Dnmt3a
and
Dnmt3b
-Decommissioned Fetal
Enhancers are Linked to Kidney Disease
Yuting Guan, Hongbo Liu , Ziyuan Ma , Szu-Yuan Li, Jihwan Park , Xin Sheng, and Katalin Susztak
Department of Medicine, Renal Electrolyte and Hypertension Division, Department of Genetics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania
ABSTRACT
BackgroundCytosine methylation is an epigenetic mark that dictates cell fate and response to stimuli. The
timing and establishment of methylation logic during kidney development remains unknown. DNA meth-yltransferase 3a and 3b are the enzymes capable of establishingde novomethylation.
MethodsWe generated mice with genetic deletion ofDnmt3aandDnmt3bin nephron progenitor cells
(Six2CreDnmt3a/3b) and kidney tubule cells (KspCreDnmt3a/3b). We characterizedKspCreDnmt3a/3bmice
at baseline and after injury. Unbiased omics profiling, such as whole genome bisulfite sequencing, reduced representation bisulfite sequencing and RNA sequencing were performed on whole-kidney samples and isolated renal tubule cells.
ResultsKspCreDnmt3a/3bmice showed no obvious morphologic and functional alterations at baseline.
Knockout animals exhibited increased resistance to cisplatin-induced kidney injury, but not to folic acid–inducedfibrosis. Whole-genome bisulfite sequencing indicated thatDnmt3aandDnmt3bplay an important role in methylation of gene regulatory regions that act as fetal-specific enhancers in the de-veloping kidney but are decommissioned in the mature kidney. Loss ofDnmt3aandDnmt3bresulted in failure to silence developmental genes. We also found that fetal-enhancer regions methylated byDnmt3a andDnmt3bwere enriched for kidney disease genetic risk loci. Methylation patterns of kidneys from patients with CKD showed defects similar to those in mice withDnmt3aandDnmt3bdeletion.
ConclusionsOur results indicate a potential locus-specific convergence of genetic, epigenetic, and
de-velopmental elements in kidney disease development.
JASN31:ccc–ccc, 2020. doi: https://doi.org/10.1681/ASN.2019080797
Cytosine methylation is erased and reestablished between generations. DNA methylation is removed from the zygote by the blastocyst stage and
rein-stated during embryonic development.1De novo
meth-yltransferases 3a (Dnmt3a) and 3b (Dnmt3b) play key
roles in establishing new methylation patterns.2
Dnmt3a-deficient animals die several weeks after birth andDnmt3b-deficient animals diein utero, indicating
the essential roles ofDnmt3a- andDnmt3b-mediated
de novomethylation in development.
Cytosine methylation has several important functions. Most cytosines in the genome are
meth-ylated for efficient silencing of transposable
ele-ments.3 Transposable elements are the footprint
of ancient integrated retroviruses, making up close
to 50% of the genome.4Recent studies indicated that
Dnmt1deletion inSix2-positive nephron progenitors
Received August 10, 2019. Accepted December 24, 2019. Y.G. and H.L. contributed equally to this work.
Published online ahead of print. Publication date available at www.jasn.org.
Correspondence: Dr. Katalin Susztak, Renal Electrolyte and Hypertension Division, Department of Medicine, Department of Genetics, Perelman School of Medicine, University of Pennsyl-vania, 12-123 Smilow Translational Research Center, 3400 Civic Center Boulevard, Philadelphia, PA 19104. Email: ksusztak@ pennmedicine.upenn.edu
resulted in a release of transposable-element silencing, endoge-nous retroviral expression, cytokine release, and a downstream
severe kidney developmental defect.5
Cytosine methylation is believed to be a key regulator of
gene expression.6Gene regulatory regions, such as promoters
and enhancers, contain cytosine-guanine (CpG)–rich regions
(islands). In general, unmethylated promoters are permissive to transcription-factor binding and are associated with active gene expression. Methylated promoters exclude transcription factors, therefore they are associated with gene repression. Methylation of promoter and enhancer regions plays a key role in stabilizing linage decisions and restricting lineage fates. In addition to promoters, enhancers are critical for
establish-ing cell type–specific gene regulation and gene expression.
Enhancers are enriched for cell type–specific transcription
factor binding sites to ensure cell-specific gene regulation.
Cell type–specific genes often have multiple enhancers that
loop around and join promoters to establish a cell type—specific
gene expression pattern.Six2is a critical transcription factor in
kidney development.Six2-positive progenitors can undergo a
symmetric and asymmetric division to renew or to commit and differentiate into specialized nephron epithelium
seg-ments.7,8 The role of cytosine methylation in this process is
poorly understood.
