4). The traced cerebral vascular system included, six main arteries: Acer, Mcer, Pcer, and anterior choroidal artery (Ach) originated from the internal carotid artery (Ictd) and the basilar artery (Bas), superior cerebellar artery (Scba) originated from the vertebral artery (Vert). Twelve main veins in four areas: Rrhv, Sss, Trs, and Crhv in dorsal area, the basal vein [Basv: anterior cerebral vein (Acerv), middle cerebral vein (Mcerv)], anterior choroidal vein (Achv) in the ventral area, the vein of Galen (Gcv), azygos internal cerebral vein (Azicv), longitudinal hippocampal vein (Lhiv), medial collicular vein (Mcolv), and lateral collicular vein (Lclov) inside the brain and the Unamed vein [Unamedv: (Unamedv1, Unamedv2)] in the caudal area. To show the positions of the vessels clearly, the contours of the corresponding brain regions were shown as well.

In the arterial system, Acer is the rostral branch of the Ictd. Its main trunks head dorsally to the pial surface along the midline area, and the branches are mainly located in the HY, striatum (CP), frontal lobe (FL), olfactory bulb (OLF), and dorsal part of the cerebral cortex. Mcer is the middle branch of the Ictd. The three branches of Mcer extend on the pial surface in the ventral to dorsal direction, together with branches of Acer in the pial surface in the dorsal to ventral direction, cover almost all of the cerebral cortex. The branches of Mcer inside the brain are mainly located in the HY, CP, and amygdala (AMY). Ach originates from the Ictd and is located near Mcer in the caudal direction. The branches of Ach head in the ventral to dorsal direction to the dorsum of the TH through the AMY, HY, TH, and choroid. Pcer is the caudal branch of the Ictd. The main trunk of Pcer heads dorsally through the area between the cerebral cortex and midbrain (MB), and four large branches extend to the medial choroid, lateral choroid, ventral HIP, and SC, respectively, while pial surface branches in the caudal to rostral direction supply the cerebral cortex not covered by Acer and Mcer. Bas originates from the fusion of the two Verts, and is located in the ventrum of the medulla (MY) and pons (P). There are also branches that penetrate the MY and P in the ventral to dorsal direction. Then, Bas extends to the ventral part of the TH to form Scba, which extends in the ventral to dorsal direction into the rostral area of the cerebellum (CB). The Scba supplies blood to the CB and is connected with the Pcer with the posterior communicating artery (Pcom), forming a cooperation between the two blood supply sources.

In addition, in the venous system, in the dorsal area, the branches of Rrhv head in the dorsal to ventral direction to supply the FL and OLF. The main trunks of the Sss travel along the midline in the pial surface, and the branches cover the dorsal cerebral cortex and subcortex. Trs is located in the caudal area of the cerebral cortex, and heads ventrally to form Crhv, which extends in the caudal to rostral direction through the lateral midline on the pial surface, together with Rrhv, Sss, and Trs and covers almost all the pial surface. The penetrating branches of Crhv extend to the CP and HIP. In the ventral area, Basv branches off Acerv and Mcerv, which are almost parallel to Acer and Mcer, respectively, but the extending distance to the dorsal area are short. Penetrating branches of Basv extend to the cerebral cortex, CP, HY, AMY, TH, HIP, and MB from the ventral area. Achv, which is connected to Basv, extends in the ventral to dorsal direction. Achv runs almost parallel to Ach. Inside the brain, Gcv produces Azicv and Lhiv. Azicv heads in the opposite direction of Acer to the corpus callosum (cc), CP, TH, and HIP, along the midline area. The main trunks of Lhiv are located between HIP and TH, and branches also penetrate HIP and TH. Lcolv covers the lower portion of HIP. The main trunks of Mcolv exist between MB and cortex, and branches will supply CB, MB, and TH. In caudal area, Unamedv contains Unamedv1, located in dorsal CB, and Unamedv2, located in dorsal P, and penetrates into cerebellum and pons.

