Figure 8 shows a temporal diagram including all the image acquisitions performed during the study.Fig. 8: Temporal diagram for image acquisitions.Diagram illustrating the timeline for acquiring in vivo CT and MRI images, along with ex vivo histological specimens and their corresponding images, following PKD induction in mouse models. CT, magnet and microscope models extracted from Vecteezy.com. Kidneys and mouse models created with BioRender.com.We have complied with all relevant ethical regulations for animal use. Experimental procedures conducted during this project adhered to the European Communities Council Directive (2010/63/EU) and national rules (RD53/2013, ECC/566/2015) for care of laboratory animals. All protocols received approval from the Ethics Committees for Animal Experimentation of each center, “Comité de Ética de Experimentación Animal, CEEA” from Hospital General Universitario Gregorio Maranon and “Comité de Ética de Experimentación Animal, CEEA” from Centro de Investigacion Medica Aplicada (CIMA) Universidad de Navarra, and followed the ARRIVE guidelines (https://doi.org/10.1371/journal.pbio.3000410). All efforts were made to minimize the number of animals used, following the 3Rs principle and based in previous studies of the group.Sample sizes were limited as well by the laboratory capacity to breed the amount of animals and the survivability of the individuals. For the histology analysis, no less than 6 mice per time step were included, attempting to maintain a balanced proportion of males and females, which provides a sufficient series of specimens so that statistical measures can be extracted. For CT and MRI, no less than 4 pathological mice were imaged at each time for statistical measures. Sample numbers were lower at later stages of the disease, due to mortality. 2 additional control specimens were allocated to the CT study to provide a baseline comparison.Generation and housing of PKD2 knockout miceThroughout the project, we utilized Pax8rtTA20; TetO-Cre21; PKD2fl/fl mice to generate a conditional inducible knockout (KO) of the polycystic kidney disease 2 (PKD2) gene. These mice were obtained from the Baltimore PKD Research and Clinical Core Center. The transgenic mice possess loxP sites flanking exons 11–13 of the PKD2 gene, which are recognized by Cre recombinase under the control of a tetracycline-responsive promoter element (tetO). Administration of the tetracycline analog, doxycycline, results in the expression of the reverse tetracycline-controlled transactivator protein (rtTA), leading to Cre-mediated recombination and excision of the floxed PKD2 sequence. In this system, the Pax8 promoter directs gene expression to the proximal and distal tubules and collecting ducts of the kidney, allowing for tissue-specific control of PKD2 gene deletion.The animals were bred in the CIMA Universidad de Navarra animal facilities. All of them were housed in ventilated plastic cages of up to 6 mice per cage in a temperature-controlled room with a 12-hour light-dark cycle. They received a standard laboratory chow diet, and water was supplied ad libitum. PKD2 was deleted in the mice when they reached four weeks of age. This was achieved by providing the mice with drinking water containing 2 mg/mL doxycycline (Sigma D9891-25G), supplemented with 2.5% sucrose, over a duration of 2 weeks. Littermates were randomly assigned to doxycycline-treated and untreated groups, and they were monitored by personnel who were blinded to the treatment assignments.When the welfare of the animals was compromised, they were sacrificed by using CO2. End point was considered when a weight loss of 20% was observed.Samples for biomarker analyzes, histological imaging and survival analysisA first set of pathological mice was selected for histology imaging, survival analysis, and molecular and urinary biomarker analyzes at CIMA Universidad de Navarra. This consisted on 26 male and 26 female pathological Pax8rtTA; TetO-Cre PKD2fl/fl mice, with doxycycline administration triggering PKD2 knockout. Additionally, 8 male and 4 female littermates were used as wild-type controls.Most of the mice were euthanized at various stages of disease progression (2, 4, 8, 12, and 16 weeks) for histology imaging. Prior to euthanasia and other measurements. Mice were anesthetized with a combination of ketamine and xylazine (90:10 mg/kg body weight) administered via a single intraperitoneal injection. The kidneys were then excised, and selected sections were fixed in 4% formaldehyde for a period of 48 h. They were then embedded in paraffin, sectioned to a thickness of 3 μm, and stained with hematoxylin and eosin (H&E). The stained sections were imaged using an Aperio CS2 slide scanner (Leica Biosystems) at 20× optical magnification.For kindey to body weight ratio measurements the following specimens were used. At 2 weeks: 3 male and 3 female; at 4 weeks: 3 male and 3 female; at 8 weeks: 3 male and 3 female; at 12 weeks: 7 male and 10 female, and at 16 weeks: 3 male and 3 femaleFor cystic index calculations, the following specimens were used. 3 male and 3 female for each of the 2, 4, 8, 12 and 16 weeks points.To later assess urinary parameters, the specimens (both pathological and control) that