Tomographic quantification of branching morphogenesis and renal development

Branching morphogenesis is a central process in renal development, but imaging and quantifying this process beyond early organogenesis presents challenges due to growth of the kidney preventing ready imaging of the complex structures. Current analysis of renal development relies heavily on explant organ culture and visualization by confocal microscopy, as a more developmentally advanced native tissue is too thick for conventional microscopic imaging. Cultured renal primordia lack vascularization and a supportive matrix for normal growth, resulting in tissue compression and distortion of ureteric branching. To overcome this, we used optical projection tomography to image and reconstruct the branching ureter epithelium of ex vivo embryonic kidneys and developed software to quantify these three-dimensional (3D) data. Ureteric branching was assessed by measuring tree and terminal branch length, tip number, branching iterations, branch angles, and inter-tip distances in 3D space. To validate this approach for analyzing genetic influences on renal development, we assessed branching and organ morphology in Tgfβ2+/− embryos from E12.5 through E15.5. We found decreased branching, contrary to previous findings using organ culture, and quantified a primary defect in renal pelvic formation. Our approach offers many advantages from improved throughput, analysis, and observation of in vivo branching states, and has demonstrated its utility in studying the basis of renal developmental disease. Branching morphogenesis is a central process in renal development, but imaging and quantifying this process beyond early organogenesis presents challenges due to growth of the kidney preventing ready imaging of the complex structures. Current analysis of renal development relies heavily on explant organ culture and visualization by confocal microscopy, as a more developmentally advanced native tissue is too thick for conventional microscopic imaging. Cultured renal primordia lack vascularization and a supportive matrix for normal growth, resulting in tissue compression and distortion of ureteric branching. To overcome this, we used optical projection tomography to image and reconstruct the branching ureter epithelium of ex vivo embryonic kidneys and developed software to quantify these three-dimensional (3D) data. Ureteric branching was assessed by measuring tree and terminal branch length, tip number, branching iterations, branch angles, and inter-tip distances in 3D space. To validate this approach for analyzing genetic influences on renal development, we assessed branching and organ morphology in Tgfβ2+/− embryos from E12.5 through E15.5. We found decreased branching, contrary to previous findings using organ culture, and quantified a primary defect in renal pelvic formation. Our approach offers many advantages from improved throughput, analysis, and observation of in vivo branching states, and has demonstrated its utility in studying the basis of renal developmental disease. The efficient delivery of nutrients and removal of toxins is essential for life, and the networks of interconnected tubes such as the vasculature, and respiratory and renal systems, which facilitate these processes, are formed largely by branching morphogenesis. Branching defects are a key feature of a number of different congenital disorders and the kidney is particularly susceptible to errors in this process, with branching defects a feature of a number of diseases, including renal coloboma syndrome, Townes–Brock syndrome, brancho-oto-renal syndrome, and Simpson–Golabi–Behmel syndrome.1.Shah M.M. Sampogna R.V. Sakurai H. et al.Branching morphogenesis and kidney disease.Development. 2004; 131: 1449-1462Crossref PubMed Scopus (125) Google Scholar Perhaps more importantly, subtle defects in branching morphogenesis are considered to have significant impacts on adult organ function. Development of the renal ureteric tree is intimately associated with the generation of nephrons in the adult organ,2.Oliver J. Nephrons and Kidneys. A Quantitative Study of Developmental and Evolutionary Mammalian Renal Architectonics. Harper & Row, New York1968: 117Google Scholar and nephron number has a defining role in the predisposition to adult defects such as hypertension3.Brenner B.M. Garcia D.L. Anderson S. Glomeruli and blood pressure. Less of one, more the other?.Am J Hypertens. 