Glis3 Is a Modifier of Cyst Progression in Autosomal Dominant Polycystic Kidney Disease

Visual Abstract

Abstract

Key Points

• Dual inactivation of Glis3 and Pkd1 exacerbated polycystic kidney disease compared with Pkd1 inactivation alone in mouse models of autosomal dominant polycystic kidney disease.

• RNA-Seq and ATAC-Seq suggested Glis3 inactivation resulted in dysregulated fatty acid metabolism and alteration of circadian regulation.

• Glis3 was involved in a transcriptional network consisting of the transcription factors HNF1 homeobox B, hepatic nuclear factor 4, alpha, and D site albumin promoter binding protein.

Background

Autosomal dominant polycystic kidney disease is caused by mutations affecting polycystin-1 or polycystin-2. The existence of a cilia-dependent cyst activation pathway has been identified by showing that structurally intact primary cilia are crucial for rapid cyst growth following loss of polycystins. We previously used translating ribosome affinity purification RNA-Seq on precystic mouse kidneys to determine a translatome that meets the criteria for cilia-dependent cyst activation. From this, we identified Glis2 as an early effector of polycystin signaling and a potential target for therapy. Here, we investigate the role of Glis3 which, while not transcriptionally altered in autosomal dominant polycystic kidney disease models, encodes a cilia-localized transcription factor belonging to the same gene family as Glis2.

Methods

We used live cell imaging along with gene and protein expression studies to determine the relationships between Glis3, Glis2, and polycystin-1 expression. We used Glis3 conditional knockout mice to investigate the in vivo genetic interaction between Glis3 and Pkd1. We used gene expression and chromatin accessibility analyses by RNA-Seq and ATAC-Seq, respectively, on an allelic series of Glis3 and Pkd1 inactivation models to explore the genetic relationships between the two genes.

Results

The ciliary localization of Glis3 was not affected by Pkd1 mutation status. Kidney selective inactivation of Glis3 by itself did not affect kidney structure or function, but dual inactivation of Glis3 and Pkd1 significantly worsened polycystic kidney disease. Integration of transcriptomic profiling and chromatin accessibility assays suggested that kidney tubule-specific Glis3 inactivation resulted in dysregulated fatty acid metabolism and alteration of circadian regulation.

Conclusions

Glis3 is a primary cilium localized transcription factor that genetically interacts with Pkd1 and modifies kidney epithelial cell metabolism and circadian function.

Introduction

Autosomal dominant polycystic kidney disease (ADPKD) is one of the most common human monogenic diseases, affecting more than 12 million people worldwide. Most cases of ADPKD result from mutations in either PKD1^1,2 or PKD2,³ encoding the proteins polycystin-1 and polycystin-2, respectively. ADPKD is mainly characterized by cysts originating from the epithelia of kidney tubules and is varyingly associated with extrarenal manifestations including bile duct cysts and intracranial aneurysms.⁴ The precise functions of polycystins and the precise molecular mechanisms underlying ADPKD remain incompletely understood. A number of known pathways have been implicated in ADPKD, including canonical Wnt signaling, mammalian target of rapamycin, cyclic AMP–dependent signaling, calcium signaling, G-protein signaling, and disordered cellular metabolism.^5–7 There is general consensus that polycystins expressed on primary cilia are an essential component of the disparate molecular signaling events associated with cyst formation. Genetic studies in mice have shown that structurally intact cilia are a critical requirement for relatively rapid cyst growth after the loss of polycystins.⁸ This has led to the inference of a cilia-dependent cyst activation (CDCA) pathway⁸ that acts as an effective gain-of-function signal downstream of polycystin inactivation when intact cilia are present. We previously performed translating ribosome affinity purification (TRAP) RNA sequencing on precystic mouse kidneys to define a translatome that is associated with CDCA pattern of expression. From this, we identified Glis2 as an early effector of polycystin signaling.⁹

Glis2 belongs to the Gli-similar Krüppel-like zinc finger transcription factor family,¹⁰ which has two other members: Glis1¹¹ and Glis3.¹² Glis1-3 share structural homology in the region of their five tandem Cys2-His2 zinc finger motifs with the Gli transcription factors in Hedgehog signaling, while the rest of protein sequences are divergent.^10–12 Glis3 is abundantly expressed in the kidney, while Glis1 is expressed at a relative low level.¹³ Glis3 has been localized in both the nucleus¹⁴ and the primary cilia.¹⁵ Recessive mutations in Glis3 are associated with a rare syndrome of neonatal diabetes and congenital hypothyroidism,¹⁶ in which some patients develop cystic kidneys.¹⁷ Two Glis3 null mice, in which either the fifth zinc finger motif¹⁵ or all five zinc finger motifs¹⁸ are deleted, were generated independently. In both, mice with homozygous Glis3 inactivation are born in correct Mendelian ratios, but die shortly after birth because of pancreatic insufficiency and neonatal diabetes; they also exhibit cyst formation in renal tubules and the glomerulus.^15,18 Glis3 function has also been linked to metabolism. During early postnatal kidney development, dysregulation of Glis3 results in altered mitochondrial function and reprogramming of glucose and lipid metabolism.¹⁹ While Glis3 is not transcriptionally altered in our precystic in vivo Pkd1 ADPKD model (⁹; https://pkdgenesandmetabolism.org/expression/Glis3), Glis3 is downregulated by in vitro depletion of polycystin-2 in immortalized cultured cells and in mouse kidneys with advanced polycystic kidney disease.²⁰ Given parallels drawn from the coordinated functioning of Gli1–3 in Hedgehog signaling, the importance of Glis2 as a CDCA effector in ADPKD and the cilia localization of Glis3, we sought to determine whether Glis3 is part of ADPKD signaling pathways.

