Open Access
Add to BibSonomy
Abstract
The kidney is an important organ for excreting metabolic waste and maintaining the stability of the body’s internal environment. The renal function involves multiple complex and fine structures in the whole kidney, and any change in these structures may cause impaired nephric function. Consequently, achieving three-dimensional (3D) reconstruction of the entire kidney at a single-cell resolution is of significant importance for understanding the kidney’s structural characteristics and exploring the pathogenesis of kidney diseases. In this paper, we propose a pipeline from sample preparation to optical microscopic imaging of the entire kidney, followed by data processing for 3D reconstruction of the whole mouse kidney. We employed transgenic fluorescent labeling and propidium iodide (PI) labeling to obtain detailed information about the vascular structure and cytoarchitecture of the kidney. Subsequently, the entire mouse kidney was imaged at submicron-resolution using high-definition fluorescent micro-optical sectioning tomography (HD-fMOST). Finally, we reconstructed the structures of interest through various data processing methods on the original images. This included detecting glomeruli throughout the entire kidney, as well as the segmentation and visualization of the renal arteries, veins, and three different types of nephrons. Our method provides a powerful tool for studying the renal microstructure and its spatial relationships throughout the entire kidney.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
The function of the kidney is to excrete metabolic waste and play a crucial role in maintaining a healthy balance of water, salts, and minerals in the body. The metabolic process of the kidney starts from the renal artery, and the blood travels through the arterial system to reach the glomeruli in the renal cortex. Filtered by the renal corpuscles, the filtered fluid enters the renal tubules, and after reabsorption in the tubules, the remaining fluid and wastes become urine. The entire metabolic process spans the whole kidney. Meanwhile, the renal structures involved in metabolism are very fine. For example, the glomerulus is a vascular bulb formed by the intertwining of capillaries branching from the afferent arteriole (AA), and its structure can only be observed at the level of capillaries [1]; the renal tubules are long and slender tubules connected to the renal corpuscles, with the narrowest part having a diameter of only 10-15 µm [2,3]. The morphological structure of these tubules can only be observed at the cellular resolution [4]. These fine structures in the kidney are closely coordinated to maintain kidney function as well as the normal operation of the urinary system. Any alterations in these structures can lead to kidney dysfunction. For example, glomerular injury leads to a range of kidney diseases, including chronic kidney disease and end-stage renal disease [5]; tubular injury leads to chronic kidney disease [6]; and renal vascular rarefaction is associated with common cardiovascular diseases such as diabetes, hypertension, and atherosclerosis [7]. Therefore, precise 3D reconstruction of the entire kidney at a single-cell resolution is of great significance for a comprehensive understanding of renal structures and function, as well as for investigating the pathogenesis of kidney diseases.
In recent decades, ultrasound, magnetic resonance imaging (MRI), and computed tomography (CT) have been the primary modalities for kidney imaging. Ultrasound is used to detect renal obstruction or renal medical diseases [8]. MRI is employed to determine the stage of renal cell carcinoma [9]. CT is utilized to assess renal vascular morphology and to locate kidney stones [10,11]. However, these traditional methods are constrained by resolution limitations, preventing them from achieving 3D imaging of the fine structures of the kidney. The optical microscope can provide extensive visualization of the kidney microstructure. However, due to light scattering caused by the refractive properties of biological tissues, image quality significantly deteriorates with increasing imaging depth. Consequently, traditional optical microscopy is unable to image the entire kidney. In recent years, many tissue optical clearing methods have been developed to alter the optical properties of tissues thereby increasing the depth of imaging [12,13]. After tissue optical clearing, optical microscopy (confocal microscopy, multiphoton microscopy, light-sheet microscopy) can be used to image the kidney and reconstruct renal structures. For example, Anika Klingberg et al. used light-sheet microscopy to analyze glomerular number and size [14]; Turgay Saritas et al. used a combination of immunolabeling, light-sheet microscopy, and confocal microscopy to assess potassium-mediated tubular remodeling in adult mouse kidneys [15]; Ebrahim Tahaei et al. used confocal microscopy to reveal sexual dimorphism in the length of the distal convoluted tubule (DCT) [16]; Thomas Blanc et al. used a combination of multiphoton microscopy and digital tracing to reconstruct single nephrons under physiological and pathological conditions and to explore the segments of cyst formation [2]. However, tissue optical clearing techniques cannot fundamentally solve the problem, resulting in poorer image quality at greater imaging depths [17]. Compared with optical microscopy, electron microscopy has a shorter wavelength of the electron beam and therefore higher resolution, making it an important tool for ultrastructural analysis of renal tissues as well as for nephropathy research [18,19]. However, due to the complexity of sample preparation, only local renal tissue can be observed with high resolution and large samples cannot be imaged. For example, the 3D reconstruction of regional glomeruli [20] and visualization of the 3D structure of podocytes [21]. Therefore, there is a lack of a method that allows high-resolution imaging of the entire kidney and reconstruction of its principal structures.