1988; 1: 335-347Crossref PubMed Scopus (1004) Google Scholar and the progression of chronic renal disease to organ failure.4.Hughson M. Farris III, A.B. Douglas-Denton R. et al.Glomerular number and size in autopsy kidneys: the relationship to birth weight.Kidney Int. 2003; 63: 2113-2122Abstract Full Text Full Text PDF PubMed Scopus (551) Google Scholar Consequently, the ability to assess subtle defects in developing organs represents a necessary prerequisite to understanding the development of adult disease. Murine renal development begins when the ureteric bud, an epithelial outgrowth of the Wolffian duct, invades the metanephric mesenchyme at E10.5, and subsequently undergoes recursive, mainly dichotomous, branching morphogenesis. This branching of the renal epithelium is not only governed in part by signals from the mesenchyme but also by complex inductive and repressive interactions between these two tissue components. Although animal models such as the mouse provide the best opportunity to study such branching in vivo, the rapid developmental increase in organ volume makes them difficult to image in three dimensions using standard microscopy systems. In particular, their size and tissue composition means that most developing organs in the mouse fall between the technological limits of confocal microscopy and magnetic resonance imaging, thus meaning that laborious serial sectioning approaches are often required to map branching morphogenesis and renal development in a native context.5.Cebrian C. Borodo K. Charles N. et al.Morphometric index of the developing murine kidney.Dev Dyn. 2004; 231: 601-608Crossref PubMed Scopus (133) Google Scholar,6.Metzger R.J. Klein O.D. Martin G.R. et al.The branching programme of mouse lung development.Nature. 2008; 453: 745-750Crossref PubMed Scopus (501) Google Scholar Renal development has been extensively studied in vitro,7.Gupta I.R. Lapointe M. Yu O.H. Morphogenesis during mouse embryonic kidney explant culture.Kidney Int. 2003; 63: 365-376Abstract Full Text Full Text PDF PubMed Scopus (32) Google Scholar, 8.Korostylev A. Worzfeld T. Deng S. et al.A functional role for semaphorin 4D/plexin B1 interactions in epithelial branching morphogenesis during organogenesis.Development. 2008; 135: 3333-3343Crossref PubMed Scopus (32) Google Scholar, 9.Sakai T. Larsen M. Yamada K.M. Fibronectin requirement in branching morphogenesis.Nature. 2003; 423: 876-881Crossref PubMed Scopus (388) Google Scholar although significant questions remain as to whether this system accurately models organogenesis. The cultured kidney is unnaturally flattened, avascular, and grows at a comparatively slow rate at an air–liquid interface. Conversely, the organ in vivo develops very rapidly, both spatially and functionally, over a 3-day period between E12.5 and E15.5. During this process, the kidney quickly reaches a size unsuited to imaging using conventional optical microscopy techniques. Although μ-magnetic resonance imaging can be used to image mouse embryos as early as E10.5,10.Turnbull D.H. Mori S. MRI in mouse developmental biology.NMR Biomed. 2007; 20: 265-274Crossref PubMed Scopus (51) Google Scholar limitations in spatial resolution preclude the ability to reproduce fine structures such as the developing ureteric tree through to late gestation. These technological limitations leave a time window of mouse renal development which cannot be imaged adequately in three dimensions, making the assessment of the in vivo contributions of a given gene both difficult and time consuming. To overcome these limitations, we present an optical projection tomography (OPT)-based approach, which permits the examination of branching and allows the study of other spatial aspects of kidney development in the mouse. OPT is a tomographic approach which can be used to image gene or protein expression in whole mount tissue samples, and which provides an opportunity to image developing organs11.Sharpe J. Ahlgren U. Perry P. et al.Optical projection tomography as a tool for 3D microscopy and gene expression studies.Science. 