Detailed spatial distributions and extending patterns of all annotated vessels are described in the Supplementary Notes. We first provide 3D distributions and annotation results for the fine vascular system of both the arteries and veins in the whole mouse brain. The annotations of arteries and the large veins are the same as those reported in the literature (Cook, 1965; Dorr et al., 2007; Lin et al., 2009; Scremin, 2012), but we find and annotate some new branches, which are presented in Figure 4, especially in veins.

Combined with the brain regions provided by the cytoarchitecture, we accurately located each traced artery and vein. The extending directions, detailed extending paths, numbers, covered areas, and the relationships with brain regions of the vascular branches could be determined. Furthermore, we could acquire quantitative morphological parameters based on the vectorized results, such as the length, diameter, order, volume, density, and bifurcation number of the vascular branches.

To create a conserved cerebral vascular system atlas, we reconstructed and annotated the cerebral vascular system in five C57BL/6 mouse brains. The reconstructed structures of Pcer, Scba, Lhiv, and Lcolv in these brains were presented to illustrate the conserved vascular distributions among different individuals (Figure 5). Trhi always originates from Pcer and penetrates the HIP in the caudal to rostral direction, and the thalamoperforating artery (Thp) always originates from Scba and emanates in the ventral to dorsal direction in the TH below the HIP. Trhiv originates from Lhiv, and Vtrhiv originates from Lcolv in the upper and lower portions of the HIP, respectively. The thalamoperforating vein (Thpv) emerges from Lhiv and penetrates into the TH from the dorsal and lateral areas to form an anastomotic distribution with Thp. The qualitative results indicate that the distributions of arteries and veins are conserved among the different mouse brains.

The 3D reconstructed cerebral vascular system in the brains mentioned above were combined to obtain conserved distribution patterns of the brain-wide arterial and venous vascular system in the form of a 2D brain map (Figure 6). We classified the arteries and veins according to the bifurcation position, starting from the origin of large vessels and the diameter of the branches from large to small. To ensure the readability of the image, Figure 6 show the distributions and extents of the arteries and veins at 3 levels (about D > 90 μm, 40 < D < 90 μm, D < 40 μm) with diameters from large to small. The results show the distribution patterns of both the arteries and veins (Figure 6) and the relationships between the arteries and veins (Figure 7). The cortical penetrating arteries and penetrating veins, Ach and Achv, and Mcer and Mcerv, are parallel in the cortical area. By contrast, in the subcortical area, Thp and Thpv and Thsv and the lateral hypothalamic artery (Lhy) anastomose in the dorsal to ventral direction; rostral septal artery (Rsa) and dorsal septal vein (Dsv) anastomose in the rostral to caudal direction; Thp and the ventral thalamic vein (Vthv) anastomose in the medial to lateral direction, and Trhi and Thriv, and Trhi and Vtrhiv distribute alternately. However, there is no relationship of the distribution patterns between most of the arteries and veins, indicating that it is necessary to study both arteries and veins. Based on our results, the locations, extents, connecting patterns, and coverage area of each artery and vein in the whole mouse brain can be observed directly. We first provide a fine mouse cerebral vascular system atlas in stereotaxic coordinates of both the arteries and veins at three levels, which will benefit the understanding of the mouse cerebral vascular system, especially to redefine the knowledge of the venous vascular system.

The brain regions of different function are correlated with neural activities. When a sensory is inputted to a brain region, the flow of blood will increase in the brain function imaging (Grinvald et al., 1986; Blinder et al., 2013). But the knowledge of real vascular distribution patterns and the vascular density within different brain regions are rather sparse (Hirsch et al., 2012). The areas that are affected when an artery or vein is infarcted, that is the distributing patterns between vessels and brain regions, provide essential information for the study of clinical and cerebral vascular diseases. Previous studies regarded the brain regions as the marks to orientate the vessels and there have been no complete distribution proportion results among blood vessels and brain regions of the whole mouse brain.