To this end, this paper proposes a pipeline for 3D reconstruction of the entire mouse kidney at single-cell resolution. The pipeline contains three steps: sample preparation, kidney imaging, and data processing (Fig. 1). First, we employed a transgenic fluorescent labeling strategy to label mouse kidney blood vessels. Subsequently, we utilized the HD-fMOST [22] to image the entire kidney at submicron-resolution. A dual-channel imaging strategy was used to obtain both vascular structure and cytoarchitecture information of the whole kidney during the imaging process. Finally, data processing was conducted on the raw images to reconstruct typical structures of the kidney. We performed 3D reconstruction of the complete renal arteries and veins. Additionally, we detected and statistically analyzed the distribution characteristics of glomeruli throughout the entire kidney. Furthermore, we reconstructed the morphological structures of the entire nephron and analyzed the morphology differences among three different types of nephrons. The results demonstrate that our proposed pipeline can achieve 3D reconstruction of the fine structures in the entire mouse kidney. This not only reveals the morphology of individual structures and their spatial relationships but also provides a powerful tool for studies of renal physiology and pathology.
Fig. 1. Kidney reconstruction pipeline at single-cell resolution. (a) Mouse kidney sample preparation. Fluorescent labeling of blood vessels using Tek-Cre::Ai47 transgenic mice. The entire kidney was removed from the mouse and embedded with HM20 resin. (b) Whole kidney imaging. HD-fMOST adopted dual-channel imaging to obtain kidney vascular and cytoarchitecture information. (c) Data processing. Target detection, image segmentation, and digital tracing methods were used to process the raw images to reconstruct the structures of the kidney in 3D.
2. Materials and methods
2.1 Animals
In this study, Tek-Cre mice were crossed with Ai47 mice to obtain Tek-Cre::Ai47 transgenic mice, which have vascular endothelial cells carrying green fluorescent protein (GFP) [23]. The mice were placed in free cages. The 12:12 hours light/dark cycle and provided with adequate food and water. We then removed the mouse kidney for imaging. The specific sample preparation process is detailed in [24]. Four 8-week-old mice were used for the experiments in this paper. All animal experiments followed procedures approved by the Institutional Animal Ethics Committee of Huazhong University of Science and Technology, and all experiments were carried out under relevant guidelines and regulations.
2.2 High-resolution imaging of the whole kidney
The dual-channel HD-fMOST was used to capture the fluorescence signal of vascular endothelial cells and the PI-labeled cytoarchitecture information. The specific imaging process is as follows: Dissected samples were post-fixed in 4% paraformaldehyde, rinsed, dehydrated with alcohol, and embedded in Lowicryl HM20 resin. The embedded samples were imaged by HD-fMOST system at a voxel resolution of 0.325 µm × 0.325 µm × 1 µm. The system utilizes the natural intensity modulation of Gaussian line illumination to achieve high-throughput line scanning with remarkable background inhibition. After acquiring the optical sectioning images of the two coronal layers by block-face imaging, a diamond knife was used to cut off the corresponding top layer of 2 µm at a feed speed of 5 mm/s. Moreover, simultaneous staining with PI (1µg / mL) was used during the imaging process to obtain cytoarchitecture landmarks. The data acquisition was completed in 4 days. The acquired dataset contains over 9,000 slices, with a single slice size of 28,000 × 28,000 pixels, and the uncompressed data volume of two channels is about 19 terabytes (TB).