Methods

Mouse Strains and Procedures

All animal procedures followed Yale Institutional Animal Care and Use Committee guidelines. Mice (>95% C57BL/6J) of both sexes were used. Pkd1^fl,²¹ Pax8^rtTA (JAX # 007176), and TetO^Cre (JAX # 006234) strains were previously described. Glis3^fl mice were generated by flanking exon 6 with loxP sites using CRISPR/Cas9. Successful editing and germline transmission were confirmed by sequencing of PCR products (Supplemental Figure 1). Conditional gene inactivation was induced with 2 mg/ml of doxycycline in 3% sucrose water given to nursing dams (P0–P14) or adult mice (P28–P42) as described previously.⁹ Copy number of Pax8^rtTA and TetO^Cre was verified as reported.⁹ Adult mice were euthanized at 7, 14, 18, and 24 weeks. Blood was collected by ventricular puncture, and serum urea nitrogen was measured by the Yale Animal Physiology Core. One kidney was snap-frozen for protein and RNA, and the other was fixed in 4% paraformaldehyde for histology. Cystic index was calculated as described.²¹

Primary Cell Culture

Primary kidney cells were isolated and cultured as described.⁹ After 48–72 hours attachment, cells were treated with 1 µg/ml doxycycline (Sigma, Cat. no. D9891-100G) for 72  hours, then serum-starved in 0.1% FBS media for 24 hours to induce cilia before protein or RNA preparation.

cDNA Constructs, Transfection, and Cell Culture

Mouse Glis3 (NM_001404124.1) was cloned into the pLenti-CMV-GFP-Blast (659-1) vector (Addgene, Cat. no. 17445). Wild-type (WT) and Pkd1^KO IMCD3 cells stably expressing the Nphp3^(1–200)-mApple cilia marker⁹ were electroporated with the Glis3 construct and selected with G418 (500 µg/ml, Sigma-Aldrich, Cat. no. G8168-100ML). Cells were cultured in 5% FBS DMEM (Thermo Fisher Scientific, Cat. no. 11965092) and serum-starved in 0.1% FBS media for 24 hours before experiments.

Immunocytochemistry

Live WT and Pkd1^KO IMCD3 cells were imaged without fixation. Kidney cryosections (5–7 µm) were processed for immunofluorescence using standard methods.⁹ Images were acquired on Nikon Eclipse Ti (Nikon Instruments Inc, Japan) equipped with Yokogawa CSU-W1 spinning disc and Andor solid state lasers (Andor Technology, United Kingdom), using the NIS-Elements AR software (Nikon, Version 4.30.02). Antibodies and lectins are listed in the Supplemental Methods.

Protein Preparation, Immunoblotting, and Immunoprecipitation

Cytosolic and nuclear fractions were prepared using the NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo, Cat. no. 78833). Protein concentrations were measured with Protein Assay Dye Reagent Concentrate (Bio-Rad, Cat. no. 5000006), and equal amounts were separated on 4%–20% Mini-Protean TGX Precast Gels (Bio-Rad, Cat. no. 4568094) and transferred to Nitrocellulose membrane (Bio-Rad, Cat. no. 1620115). Membranes were stripped using Restore PLUS Western Blot Stripping Buffer (Thermo Scientific, Cat. no. 46430). Images were acquired with LI-COR Odyssey Fc Imaging system. Antibodies are listed in the Supplemental Methods.

RNA Isolation, RT-qPCR, and RNA-Seq Analysis

Total RNA from cells or kidneys was isolated using the Trizol (Themo Fisher Scientific, Cat. no. 15596026) and RNeasy Mini Kit (Qiagen, Cat. no. 74104) and used for cDNA synthesis with the iScript cDNA synthesis kit (Bio-Red, Cat. no. 1708890) or used for RNA sequencing. RT-qPCR was performed using iTaq Universal SYBR green Supermix (Bio-Rad, Cat. no. 1725121) in CFX96 Touch Real-Time PCR detection system (Bio-Rad). Primers are listed in the Supplemental Methods. RNA sequencing libraries were prepared using the ribosomal depletion method (KAPA RNA HyperPrep Kit with RiboErase, Roche) and sequenced on the Illumina NovaSeq 6000 platform (150  bp paired-end, 100 million reads per sample). RNA sequencing analysis was performed on the basis of previously published pipeline.⁹ Briefly, reads were processed, aligned to GRCm38.p6 with vM25 annotation using STAR²² (version 2.7.9), and filtered for protein coding genes. Counts were normalized by the transcripts per million method. Detection of differentially expressed genes (DEG) was performed using R package DESeq2²³ with false discovery rate (FDR) ≤0.05 as the significance threshold. Heatmaps were generated using the pheatmap package in R with input matrix consisted of log2(transcripts per million+1) transformed gene expression values. Gene ontology (GO) analyses and gene set enrichment analyses were performed using the clusterProfiler²⁴ R package.

ATAC-Seq Processing and Data Analysis

ATAC-Seq²⁵ was performed using a modified ActiveMotif (ActiveMotif, Cat. no. 53150) and Omni-ATAC protocol.^26,27 Briefly, snap-frozen kidney tissues were resected and transferred to a homogenization tube with three CK28 beads (Bertin Crop, Cat. no. P000911-LYSK0-A) and 500 μl of ice-cold nuclei preparation buffer (10 mM Tris-HCl pH 7.5, 3 mM MgCl₂, and 10 mM NaCl). Homogenization was perfomed three times at 4500 rpm for 10 seconds each time with the Precellys evolution homogenizer (Bertin Crop). Nuclei suspension was passed through a 20-μm strainer, washed twice with buffer at 2000 rpm for 1 minute, and resuspended in 500 μl buffer. Nuclei were checked under microscope and counted for 50,000 nuclei per sample. Tagmentation step follows the ActiveMotif protocol with modifications of using half the amount of tagmentation mix and incubating at 37°C for 20 minutes. The tagmented DNA was purified and amplified following the ActiveMotif protocol. The final ATAC-Seq library was analyzed by the Yale Center for Genome Analysis using TapeStation to assess size distribution and followed by sequencing at 50 bp paired-end reads for 100 million reads per sample.