2002; 296: 541-545Crossref PubMed Scopus (936) Google Scholar at sufficient resolution so as to discern and identify organogenesis events in three dimensions using fluorescence or bright-field illumination. We applied this technology to image the development of the embryonic mouse ureteric tree from early kidney (E12.5) to a time point at which nephrogenesis is well advanced (E15.5). We developed software that uses OPT scans and their derivative volumetric data to automatically segment, skeletonize, and quantitatively analyze a scanned tree to provide ureteric tree length, terminal branch lengths, tip number, and branching iterations. The skeletal data produced by our software can be combined with whole kidney scan data to separately assess several other parameters, including surface area, surface tip packing, volume, branching angle, and comparative morphology. Finally, we show the utility of this system by characterizing aberrant branching and development in a genetically modified kidney in vivo, which are contrary to results derived from renal culture. To develop an OPT-based imaging and analysis system to examine and quantify branching morphogenesis, fetal mouse kidneys (including their ureters) were dissected from the progeny of timed matings and stained with an antibody to the renal epithelial marker, cytokeratin. To stage the developmental time points in these mid-gestation embryos more accurately, we used a limb staging system.12.Wanek N. Muneoka K. Holler-Dinsmore G. et al.A staging system for mouse limb development.J Exp Zool. 1989; 249: 41-49Crossref PubMed Scopus (180) Google Scholar The fluorescently labeled kidneys were then embedded in agarose, dehydrated in methanol, cleared in benzyl alcohol/benzylbenzoate, and OPT imaged.11.Sharpe J. Ahlgren U. Perry P. et al.Optical projection tomography as a tool for 3D microscopy and gene expression studies.Science. 2002; 296: 541-545Crossref PubMed Scopus (936) Google Scholar This process involves the generation of emission images of the organ through 360° rotation and the reconstruction of this data set using a backprojection algorithm to produce a model that reproduced the native elaborating epithelial tree (Figure 1). A 'background' scan was also performed using transmitted white light or emitted autofluorescence to define the surface of the organ. Imaging of these stained kidneys showed that OPT has sufficient resolution to delineate the branching of the renal epithelium from a relatively simple structure (at E11.5–E13) (Figure 1a–c) through to an elaborate, complex three-dimensional (3D) tree comprising >500 terminal branch tips (at E15.5) (Figure 1d–f). Although these reconstructed structures are an elegant 3D representation of the developing kidney, they do not provide specific information for quantifying the branching process or for the generation of branch tips, which are associated with subsequent nephrogenesis. To address this, we developed software that uses information produced during the OPT scan (Figure 2a) as well as associated reconstructed data to process, automatically segment, and quantitatively analyze a scanned tree. The software (Figure 2b) systematically processes the data (Figure 2c), reading in the dimensions of the reconstructed data set and its real-world pixel size, meaning that throughout subsequent image processing, the size of the actual imaged object remains unchanged. Z-stack images are then loaded into a 3D grid of intensity values, and image processors manipulate the data to a point wherein a 3D skeleton can be extracted. During this process, a median filter is applied to remove noise from reconstructed OPT slices by analyzing the intensity of staining in its direct neighborhood, replacing the center voxel (a pixel in 3D space) with a median gray-scale intensity of its neighbors (Supplementary Figure 1). On completion, a threshold is performed which turns gray-scale intensity data into binary images for analysis. Either manual or automatic global thresholding is applied, in which the automated system assesses staining across the whole data set and chooses the threshold value using the method proposed by Otsu.13.Otsu N. A threshold selection method from gray-level histograms.IEEE Transactions on Systems, Man and Cybernetics. 