2.3 Glomeruli detection model based on deep learning
We proposed an end-to-end convolutional neural network (CNN) for detection of glomeruli in the whole kidney. The CNN extracts glomerular features from the input images and predicts the precise coordinates of the glomeruli. Since glomeruli are observed in both the vessel channel and PI channel with different features in morphology, we used images from both channels to learn glomerular features simultaneously. Firstly, we down-sample the original images to 2µm × 2µm × 2µm and then cropped 125 pairs 3D images of 160 × 160 × 160 pixels for network training. The network uses 3D-CNN to extract features from the input images. Then, it obtains five parameters: the coordinates of the prediction point (x, y, z), the side length of the prediction box (fixed value), and the confidence. Finally, non-maximum suppression and distance suppression algorithms are used to obtain the final prediction result, which is the precise position of the glomeruli. The detailed network structure and implementation process are described in our previous work [25,26]. Since previous work focused on glomeruli detection in the local region, the data blocks used in the network training and prediction process all contain glomeruli. However, the glomeruli are only present in the cortex of the kidney. If this method is directly applied to detect glomeruli across the entire kidney, the network will erroneously detect some bright vessels as glomeruli, resulting in false positives. Therefore, based on previous efforts, we introduced additional data blocks that do not contain glomeruli during the network training process to enhance the generalization performance of the network. We used a total of 215 (125 with glomerulus and 90 without glomerulus) 3D blocks with a size of 160 × 160 × 160 pixels for network training. Then, we split these blocks into the training set (180 blocks), validation set (10 blocks), and test set (25 blocks). Since the whole kidney image cannot be read into memory at one time for prediction, we divided the entire kidney image into multiple 160 × 160 × 160 pixels blocks for detection respectively. To avoid inaccurate detection at the block boundary, there is a 32-pixel overlapping region between adjacent data blocks. After all block detection is completed, the results are merged to obtain the glomeruli detection results of the whole kidney.
2.4 Vessel reconstruction
The morphological characteristics of the renal arteries and veins can be observed in both the vessel channel and PI channel. However, in the vessel channel, arteries and veins appear as bright lumens without clear cellular boundaries, making 3D reconstruction challenging. Therefore, we reconstructed the kidney vascular system on PI channel images using a threshold segmentation method. First, we down-sampled the original images to 6µm × 6µm × 6µm. Then, the Segmentation Editor of the Amira software was used to segment the vessels in 2D mode. The segmentation threshold was manually adjusted to select a region containing any vessel. Through connectivity analysis, the software automatically segments pixels connected to this region. Manually inspecting the segmented images and correcting any inaccuracies. We repeated these steps until all the vessels were reconstructed. To better observe the spatial position of vessels, we reconstructed the renal pelvis in the same way.
2.5 Nephron reconstruction
The TDat-Amira plug-in [27] was used to trace nephrons. To facilitate the processing of TB-scale whole kidney images, the original image of the PI channel was reformatted into TDat format (a multi-level resolution image format). Then, a region of interest (ROI) containing glomeruli was selected. Subsequently, this ROI was displayed in 2D series images in the Filament Editor module of Amira. We select a glomerulus from the image and semi-automatically trace it along the direction of the nephron from the urine pole of the glomerulus. When the tracing reaches the boundary of this ROI, we adjust the coordinates and repeat the above operation until tracing the end of the collecting duct (CD).
2.6 Visualization and analysis
All 2D images are visualized in ImageJ. For 3D reconstruction results, Amira software was used for visualization. The results of renal vessel and renal pelvis reconstruction were displayed in different colors by the SurfaceGen module. The results of whole kidney glomeruli detection and nephrons reconstruction were visualized using the SpatialGraphView module. In addition, in order to demonstrate the distance of the glomeruli from the kidney surface, we used VTK to display the glomeruli at different locations in different colors.
3. Results
3.1 Results of precise dual-channel imaging of the entire kidney
Using the HD-fMOST, we can image the entire mouse kidney at a single-cell resolution and simultaneously obtain images of both the vascular structure and cytoarchitecture of the kidney. As shown in Fig. 2(a) and Visualization 1, the morphology of vessels and cells and their distribution in the kidney can be clearly observed. At the same time, it is possible to distinguish between arteries and veins in the kidney based on the characteristics of the vascular walls. Arteries are identified by thicker vessel walls that appear brighter, while veins, accompanying arteries with thinner vessel walls, appear darker (Fig. 2(b)). Cellular morphology also exhibits variations in different regions. Figure 2(c), (d) respectively illustrate the cytoarchitecture characteristics in the inner medulla (IM) and cortex regions. Based on the vascular and cytoarchitecture features in the images, we can obtain macroscopic-level information about the kidney. As shown in Fig. 2(e), we are able to obtain a 3D image of the complete kidney, in which the kidney morphology and major blood vessel branches are visible. In the vessel channel of the coronal section of the kidney (Fig. 2(f)), it is evident that there are significant differences in the morphology and density of vessels in different regions, which could clearly distinguish the four subregions of the kidney: the cortex, the outer stripe of the outer medulla (OSOM), the inner stripe of the outer medulla (ISOM), and IM. Similarly, in the PI channel, the differences in the morphology and density of cells in different regions can also be observed (Fig. 2 (g)).