ATAC-Seq data were processed based on a published pipeline for mouse kidney.²⁸ In brief, raw sequencing reads were trimmed using cutadapt and mapped to mm10 using bowtie2.²⁹ Duplicates and mitochondrial reads were marked by Picard and removed by Samtools. Peak calling was performed using MACS2,³⁰ and peaks falling into blacklist regions (mm10 blacklist v2) were removed. One Pkd1^KO+Glis3^KO sample was an outlier and was removed from downstream analyses. Transcription start sites (TSS) enrichment score showing the aggregate distribution of ATAC-Seq peaks relative to TSS with flanking regions spanning ±2 kb upstream and downstream of TSS was performed by ChIPseeker.³¹ Consensus peaks were generated by requiring overlap of MACS2-called peaks in at least three biological replicates (minOverlap=3). Peak annotation was subsequently performed using ChIPseeker in R. Differentially accessible regions (DAR) were identified using the Diffbind R package and FDR ≤0.05 as the threshold for significance. De novo motif analysis was perfomed by HOMER³² (version v4.11). GO analysis was performed by Genomic Regions Enrichment of Annotations Tool.³³ Footprinting analysis was performed using Transcription factor Occupancy prediction By Investigation of ATAC-Seq Signal (TOBIAS).³⁴

Sample Size and Statistics

Sample size and power calculations were performed prospectively using STPLAN (version 4.5; University of Texas, MD Anderson Cancer Center). For the early-onset model, we used a two-sided significance threshold of 0.05 and 95% power. A sample size of ten mice would detect at least a 30% change in Pkd1^KO+Glis3^KO, on the basis of the mean kidney-to-body weight ratio (7.89%±1.84%) observed in Pkd1^KO mice at P14. For the adult-onset model, we used a two-sided significance threshold of 0.05 and 80% power. A sample size of 19 mice would detect at least a 50% change in Pkd1^KO+Glis3^KO, on the basis of the mean kidney-to-body weight ratio (7.09%±5.35%) observed in Pkd1^KO mice at P14.

Quantitative data were analyzed using either one-way ANOVA followed by Tukey multiple-comparison test or two-tailed, unpaired Student’s t test as indicated in figure legends. All data are presented as mean±SEM, and P ≤ 0.05 was used as the threshold for statistical significance. GraphPad Prism (v. 10.3.0) software was used to perform statistical analyses.

Results

We first determined whether the cilia localization of Glis3 was altered by inactivation of Pkd1. We stably expressed C-terminal EGFP-tagged Glis3 in isogenic WT and Pkd1 knockout (Pkd1^KO) IMCD3 cells³⁵ that have stable expression of Nphp3^(1–200)-mApple cilia marker.⁹ In WT IMCD3 cells, Glis3 showed the expected localization in primary cilia (Figure 1A). The pattern of expression was not changed in Pkd1^KO IMCD3 cells (Figure 1A), indicating that at the level of light microscopy, the cilia localization of Glis3 is not affected by Pkd1 inactivation in this epithelial cell line. We next sought to determine whether there is functional interaction between Glis3 and Pkd1. The neonatal lethality and extrarenal phenotypes of Glis3 null mice prevented us from using these mice to study these potential functional interactions. Therefore, we generated a Glis3 conditional allele (Glis3^fl) using CRISPR/Cas9 to insert loxP sites flanking exon 6 to enable deletion of the fifth zinc finger domain (Supplemental Figure 1). This has been shown to be sufficient to abolish the DNA-binding ability and transcriptional activity of Glis3.¹⁵ We crossed the Glis3^fl and Pkd1^fl alleles with the digenic Pax8^rtTA; TetO^Cre system to achieve doxycycline inducible deletion of the target genes selectively in kidney tubule epithelial cells. We have previously shown that Glis2 upregulation is a reliable in vitro indicator of CDCA in vitro.⁹ We began by examining whether Glis3 inactivation affects polycystin-dependent Glis2 expression in primary kidney cell cultures from mice with Pkd1^fl/fl; Pax8^rtTA; TetO^Cre and Glis3^fl/fl; Pkd1^fl/fl; Pax8^rtTA; TetO^Cre genotypes without or with doxycycline treatment in vitro. RT-qPCR confirmed Glis3 and Pkd1 knockout in the corresponding samples (Figure 1B). Glis2 mRNA levels were significantly higher following inactivation of Pkd1 (Figure 1B, left) and remained elevated in Glis3 and Pkd1 double mutants (Figure 1B, right). Glis2 protein expression was also unchanged in Glis3 and Pkd1 double mutant cells compared with Pkd1 single knockouts (Figure 1C). Glis3 is unlikely to function as a regulator of Glis2 expression in the setting of Pkd1 inactivation.

Figure 1.

Glis3 does not affect polycystin-dependent Glis2 expression. (A) Isogenic WT or Pkd1^KO IMCD3 cells stably expressing Glis3-EGFP and the Nphp3^(1–200)-mApple cilia marker cultured under conditions to form cilia and imaged live by confocal microscopy. 3D reconstructions and z-stack projections with magnified views of boxed regions. Scale bar, 10 μm. (B) RT-qPCR of renal primary cell culture after in vitro doxycycline induced inactivation of either Pkd1 alone (left panel) or Glis3 and Pkd1 together (right panel). Fold change of each gene following doxycycline induction (closed symbol) is shown relative to the expression of same gene without doxycycline (open symbol), which is set to 1.0. Each data point is an independent primary cell culture from a single mouse with the indicated genotype. Statistical significance for each gene is determined by unpaired two-tailed Student’s t test and presented as mean±SEM. (C) Immunoblots of nuclear fraction of renal primary cell after in vitro doxycycline induced inactivation of either Glis3 only, Pkd1 only, or Glis3 and Pkd1. WT, wild-type.

