FAT1 mutations cause a glomerulotubular nephropathy

Steroid-resistant nephrotic syndrome (SRNS) causes 15% of chronic kidney disease (CKD). Here we show that recessive mutations in FAT1 cause a distinct renal disease entity in four families with a combination of SRNS, tubular ectasia, haematuria and facultative neurological involvement. Loss of FAT1 results in decreased cell adhesion and migration in fibroblasts and podocytes and the decreased migration is partially reversed by a RAC1/CDC42 activator. Podocyte-specific deletion of Fat1 in mice induces abnormal glomerular filtration barrier development, leading to podocyte foot process effacement. Knockdown of Fat1 in renal tubular cells reduces migration, decreases active RAC1 and CDC42, and induces defects in lumen formation. Knockdown of fat1 in zebrafish causes pronephric cysts, which is partially rescued by RAC1/CDC42 activators, confirming a role of the two small GTPases in the pathogenesis. These findings provide new insights into the pathogenesis of SRNS and tubulopathy, linking FAT1 and RAC1/CDC42 to podocyte and tubular cell function.

N ephrotic syndrome (NS) results in proteinuria, hypoalbuminemia and oedema, and is classified by its response to steroid treatment into steroid-sensitive NS and steroid-resistant NS (SRNS). SRNS leads to chronic kidney disease (CKD) within a few years of onset, requiring renal replacement therapy for survival. It causes 15% of all end-stage kidney disease that manifests by 25 years of age 1 . SRNS is considered one of the most intractable kidney diseases, because it has a 30% risk of recurrence after renal transplant if SRNS is idiopathic. However, in SRNS with a monogenic cause the risk of recurrence is reduced from 35 to 8% (ref. 2). Histologically, SRNS manifests as 'focal segmental glomerulosclerosis' (FSGS) or as the early-onset developmental variant 'diffuse mesangial sclerosis' (DMS) 3 . The first insights into the pathogenesis of SRNS were gained by the discovery of single-gene (monogenic) causes of SRNS, revealing that the encoded proteins are essential for the function of the glomerular podocyte 4,5 . Podocytes are neuron-like cells that extend multiple tubulin-based primary processes that branch into secondary actin-based foot processes. Foot processes interdigitate with those of neighbouring podocytes, forming between them the glomerular slit diaphragm, which is critical for the filtering process. Loss of foot process and slit membrane integrity causes glomerular protein leakage and SRNS.
Advances in molecular diagnostics enables identification of the primary cause of disease in a large fraction of individuals with SRNS. In a worldwide cohort of 1,783 families, a monogenic cause of SRNS was detected in 1 of 27 genes in B30% of SRNS cases manifesting before age 25 years 6 . Gene identification also enables disease modelling in cell-based and animal models of gene knockdown or knockout, facilitates screening for small molecule therapeutics and permits an aetiologic classification of SRNS for therapeutic trials.
Although disease mechanisms of SRNS are not well-understood, altered podocyte cell-migration rate represents a relevant assay of pathogenicity for most SRNS disease-causing alleles 16,30 . Actin remodelling by members of the 'RHO family of small GTPases' RHOA, RAC1 and CDC42 (henceforth 'RHO GTPases') regulates podocyte cell-migration rate 30 .
Because genetic mapping data indicate a multitude of additional loci-bearing mutations that might cause monogenic forms of SRNS, here we perform homozygosity mapping (HM) and whole-exome sequencing to identify novel monogenic SRNS genes. We identify mutations in FAT1 that cause a human glomerulotubular disease with features of both SRNS and tubular ectasia. Knockdown of Fat1 in renal tubular epithelial cells decreases cell-cell adhesion, cell migration and activity of small RHO-like GTPases RAC1 and CDC42, but also induces defects in the formation of the tubular cell lumen. Podocyte-specific Fat1 loss-of-function recapitulates the NS phenotype in a mouse model. Furthermore, knockdown of fat1 in zebrafish causes formation of pronephric cysts, which is partially rescued by RAC1/CDC42 activators, confirming a role of the two small GTPases in the pathogenesis of this renal disease.