1979; 9: 62-66Crossref Google Scholar To adapt to variations in staining of a given sample and local differences in contrast to a given tissue, we developed a 'localized thresholding' tool which applies Otsu's method in a definable spherical region surrounding each candidate voxel. A detailed technical description of the image manipulation is presented in the accompanying Supplementary Methods. Download .jpg (.04 MB) Help with files Supplementary Figure 1 Once the program has thresholded the OPT data, a solid structure is required for skeletonization and subsequent quantification to proceed. The hollow, branched ureteric tree is 'filled' in x, y, and z cross-sections and skeletonization proceeds by reducing this filled tree structure to a single-voxel wide skeleton. The skeleton is further simplified by removal of spurious, definable, microscopic branches. As the length of the ureter is variable, it is important to exclude this element when quantifying the tree structure. By measuring the distance of each terminal voxel to the nearest branch point, the program automatically classifies the longest distance as belonging to the ureter 'root' and erodes this back to the initial branch point. The finalized structure then undergoes a smoothing process adapted from the study by Palágyi et al.14.Palágyi K. Tschirren J. Sonka M. Quantitative analysis of three-dimensional tubular tree structures.in: Milan S. Fitzpatrick J.M. SPIE. 2003: 277-287Google Scholar With the processing complete, the skeleton, thresholded region, and raw sample data are overlayed and displayed in 3D to allow the user to confirm that the constructed skeleton accurately maps the branch pattern of the ureteric tree (Figure 2b; also presented as rendered cutaways in Figure 2d and e and Supplementary Video 1). The program then measures the total (summed) length of the tree, the number of branches, the number of terminal nodes, and the length of the terminal branches. As the data sets generated by the program maintain a register with the size parameters of the original scan, voxel distances can be used to estimate real-world length. To avoid overestimation of length due to voxel 'stepping', a piecewise linear approximation is used (described in Supplementary Information) to approximate the direction of the branches and to measure the length of the slope.eyJraWQiOiI4ZjUxYWNhY2IzYjhiNjNlNzFlYmIzYWFmYTU5NmZmYyIsImFsZyI6IlJTMjU2In0.eyJzdWIiOiIyZWFmZGFlNmJiNjc0NzQxZTI1ZjM2OWNhMjE3ZGRjMyIsImtpZCI6IjhmNTFhY2FjYjNiOGI2M2U3MWViYjNhYWZhNTk2ZmZjIiwiZXhwIjoxNjM0NTQyNDE3fQ.L1KNA0-5EqXsvZ1ghDVb_cZsyW4MfG55ExWeDg0qQ-nV1G0-sK3wgxCc7rxqEHefJW8w1ZUari53MsyXh6RPKD1bUBlr21nM14zOigK8rpyF3vb4lXk50CY3HwjqtWLJih7eEXUUzBr2ZpVhuszmk4Js5tatdB24472cmvwLu560HmolyUm8K2w0AGAajQN7-wVWBdeBLfQdu26dbhVvpC98u7TgW81vl35uSS4zTn8i84kxpwho1xRopYcJWQlsDXovCooTAv_DyzNS5KIGps9DDdZYLogvt25q2pyhXcumUr4vXWiAFAtKQ8PrCmBQORaY2ZkY6iQfR3gyAydTMg Download .mp4 (3.16 MB) Help with .mp4 files Supplementary Video 1 Download .doc (.05 MB) Help with doc files Supplementary Information To verify that the program provides an accurate description of ureteric branching, we manually assessed the differences between tips with and without skeletal termini superimposed in our software for 10 samples after 3 rounds of processing optimization. We also used Amira (Visage Imaging, San Diego, CA, USA) and manually curated, counted, and assessed tip numbers for six skeletons. Although this program failed to trace all limb stage 12 (∼E15.5) and some limb stage 11 (∼E14.5) kidneys, after significant manual skeletal curation and counting of tips, the estimates of tip numbers from 6 kidneys at limb stages 9–11 were within 10% of the estimates from our software. In addition, the numbers of tips estimated compared favorably with those previously reported using serial sectioning and manual counting,5.Cebrian C. Borodo K. Charles N. et al.Morphometric index of the developing murine kidney.Dev Dyn. 2004; 231: 601-608Crossref PubMed Scopus (133) Google Scholar even in our oldest samples containing >500 tips (Figures 1f and 2d). Indeed, the only limitation to the scope of our approach was the resolution capacity of the OPT scanner itself (and hence the ability to resolve neighboring ureteric components). Thus, we conclude that this