To determine the role of Glis3 in polycystic kidney disease in vivo, we generated mice with the following four genotypes: WT, Glis3^fl/fl; Pax8^rtTA; TetO^Cre (Glis3^KO), Pkd1^fl/fl; Pax8^rtTA; TetO^Cre (Pkd1^KO), and Glis3^fl/fl; Pkd1^fl/fl; Pax8^rtTA; TetO^Cre (Pkd1^KO+Glis3^KO). We started with the early-onset model where doxycycline in drinking water was administrated to the nursing dams from postnatal day 0 (P0) to postnatal day 14 (P14), and the experimental pups were examined at P14. Compared with the Pkd1^KO cystic controls, kidney-to-body weight ratio and cystic index were significantly higher in Pkd1^KO+Glis3^KO, indicating a significant worsening of the polycystic phenotype (Figure 2, A–E, and Supplemental Figure 2), regardless of sex (Supplemental Figure 3). Glis3^KO mice showed very occasional and sporadic tubule dilation (Figure 2B), but there was no significant abnormal phenotype in the P14 kidney (Figure 2, C–E). All kidneys showed intact cilia detected by immunofluorescence microscopy using antibodies to acetylated α-tubulin (Figure 2F), confirming that Glis3 inactivation does not affect gross cilia structure. We next examined the role of Glis3 in adult-onset models of ADPKD where experimental animals were administrated doxycycline in drinking water from P28 to P42 and were examined at age 14 weeks for the polycystic phenotype. Similarly to the early-onset model, Pkd1^KO+Glis3^KO mice at 14 weeks showed worse polycystic kidney disease progression with significantly higher kidney-to-body weight ratio, cystic index, and BUN level compared with Pkd1^KO (Figure 3, A–E, and Supplemental Figure 4). The 14-week end point was chosen because Pkd1^KO+Glis3^KO, but not Pkd1^KO, mice have decreased survival after that timepoint. Of note, in distinction from the early-onset model, Glis3^KO kidneys at 14 weeks did not exhibit any histological changes (Figure 3B). Glis3^KO kidneys were also completely normal when they were further aged and evaluated at 18 and 24 weeks (Figure 3F). These results indicate that the worsening phenotype observed in Pkd1^KO+Glis3^KO is not due to additive effect of the two single mutants but suggests that Glis3 genetically interacts with Pkd1. Sex dimorphism in severity of adult Pkd1 mouse models has been previously reported.³⁶ We therefore examined the effect of Glis3 inactivation in both sexes (Supplemental Figure 5). Male Pkd1^KO mice developed severe polycystic kidney disease with a mean kidney-to-body weight ratio of approximately 10.3%, mean cystic index of approximately 49%, and average BUN of 66 mg/dl (Supplemental Figure 5, A–C). All these parameters were further significantly higher in male Pkd1^KO+Glis3^KO mice (approximately 15.8%, approximately 72%, and 143 mg/dl, respectively; Supplemental Figure 5, A–C). Female Pkd1^KO mice exhibited a mild phenotype (kidney-to-body weight ratio approximately 3.0%, cystic index approximately 43%, and BUN 9 mg/dl), which was higher in Pkd1^KO+Glis3^KO (approximately 5.3%, approximately 54%, 35 mg/dl, respectively) but did not reach statistical significance likely because of the relative protection from polycystic kidney disease in female mice (Supplemental Figure 5, D–F). Glis3 is a modifier of cyst progression in early and late models of ADPKD in both sexes.

Figure 2.

Glis3 inactivation worsens ADPKD in an early model of Pkd1. (A) Representative images of kidneys from mice at postnatal day 14 (P14) with indicated genotypes. Oral doxycycline was administrated to the nursing dams from P0 to P14, and the kidneys of the pups were examined at P14. Scale bar, 2 mm. (B) Representative images of H&E staining for the corresponding genotype. Scale bar, 50 μm. (C–E) Aggregate quantitative data for kidney-to-body weight ratio (C), cystic index (D), and BUN (E). Colors and symbol shapes correspond to genotype defined in (A). Sex of mice are shown for male (closed symbols) and female (open symbols). Multiple-group comparisons were performed by one-way ANOVA followed by Tukey multiple-comparison test, presented as mean±SEM. (F) Immunofluorescence of LTL (proximal tubule), DBA (collecting duct), and acetylated α-tubulin on cryosections from kidneys with genotypes indicated by color symbols in (A). Boxed regions show magnified views. Scale bar, 10 μm. n, number of mice in each group. αTub, α-tubulin; ADPKD, autosomal dominant polycystic kidney disease; DBA, dolichos biflorus agglutinin; H&E, hematoxylin and eosin; LTL, Lotus tetragonolobus lectin.

Figure 3.

Glis3 inactivation worsens ADPKD in an adult model of Pkd1. (A) Representative images of kidneys from mice at age 14 weeks with indicated genotypes. All mice were administrated with oral doxycycline from P28 to P42 and examined at 14 weeks. Scale bar, 2 mm. (B) Representative images of H&E staining for the corresponding genotype. Scale bar, 50 μm. (C–E) Aggregate quantitative data for kidney-to-body weight ratio (C), cystic index (D), and BUN (E). Colors and symbol shapes correspond to genotype defined in (A). Sex is shown for male (closed symbols) and female (open symbols) mice. Multiple-group comparisons were performed by one-way ANOVA followed by Tukey multiple-comparison test, presented as mean±SEM. (F) Representative H&E staining images for WT and Glis3^KO mice aged to 18 and 24 weeks. Scale bar, 50 μm.