Identification of this new glomerulotubular disease entity demonstrates that FAT1 is necessary for glomerular as well as tubular function. These findings provide new insights into the pathogenesis and therapeutic approaches for SRNS and tubular ectasia, and link FAT1 and RAC1/CDC42 to podocyte and tubular cell function.

FAT1 mutations cause a glomerulotubular nephropathy.
We previously showed that a causal mutation in 1 of 27 different single genes can be identified in B30% of cases with SRNS that manifest before 25 years of age 31 . However, genetic mapping data strongly suggest a multitude of additional monogenic causes of SRNS 32  A4623, a Turkish boy from consanguineous parents, was diagnosed with intellectual disability, pulmonary artery stenosis and bilateral blepharoptosis in early childhood. Magnetic resonance imaging showed pachygyria and Virchow-Robin spaces. He was admitted to the hospital at the age of 15 years because of proteinuria and haematuria. Kidney biopsy showed a glomerulotubular nephropathy. Typical features of SRNS, including glomerular podocyte foot process effacement (Fig. 1a), were seen together with tubular dilation, tubulointerstitial infiltrations and irregular tubular basement membranes (Fig. 1b). We performed HM in A4623. HM yielded 13 regions of homozygosity by descent that represents candidate regions for a recessive disease gene (Fig. 1g)  One child of Arab origin (A3027) presented with proteinuria and haematuria, and renal ultrasound showed increased echogenicity ( Supplementary Fig. 1). A3027 was diagnosed with Ewing sarcoma, and lung and spinal metastasis at the age of 15 years. He rapidly progressed to renal failure and despite radiotherapy, chemotherapy and haemodialysis he died at the age of 19 years. Because of the rapid onset of renal failure, kidney biopsy was not performed but ultrasound showed echogenicity and urine analysis showed haematuria and proteinuria. There was no other affected individual in the family. HM of this individual yielded segments of homozygosity by descent with a cumulative genomic length of 215 Mb (Supplementary Table 1). Following variant filtering by HM and WES, two rare missense variants in FAT1 and EHD1 remained (Supplementary Table 1). Mutations in the 27 known SRNS genes were excluded by evaluation of the WES data. The variant (c.857A4F;p.N286S) in FAT1 is reported as a SNP in the dbSNP database, however, its minor allele frequency is 0.0002 and it never occurred in the homozygous state (Table 1 and Fig. 1h-j). The FAT1 variant alters an amino-acid residue conserved throughout evolution down to Drosophila melanogaster (Table 1).
When performing highly parallel sequencing 37 of all FAT1 exons in 1,500 additional individuals with features of NS and 800 individuals with features of tubulointerstitial nephroapathy, we detected in 2 additional families 4 different recessive (biallelic) mutations of FAT1 (Table 1, Fig. 1j). In a female African-American girl (A789) from non-consanguineous parents, another compound-heterozygous mutation was detected (c.3008C4T, p.A1003V and c.9259C4G, p. R3087G). At the age of 9 years, kidney biopsy revealed minimal change NS.
Urine analysis showed haematuria and proteinuria. After steroid resistance, cyclosporine A was started. One year after onset of NS, she was diagnosed with Morbus Hodgkin and chemotherapy was performed. Further information was not available and segregation analysis was not performed since the girl was lost for follow-up. There was no other affected individual in the two families. A3507, an African girl from non-consanguineous parents, showed haematuria and proteinuria at the age of 2 years. Kidney biopsy revealed DMS at the age of 2 years and she received unilateral    nephrectomy at the age of 2 years. In addition, a mild hydrocephalus was drained with a ventriculoperitoneal shunt.