method is an accurate tool for analyzing the complexity of the developing ureteric tree. Although we describe the application of this approach to kidneys, preliminary experiments indicate that it is suited to the analysis of other branching organs, such as the lung (Supplementary Figure 2, Supplementary Methods). Download .jpg (.03 MB) Help with files Supplementary Figure 2 To demonstrate the utility of our approach, and to explore its potential to uncover novel phenotypic variations, we tested its ability to assess tree length as well as branch and tip numbers in the kidneys collected from Tgfβ2+/− embryonic mice and littermate wild-type controls. Previous studies have shown differences in the branching of kidneys from these mice in explant culture in which heterozygous kidneys develop more branch tips after 48 h in culture.15.Sims-Lucas S. Caruana G. Dowling J. et al.Augmented and accelerated nephrogenesis in TGF-beta2 heterozygous mutant mice.Pediatr Res. 2008; 63: 607-612Crossref PubMed Scopus (38) Google Scholar We harvested kidneys from embryonic mice of equivalent limb developmental stages and analyzed both wild-type and heterozygous kidneys from limb stages 7 to 12 (E12.5–E15.5). The younger developmental stages assessed by our system showed no significant length or tip number difference between the wild-type and Tgfβ2+/− mice at limb stages 7–10 (approximately E12.5–E13.5) (Figure 3a and b). However, after this early period of normal growth, older Tgfβ2+/− kidneys developed with a significantly reduced number of tips than age-matched Tgfβ2+/+ kidneys (Figure 3a). In addition, the overall length of the ureteric tree was significantly reduced (Figure 3b). Frames from the tomographic scans of two representative stage 12 kidneys show a smaller Tgfβ2+/− ureteric tree with thickened branches (Figure 3c). To further understand the branching and morphological defects in our mutant kidneys, we investigated the 'space-filling' properties of the ureteric tips during the period of phenotypic divergence in Tgfβ2 kidneys, by relating tip number to organ surface area (calculated from our 'background' whole kidney scan). This showed a significant increase in tip packing at limb stage 12 (∼E15.5) in wild-type but not Tgfβ2+/− kidneys (11,135 versus 22,655 μm2 per tip, P=0.0002, Figure 4a). The ureteric skeleton provided by our analysis tool allows the subsequent assessment of many key parameters of renal development. We used filament tracing in the Imaris software suite (Bitplane A.G., Zurich, Switzerland) to further interrogate our skeletal data sets to examine the phenotypic differences apparent in Tgfβ2+/− kidneys as they develop from limb stages 10–12 (approximately E14–E15.5). We first measured the inter-tip distance from the most peripheral tip at each pole (anterior, posterior, lateral) to its five nearest neighbors (Figure 4b). In wild-type kidneys, this average 'neighborhood' inter-tip distance decreases with developmental age (143, 142, and 105 μm at limb stages 10, 11, 12, respectively), whereas in Tgfβ2+/− kidneys, the inter-tip distance is comparatively invariant (139, 142, and 145 μm, respectively). The differences were significant at limb stage 12 (∼E15.5) in the organ as a whole (P=0.001), and between each of the anterior, posterior, and lateral poles (P=0.015, P=0.039, and P=0.023, respectively) (Figure 4c). This comparative increase in inter-tip distance was also accompanied by an increase in the incidence of longer terminal branch lengths calculated automatically by our analysis tool (Figure 4d, P=2.30 × 10−5). To try to understand the link between branching in the kidney and tip density at limb stage 12, we also used these skeletal data to assess the angles of terminal branching in the organ. Although branch angles varied within a given kidney (presumably reflecting the stage of a given branching event), we observed a significant increase in the branching angle of wild-type kidneys versus Tgfβ2+/− kidneys from littermate controls (average of 126° versus 102°, P=3.14 × 10−4), indicating that the loss of a single copy of Tgfβ2 leads to both numerical and spatial differences in the pattern of renal branching during development (Figure 4e). In the course of reconstructing the OPT data set samples, we noted differences in the development of the renal pelvis, a structure known to be enlarged in a proportion of Tgfβ2−/− kidneys at birth.16Sanford L.P. Ormsby I. Gittenberger-de Groot A.C. et al.TGFbeta2 knockout mice have multiple developmental defects that are non-overlapping with other TGFbeta knockout phenotypes.Development. 