To begin to understand how Glis3 inactivation exacerbates cyst progression in ADPKD, we performed RNA-Seq and ATAC-Seq across the four genotypes: WT, Glis3^KO, Pkd1^KO, and Pkd1^KO+Glis3^KO. All samples were from male mice to limit confounding by sex-dependent genomic effects and gene expression changes. All animals were induced with doxycycline from P28 to P42. RNA and Tn5 transposase tagmented DNA were both prepared from the same kidneys at age 7 weeks (P49)—the same age used in our original TRAP RNA-Seq studies.⁹ Principal component analysis for the RNA-Seq showed promising clustering of samples by genotype (Figure 4A). Pkd1^KO and Pkd1^KO+Glis3^KO clustered near each other, suggesting Pkd1 inactivation is a major driver of the mRNA expression changes (Figure 4A). To determine whether the core transcriptomic signature of cyst initiation after Pkd1 inactivation is affected by Glis3, we examined the differential expression of the previously published 73 CDCA pattern genes⁹ across the four genotypes. Heatmaps with unsupervised hierarchical clustering showed clear separation of genotypes that do not form cysts (WT and Glis3^KO) from genotypes that will form cysts (Pkd1^KO and Pkd1^KO+Glis3^KO) (Figure 4B). These data confirm the specificity of the 73 gene CDCA translatome expression signature as an early in vivo bulk RNA-Seq readout for polycystic kidney disease before overt cyst formation. The CDCA gene pattern was not affected by Glis3^KO in Pkd1^KO+Glis3^KO, supporting the hypothesis that the effect of Glis3^KO on the polycystic kidney phenotype likely involves pathways other than CDCA (Figure 4B). We next examined DEG across the entire transcriptome by applying a FDR ≤0.05 as the significance threshold. To examine the Glis3-dependent transcriptome changes relative to both WT and Pkd1^KO conditions, we compared Glis3^KO with WT and Pkd1^KO+Glis3^KO with Pkd1^KO. This yielded 2404 and 537 DEG, respectively (Table 1 and Supplemental Data 1). GO analyses of the 2404 DEG in Glis3^KO compared with WT showed strong signals for changes in metabolic processes associated with Glis3^KO (Figure 4C). The 537 DEG in the Pkd1^KO+Glis3^KO to Pkd1^KO comparison, presumably enriched for the genes most responsible for the phenotypic worsening because of Glis3^KO, highlighted a striking enrichment of fatty acid metabolic process (Figure 4D). Fatty acid metabolism process genes in Pkd1^KO+Glis3^KO showed a significant skew toward downregulation relative to Pkd1^KO alone (42 downregulated, 18 upregulated; P < 1.3e⁻³ by exact binomial test; Figure 4E). Gene set enrichment analysis in Pkd1^KO+Glis3^KO compared with Pkd1^KO confirmed the strong enrichment for fatty acid metabolism and predominance of downregulated genes (Figure 4F). There was a significant representation of downregulated genes involved in fatty acid β-oxidation (e.g., Acsm2, Acsm3, Acadl, Ech1, Hadh, Slc27a2, Acox1, and Amacr; adjusted P < 1.0e⁻⁴ by the Fisher exact test with Benjamini–Hochberg adjustment for multiple comparisons; Figure 4E). Genes involved in fatty acid biosynthesis (e.g., Acaca, Fasn, and Scd1/2) were largely unchanged, and elongases (e.g., Elovl2 and Elovl5I) were upregulated. Dysregulated fatty acid metabolism, and more specifically downregulation of β-oxidation, is a compelling candidate process for contributing to the worsening of polycystic kidney disease in Pkd1^KO+Glis3^KO mouse models compared with Pkd1^KO alone.

Figure 4.

Dysregulated metabolism gene expression in kidneys with Glis3 inactivation. (A) PCA plot of kidney tissue RNA-Seq showing the global transcriptomic difference among the indicated genotypes. Each symbol represents a kidney from a different mouse. (B) Heatmap with unsupervised hierarchical clustering showing relative gene expression of the 73 CDCA signature genes with the indicated genotypes. (C and D) GO analysis performed using the clusterProfiler R package with gene lists with FDR ≤0.05 threshold for significance in Glis3^KO compared with WT (C) and Pkd1^KO+Glis3^KO compared with Pkd1^KO (D). (E) Volcano plot showing the expression of the 60 genes in the GO term fatty acid metabolic process in Pkd1^KO+Glis3^KO to Pkd1^KO comparison. Red, upregulated; blue, downregulated. Genes involved in fatty acid β-oxidation and labeled. (F) Gene set enrichment analysis highlighting the strong enrichment for fatty acid metabolism in Pkd1^KO+Glis3^KO compared with Pkd1^KO. BP, biological process; CC, cellular component; CDCA, cilia-dependent cyst activation; FDR, false discovery rate; GO, gene ontology; MF, molecular function; PC1, polycystin-1; PC2, polycystin-2; PCA, principal component analysis.

Table 1.

Numbers of differentially expressed genes (DEG) and differentially accessible regions (DAR) in pairwise comparisons of RNA-Seq and ATAC-Seq

Differential Analyses	RNA-Seq DEG	ATAC-Seq DAR		Number of DAR (FDR ≤0.05 and Promoter) Overlapped with DEG (FDR ≤0.05)
Pairwise comparison	FDR ≤0.05	FDR ≤0.05	FDR ≤0.05 and promoter	Total	Same direction change
Glis3KO versus WT	2404	3776	697	234	216 (26a/190b)
Pkd1KO versus WT	1361	3147	445	68	62 (23a/39b)
Pkd1KO+Glis3KO versus WT	2053	3679	499	91	73 (27a/46b)
Pkd1KO+Glis3KO versus Pkd1KO	537	330	44	6	6 (2a/4b)

Table showing numbers of DEG and DAR identified in the indicated pairwise comparisons. In the rightmost columns, DAR in promoter regions are overlapped with DEG and filtered for same direction change. FDR, false discovery rate; WT, wild-type.

^aMore accessible.

^bLess accessible.

