Unbiased single-nucleus RNA sequencing data (snRNA-seq) on cryopreserved human diabetic kidney samples were obtained from NCBI, GEO accession number GSE131882. The data were processed using the Scanpy (30) pipeline. Genes expressed in >3 nuclei and nuclei with at least 500 genes were retained. The normalization of the data was conducted using the Scran (31) package, which included assuming equal size factors, normalizing the library size to counts per million, and log-transforming the count data. The Harmony (32) integration pipeline was employed to reduce data dimensions and remove batch effects. The 3,000 most variable genes were used for the principal component analysis. The Uniform Manifold Approximation and Projection (UMAP) was applied for unsupervised clustering based on the first 50 integrated principal components. Cell annotation was completed manually based on mRNA-marker expression profiles from the original article (33). The results of this re-analysis are shown in Figure 1A . Data shown in Figure 2C are derived from a reanalyzed and previously published scRNAseq dataset (GSE195797) containing data generated on human RPCs treated with vehicle or panobinostat (34). We handled the dataset in the Scanpy environment and evaluated the expression of the genes of interest, as shown in Figure 2C .

A previously published dataset (GSE212273) containing single-cell data generated from healthy male mouse kidney (35) was reanalyzed to investigate the role of the NLRP3 inflammasome in the different kidney cell types. As the dataset (GSE212273) integrates samples from three experimental points t0, t2, and t30, we loaded the dataset using the Scanpy framework, removed the t2 and t30 samples, and clustered and annotated the data after recalculating the neighborhood graph on the latent space. After filtering, we obtained 3,766 cells with less than 40% of mitochondrial read rate and expressing more than 200 genes. The scran computeSumFactors method, which implements the deconvolution strategy for scaling normalization, was used to normalize cell-specific biases. After the addition of a pseudocount of 1, all counts were log-transformed. As a next step, we kept only “informative” genes and those were further used for dimensional reduction by principal component analysis. A neighborhood graph of observations was constructed using the first 50 principal components. Therefore, the pp.neighbors function, which estimates the connectivity of data points (Uniform Manifold Approximation and Projection (UMAP) algorithm), was used. To annotate the clusters we ran the tl.rank_gene_groups function (Wilcoxon rank-sum method) to define the marker genes of each cluster as previously shown (35). The results of this reanalysis are shown in Figure 1B .

Human renal progenitor cells (hRPC) were obtained and cultured as described in a previous study (36). The cells were used in passages 1-3 and cultured in EGM™-MV medium + 20% FCS (Hyclone™ Defined Fetal Bovine Serum, Cytiva) under sterile conditions at 37°C and 5% CO₂ (the culture medium recipe can be found in Supplementary Table 1 ). To differentiate hRPCs into podocyte-like cells, hRPCs were cultured in Endothelial Cell Growth Medium (ECGM) + FCS. At 80% to 100% cell confluence, the medium was changed to ECGM without FCS for 24 hours. Cells were then incubated using filtered 100 ml ddH₂O, 1.56 g Dulbecco’s Modified Eagle´s Medium/Nutrient Mixture F-12 Ham (DMEM-F12), 0.12 g sodium-bicarbonate, and 0.1 µM panobinostat. After 48 hours of incubation, the differentiation to podocyte-like cells was completed.

Human-derived THP-1 (DSMZ# ACC16) were cultivated in RPMI medium 1640 with GlutaMAX^TM-I (Gibco) enriched with 20% FCS and 100 µM penicillin-streptomycin. Differentiation into a macrophage-like phenotype was achieved by 72h treatment with 50 ng/mL phorbol myristate acetate (PMA).

Native RPCs, panobinostat-treated RPCs, and PMA-treated THP-1 cells were stimulated with 100 ng/mL of lipopolysaccharides (LPS, Sigma Aldrich) for 3h in their respective (differentiation) media. Subsequently, the media were aspirated, and cells were stimulated with 5 mM of Adenosintriphosphate (ATP, Sigma Aldrich). Media were harvested, and secreted IL-1β and IL-18 were detected by ELISA (#88-7261-88, #BMS267INST, Invitrogen), according to the manufacturer’s protocol.