1997; 124: 2659-2670Crossref PubMed Google Scholar Therefore, we used a semi-automated segmentation tool (CTan, SkyScan, Kontich, Belgium) to virtually 'extract' the renal pelvis from the volumetric cytokeratin-stained OPT data from limb stage 12 kidneys (∼E15.5, Figure 4f). The general size and morphology of the pelvis was significantly affected (Figure 4g and h) and analysis of these volumes show a roughly 10-fold decrease in the volume of the Tgfβ2+/− renal pelvis (Figure 4i), which was also evident after normalization for organ size (Figure 4j). This decrease in size was concomitant with a roughly two-fold reduction in pelvis height and width (Figure 4k and l). To date, the analysis of genetic and environmental effects on kidney development has been limited to studies of growth and patterning using two-dimensional organ culture or through laborious serial sectioning of embryonic kidneys, followed by re-assembly and then manual segmentation of the branching ureteric tree. To obviate these difficulties, we demonstrated the application of OPT as a tool to image and reconstruct the developing kidney and ureteric tree in three dimensions. To achieve this, we optimized staining of renal primordia to map out the branching epithelia, and developed software, which assesses and quantitatively measures the process of branching morphogenesis with high accuracy and reproducibility. As a result, it is now possible to analyze developmental in vivo branching defects in the kidney and other developing organs (such as lung). Data captured from the OPT scan also allow the quantitative study of the formation of different structures within the kidney. To test the robustness of our technique and its ability to delineate developmental defects in the context of aberrant embryogenesis, we re-examined a previously studied model of dysplastic kidney development, the Tgfβ2 mutant mouse. We selected Tgfβ2 as a model system because previous studies using organ culture have shown that perturbations in Tgfβ2 signaling led to defects in branching morphogenesis in the kidney. In particular, cultures of Tgfβ2+/− E12.5 renal primordia display a more elaborate branching morphogenesis (as assessed by tip number) than their wild-type littermates.15.Sims-Lucas S. Caruana G. Dowling J. et al.Augmented and accelerated nephrogenesis in TGF-beta2 heterozygous mutant mice.Pediatr Res. 2008; 63: 607-612Crossref PubMed Scopus (38) Google Scholar,16Sanford L.P. Ormsby I. Gittenberger-de Groot A.C. et al.TGFbeta2 knockout mice have multiple developmental defects that are non-overlapping with other TGFbeta knockout phenotypes.Development. 1997; 124: 2659-2670Crossref PubMed Google Scholar In our OPT-based ex vivo studies, we instead describe a marked decrease in the branching of Tgfβ2+/− kidneys, highlighting the potential of our approach to better model in vivo organ development. Although the culture of fetal kidney primordia remains one of the most widely used approaches to examine renal branching morphogenesis and represents an excellent system for assessing the effects of exogenous compounds on this process, the resulting organ is avascular and thin. This is especially problematic with regard to the in vivo spatial relationships between epithelial and mesenchymal components of the organ during its development. The volumetric skeletal and primary data sets generated by our analysis also provide the opportunity to examine and quantify other differences in renal development driven by the loss of Tgfβ2. We show that during normal development, the density of tips on the surface of the kidney increases rapidly during development from E13.5 to E15.5. This process is abrogated in the Tgfβ2+/− kidney, such that a given tip occupies a significantly greater surface area (an observation corroborated by direct tip–tip measurements). Although branching defects in Tgfβ2+/− kidneys are apparent by limb stage 12 (∼E15.5), this does not