We next sought to integrate the RNA-Seq and ATAC-Seq data obtained from the same biological samples. Principal component analysis of ATAC-Seq showed clustering of samples by genotype (Figure 5A). We observed strong signal to noise at TSS across all samples (Figure 5B), highlighting successful enrichment of open chromatin accessibility at promoter-proximal regions. In total, we detected 129,765 consensus peaks, of which 21.49% were annotated to promoter regions defined as ±2 kb from TSS and 33.45% were mapped to distal intergenic regions (Figure 5C). This distribution aligns with a previous report of transcriptional regulatory element localization in the kidney.²⁸ We identified DAR in pairwise comparisons among the four genotypes with FDR ≤0.05 as the threshold for significance (Table 1 and Supplemental Data 2). We further refined this to select DAR located in the promoter regions near TSS that also overlapped with DEG identified in the respective pairwise genotype group comparisons for RNA-Seq (Table 1 and Supplemental Data 3). Within these overlap groups, we further constrained analysis to DAR in TSS regions that overlapped with DEG with the same direction of change, meaning that more accessible TSS region DAR in a pairwise comparison had upregulated gene mRNA expression and less accessible TSS region DAR had downregulated gene mRNA expression. For the Glis3^KO compared with WT, 234 of 697 DAR in the region of TSS overlapped DEG from the corresponding RNA-Seq comparison. Of these, 216 DAR (>90%) had the same direction change as the corresponding 190 DEG (Figure 5D and Table 1). There were fewer DEG than DAR because several genes has more than one TSS region DAR with same direction change associated with it. By comparison, Pkd1^KO compared with WT had 62 of 68 DAR overlapping with 55 DEG with the same direction change (Figure 5E and Table 1). Notably, these 55 DEG included Lad1, Ntn4, and Spns2, three genes significantly downregulated by Pkd1 inactivation in the TRAP RNA-Seq.⁹ All three genes showed downregulated gene expression and less accessible chromatin in the TSS region in Pkd1^KO compared with WT. Overall, these data suggest an informative correlation between overlapping changes chromatin accessibility in TSS regions and gene expression from bulk RNA-Seq.

Figure 5.

Genomic accessibility and transcription factor footprinting in kidneys with Glis3 inactivation. (A) PCA plot of ATAC-Seq showing the global chromatin accessibility differences among the indicated genotypes. Each symbol is an independent biological sample. (B) TSS enrichment profile of ATAC-Seq samples. The aggregate distributions of ATAC-Seq consensus peaks for each genotype are shown relative to TSS and flanking regions spanning ±2 kb upstream and downstream of TSS. (C) Genomic annotation of consensus peaks. 21.49% of the peaks fall into promoter regions. (D) Heatmaps illustrating the fold change of 216 promoter region ATAC-Seq DAR peaks (left panel) and the corresponding overlapping 190 RNA-Seq DEG (right panel) with the same direction change in expression in Glis3^KO versus WT. (E) Heatmaps of fold change in 62 promoter region ATAC-Seq DAR peaks (left panel) and overlapping 55 RNA-Seq DEG (right panel) with the same direction change in Pkd1^KO versus WT. (F) Bar plot showing GO analysis of the 216 TSS-associated DAR with same direction change in associated DEG in the Glis3^KO versus WT comparison using the GREAT tool. (G) Significant results for de novo motif analysis for the 190 downregulated TSS region DAR overlapped with DEG in Glis3^KO versus WT. (H and I) Volcano plots showing differential binding score distribution of the 841 transcription factor binding sites scanned by TOBIAS in Glis3^KO versus WT (H) and Pkd1^KO versus WT (I). Blue, less occupancy in Glis3^KO (H) or Pkd1^KO (I); red, more occupancy in Glis3^KO (H) or Pkd1^KO (I). (J–O) Footprint comparisons for transcription factors Dbp (MA0639.1; J and M), Hnf4a (MA0114.4; K and N), and Hnf1b (MA0153.2; L and O) in Glis3^KO versus WT (J–L) and Pkd1^KO versus WT (M–O). Dbp, D site albumin promoter binding protein; DEG, differentially expressed gene; GREAT, Genomic Regions Enrichment of Annotations Tool; Hnf1b, HNF1 homeobox B; Hnf4a, hepatic nuclear factor 4, alpha; Klf15, Krüppel-like transcription factor 15; TOBIAS, Transcription factor Occupancy prediction By Investigation of ATAC-Seq Signal; TSS, transcription start sites.

We next explored the potential functional significance of the 216 DAR in the Glis3^KO versus WT comparison that showed overlap with DEG to gain insight into the consequences of the genomic changes after Glis3 inactivation. These 216 DAR had a significant preference for less accessible chromatin (190 compared with just 26 that were more accessible; P < 2.2e⁻¹⁶ by exact binomial test; Table 1), suggesting that Glis3 is an activating transcription factor that normally promotes chromatin accessibility. Analysis using the Genomic Regions Enrichment of Annotations Tool³³ suggested that the cis functions of the 216 DAR are related to circadian regulation and metabolic processes (Figure 5F). De novo motif analysis with HOMER for the 190 DAR that were less accessible in Glis3^KO yielded significant signals for only four transcription factors: the two kidney epithelial-related transcription factors, HNF1 homeobox B (Hnf1b)^37,38 and hepatic nuclear factor 4, alpha (Hnf4a)^39–41; the circadian-related transcription factor, D site albumin promoter binding protein (Dbp)^42,43; and the metabolism and circadian-related transcription factor, Krüppel-like transcription factor 15 (Klf15)^42,44,45 (Figure 5G and Supplemental Figure 6). Glis3^KO results in downregulation of chromatin access for epithelial- and circadian-related transcriptional programs.