Fig. 2. The results of dual-channel imaging of the whole kidney. (a) Horizontal section of the merged dual-channel image of the kidney. The vessel channel is in green and the PI channel is in red. (b-d) Enlarged view of the region indicated by the white box in (a). A represents an arterial vessel, and V represents a venous vessel. (e) 3D volume rendering of the kidney. (f) Coronal projection of the kidney in the vessel channel. A maximum intensity projection with a thickness of 100µm. The section position is marked by the yellow dashed line in (e). Dashed lines delineate different subzones of the kidney. (g) Coronal projection of the kidney in the PI channel. An average intensity projection with a thickness of 100 µm, and the section position is the same as in (f). Scale bar, (a) 1000µm; (b-d) 200µm; (f,g) 1000µm.
Furthermore, our imaging method could observe the typical features of the microstructure of the kidney. As shown in Fig. 3(a), (d), the structure of the glomerulus can be observed in the vessel channel. The spherical mass in the center is the vascular tuft formed by the winding capillaries. The larger vessels on the upper and lower sides of the glomerulus represent the AA and efferent arteriole (EA), respectively. It is worth noting that we only show the details of the glomerulus. The EA and AA is distinguished by whether they are connected to the interlobular arteries. The AA is formed by a branch of interlobular arteries (see Fig. 5(f)). In the PI channel, the cellular morphology of the nephron can be clearly observed. Figure 3(b), (e) shows a typical structure of the nephron in the cortex region. The structure indicated by the blue box represents a renal corpuscle, and the tubule emanating from the glomerulus and connecting with the Bowman’s capsule is the proximal tubule (PT). It can be seen that the tubules in the cortex region with thicker cell walls and larger tube diameters. Figure 3(c), (f) shows the structure of the renal tubules in the IM region. Compared to the tubules in the cortex region, the tubules in the IM region have thinner cell walls and smaller tube diameters and are located closer to the renal pelvis. Most of the tubules in this area are CD.
Fig. 3. 2D and 3D visualization of typical kidney structures. (a) Glomerulus in the 20µm projection image of vessel channel. AA represents the afferent arteriole, EA represents the efferent arteriole, and the red arrows represent the direction of blood flow. (b) The nephron of the cortex region in the PI channel. The blue dashed box is a renal corpuscle, BC is the Bowman’s capsule, and PT is the proximal tubule. (c) The renal tubules of the IM region in the PI channel. CD is the collecting duct. (d-f) 3D visualization of (a), (b), and (c), respectively. Scale bar, (a-c) 80µm. block size, (d) 246µm × 227µm × 115µm; (e) 340µm × 323 µm × 193µm; (f) 232µm × 220µm × 176 µm.
Fig. 4. 3D reconstruction of renal arteries and veins. (a) Visualization of the whole renal arteries, veins, and renal pelvis. Red represents renal arteries, blue represents renal veins, and yellow represents the renal pelvis. The renal pelvis terminal tubule is the ureter. (b) Reconstruction result of renal arteries. White arrows represent the anterior trunk, from top to bottom: superior segment, anterior superior segment, anterior inferior segment, and inferior segment. The yellow arrows represent the posterior segmental artery. (c) Renal veins reconstruction results. (d) Interlobular artery. An enlarged image of the area is shown by the green box in (b). (e) The original image in the vessel channel with the white box in (d) represents the AAs and their capillary branches entangled to form the glomeruli. (f) 3D reconstruction results of (e). (g) Comparison of renal artery reconstruction results in three mice. scale bar, (a-c) 1000µm; (d) 400µm; (g) 2000µm. block size, (e, f) 365µm × 315µm × 302µm.
Fig. 5. Glomeruli detection results and quantitative analysis. (a) Schematic representation of the glomerular detection neural network. CH1 is the vessel channel image. CH2 is the PI channel image. (b, c) The glomeruli detection results were overlaid with the vessel channel image projection. The green dots represent the detected glomeruli. (d) Quantitative statistical results of glomeruli detection model. (e) Glomeruli detection across the entire kidney. (f) Color-coded rendering of glomeruli based on their distance from the kidney surface. (g, h) Horizontal sections of the upper and middle parts of (f). (i) Distribution of the glomeruli number and the distance from the kidney surface. The three curves represent three different mice, and the n value represents the total number of glomeruli. Scale bar, (b, c) 200µm; (e-h) 1000µm.