immediately correlate with defects in tip density, suggesting that these phenotypes are only partially linked. The increase in tip density at limb stage 12 also correlates with increased terminal branch angles, suggesting that this mechanism may drive the intercalation of tips from neighboring branches. Our analysis also shows primary defects in renal pelvis formation in Tgfβ2 kidneys, with the volume and shape of this structure smaller than wild-type controls. As this difference is not relative to the size of the organ, we propose a primary role for Tgfβ2 in positively driving the genesis of the pelvis. The ability of OPT (and consequently our branching software) to map epithelial morphogenesis is limited solely by the ability to spatially resolve the increasingly dense peripheral 'tip' epithelia in the kidney. Although the upper limit of this resolution in mouse kidneys is estimated to be approximately E15–16 using our commercial OPT machine, there is no theoretical reason why further developments of this technology should not be able to visualize later time points. The staining approaches developed in this study, and the results of other groups,17.Alanentalo T. Loren C.E. Larefalk A. et al.High-resolution three-dimensional imaging of islet-infiltrate interactions based on optical projection tomography assessments of the intact adult mouse pancreas.J Biomed Opt. 2008; 13: 054070Crossref Scopus (41) Google Scholar indicate that antibody penetration to the renal medulla is certainly possible at such time points. The morphogenic differences identified using our experimental approach underscore its potential to quantitatively measure key developmental milestones in kidney development and to uncover novel aspects of the genetic regulation of organ development. Our findings prompt a reassessment of some conclusions drawn from two-dimensional renal culture experiments when measuring the genetic and environmental effects on gross branch patterning. The extent to which these differences apply to different developmental mechanisms remains to be established. Visualization and quantification of organogenesis in a developmentally relevant context provides an enhanced understanding of the primary mechanisms, which underlie organ formation as well as those that shape developmental disease. This study highlights the utility of this approach and demonstrates its potential to uncover fundamental in vivo aspects of gene function not easily achieved using current methods. Tgfβ2tm1Doe heterozygous males from a 129/C57BL6/Swiss background were backcrossed to C57BL/6 females, and embryos were collected at daily intervals from ∼E12 to ∼E15.5 (the day of plug detection was considered as E0.5). Limb morphology staging was used to more accurately developmentally match mid-gestation embryos.12.Wanek N. Muneoka K. Holler-Dinsmore G. et al.A staging system for mouse limb development.J Exp Zool. 1989; 249: 41-49Crossref PubMed Scopus (180) Google Scholar The kidneys were dissected and fixed on a stage-dependent basis in 4% paraformaldehyde in phosphate-buffered saline at 4°C (ranging from 5 and 15 min) and rinsed with phosphate-buffered saline. Older kidneys with evidence of capsule formation were treated for 5 min with 50 μg/ml Proteinase K to assist antibody penetration and washed three times in 5 mmol/l PMSF (phenylmethanesulfonylfluoride) in ice-cold phosphate-buffered saline to stop digestion, followed by two room temperature washes in TBS-Tx (Tris-buffered saline-0.1% Triton X-100). Samples were blocked with 1% serum in TBS-Tx and were then incubated with a pan-cytokeratin primary antibody (Sigma C2562, St Louis, MO, USA). Subsequent TBS-Tx washes removed excess primary antibody, and samples were then incubated with an anti-mouse Alexa Fluor 555-conjugated secondary antibody (Invitrogen, A-31570, Carlsbad, CA, USA). All animal experiments were assessed and approved by the Monash University Animal Ethics Committee and were conducted under applicable Australian laws governing the care and use of animals for scientific purposes. Stained embryonic kidneys were embedded in 1% low melting point agarose, such that the tissue was completely surrounded by agarose. Blocks were then affixed to metal OPT mounts, trimmed to remove excess