Transcription factors directly bound to DNA can protect DNA bases involved in binding from Tn5 transposase activity resulting in depletion of signals in those regions. These are known as transcription factor footprints.^34,46 Genome-wide differential footprinting analysis in ATAC-Seq reflects changes in transcription factor binding across biological conditions.⁴⁷ We applied TOBIAS³⁴ to detect transcription factor footprints and compare transcription factor occupancy among the ATAC-Seq data for different genotypes. Of 841 transcription factor binding sites from the JASPAR database scanned by TOBIAS,⁴⁸ 577 trended toward reduced occupancy and 264 showed increased occupancy in Glis3^KO relative to WT (Figure 5H and Supplemental Data 4). Pkd1^KO showed more balanced trends, with 455 toward reduced and 386 toward greater occupancy (Figure 5I and Supplemental Data 4). These data show that Glis3^KO not only results in less accessible chromatin (Figure 5D and Table 1) but also results in reduced transcription factor occupancy at motif binding sites (Figure 5H), reinforcing the hypothesis that Glis3 normally functions as an activating transcription factor. The top 5% footprint scores assigned by TOBIAS represent high confidence differential motif binding sites in the dataset.³⁴ Glis3^KO had 50 less occupied and 42 more occupied instances meeting this confidence threshold (Figure 5H). Among these, Dbp showed the greatest reduced occupancy in Glis3^KO with a differential binding score of –0.8054, which is further supported by the aggregate footprint plot showing a greatly reduced footprint depth of Dbp in Glis3^KO (Figure 5, H and J). Similarly, the occupancy status of Hnf4a binding sites were also greatly decreased in Glis3^KO compared with WT (Figure 5, H and K, and Supplemental Figure 7A). Neither of the other two motifs identified by the de novo motif analysis, Hnf1b and Klf15 (Figure 5G), were present in the 95% quantile in the footprinting analysis although the footprint depth of Hnf1b did show a large reduction in Glis3^KO (Figure 5, H and L). By comparison, footprinting analysis of Pkd1^KO compared with WT found that Dbp, Hnf4a, and Klf15 were not present above the 95% quantile (Figure 5, I, M, and N, and Supplemental Figure 7B). Hnf1b did appear within the top 5% differentially bound transcription factors in Pkd1^KO (Figure 5I); however, the aggregate footprint plot only displayed subtle changes in footprint depth (Figure 5O), suggesting a likelihood of a false positive. We next evaluated the footprints for the three published Glis3 binding motifs.^19,32,48 The Glis3 motifs from either the JASPAR or HOMER databases did not show footprints of transcription factor occupancy in WT or Glis3^KO (Figure 5H and Supplemental Figure 7, C and D). The footprint motif for Glis3 identified by a recent ChIP-Seq study¹⁹ was readily identified, but there was no difference in binding occupancy for this motif when comparing Glis3^KO with WT (Supplemental Figure 7E). Glis3^KO showed overall reduced transcription factor binding occupancy that included significant reductions in binding by Dbp and Hnf4a.

Finally, we integrated ATAC-Seq DAR data for Glis3^KO versus WT with the recent Glis3 ChIP-Seq dataset.¹⁹ We identified 371 of 3776 DAR (9.80%) that overlapped with Glis3 ChIP genomic peaks (Supplemental Data 5). This suggests that Glis3 may in large part affect transcriptional regulation through coordination with other transcription factors, rather than changing chromatin accessibility by directly binding to DNA. To explore potential direct targets of Glis3, we intersected the ChIP-Seq peaks that fall into the promoter regions with the 190 downregulated DAR in promotor regions that overlapped with DEG in Glis3^KO versus WT. We identified 23 genes (Supplemental Data 6) that were transcriptionally downregulated and had less open chromatin in the promoter region in Glis3^KO versus WT and whose promoter directly bound to Glis3 on the basis of published ChIP-Seq data. This overlapping dataset gene list is notable for the circadian transcription factor Dbp. Dbp and perhaps other genes on this list are likely candidates for direct regulation by Glis3 transcription factor activity.

Discussion

Glis2 is a necessary effector of the cyst-promoting CDCA pathway downstream of polycystin inactivation in the setting of intact cilia. Within the same transcription factor family, Glis3 is known to localize in the primary cilia¹⁵ and is potentially involved in polycystin-2 signaling.²⁰ On the basis of this, we tested a hypothesis, analogous to the functioning of Patched and the Gli transcription factors in Hedgehog signaling, that the Glis3 transcription factor located in cilia functions as the polycystin and cilia-dependent upstream modulator of Glis2 upregulation in ADPKD. We found this not to be the case. Glis3 inactivation did not change Glis2 expression or expression of other CDCA pattern genes. Notably, adult inactivation of Glis3 in kidney tubules did not lead to cyst formation as has been reported for Glis3 germline mutant mice.^15,18,49 Glis3 inactivation did, however, interact genetically with Pkd1 to exacerbate the polycystic kidney disease phenotype in ADPKD mouse models with selective kidney tubule inactivation of both genes. We propose that this effect of Glis3 is related to activity in pathways other than CDCA.