3.2 Results of renal arteries and veins reconstruction
Due to our imaging method’s ability to capture the entire kidney at high resolution, we can perform 3D reconstruction of the arteries and veins that span the whole kidney. Figure 4(a)-(c) and Visualization 2 show the 3D reconstruction results of the renal arteries, veins and renal pelvis. It can be seen that both the arteries and veins exhibit a fan-shaped pattern. Their morphology and spatial orientation are generally similar, but the veins have a thicker trunk. The renal pelvis is located in the middle of the kidney and is shaped like a funnel, with the lower end connected to a thin cylindrical ureter. Figure 4(d) shows the interlobular arteries formed by arterial branches, and their terminals further branch to form AAs. To show this structure, we reconstructed the AA region, and the original image and reconstruction results are shown in Fig. 4(e), (f). This demonstrates that our method is capable of not only reconstructing large vascular branches but also conducting reconstruction of micro-vessels.
Additionally, in order to study the vascular structure characteristics among different individuals, we reconstructed the renal arteries of three mice (Fig. 4 (g)) to reveal the commonalities of renal vessel distribution in mice. According to the classification standards for human renal arteries [28,29], we divided the mouse renal arteries into five segments: superior segment, anterior superior segment, anterior inferior segment, inferior segment, and posterior segment (Fig. 4(b)). Subsequently, we segmented the three arteries (Fig. 4(g)) according to this standard and obtained the number of branches for each segment, as shown in Table 1. Statistics showed that the posterior segment had three branches in all three mice, whereas the superior, anterior superior, anterior inferior, and inferior segments each had one branch (except for the absence of the superior segmental artery in data 3). This suggests that although the morphology of the mouse renal vessel is different, there is still uniformity in the distribution of each segment in the kidney.
3.3 Results of glomeruli detection in whole kidney
We used the proposed 3D-CNN to detect glomeruli in the whole kidney and visualized and analyzed the results (Fig. 5(a)). First, we evaluated the efficacy of glomeruli detection. We selected two vessel channel data blocks with sizes of 1000 µm × 1000 µm × 200 µm overlaid the glomeruli detection results (Fig. 5(b), (c)). Almost all glomeruli in the image were detected and their locations are also accurate (Visualization 3). We randomly selected 25 data blocks with sizes of 320 µm × 320 µm × 320 µm for prediction and used the F1-score, Precision, and Recall (the specific formula for Ref. [26]) metrics to evaluate the effect of glomeruli detection quantitatively (Fig. 5(d)). Most of the data blocks achieved 100% accuracy, with an average F1-score of 0.961, an average Precision of 0.946, and an average Recall of 0.98. We then applied this network model to detect glomeruli within the entire mouse kidney. The results are shown in Fig. 5(e) and Visualization 4. To understand the distribution of glomeruli, the glomeruli were visualized using different colors according to their distance from the kidney surface. As shown in Fig. 5(f)-(h), it is evident that glomeruli exist only in the cortex, and there are no glomeruli in the OM and IM of the kidney. Finally, we conducted a statistical analysis of the total number and distribution of glomeruli in the three datasets (Fig. 5(i)). The results showed that the total number of glomeruli in a single mouse kidney is around 15,000. Furthermore, the distribution of glomeruli exhibits a certain consistency, with the peak number of glomeruli occurring at a distance of 200-300 µm from the kidney surface, and then gradually decreasing with the increase of the distance from the surface of the kidney.
Table 1. The number of segments on each mouse renal artery
3.4 Results of nephron reconstruction
Benefiting from the high-resolution imaging of HD-fMOST, we were able to reconstruct the complete nephron structure from the original image (Visualization 5). Based on the location of the glomeruli in the cortex, we reconstructed three different types of nephrons [2]. These include the superficial nephrons (SN) located in the upper cortex and close to the surface of the kidney; the middle nephrons (MN) located in the middle of the cortex; the juxtamedullary nephrons (JN) located in the deepest part of the cortex, close to the medulla. As shown in Fig. 6(a), the position and spatial direction of each nephron in the kidney are clearly visible. It can be seen that the Helen loop (HL) of SN is shorter and only reaches the OSOM, the HL of JN is longer and can reach the IM, while the length of HL of MN is in the middle. For better visualization, we have magnified each nephron for display (Fig. 6(b)).