One of the compound-heterozygous mutation (c.4517G4A, p.R1506H) was detected in her mother, but not the other. DNA from the father was not available. Three out of four mutations are reported as a SNP in the dbSNP database, but their allele frequencies are either not available or very rare (Table 1). For exclusion of known genetic causes of SRNS, 27 known genes previously linked to SRNS were screened in these individuals, but no explanatory mutations were detected 6 . Interestingly, all affected individuals exhibited a glomerulotubular nephropathy of SRNS, tubular ectasia and microscopic haematuria (Fig. 1a-f). In addition, two individuals had a central nervous system phenotype, including hydrocephalus and pachygyria (with intellectual disability and bilateral blepharoptosis) ( Table 1, Supplementary Fig. 1).
FAT1 is a member of a small family of vertebrate cadherin-like genes, designated FAT1-FAT4 in humans, whose orthologues were first recognized in Drosophila 38 . The FAT1 protein contains 33 cadherin repeats, followed by 5 epidermal growth factor (EGF)-like repeat domains, a laminin G domain, a transmembrane domain and an intracellular domain (Fig. 1i).
FAT cadherins play a role in cell migration, lamellipodia dynamics, cell polarity and cell-cell adhesion. Fat cadherins have been reported to interact with Ena/VASP proteins, atrophins, b-catenin, scribble and HOMER1-HOMER3, thereby influencing Wnt and Hippo signalling and the regulation of planar cell polarity (PCP), the process by which cells become polarized and organized within the plane of an epithelial sheet. The Fat1 À / À mouse displays abnormal podocyte foot processes, brain developmental defects and eye abnormalities 39 .  (Fig. 1i) we demonstrated absence of FAT1 protein in fibroblasts of individual A4623 (Fig. 2a), who carries a homozygous truncating mutation of FAT1 (p.P1032Cfs*11) ( Table 1, Fig. 1j).
To further characterize cellular defects in patient fibroblasts, we employed a cell-migration assay (Fig. 2b), which has been previously used to demonstrate deficiencies of cell migration rate in models of multiple monogenic SRNS genes, including 15) and FAT1 (refs 40,42). We found that fibroblasts from individual A4623 with a homozygous truncating mutation of FAT1 p.P1032Cfs*11 exhibited decreased migration rate (Fig. 2b,c). The decreased migration of fibroblasts from A4623 with FAT1 mutation was partially rescued by RAC/CDC42 activator II, indicating that RHO-like small GTPase activity may be relevant to the pathogenesis of the disease caused by loss of FAT1 (Fig. 2b,c). Similarly, knockdown of FAT1 in differentiated cultured podocytes led to decreased migration rate (Fig. 2d,e and Supplementary Fig. 2). This effect was likewise mediated by RHO GTPases, because decreased migration was partially mitigated by treatment with RAC/CDC42 activator II (Fig. 2d,e). To confirm whether knockdown of FAT1 affects RHO GTPases, we used GST-rhotekin to assay the active GTP-bound state of RHOA and used GST-PAK1 to assay the active states of RAC1 and CDC42. Knockdown of FAT1 in differentiated podocytes decreased the active state of RAC1 and CDC42 ( Supplementary Fig. 3a,c,d) but had no effect on RHOA ( Supplementary Fig. 3b,e), indicating RAC1 and CDC42 are more relevant to the pathogenesis of SRNS caused by loss of FAT1. In addition, overexpression of dominant negative CDC42 (T17N) decreased cell migration in control differentiated cultured podocytes, but failed to further decrease migration in podocytes with FAT1 knockdown, suggesting that CDC42 is a downstream mediator of FAT1-mediated cell migration ( Supplementary Fig. 4).
We then tested whether there was decreased cell-cell adhesion, which was recognized previously as a feature of FAT1 loss-offunction 43 . We found that fibroblasts from individual A4623 with the homozygous truncating mutation exhibited decreased cell-cell adhesion (Fig. 2f). Moreover, specific knockdown of FAT1 in differentiated cultured podocytes also impaired cell-cell adhesion, confirming that this effect is due to loss of FAT1 expression (Fig. 2g).