Total RNA was extracted from hRPC cultures using the pure Link RNA Mini Kit and samples were preserved in RNAlater, according to the manufacturer’s instructions. A total of 2 µg RNA was used to transcribe cDNA by reverse transcription polymerase chain reaction (PCR) using the reagents listed in Supplementary Table 2 . For each sample, the reaction was run both in the presence and absence of reverse transcriptase in order to adjust for residual genomic DNA contamination during qPCR. Using a Mastercycler Pro (Eppendorf, Germany), first a denaturation was performed at 65°C for 10 minutes, followed by reverse transcription at 42°C for 90 minutes. For quantitative real-time PCR (qPCR) from cDNA, the reagents listed in Supplementary Table 2 and primers listed in Supplementary Table 3 were used. As we used 18s rRNA as a reference transcript for relative quantification, qPCR data for genes of interest were normalized to 18s. Controls (ddH2O) were negative for target and reference genes. Each amplification step included the following: initiation phase at 95°C, annealing phase at 60°C, and amplification phase at 72°C. This procedure was repeated for 40 cycles (37). We designed cDNA-specific primers that targeted the most CCDS-approved transcripts. All samples not exceeding background fluorescence were classified as undetectable. Melting curve profiles of every sample were analyzed to identify unspecific products and primer dimers. If necessary, agarose gels for further verification of PCR products were used.

Nphs2 Cre; Nlrp3^WT/A350V (Podo-Cre; Nlrp3^WT/A350V ) mice in the C57BL/6J background were generated by crossing Nphs2 Cre mice (B6.Cg-Tg(NPHS2-cre)295^Lbh/J (38), Jax strain ID #008205) with Nlrp3^WT/A350V mice (B6.129-Nlrp3^tm1Hhf/J (2), Jax strain ID #017969 (39), Supplementary Figure 1A ). We used littermate mice for all experiments ( Supplementary Figure 1B ). Nlrp3 KO mice were provided by J. Tschopp (University of Lausanne, Lausanne, Switzerland) and by V. Dixit (Genentech, San Francisco, California, USA). We co-housed male or female mice from 3 weeks of age onwards, in groups of 4-5 in cages under specific-pathogen-free conditions with unlimited access to standard chow (ssniff, Germany) and desalinated water. We sterilized cages, nestlets, food, and water by autoclaving before use. All experimental procedures were approved by the local government authorities according to the European equivalent of the NIH’s Guide for the Care and Use of Laboratory Animals (directive 2010/63/EU).

Paramagnetic isolation of glomeruli from both genotypes was performed as described in a previous study (40). Genomic DNA was isolated from harvested glomeruli by the addition of Direct PCR Tail Peqgold (Peqlab) and proteinase K (Qiagen), according to the manufacturer’s protocol, and incubation for 3h at 55°C and 800 rpm. Subsequent inactivation of proteinase K was achieved by heating the sample to 85°C for 45 min. End point PCR using Taq DNA Polymerase (M0273X, New England Biolabs) and primers indicated by The Jackson Laboratories for the B6.Cg-Tg(NPHS2-cre)295^Lbh/J strain verified successful excision of the floxed transgene by Nphs2-driven Cre recombinase ( Supplementary Figure 1D ).

We based group size calculation on numeric assumptions derived from our pilot studies ( Supplementary Figure 2 ). The total numbers of animals were 5 per group for the studies presented in Figure 3 and Supplementary Figure 2 (C57BL/6J wildtype, STZ/uNX treatment), and 11-13 per group for the studies presented in Figures 4 – 6 . We collected data for each treatment group from one ( Figure 3 and Supplementary Figure 2 ) or at least seven independent experiments ( Figures 4 – 6 ).

We injected female C57BL/6J mice at 8 weeks of age (Charles River, Sulzfeld, Germany) intraperitoneally with 10 mg/kg/day oridonin (Cayman/Biomol, 25665) for 14 consecutive days, starting 1 week after uninephrectomy. Control animals of the same background, age, and sex received 10% DMSO/90% PBS (vehicle) at the same volume and time points. To treat either transgenic or wildtype animals with β-hydroxybutyrate (BHB), a BHB precursor, 1,3-butanediol (Sigma Aldrich, Taufkirchen, Germany), was added to 20% mass fraction of standard chow and supplied ad libitum.