Fig. 6. Nephron reconstruction results and visualization. (a) Reconstruction results of three different types of nephrons, and their locations in the horizontal section of the kidney. Blue arrows indicate the bend of HL in SN, yellow arrows indicate the bend of HL in MN, and red arrows indicate the bend of HL in JN. The white box in the upper right corner shows the 3D position of these nephrons in the kidney. (b) Enlarged display of SN, MN, and JN. Green dots indicate the positions of their respective glomeruli. (c) Segmentation of the three different types of nephrons and the morphological characteristics of each segment in the 2D PI channel image. The green point in the 2D image indicates the centerline during nephron tracing. (d) Reconstruction results of nephrons connected to the same CD, as well as a 3D magnified view of the connection and the bifurcation point when tracing nephrons on the 2D PI channel image. Scale bar, (a) 1000µm; 2D images in (c) 30µm; 2D images in (d) 50µm.
During nephron tracing, we can distinguish different segments of the nephron based on the differences in cell morphology (Visualization 5). Each nephron (renal tubule) can be divided into six segments, namely PT, thin limb (TL) of HL, thick ascending limb (TAL) of HL, distal convoluted tubule (DCT), connecting tubule (CNT), and CD (Fig. 6(c)). The PT starts from the urinary pole of the glomerulus. The tubular wall is thicker and the lumen width is larger. TL has the thinnest tube wall. TAL is characterized by the thickening of the tube wall. DCT has an increase in lumen width compared to TAL. CNT has a reduced lumen width compared with DCT. The CD is characterized by a sudden increase in lumen width. In addition, we visualized the 3D morphology of each segment in different colors. It can be seen that the differences in shape among the three types of nephrons are mainly derived from the variations in PT. The PT is highly convoluted and located near the corresponding glomeruli. In SN, PT forms small and tight convolutions; in JN, it forms large and loose convolutions, occupying a larger space. The PT in MN represents a transition from SN to JN, with convolutions larger than SN but smaller than JN. The length of TL was significantly different among the three kinds of nephrons, with the JN having the longest length, followed by the MN, and SN having the shortest length. The transition position from TL to TAL also differs among the three types of nephrons, with SN and MN occurring before the HL turnback, and JN occurring after the HL turnback. DCT, CNT, and CD showed no significant difference in the three types of nephrons.
Moreover, we also reconstructed the nephrons connected to the same CD. As shown in Fig. 6(d), six nephrons connected to the CD. Although these nephrons are relatively close to each other and have similar shapes, each nephron occupies its own independent space, never mixing, and finally gathers in the same CD.
4. Discussion
In this study, we introduced a cell-resolution reconstruction pipeline for the entire kidney. The pipeline utilized TekCre::Ai47 transgenic fluorescent mice to label the kidney vasculature. The HD-fMOST was employed to acquire vascular and cytoarchitecture images simultaneously at a single-cell resolution. Afterward, the original images were used for data processing and digital reconstruction to obtain the entire kidney glomeruli and the 3D reconstruction results of renal arteries, veins, and nephrons.
There are several challenges to the 3D reconstruction of the mouse kidney at single-cell resolution. First, how to image the whole mouse kidney at single-cell resolution? The vasculature of the kidney spans the entire organ, and nephrons traverse from the cortex to the end of the IM. To reconstruct these structures in 3D, it is essential to image the entire kidney with high resolution. Traditional imaging methods cannot simultaneously capture both the whole-organ and single-cell resolution images of the kidney. However, with the HD-fMOST, we could achieve imaging of the entire kidney at a voxel resolution of 0.325 µm × 0.325 µm × 1 µm, obtaining both vascular and cytoarchitecture images of the kidney. This allowed us to present the fine structure of the kidney at the level of individual cells and capillaries, providing a foundational dataset for kidney reconstruction. Second, how to reconstruct multiple kidney structures? As far as we know, most studies can only reconstruct a certain structure in the kidney (such as nephrons) [30–32] and rarely reconstruct the complete vascular system and nephrons of the kidney at the same time. The metabolic function of the kidney involves multiple structures. Blood flows from the arteries, reaches the glomeruli for filtration through the AA, and then reabsorbs through the renal tubules, finally transfers metabolic waste into the renal pelvis through the CD. We employed various image processing methods to simultaneously reconstruct the complete arteries and veins, locate all the glomeruli in the entire kidney, and track the whole nephron morphological structure. These structures comprehensively cover the metabolic processes and pathways of the kidney, providing valuable insights into understanding the process of urine formation and identifying the occurrence of lesions.