Podocyte-specific Fat1 loss-of-function leads to FSGS. In addition to in vitro cell-based studies, we examined the effect of  C.e. C.e.
- Fat1 loss-of-function on glomerular integrity in vivo. Loss of Fat1 in a constitutive Fat1 À / À mouse model was previously described to cause perinatal death, features of NS with podocyte foot process effacement and brain developmental defects 39 . Thus, we generated a mouse model of podocyte-specific loss of Fat1 function, in which loss of Fat1 is driven by the podocyte NPHS2/podocin promoter (Pod-Cre) ( Supplementary Fig. 5a,b) 44 .
To examine mechanisms of tubular defects of renal tubule architecture, we performed transient Fat1 knockdown in mouse inner medullary collecting duct (IMCD3) cells, a renal tubule cell line that has broadly been used to study tubular defects 35,45 . We found that knockdown of Fat1 perturbed lumen formation of IMCD3 spheroids (Fig. 4a,b), suggesting that FAT1 is involved in the regulation of apicobasal polarity. However, this lumen formation did not result from ciliogenesis because cells with FAT1 knockdown did not have a problem in ciliation (Fig. 4a). Similarly, fibroblasts from A4623 with a homozygous truncating mutation did not show any defect in number or structure of cilia compared with control fibroblasts (Supplementary Fig. 6), suggesting that FAT1 is not directly involved in ciliogenesis. Because fibroblasts of individuals with FAT1 mutation had cell migration defects (Fig. 2b), and because Fat1 also regulates cell migration in many different cell types including podocytes, we examined whether Fat1 also regulates cell migration in IMCD3 renal tubular cells. We found that knockdown of Fat1 using two different short hairpin RNAs (shRNAs) did in fact decrease cell migration rate in IMCD3 cells (Fig. 4b,c and Supplementary Fig. 7a).
Apicobasal cell polarity is controlled by the PCP pathway and FAT family proteins are involved in this pathway 46 . FAT family proteins are upstream PCP proteins, Scribble is one of the PCP core proteins and RHO GTPases are downstream PCP effector proteins 47 . Therefore, we hypothesized that the observed defects in both lumen formation and cell migration may be mediated by disruption of RHO/RAC/CDC42 signalling. When measuring active RHOA by a GST-Rhotekin pull-down assay in IMCD3 cells, we found that on knockdown of Fat1active GTP-bound RHOA was unaltered ( Fig. 4e and Supplementary Fig. 7b). In contrast, when measuring active CDC42 and RAC1 by GST-PAK1 pull-down assay, we found that on knockdown of Fat1 active CDC42 and RAC1 were both decreased ( Fig. 4f and Supplementary Fig. 7c,d). These results in IMCD3 cells are congruent with those in cultured podocytes ( Supplementary  Fig. 3) and indicate that CDC42 and RAC1 are more relevant than RHOA for the pathogenesis of the glomerulotubular disease that is caused by loss of FAT1. ARTICLE We then studied in vivo the effects on renal tubule cell architecture that we had observed in cell-based systems in vitro. We chose zebrafish as a model organism, because zebrafish embryo is amenable to genetic manipulation such as injection of morpholino (MO) and knockdown of fat1 in zebrafish is known to cause pronephric cysts 47 . We performed MO knockdown of fat1 in Wt1b::GFP transgenic zebrafish, in which pronephric kidney cysts can readily be visualized. We found that, whereas zebrafish injected with control MO did not produce any phenotype (Fig. 4e), depletion of fat1 caused pronephric cysts in 24% of zebrafish embryos (Fig. 4f). To further corroborate the pathogenic role of RAC1/CDC42 signalling, we tested the effect of two different RHO/RAC/CDC42 activators on rescue of the zebrafish pronephric renal cystic phenotype. We found that the chemically distinct Rho GTPase activators I and II reduced cyst formation from 31% to 9.7% and 12.0%, respectively (Fig. 4i).