We conducted uninephrectomy in animals at 10 weeks of age as described in a previous study (41). Briefly, anesthesia included a mixture of medetomidine, midazolam, and fentanyl to achieve analgesia, amnesia, and hypnosis prior to surgery. Animals were heated during all phases of surgery as described in a previous study (42). Core body temperature maintenance between 36.5 and 38.5°C during the procedure was supervised by using online rectal temperature recording probes. Fully anesthetized animals received a left flank incision of 1 cm in length. The kidney was decapsulated and the adrenal gland was preserved before ligating the renal pedicle using non-absorbable sutures. After resection of the kidney, wounds were closed using absorbable sutures for the peritoneal and cutaneous layers. Buprenorphine pain control was started 30 min prior to the antagonization of narcosis. Dosages and treatment regimens of narcosis, antagonization, and analgesia are listed in Supplementary Table 4 . For adjustment of fluid losses, we administered 200 µL of saline into the peritoneal cavity. We handled sham-operated mice in the same manner, except for pedicle ligation and kidney resection.

Blood was collected at the indicated time points. Blood glucose was determined using an Accu Chek blood glucometer (range 11-600 mg/dL; Roche, Mannheim).

Urine was collected at predetermined time points, centrifuged at 8000g and 4°C for 6 min, and stored at -80°C until analysis. Reagents, working standards, and diluted samples (dilution depending on the respective blood glucose: < 200 mg/dL: 1:150; 200-350 mg/dL: 1:250; >350 mg/dL: 1:500) were prepared according to the manufacturer’s instructions (Mouse Albumin ELISA Quantitation Set, Bethyl Laboratories). Measurements of absorption (450 nm) were performed using a photometer (GENios Plus, Tecan Austria GmbH). Mean values were built from duplicates and the blank was subtracted. Albumin values (µg/dL) were calculated using a standard curve.

At the indicated time points, plasma was collected in the presence of heparin, centrifuged at 8000 g and 4°C for 6 min, and stored at -80°C until analysis. Reagents, working standards, and undiluted samples were prepared according to the manufacturer’s instructions (Urea FS, DiaSys Diagnostic Systems GmbH). Measurements of absorption (340 nm) were performed after 60 and 120 seconds using a photometer (GENios Plus, Tecan Austria GmbH). Values were calculated as follows: BUN=((value_60s-value_120s)-Blank) * 0.467. Blood urea nitrogen values (mg/dL) were determined using linear regression.

Urine was collected at predetermined time points, centrifuged at 8000g and 4°C for 6 min, and stored at -80°C until analysis. Reagents, working standards, and diluted samples (1:9) were prepared according to the manufacturer’s instructions (Creatinine FS, DiaSys Diagnostic Systems GmbH). Measurements of absorption (492 nm) were performed after 60 and 120 seconds using a photometer (GENios Plus, Tecan Austria GmbH). Values were calculated as follows: CREA=((value_120s-value_60s)-Blank). Creatinine values (mg/dL) were determined using linear regression. Albumin-Creatinine-Ratio (ACR) was determined by dividing the respective albumin value by the respective creatinine value.

Kidneys from T2, T6, and T12 were stored in 4% buffered formalin for 18-24 h, embedded in paraffin, and 2–4 µm sections were prepared for periodic acid–Schiff (PAS) and Picro-Sirius Red or immunostaining for WT-1, nephrin, and Nlrp3 by deparaffinization (xylene), rehydration (ethanol series), and blockade of endogenous peroxidase (3% H₂O₂ in methanol). For antigen retrieval, we used either proteinase K or microwave-heated unmasking solution at pH 6. Sections were blocked with avidin and biotin for 20 min each prior to staining. After the slides were washed in PBS, they were incubated with the primary and secondary antibodies as indicated in Supplementary Table 5 . Low-power field (40x) images were taken with a Leica DMRBE Research Microscope for each staining. Sections were analyzed as follows: PAS: Glomerular hypertrophy was determined by analyzing the size of both 10 randomly selected cortical and juxtamedullary glomeruli. Sirius Red: Both 10 cortical and juxtamedullary glomeruli were analyzed regarding glomerulosclerosis as described in a previous study (41). The positive signal was measured in pixels and normalized to the whole glomerular area section using ImageJ 1.53f52 (43). Nlrp3: Sections were stained with Nlrp3 for identification of the inflammasome complex. Nlrp3-positive cells were quantified in 10 randomly selected glomeruli. The positive signal was measured in pixels and normalized to the whole glomerular area section using ImageJ 1.53f52 (43).