agarose, dehydrated in 100% methanol for 2–24 h depending on block size, cleared overnight in a 2:1 mixture of benyl alcohol/benzyl benzoate, and imaged in a Bioptonics 3001 OPT scanner (Bioptonics, Edinburgh, UK), at maximum resolution of 3.2 μm per pixel zoom. Images were acquired at 0.9° intervals, with each frame averaged over 4 images. A total of 31 wild-type and 27 heterozygote kidneys were analyzed (GraphPad Prism 5, GraphPad Software, La Jolla, CA, USA). A two-way ANOVA (analysis of variance) was used to analyze the data collected as both genotype and the six developmental stages had the potential to affect the samples collected. For tip number, it indicated that 67% of total variation came from limb stage alone, 9% from genotype alone, and 24% from an interaction between genotype and limb stage. For tree length, these measures were 68.5, 8, and 23.5%, respectively. The P-value for all these sources of variation was P<0.0001. Bonferroni's posttests indicated no significant differences between the tips or lengths of Tgfβ2+/+ and Tgfβ2+/− kidneys measured from stages 7 to 10. A significant difference was identified at stages 11 and 12 with P<0.001 for both tip number and length. A total of 58 samples were analyzed in total, and a Gaussian distribution was assumed for all samples collected from both genotypes at all stages. Sample number, listed as (Tgfβ2+/+:Tgfβ2+/−), stage 7 (4:3), stage 8 (6:4), stage 9 (6:7), stage 10 (6:6), stage 11 (5:4), stage 12 (4:3). Kidney surface area was calculated using Imaris 6.4 (Bitplane A.G.) surface rendering software and background fluorescence or visible light scans for individual organs. For tip–tip distance measurements, individual skeletal data sets produced by analysis of kidneys through our software were analyzed using the filament-tracing tool in Imaris. Branch angles were measured using the Branching angle B measurement (which measures the angle from a branch point to terminal points in the skeleton, relative to the root branch). Tip–tip distance was assessed from this same data set using the filament-tracing tool. To assess tip packing, the most apical tip from the anterior, posterior, and lateral poles was used as a basis for measurement of the next five nearest tips in 3D space (three kidneys/stage/genotype). In limb stage 12 (∼E15.5) kidneys, a '10 nearest tip' neighborhood to these apical poles was used to assess the terminal bifurcated branching angle (3 kidneys/genotype). Significant differences in angle and distance measurements were assessed by Student's t-tests on biological replicates. Spread of terminal branch length measurements was assessed using a Siegel–Tukey test. Renal pelvis volume was assessed by semi-automated segmentation from primary OPT data sets using CTan software (Skyscan), and volumes and dimensions were directly assessed by the program in limb stage 12 kidneys (n=3 Tgfβ2+/− and n=5 Tgfβ2+/+). When volumetric data were not rendered using our software, this was performed using Drishti (Australian National University, ANUSF VizLab, Canberra, ACT, Australia, http://anusf.anu.edu.au/Vizlab/drishti/). The final figures were assembled using ImageJ 1.41, Adobe Photoshop, and Adobe Illustrator (Adobe Systems, San Jose, CA, USA). The graphs in Figure 2 were prepared using GraphPad Prism 5 (GraphPad Software). All the authors declared no competing interests. We thank John Bertram for the Tgfβ2 mice, John Bertram and Luise Cullen-McEwen for helpful discussions and comments on this paper, and Malinee Sirikhant for preliminary studies on the fetal lung. We thank the staff of Monash MicroImaging and Monash e-Research for assistance in volumetric analysis. This work was supported by a National Health & Medical Research Council (NH&MRC) Project grant to IMS and Fluorescent Applications in Biotechnology and Life Sciences support grants to KMS and IMS. IMS acknowledges the support of NH&MRC R. Douglas Wright and Monash University Fellowships. Figure S1. Major processing stages used for skeletonization. Figure S2. Skeletonized E13.5 embryonic mouse lung. Video S1. Juxtaposition of thresholded and skeletonized data sets shows accurate mapping of scanned kidney data. Supplementary material is linked to the online version of the paper at http://www.nature.com/ki