Discussion
We here identify recessive mutations of FAT1 as causing a glomerulotubular nephropathy. Four families with FAT1 mutations presented with a combination of SRNS, tubular ectasia, haematuria and facultative neurological involvement. Mechanistically, our studies implicate RHO-like small GTPase signalling in the pathogenesis of both, the glomerular and the tubular phenotypes.
In a global Fat1 knockout mice, Fat1 null pups died within 48 h of birth, which was attributed to abnormal podocyte foot process fusion with obliteration of the slit membranes 39 . These findings are consistent with the concept that FAT1 provides spacing between cells in the kidney and plays a role in the formation of intercellular junctions in the kidney 60 . While our model of podocyte-specific Fat1 deletion caused identical podocyte foot process fusion at birth, we did not observe perinatal lethality, suggesting that perinatal lethality with global Fat1 loss reflects a different developmental problem.
Fat1 null mice also exhibited partially penetrant midline developmental defects including holoprosencephaly, as expected from strong Fat1 expression within the neuroepithelium. 39 This mouse model is consistent with our finding of a combined renal and neurologic disease phenotype in human recessive FAT1 mutation. Our observation of incomplete penetrance of the neuronal involvement in individuals with FAT1 mutation parallels the findings in the mouse model and is most likely due to compensation by the other FAT proteins (FAT2, 3 or 4) 38 , and to the fact that all of the individuals with two recessive FAT1 mutations (except A4623) carried at least one hypomorphic allele, that is, an allele that is expected not to convey full loss-of-function (Table 1). It is also consistent with the finding that individual A4623, who carries the homozygous truncating mutation, has a disease phenotype with the most severe neuronal involvement ( Table 1).
Cooperation of different members of FAT cadherins was demonstrated in mouse models of combined loss-of-function of Fat4 and Fat1, where FAT proteins cooperate in regulating multiple aspects of tissue morphogenesis in the kidney and other organs. Fat function affected renal tubular elongation, neural tube formation and cochlear morphology via its effects on mechanisms of PCP. In this context, removal of one copy of the murine Fat1 gene exacerbated the renal tubular cyst phenotype of mice 48 .
Taken together, our cell-based and in vivo studies using mouse and zebrafish suggest that the glomerular SRNS-like disease phenotype results from impaired cell migration of glomerular podocytes. Indeed, defective podocyte migration is a disease mechanism common to multiple different monogenic forms of SRNS 30 . Likewise, defective cell migration was recently recognized as a central mechanism of renal tubular regeneration 49 . Interestingly, Fat1 knockdown limited migration of renal tubule cells and also impaired their ability to form lumens in cultured spheroids, suggesting that also the renal tubular defect may be mediated by impaired cell migration.
We further demonstrated that RAC1 and CDC42 activity are impaired in both cultured podocytes and tubular cells with decreased Fat1 expression, and that activation of these small GTPases partially corrects defective cellular migration that occur with Fat1 deficiency. This suggests that the pathogenesis of both the renal glomerular and tubular defects is mediated by decreased RAC1/CDC42 as a potential unifying mechanism underlying both the glomerular and tubular disease phenotypes. We confirm the relevance of this disease mechanism by showing that two different RAC1/CDC42 activators mitigate the pronephric cyst phenotype in zebrafish.