would be later used for the 12 week point histology images were housed at that moment of time in metabolic cages for 24 h (1–3 mice per cage) for urine collection, with unrestricted access to food and water. The collected urine samples were then frozen at −80 °C until further analysis. Concurrently, whole blood was drawn at the 12-week euthanasia point via retro-orbital bleeding, followed by centrifuging to separate the serum, which was then stored in new tubes at −80 °C.The specimens used for the PKD2 vs histone expression analysis were as follows. Pathological mice: 7 male and 10 female; control mice: 8 male and 4 female.The specimens used for the urinary creatinine analyzes were the following. Pathological mice: 6 male and 6 female; control mice: 5 male and 4 female.Some of the pathological mice that were not used for histopathology collection were separated for a survival analysis (7 males and 4 females).Samples for CT and MRI ImagingA second set of mice was used for CT and MRI image acquisition at Hospital General Universitario Gregorio Marañón. This set included 11 pathological animals (5 females, 6 males) and 2 controls (1 female, 1 male). Images were acquired at 4, 6, 8, and 10 weeks after the disease induction.For both CT and MRI, the female and male mice were observed together as a single group. With these techniques we were looking to extract overall trends of our mouse population, for which this arrangement would be adequate. The number of images in some of the time points was also not enough to perform two separate analyzes.For the CT study, Iopamiro (300 mg/ml, Iopamidol, Bracco Imaging S.p.A, Italy) was administered intravenously via tail vein (0.25 ml). After 5 minutes of its distribution, CT images were acquired under sevoflurane-inhaled anesthesia using a small-animal ARGUS PET/CT scanner (SEDECAL, Spain), with the following parameters: 340 mA, 40 kV, 360 projections, 8 shots and 200 μm of resolution. Image acquisition was performed at the nephrographic phase of the contrast agent, aiming to enhance the visualization of kidney structures. CT images were reconstructed using a Feldkamp algorithm. The resolution was close to identical in all three dimensions, resulting in images of 514 x 514 x 469 voxels, with a voxel size of 0.00187 cubic mm.For MRI, animals were scanned using a 7-Tesla Biospec 70/20 scanner (Bruker, Germany) under sevoflurane inhaled anesthesia. Axial T2-weighted images of the abdomen were acquired with a Fast Spin-Echo sequence with TE = 29.73 ms, TR = 1500 ms, RARE factor = 8, averages = 6, a slice thickness of 0.5 mm (15 slices), and the matrix size was 256 × 256 pixels. The acquired MRI images were centered around the kidney area, with a lower number of coronal planes. The pixel resolutions for the X and Y dimensions were almost the same, resulting in approximately 0.01 squared mm per pixel when considering these two dimensions.Table 1 shows the number of CT and MRI images taken at different time points, both for pathological and control specimens.Table 1 Number of CT and MRI images at each time pointMolecular and urinary biomarker analysis in PKD2 knockout miceWe conducted an analysis in the first set of mice at CIMA Universidad de Navarra to compare the expression levels of PKD2 against Histone and assess the impact of our disease induction protocol. Kidney tissue samples from the 12 week euthanized specimens were processed to isolate RNA using the Maxwell® RSC simplyRNA Tissue Kit (Promega; AS1340), with the yield and purity of the extracted RNA being quantified through spectrophotometry (Nanodrop ND 1000). Complementary DNA (cDNA) was synthesized from 1 μg of total RNA in a 21 μL reaction volume using M-MLV Reverse Transcriptase (Invitrogen; 28025-013). We then performed real-time quantitative PCR (qPCR) to measure PKD2 expression, utilizing iQTM SYBR® Green Supermix (BioRad; 1708882) on a CFX96 Real-Time System (BioRad), with PKD2 primer sets referenced against Histone levels.Additionally, DNA was extracted from ear tissue samples of this set of specimens for genotyping PCR, conducted with the KAPA HotStart Mouse Genotyping Kit (Kapa Biosystems; KK7352). The primer sets designed for amplifying PKD2, rtTA, and Cre recombinase sequences are detailed in Table 2. Electrophoresis of the PCR products was performed on 2% agarose gels for PKD2 and 1.5% for rtTA and Cre, with the resulting bands visualized using the GelDoc XR Imaging System (Bio-Rad). Serum urea nitrogen levels and urinary creatinine concentrations were determined from the available samples at the 12 week time point using an autoanalyzer (Cobas C311, Roche).Histological analysis of ADPKD progression and cyst quantificationThe histology images acquired were analyzed using Aperio ImageScope software (Leica Biosystems) at 2, 4, 8, 12, and 16 weeks post-doxycycline administration. Figure 1 showcases selected histological images from these intervals, featuring detailed views to illustrate the textural changes in kidney architecture as ADPKD advances.Representative images of H&E stained longitudinal kidney sections were analyzed to determine the cystic index, which reflects the proportion of the total kidney parenchyma occupied by cysts. The quantification