To acquire the distributing patterns of the arteries and veins with brain regions, we obtained quantitative statistics of the mean volume distribution proportions (Supplementary Figure 7) of each artery and vein within different brain regions (see the section Materials and Methods) (n = 5). The quantitative statistical results for Mcer and Lhiv (Figures 8A,B) presented as samples of arteries and veins, respectively. The statistical results show that Mcer is mainly distributed in the FL, parietal temporal lobe (PTL) and CP, in which the cortex represents 80.6% in all and the CP represents the maximum portion in the subcortex of 3.8%. Lhiv is mainly distributed in the HIP, TH, MB, and cc; these 4 regions account for ~80% of Lhiv. Based on the distributions of different arteries and veins in different brain regions, we constructed the distributing patterns between the vessels and brain regions (Figures 8C,D). Only the distributing patterns results of Ach and Achv with the brain regions are consistent (11 brain regions with one difference), while the results for the other arteries and veins are not. Each artery or vein connects to one or more brain regions, and there are weakly connected brain regions. The distributing patterns of individual vessels different from each other. Each brain region is connected to many different arteries and veins to ensure the blood supply, indicating that the fine and accurately located vascular system of both the arteries and veins and the distributing patterns between cerebral vessels and brain regions in whole mouse brain are of great importance to the study of vascular distributions and brain function. Here we provide the relation between the vessels and brain regions, and we will provide the vascular density in different brain regions in the next part.

To quantitatively study the conserved distributions of both arteries and veins among different individuals, we counted the branch numbers of 6 large arteries and large veins in 4 areas of 5 mouse brains. The statistical results show that the standard deviation values are small for both arterial and venous branch numbers at the first and second levels with diameter D > 90 μm and 40 < D < 90 μm, respectively (Table 1). The mean values are similar between the left and right brain of both arteries and veins at the first and second levels, indicating that the branches of arteries and veins at the first and second levels are conserved both among different mouse brains and between the left and right brain. No correlation is observed between the statistical results for the arteries and veins. The branch numbers increase with the level increases in both arteries and veins. The maximum branch number is observed in both arteries and veins at the third level with diameter D < 40 μm. The differences in the mean and standard deviation values increase at the third level, reflecting differences in vascular distribution among the different mouse brains and between the left and right brains. These findings confirm traditional beliefs about mouse brain vessels that the large vessels are more conserved than the small vessels (Semenza, 2007; Geudens and Gerhardt, 2011). Detailed statistical results are listed in Table 1.

Functional interaction between the cerebral vessels and brain tissues occurs in the microvessels, where the neurovascular unit is generated (Zhang et al., 2012) that maintains the energy metabolism of neurons and glial cells. To facilitate functional study of the cerebral vessels, we further analyzed the microvessels (Figure 8C, Supplementary Figures 8a3–e3). The capillaries were separated with the diameter size of 8 μm (Michaloudi et al., 2006).

We provided fine microvascular structures (Figures 9A–C) and statistical vascular densities in five typical brain regions, as calculations at the microvessels are very time consuming. Based on the cytoarchitecture (Supplementary Figures 8a–e), we selected the S1BF, SC, TH, HIP, and HY as five brain regions in the cortical and subcortical areas as study subjects in three mouse brains. One image stack with a size of 400 × 400 × 400 μm (about 1.4 GB) was chosen in both the left and right brains of three mice (n = 6). Fractional vascular volume (Fv) and normalized vascular length (Nl) were calculated in small-sized vessels (capillaries) (SD, D < 8 μm), medium-sized vessels (MD, 8 < D < 20 μm), and large-sized vessels (LD, D > 20 μm; Michaloudi et al., 2006), and the difference in distribution of the vascular network among five brain regions was analyzed by One-way ANOVA. The results (Figures 9D,E) show that the Fv and Nl of the capillaries are both high in the TH because all information transmission to the cortex passes through the TH (Sherman and Guillery, 2001). The density values of medium-sized and large-sized vessels are lowest in the HY, which contains mainly small-sized and medium-sized vessels. The statistical results reveal that the differences of vascular density mainly exist mainly in small-sized vessels and large-sized vessels among different brain regions (Figures 9F–K), especially in small-sized vessels (capillary). And the connected paths between the arterioles and venules, have the potential to be used for calculating blood flow in the brain.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.