Data from human genetics and animal models have shown the importance of RHO GTPase signalling in CKDs 15,16,[50][51][52] . ARHGAP24 or ARHGDIA mutations identified in human SRNS cause upregulation of both RAC1 and CDC42 and lead to increased podocyte migration 15,16 . This is congruent with the finding that active RAC1 (but not RHOA) is upregulated in Arhgdia À / À mice. However, CDC42 activity has not been examined in the Arhgdia À / À mouse model 53 . Scott et al. 50 showed that podocyte-specific deletion of CDC42 (but not RHOA and RAC1) in mice results in foot process effacement and proteinuria. INF2 mutations also decrease active CDC42 at the plasma membrane and cause mislocalization of CDC42 within the cytoplasm 52 . Therefore, either aberrant activation or inactivation of CDC42 interferes with podocyte function. In addition, cdc42 knockdown in zebrafish led to hydrocephalus, body oedema and pronephric cysts, and kidney tubule-specific deletion of CDC42 resulted in renal cysts and renal failure within weeks of birth 51 , indicating that CDC42 is also important for proper tubular function. In this study, we show that the impaired podocyte migration and the pronephric cysts in fat1-knockdown zebrafish were partially rescued by a RAC/CDC42 activator, suggesting that the combined glomerulotubular phenotype resulting from FAT1 mutations is mediated by decreased RAC1 and CDC42 in both glomeruli and tubules.
Recently Morris et al. 43 identified FAT1 mutations in human cancers including glioblastoma and colorectal cancer, and showed that loss of FAT1 leads to aberrant activation of Wnt signalling. The FAT1 mutations identified by Morris et al. 43 were somatic and mostly heterozygous. However, in the glomerulotubular disease described here renal disease resulting from FAT1 mutations is recessive, requiring two mutant alleles, which are inherited from the heterozygous parents. It is not clear whether individuals with recessive FAT1 mutations will develop cancer later in life. However, it is important to note that two individuals (A3027 and A789) had Ewing sarcoma and Hodgkin's lymphoma, respectively (Table 1). How the respective FAT1 mutations identified in these individuals might contribute to neoplastic disease will require further focused study.
FAT1 is implicated in Hippo signalling through the interaction with Scribble and Hippo signalling is required for normal pronephros development in zebrafish 47 . Yes-associated protein (Yap) is a downstream transcriptional effector of Hippo signalling and is essential for nephron induction and morphogenesis 54 . Recently, it was shown that Cdc42 acts upstream of Yap during development to promote Yap-dependent gene expression and shape functioning nephrons 54 . Therefore, it is also conceivable that the glomerulotubular nephropathy caused by loss of FAT1 may result from defective Hippo signalling due to decreased Cdc42 activity, which we have shown in this study ( Fig. 4f and Supplementary Fig. 3a).
Because SRNS cause B15% of all CKD in the first two decades of life 6 , the balance of RAC1/CDC42 signalling may be a worthwhile target when developing drugs to prevent ESKD that is caused by SRNS.

Methods
Study participants. Following informed consent for WES, we obtained clinical data and blood samples from individuals with SRNS or NPHP from worldwide sources. Approval for human subjects' research was obtained from the University of Michigan and Boston Children's Hospital Institutional Review Boards and from relevant local Review Boards. Deposition of WES data in public databases was not obtained at the time. The diagnosis of NS or NPHP was made by (paediatric) nephrologists based on standardized clinical and renal histologic criteria 55 . Renal biopsies were evaluated by renal pathologists. Clinical data were obtained using a standardized questionnaire (http://www.renalgenes.org).