of the cysts was conducted using CystAnalyser software22, an automated tool designed for cyst recognition. For each time point (2, 4, 8, 12, and 16 weeks) and each sex, 3 different histology cuts were chosen. The cystic index of each cut was derived from the average measurements of four randomly chosen fields per sample, with two fields each from the cortex and medulla. In these fields, the total area occupied by cysts was measured and expressed as a percentage of the overall kidney tissue area to calculate the cystic index.CT and MRI Imaging Techniques for ADPKD Progression AnalysisCT and MRI scans were obtained in vivo from the second set of mice, with a focus on segmenting both kidneys and specifically the areas impacted by the disease. In the course of ADPKD, our CT images reveal the progression of small cysts merging into larger affected regions, disrupting normal kidney functions. This is caused by groups of nearby cysts that grow adjacent to each other, creating areas in the kidney that are formed by only cysts, separated by thin walls. These are detected as homogeneous areas in our CT analysis, which is also supported by observations in histology images of cystic walls breaking so that several nearby cysts become combined to form a larger one (Supplementary Fig. 1). Given the expansive nature of these diseased areas across the kidney, quantifying the volume of affected tissue provides a more comprehensive indicator of the impact of the disease rather than concentrating solely on individual cysts. Therefore, the analysis of CT scans yields the proportion of the volume of the kidney that is diseased, offering a measure of the extent of the effects of ADPKD on the organ.In recent years, the application of automatic methods for full kidney segmentation has experienced a significant increase, especially regarding deep learning algorithms. This stems from the aim to avoid completely manual segmentation methods, which can be very time-consuming and less reproducible compared to their automatic counterparts. For instance, the application of convolutional networks trained on 2D slices of human abdominal CT images has been proven as an accurate method to segment full PKD-affected kidneys5. Training three-dimensional networks for this task has also provided promising results. For instance,23 presents a 3D fully convolutional neural network with pyramid pooling, which applies pooling operations at different levels of the network to capture local features. Other approaches include the SE-ResNeXT U-Net (SERU)24, a combination of the U-net encoder-decoder main structure25 with SE-Net blocks26, used to recalibrate channel-wise feature responses, and ResNeXT layers27, which use aggregated residual transformations to improve feature extraction. In28, a U-net is trained on 2D CT cuts to segment the kidneys and mark the Region of Interest (ROI), after which a second U-net is applied to segment Renal Masses (RM) inside them. Similarly,29 presents the initial application of an AlexNet classification network30 to keep only the relevant slices from an abdominal CT, to be later segmented using a 2D U-net architecture. All of these have been applied on human CT scans.Regarding mouse CT processing,19 presents a comparison of U-net derived methods for kidney and cyst segmentation on very high resolution μCT images depicting ex vivo ADPKD rat kidneys. On a more comparable scale to our images,31 shows a pipeline for full organ segmentation on full body mouse CT, based again on a U-net-like network applied slice by slice. On the other hand, in32 a semiautomatic method is presented as a way to compensate for the lower resolution of mouse full body CT when compared to human CT. In general, most of the examples mentioned show the prevalent use of different variations of the U-net or similar arquitectures throughout the literature for kidney segmentation.In the case of our study, a semi-automatic approach was selected. The first step was the use of a 2D U-net for full kidney segmentation25,33, as it was deemed a reliable method for this task. More complicated architectures, as presented in the previous examples, were normally aimed at human CTs, while the U-net would suit our aim to evaluate kidney volume evolution at the available resolution. The implemented network was based on the ZeroCostDL4Mic Colab Notebooks34, and was utilized to generate preliminary kidney masks. This U-net was trained on manually segmented 2D slices from CT scans of five distinct mice, selecting only those slices featuring visible kidney regions. The training dataset comprised 624 images, with a subset reserved for validation. The model underwent training over 257 epochs, with a batch size of 4 and an initial learning rate set at 0.0003. Its performance was assessed on manually segmented 2D sections of a CT scan from another mouse, achieving an average Intersection over Union (IoU) score of approximately 0.7. The trained model was then applied to all 2D slices from new CT scans to create 3D kidney masks. The scripts for this automatic processing are available online:35 (https://github.com/BSEL-UC3M/CT-PKD-processing). After obtaining these masks, the two largest connected components (corresponding to