We applied an integrated analysis of in vivo transcriptomic and chromatin accessibility studies to understand the effects of loss of Glis3 in kidney tubules in the presence and absence of Pkd1 inactivation. Several features of this study are relevant to interpreting the outcomes. Inducible inactivation of Glis3 or Pkd1 was confined to tubule cells in the mature kidney; both RNA-Seq and ATAC-Seq were performed using the same biological samples; and the selected timepoint was after there was effective tubule-cell autonomous inactivation of Glis3 or Pkd1 but before any overt cystic or other secondary phenotypes in the kidney or systemically.⁹ In this setting, Glis3 inactivation largely resulted in decreased chromatin accessibility in promoter regions, and this was associated with downregulation of gene expression from those regions. While the activating role of WT Glis3 expression implied by this knockout result is consistent with ChIP-Seq studies,¹⁹ the reduction in chromatin accessibility largely occurred in regions where Glis3 was not known to bind. Glis3 may therefore regulate promoter regions indirectly by interaction with other transcription factors or by association with enhancer or other genomic sites remote from promoter regions. The effects of Glis3 inactivation on mRNA levels in RNA-Seq from adult kidneys showed strong association with metabolic processes, and these effects seem to be related to differential expression of transmembrane transporter activities. This is generally consistent with a recent study of Glis3 germline knockout mice that showed suppression of genes critical for mitochondrial biogenesis, oxidative phosphorylation, fatty acid oxidation, and the tricarboxylic acid cycle.¹⁹ Dual inactivation of Glis3 with Pkd1 accentuated a strong signal for downregulation of fatty acid metabolism processes. More specifically, several key genes involved in fatty acid β-oxidation—including Acsm2, Acsm3, Acadl, Ech1, Hadh, Slc27a2, Acox1, and Amacr—were significantly downregulated in Pkd1^KO+Glis3^KO compared with Pkd1^KO. This is in keeping with the phenotypic shift of cyst cells toward aerobic glycolysis and away from mitochondrial oxidative metabolism.⁵⁰ Pkd1 mutant kidneys have defects in fatty acid oxidation, and exacerbations of these defects may be the cause of increased lipid peroxidation⁵¹ and progression of ADPKD.^36,52 Dysregulation of fatty acid metabolism by Glis3 inactivation is the strongest candidate biologic process from our study to explain the worsening of the Pkd1 ADPKD phenotype with Glis3 inactivation.

Hnf1b and Hnf4a encode important transcription factors for kidney epithelial differentiation and function.^40,53 Motifs for both were identified in our ATAC-Seq and showed reduced chromatin accessibility in Glis3^KO; however, Glis3 inactivation was most significantly associated with reduction of Hnf4a binding occupancy. The data suggest a context-dependent role in the kidney for Glis3 as part of a transcriptional network that affects the accessibility of sites typically occupied by the Hnf1b and Hnf4a. Downregulation of Hnf4a has been shown to be a disease modifier associated with more rapid progression of ADPKD in vivo.⁵⁴ HNF4a knockout in human kidney organoids is associated with downregulation of lipid metabolism,³⁹ and dual inactivation of Hnf4a and Pkd1 exacerbates polycystic kidneys in mice compared with Pkd1^KO alone.⁵⁴ Our study identifies Hnf4a-accessible regions as a significant target of Glis3 inactivation that was not previously known.¹⁹ This Hnf4a-related function may be the mechanistic connection with the alterations in lipid metabolism associated with worsening ADPKD in Glis3 and Pkd1 dual inactivation kidneys. The de novo transcription factor binding site analysis for Glis3 also identified Dbp and Klf15. Of these, the signal for Dbp was the most significant across the analyses and overlapped with direct Glis3 binding data identified from ChIP-Seq.¹⁹ Dbp is a transcription factor whose expression is controlled by the circadian clock genes and whose function is to effect transcriptional responses to circadian rhythm.^42,55–57 Recent data suggest disruption of circadian clocks may accelerate progression in ADPKD,⁵⁶ so the effect of Glis3 on ADPKD progression may also be mediated by the downregulation of Dbp. Glis3 may indirectly affect kidney metabolic programs, including fatty acid β-oxidation through Hnf4a and kidney circadian programs through Dbp. The consequent metabolic and circadian perturbations are how loss of Glis3 results in worsening of ADPKD severity.

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/F611.

Author Contributions

Conceptualization: Stefan Somlo, Zemeng Wei.

Data curation: Jianlei Gu.

Formal analysis: Jianlei Gu, Zemeng Wei.

Funding acquisition: Stefan Somlo.

Investigation: Zemeng Wei.

Methodology: Stefan Somlo, Zemeng Wei.

Resources: Xin Tian, Chao Zhang.

Software: Jianlei Gu.

Supervision: Stefan Somlo, Hongyu Zhao.

Visualization: Zemeng Wei.

Writing – original draft: Stefan Somlo, Zemeng Wei.

Writing – review & editing: Stefan Somlo.

Funding

S. Somlo: National Institute of Diabetes and Digestive and Kidney Diseases (DK120911, DK100592, and DK120534) and Amy P. Goldman Foundation (grant).

Declarative Statements

All animal experiments were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals or an equivalent standard that meets or exceeds the ethical and welfare requirements outlined in the NIH Guide. All protocols were approved by the appropriate Institutional Animal Care and Use Committee. This research was posted on a preprint server. https://doi.org/10.1101/2025.06.08.658051.

Data Availability Statements

Original data generated for the study are available in a public access repository. Data Type: Raw Data/Source Data. Repository Name: Gene Expression Omnibus. All the raw sequencing data and processed data have been deposited in the Gene Expression Omnibus with the following accession number: GSE299238. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE299238.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/JSN/F604, http://links.lww.com/JSN/F605, http://links.lww.com/JSN/F606, http://links.lww.com/JSN/F607, http://links.lww.com/JSN/F608, http://links.lww.com/JSN/F609, http://links.lww.com/JSN/F610.

Supplemental Methods

Supplemental Figure 1. Generation of conditional Glis3^fl allele.

Supplemental Figure 2. Images of all the histological sections used in Figure 2.

Supplemental Figure 3. Aggregate quantitative data in Figure 2 separated by sex.

Supplemental Figure 4. Images of all the histological sections used in Figure 3, A–E.

Supplemental Figure 5. Aggregate quantitative data in Figure 3 separated by sex.

Supplemental Figure 6. Full list of de novo motif analysis shown in Figure 5G.

Supplemental Figure 7. Footprinting analyses of Hnf4a and Glis3 binding motifs.

Supplemental Data 1. RNA-Seq DEG.

Supplemental Data 2. ATAC-Seq DAR.

Supplemental Data 3. Overlap of DAR in promoter region and DEG with same direction change.

Supplemental Data 4. Full list of TOBIAS analyses.

Supplemental Data 5. Overlap of DAR with published ChIP-Seq.

Supplemental Data 6. Potential transcription targets of Glis3.