Whole exome sequencing. We combined WES with HM as established previously 32,36 . For HM 32 the 'Human Mapping 250 k StyI' array (Affymetrix) was utilized. Genomic DNA samples were hybridized, and scanned using the manufacturer's standard protocol at the University of Michigan Core Facility (www.michiganmicroarray.com). Non-parametric logarithm of odds (LOD) scores were calculated using a modified version of the program GENEHUNTER 2.1 (refs 56,57) through stepwise use of a sliding window with sets of 110 SNPs using the program ALLEGRO 58 . Genetic regions of homozygosity by descent ('homozygosity peaks') were plotted across the genome as candidate regions for recessive diseasecausing genes as described 32,35 . Disease allele frequency was set at 0.0001, and Caucasian marker allele frequencies were used. Variant burden analysis was performed as previously described 59 using Agilent SureSelect human exome capture arrays (Life technologies) with next-generation sequencing on an Illumina sequencing platform. Sequence reads were mapped against the human reference genome (NCBI build 37/hg19) using CLC Genomics Workbench (version 6.5.1) software (CLC bio, Aarhus, Denmark). Mutation calling (Supplementary Table 1) was performed by geneticists/cell biologists, who have knowledge of clinical phenotypes, pedigree structure, HM and WES evaluation. All the coding exons of FAT1 were examined by applying a high-throughput mutation analysis method of microfluidic array-based muyltiplex PCR (Fluidigm 48.48 Access Array) and consecutive barcoded next-generation sequencing (MiSeq, Illumina) 37 .
Zebrafish maintenance and injections. Approval for zebrafish research was obtained from the University Committee on the Use and Care of Animals (UCUCA) of the University of Freiburg. Transgenic wt1b::GFP zebrafish line was raised and maintained as described 60 . Fertilized eggs were microinjected with 4 nl of injection solution at the one-to-two-cell stage with MOs (Gene Tools LLC) diluted in 200 mM KCl, 0.1% Phenol Red (Sigma-Aldrich Corporation) and 10 mM HEPES. zFat1 MO has been previously described 47 . About 0.5 pmol of zebrafish p53 MO (5 0 -GCGCCATTGCTTTGCAAGAATTG-3 0 , Gene Tools) was always co-injected to reduce the unspecific effects of the MOs 61 . About 1.6 pmol zFat1 MO was injected either alone or together with either 1 mg ml À 1 of Rho/Rac/Cdc42 Activator I or 0.5 U ml À 1 of Rac/Cdc42 activator II (Cytoskeleton). Scoring of percentage of cyst formation and zebrafish embryo imaging was done on mixed male and female embryos at 48 h post fertilization under a Leica MZ16 stereo microscope (Leica).
Podocyte conditional Fat1-deficient mice. To develop the Fat1 conditional allele in mice, homologous recombination was used to introduce LoxP sites flanking exon 2. The targeting construct for the Fat1 locus was generated from a C57/BL6 strain bacterial artificial chromosome (BAC) clone (Children's Hospital Oakland Research Institute) by homologous recombination BAC engineering techniques (Open Biosystems). All junctions and overall structure were confirmed by sequencing and long-range PCR. LoxP sites were introduced on each side of exon 2, which encodes the signal peptide and N-terminal 1,089 aa of Fat1. The neomycin selection cassette was flanked by Frt sequences for removal by Flp recombinase.
Electroporation of the linearized construct into WW6 embryonic stem (ES) cells, followed by selection with G418 and ganciclovir, yielded a high percentage (B60%) of homologous recombinants. A Fat1 þ /lox ES clone injected into C57/BL6 blastocysts yielded three highly chimeric mice that transmitted the targeted allele to offspring. The Frt-flanked neo cassette was removed in a cross with a FLPeR mouse (129S4/SvJaeSorGt (ROSA) 26Sor tm1(FLP1)Dym /J, Jackson Labs). Heterozygous mice were interbred to generate homozygous conditionally targeted Fat1 mice (Fat1 loxP/loxP ) in a mixed 129SV/BL6 genetic background. To delete fat1 selectively in podocytes, female Fat1 þ /fl mice were bred with male podocin-cre mice 44 and double heterozygous were backcrossed to obtain pod-cre:Fat1 fl/fl . Podocin-cre mice were provided by Dr Holzman. All animals were housed in pathogen-free conditions, and protocols approved by the standing committees of the Albert Einstein Institute for Animal Studies and the Yale Animal Resources Center. Urinary albumin was measured using mouse albumin ELISA (Bethyl Laboratories) and SDS-PAGE/Coomassie Blue staining as previously described 62 . Plasma and 24 h urine creatinine were measured by HPLC 62 . Renal morphology was examined in newborn (1-3-day-old) mice and in 4-8-month-old mice by light microscopy (haematoxylin and eosin-and periodic acid-Schiff-stained sections) and by TEM using standard techniques 62 .