the kidneys) were kept, and morphological opening and closing operations were performed to remove small oversegmented areas and fill holes. Since our dataset was not overly extensive, we decided to manually check all the produced masks after processing to correct any remaining mis-segmentation and ensure the masks were as accurate as possible, providing a human component to the process, as in32.Regarding the diseased area segmentation, some automatic methods have been proposed for similar tasks on human CT images, like cortex/medulla segmentation in36 or RM and tumor segmentations in23,24,28. However, to our knowledge, the segmentation of the ADPKD damaged area in mouse CT has not been widely studied. In our case, the observable effects of the disease in our CT images were depicted as low-intensity areas (cysts) growing through the organ and covering a larger area as ADPKD progressed. Therefore, the use of a U-net or similar network was deemed unnecessary. Instead, we applied a nested Otsu filter37. Within the voxels marked by these full kidney masks, two Otsu thresholding steps were implemented to delineate the damaged regions and track their progression over time. Otsu’s method finds the optimal threshold value that maximizes inter-class variance between both classes of voxels, separated by being smaller than or larger or equal than that specific threshold. First, Otsu’s method was applied to differentiate the high-intensity areas, marked by the contrast agent mainly in the renal pelvis. These high intensity areas were discarded. On the remaining voxels, a second Otsu was applied to isolate the darkest parts of the image, which corresponded to the cystic areas, damaged by the disease. This process facilitates the quantification of the evolving affected volume. The accuracy of the segmented diseased regions was further verified and adjusted manually to ensure precise measurement of disease progression. Th analysis encompassed the changes in both the overall kidney volume and the volume of the diseased areas over time in both the pathological and control mice. This approach allowed for an assessment of the rate and magnitude of ADPKD’s impact on the kidneys.Given the sparse z-slices in MRI scans, volumetric assessments were not possible through this technique. Instead, the focus was on manually identifying individual cysts, which were more discernible in MRI than in CT scans, appearing as distinct, bright, rounded structures that sometimes merged together. This enabled the counting of individual cysts at various stages, offering an alternative metric to track ADPKD progression.Kidney cyst segmentation in MRI has also been studied, mostly on human patients, by training convolutional neural networks on a set of manual annotations38,39. Previous studies, however, show the use of semiautomatic approaches to evaluate specific MRI slices40. As for CT, these methods are normally developed for human MRI scans. In our case, the limited number of slices and cysts per slice allowed us to manually mark the cysts without the need of performing training, ensuring a more accurate result that would otherwise be acquired with alternative methods.The 3DSlicer software41 was employed for generating manual segmentations and for the manual correction of 3D masks.Statistics and reproducibilityStatistical analyzes were applied to all histological data to identify significant differences, with a significance threshold set at (p < 0.05). Survival rates for male and female PKD2 KO mice were compared using Kaplan–Meier survival analysis and the log-rank (Mantel-Cox) test. Disease progression in PKD2 KO and control mice at 12 weeks was assessed with a two-factor ANOVA (phenotype x sex). When the interaction term was not significant, a main effects analysis was conducted. Significant interactions led to a simple effects analysis, with Bonferroni adjustments for pairwise comparisons. Outliers were identified using Grubb’s test, the normality of residuals was checked with the Shapiro-Wilk test, and variance equality was verified using Levene’s test. For data with unequal variances, a logarithmic transformation was applied, and tests were performed on both original and transformed data to ensure consistent results. In instances of significant outliers, analyzes were conducted with and without these data points to ascertain their impact on the results. Outliers were excluded only if their removal improved normality and had a significant effect on the statistical outcomes. For the survival analysis, n = 6 for males and n = 5 for females; for PKD2 expression levels, n = 4–10; for kidney to body weight ratio, n = 3–10; for cystic index, n = 3; for urinary creatinine, n = 4–6, and for serum urea nitrogen, n = 4–10.Total kidney volume and ratios of diseased volume over total kidney volume from CT images were assessed via a one-factor ANOVA with significance threshold (p < 0.05), with pathological n = 4–11. The evolution of the number of cysts and total cystic area in MRI images were evaluated using a Welch test due to the differences in variance across different time points, with the same significance threshold (p < 0.05) and with n = 4–7.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