However, our method has some limitations. Firstly, due to the limitations of optical imaging, we can only observe the rough shape of the glomeruli from the image. The finer structures, such as podocytes, macula densa, and glomerular basement membrane, remain invisible. We can only locate the position of the glomeruli and count the number of glomeruli, but we cannot precise reconstruction of the internal structures of glomeruli. Secondly, the 3D reconstruction methods of vessels and nephrons are based on semi-automatic segmentation and tracing methods, which require a huge time cost. Therefore, we were unable to reconstruct a large number of nephrons. Thirdly, because the capillarity network is very complex and dense throughout the kidney, we currently lack an effective method for its 3D reconstruction.
5. Conclusion
In this paper, we proposed a mouse kidney reconstruction pipeline from sample preparation to imaging and data processing. It not only enables detection of glomeruli throughout the whole kidney, but also can precisely reconstruct the spatial structure of the renal arteries, veins, renal pelvis, and nephrons with single-cell resolution. Our method reveals the positions, shapes and spatial relationships among various structures in the mouse kidney, providing crucial insights into renal function and physiology. In the future, our method can be applied to the 3D reconstruction of the mouse kidney in pathological environments to explore the pathogenesis of the kidney, and extended to the reconstruction of other organs, offering a powerful tool for medical diagnostics.
Funding
STI2030-Major Projects (2021ZD0201002); National Natural Science Foundation of China (82102137); National Natural Science Foundation of China (T2122015); Open Project Program of Wuhan National Laboratory for Optoelectronics (2021WNLOKF006).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. M. R. Pollak, S. E. Quaggin, M. P. Hoenig, et al., “The glomerulus: the sphere of influence,” Clin. J. Am. Soc. Nephrol. 9(8), 1461–1469 (2014). [CrossRef]
2. T. Blanc, N. Goudin, M. Zaidan, et al., “Three-dimensional architecture of nephrons in the normal and cystic kidney,” Kidney Int. 99(3), 632–645 (2021). [CrossRef]
3. A. Schedl, “Renal abnormalities and their developmental origin,” Nat. Rev. Genet. 8(10), 791–802 (2007). [CrossRef]
4. W. Kriz and H. Koepsell, “The structural organization of the mouse kidney,” Anat. Embryol. 144(2), 137–163 (1974). [CrossRef]
5. A. Greenberg and A. K. Cheung, Primer on Kidney Diseases (Elsevier Health Sciences, 2009).
6. B.-C. Liu, T.-T. Tang, L.-L. Lv, et al., “Renal tubule injury: a driving force toward chronic kidney disease,” Kidney Int. 93(3), 568–579 (2018). [CrossRef]
7. A. R. Chade, “Renal vascular structure and rarefaction,” Compr. Physiol. 3, 817–831 (2013). [CrossRef]
8. G. H. Mostbeck, T. Zontsich, and K. Turetschek, “Ultrasound of the kidney: obstruction and medical diseases,” Eur. Radiol. 11(10), 1878–1889 (2001). [CrossRef]
9. N. S. Renken and G. P. Krestin, “Magnetic resonance imaging of the kidney,” Semin. Ultrasound CT MRI 26(3), 153–161 (2005). [CrossRef]
10. S. von Stillfried, J. C. Apitzsch, J. Ehling, et al., “Contrast-enhanced CT imaging in patients with chronic kidney disease,” Angiogenesis 19(4), 525–535 (2016). [CrossRef]
11. G. E. Tasian, J. E. Pulido, R. Keren, et al., “Use of and regional variation in initial CT imaging for kidney stones,” Pediatrics 134(5), 909–915 (2014). [CrossRef]
12. V. G. Puelles, A. N. Combes, and J. F. Bertram, “Clearly imaging and quantifying the kidney in 3D,” Kidney Int. 100(4), 780–786 (2021). [CrossRef]