Cell culture and transfection. The immortalized human podocytes 63 were kindly provided by Dr Saleem (the University of Bristol) and maintained at the permissive temperature of 33°C in RPMI þ GlutaMAX-I (Gibco) supplemented with 10% fetal bovine serum, penicillin (50 IU ml À 1 )/streptomycin (50 mg ml À 1 ) and insulin-transferrin-selenium-X. To differentiate podocytes, they were cultured at 37°C for 14 days. The FAT1-specific and control scrambled short interfering RNAs (siRNAs) were purchased from Dharmacon. siRNAs were transfected into podocytes using Lipofectamine RNAiMAX (Invitrogen). Mouse kidney collecting duct IMCD3 cells were obtained from the ATCC (CRL-2123) and cultured in DMEM/F12 with 10% fetal bovine serum and penicillin (50 IU ml À 1 )/ streptomycin (50 mg ml À 1 ). To achieve stable knockdown of FAT1, IMCD3 cells were transducted with lentivirus, which contained shRNAs. The target sequences of siRNAs and shRNAs used in this study are in Supplementary Table 2. The shRNAstable IMCD3 cells were selected and maintained with 8 mg ml À 1 puromycin. Human fibroblasts were grown in DMEM supplemented with 15% FBS, penicillin (50 IU ml À 1 )/streptomycin (50 mg ml À 1 ) and non-essential amino acids (Invitrogen). Rho/Rac/Cdc42 activator I and Rac/Cdc42 activator II were purchased from the Cytoskeleton, Inc.
Migration assay. Real-time migration assay was performed using the xCELLigence system (ACEA Biosciences) in CIM-plate 16 according to the manufacturer's instruction. Briefly, 4 Â 10 4 cells of fibroblasts, cultured human podocytes or IMCD3 were plated in serum-free media in the upper chamber. The lower chambers were with 10% FBS for chemoattraction or with serum-free media. The obtained data were analysed using the RTCA software. Results are presented as the time versus cell index curve.
Spheroid assay. IMCD3 cells were trypsinized and resuspended cells were then mixed 1:1 with growth factor-depleted matrigel (BD Bioscience). This mixture was then moved to the well of a Nunc Lab-Tek II chambered coverglass (Thermo Fisher) and allowed to polymerize (45 min to 1 h) at 37°C. Once the matrigel was fully solidified, warm medium was added drop-wise to the matrix until covered. After 72 h, the cells had formed spheroids with visible cleared lumens. The matrigels were then fixed in fresh 4% PFA for 30 min at room temperature and subsequently permeabilized for 15 min in gelatin-dissolving PBS and 0.5% Triton X-100. The primary antibodies (Anti-acetylated-a-tubulin 1:2,000 and b-catenin, 1:500) were incubated at 4°C overnight. Fluorescent images were obtained with a Leica SP5X laser scanning microscope.
Fluorescence-based cell-cell adhesion assay. This assay was performed as described previously 43 . Fibroblasts or cultured human podocytes were resuspended at 1 Â 10 6 cells per ml in serum-free medium supplemented with 5 mM calcein AM and incubated at 37°C for 30 min. Cells were washed twice with serum-free medium, and 1 Â 10 5 cells were added to microplate wells containing confluent (unlabelled) cells. Calcein-labelled cells were allowed to adhere for 45 min at 37°C. Non-adherent calcein-labelled cells were washed away with medium, and PBS was added to each well. Fluorescence was measured at an absorbance of 494 nm and emission of 517 nm using a SpectraMax Multilabel Microplate Reader (Molecular Devices).