To further elucidate differences between genotypes at the cellular level, we performed quantitative image analysis using a U-Net deep learning segmentation algorithm. To analyze podocyte morphometry, we used WT-1-stained slides specific for podocyte nuclei and nephrin-stained slides highlighting the podocyte cytoplasm. Glomeruli were imaged and saved in 2048 x 2048-pixel frames, at a resolution of 8.8 pixels per micron. Glomeruli (WT-1 and nephrin staining) and podocyte nuclei (only WT-1 staining) in 400 images were manually annotated to train and validate the algorithm. Subsequently, we implemented a deep learning algorithm inspired by U-Net (44). To ensure decent segmentation quality, we used data augmentation and weight map techniques (45). The U-net was then trained separately on each task (glomerular tuft or podocyte nuclei) and segmentation quality was evaluated with validation subsets that were left out during training time. Finally, segmentations were post-processed to exclude non-representative structures, such as glomeruli, that were cut at an edge or merged structures. To obtain the nephrin area, the Otsu threshold filter was used on the glomerular tuft segmentations (46). Finally, the processed segmentations were used to automatically compute morphometrics (47, 48). These included glomerular tuft area, number of podocytes, average podocyte nuclear area, area density of podocytes, ratio podocyte nuclei/glomerular tuft, podocyte cytoplasm area, and ratio of podocyte cytoplasm/glomerular tuft. Additionally, the 2-dimensional readouts were extrapolated into 3d using model-based stereology (49–51). Using this, the volume of the glomerular tuft, volume density of podocytes, and total count of podocytes in the whole glomerulus were estimated. Finally, statistical morphometry analysis was performed with R.

Optical clearing of the kidneys was employed to quantify filtration slit density. Therefore, kidneys were dissected and incubated at 4°C in a hydrogel solution (4% v/v acrylamide, 0.025% v/v bisacrylamide, 0.25% w/v VA-044 initiator, 4% PFA) for 24 hours. The gel was polymerized at 37°C for 3 h, and oxygen within the tubes was minimized by filling the hydrogel solution to the top. The samples were removed from the hydrogel solution, cleared (200 mmol/l boric acid, 4% SDS, pH 8.5), and incubated at 50°C for 24 hours. Kidneys were dissected in 300 µm thick slices (Vibratome) and incubated at 50°C for 5 days. The clearing solution was replaced every day. Then, samples were incubated in PBST (0.1% Triton-X in 1 × PBS) for 24 hours. Immunolabeling was conducted as follows: samples were incubated in primary antibody (anti-NPHS1, R&D, Cat# AF4269, RRID: AB_2154851) at 37°C for 24 h, washed in PBST at 37°C for 4 h, incubated in secondary antibody (donkey anti-sheep IgG-TRITC, Thermo Fisher Scientific, Cat# A24566, RRID: AB_2536034) at 37°C for 24 h, and washed at 37°C for 4 h. Dilutions were made using PBST in all steps. Samples were mounted in 80,2% fructose and 0.5% (v/v) 1-thioglycerol.

An SP8 STED 3X confocal microscope (Leica Microsystems) was used to take STED xyz images (i.e., z-stacks acquired along 3 directions: x, y, and z axes). The TRITC-labeled secondary antibody was exited at 555 nm with a white light laser. The respective emission was recorded from 565 to 605 nm. The 660 nm pulsed-depletion laser had a gating of 0.7 to 6 ns. Images were captured with a Leica HC PL APO CS2 100x/1.40 oil STED White objective. The images were further analyzed using Huygens Professional software version 18.04. Z stacks of the NPHS1 signal were acquired (at least 5 μm thick) to quantify the filtration slit coverage. Thus, all images of each z stack were merged. The foot process length as well as the area of the field were manually plotted and analyzed with Image J. The total length of the foot processes was divided by the total area of interest. An evaluation was conducted for five randomly selected areas of each glomerulus of at least five glomeruli per mouse.