13. J. Xu, Y. Ma, T. Yu, et al., “Quantitative assessment of optical clearing methods in various intact mouse organs,” J. Biophotonics 12(2), e201800134 (2019). [CrossRef]
14. A. Klingberg, A. Hasenberg, I. Ludwig-Portugall, et al., “Fully automated evaluation of total glomerular number and capillary tuft size in nephritic kidneys using lightsheet microscopy,” J. Am. Soc. Nephrol. 28(2), 452–459 (2017). [CrossRef]
15. T. Saritas, V. G. Puelles, X.-T. Su, et al., “Optical clearing in the kidney reveals potassium-mediated tubule remodeling,” Cell Rep. 25(10), 2668–2675.e3 (2018). [CrossRef]
16. E. Tahaei, R. Coleman, T. Saritas, et al., “Distal convoluted tubule sexual dimorphism revealed by advanced 3D imaging,” Am J Physiol Renal Physiol. 319(5), F754–F764 (2020). [CrossRef]
17. K. R. Weiss, F. F. Voigt, D. P. Shepherd, et al., “Tutorial: practical considerations for tissue clearing and imaging,” Nat. Protoc. 16(6), 2732–2748 (2021). [CrossRef]
18. G. A. Herrera, J. Isaac, and E. A. Turbat-Herrera, “Role of electron microscopy in transplant renal pathology,” Ultrastruct. Pathol. 21(6), 481–498 (1997). [CrossRef]
19. B. Ivanyi, E. Kemeny, E. Szederkenyi, et al., “The value of electron microscopy in the diagnosis of chronic renal allograft rejection,” Mod. Pathol. 14(12), 1200–1208 (2001). [CrossRef]
20. C. R. Neal, H. Crook, E. Bell, et al., “Three-dimensional reconstruction of glomeruli by electron microscopy reveals a distinct restrictive urinary subpodocyte space,” J. Am. Soc. Nephrol. 16(5), 1223–1235 (2005). [CrossRef]
21. K. Ichimura, N. Miyazaki, S. Sadayama, et al., “Three-dimensional architecture of podocytes revealed by block-face scanning electron microscopy,” Sci. Rep. 5(1), 8993 (2015). [CrossRef]
22. Q. Zhong, A. Li, R. Jin, et al., “High-definition imaging using line-illumination modulation microscopy,” Nat. Methods 18(3), 309–315 (2021). [CrossRef]
23. D. Dumont, T. Yamaguchi, R. Conlon, et al., “tek, a novel tyrosine kinase gene located on mouse chromosome 4, is expressed in endothelial cells and their presumptive precursors,” Oncogene 7, 1471–1480 (1992).
24. Q. Zhang, A. Li, S. Chen, et al., “Multiscale reconstruction of various vessels in the intact murine liver lobe,” Commun. Biol. 5(1), 260 (2022). [CrossRef]
25. Y. Li, J. Cao, A. Li, et al., “3D visualization and detection of glomeruli in whole mouse kidney,” in Sixteenth International Conference on Photonics and Imaging in Biology and Medicine (PIBM 2023), (SPIE, 2023), 221–228.
26. Y. Li, T. Ren, J. Li, et al., “VBNet: An end-to-end 3D neural network for vessel bifurcation point detection in mesoscopic brain images,” Comput. Methods Programs Biomed. 214, 106567 (2022). [CrossRef]
27. Y. Li, H. Gong, X. Yang, et al., “TDat: an efficient platform for processing petabyte-scale whole-brain volumetric images,” Front. Neural Circuits 11, 51 (2017). [CrossRef]
28. F. Sampaio, J. Schiavini, and L. Favorito, “Proportional analysis of the kidney arterial segments,” Urol. Res. 21(6), 371–374 (1993). [CrossRef]
29. E. Daescu, D. E. Zahoi, A. Motoc, et al., “Morphological variability of the renal artery branching pattern: a brief review and an anatomical study,” Rom. J. Morphol. Embryol. 53, 287–291 (2012).
30. E. I. Christensen, B. Grann, I. B. Kristoffersen, et al., “Three-dimensional reconstruction of the rat nephron,” Am. J. Physiol. Renal Physiol. 306(6), F664–F671 (2014). [CrossRef]
31. X.-Y. Zhai, J. S. Thomsen, H. Birn, et al., “Three-dimensional reconstruction of the mouse nephron,” J. Am. Soc. Nephrol. 17(1), 77–88 (2006). [CrossRef]
32. X. Y. Zhai, H. Birn, K. B. Jensen, et al., “Digital three-dimensional reconstruction and ultrastructure of the mouse proximal tubule,” J. Am. Soc. Nephrol. 14(3), 611–619 (2003). [CrossRef]