Kidney disease is a complex gene environmental disease, affecting 800-million people worldwide. Genome-wide asso-ciation analyses have been conducted to understand the her-itability of kidney function, which uncovered close to 300 loci
associated with disease risk.9–11Each nucleotide variation only
increases disease risk by a minuscule amount, however, they should explain close to 50% of disease risk in aggregate. It has been proposed that human disease-associated genetic variants
are enriched on cell type–specific enhancer regions.12
Nucleotide-sequence changes at enhancer regions could alter transcription-factor binding, leading to quantitative differ-ences in gene expression contributing to disease development. Upon analyzing kidney disease risk loci, we found that only
20%–30% of identified loci are located in regions annotated as
enhancers in adult kidney samples. The underlying mecha-nism explaining the disease development that is associated with regions with no detectable regulatory function in the adult human kidney remains unknown. These regions might
be specific for rare kidney cell types or a disease or
develop-mental stage that is not captured by bulk analysis of adult human kidney tissue samples.
Environmental and nutritional alterations play equally
im-portant roles in kidney disease development.13–15Intrauterine
nutrient availability is known to be an important determinant of hypertension and kidney disease development, so called
“prenatal programming.”16,17Because epigenome-editing
en-zymes need substrates from the intermediate metabolism, it has been proposed that the epigenome might play a key role in prenatal programming. Nutrient availability, such as the
pres-ence of diabetes, remains the most significant risk factor for
kidney disease development.18 Indeed, the effect of poor
glycemic control on diabetic kidney disease development can be observed even decades after improved metabolic
control.19,20 It also remains unclear how environmental
and genetic factors interact and lead to kidney disease development.
To understand the role ofde novomethylation in kidney cell
differentiation, we generated mice with genetic deletion of
Dnmt3a andDnmt3b in nephron progenitor cells (NPCs)
and tubule cells, usingSix2CreandKspCre(Six2CreDnmt3a/3b
andKspCreDnmt3a/3b), respectively. Whole-genome bisulfite
sequencing (WGBS) and reduced representation bisulfite
se-quencing (RRBS) identified significant changes in the
meth-ylome of kidney tubule cells. We showed that Dnmt3aand
Dnmt3bplay important roles inde novomethylation of fetal
enhancers that were initially bound bySix2. The decline in
Six2expression during development was associated with a loss
of H3K27ac and an increase in methylation. These fetal
en-hancers decommissioned byDnmt3aand Dnmt3bwere
en-riched for kidney disease risk loci. Diseased kidney samples
showed a methylation pattern that was similar to theDnmt3a/
3bknockout animals. Overall, our data suggest that changes
brought on byDnmt3a/3b might be important for human
kidney disease.
METHODS
Animal Strains
Mice were raised and maintained in a barrier facility. Exper-iments were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Penn-sylvania and were performed in accordance with the
institu-tional guidelines. For folic acid–induced nephropathy mouse
models, 8-week-old male mice were injected with folic acid (250 mg/kg, dissolved in 300 mM sodium bicarbonate) intra-peritoneally and euthanized on day 7. For the cisplatin-induced injury model, 8-week-old male mice were injected with cisplatin (25 mg/kg) intraperitoneally and euthanized on day 3.
Significance Statement
Real-Time RT-PCR
RNA was isolated from mouse kidney using Trizol (Invitro-gen) and was reverse transcribed using the cDNA Archival Kit (Life Technologies). Real-time RT-PCR was performed using the SYBR Green Master Mix (Applied Biosystems). Primer pair sequences are shown in (Supplemental Table 1).
BUN and Creatinine Level
Serum creatinine was measured using Creatinine Enzymatic
and Creatinine Standard (Pointe Scientific). Serum BUN was
measured using Infinity Urea Liquid Stable Reagent (Thermo
Scientific). Both measurements were performed according to
the manufacturers’instructions.
Staining
Kidneys were harvested from mice, rinsed in PBS,fixed in 10%
formalin, and embedded in paraffin. Tissue sections were
stained with hematoxylin and eosin.
Isolation of CDH161Cells
Kidneys were harvested from control and KspCreDnmt3a/3b
double knockout mice and minced using a razor blade. About
0.25 g tissue was digested in 1.17 ml RPMI plus 50ml Enzyme
D, 25 ml Enzyme R, and 6.75ml Enzyme A from the Multi
Tissue Dissociation Kit 1 (Miltenyi Biotec) and incubated for 10 minutes at 37°C. Kidney was then dissociated using 21-gauge and 26.5-21-gauge needles and incubated for 10 minutes at 37°C. The dissociation step was repeated twice. To
neutral-ize the enzymes, 10% serum was added. Cells were thenfi
l-tered through the 70-mm nylon mesh to isolate single cells and
they were then centrifuged at 1000 rpm for 5 minutes. The pellet was treated with red blood cell lysis buffer and washed with PBS. The cell suspension was incubated with CDH16 antibody (Santa Cruz). CDH16-positive cells were magneti-cally isolated using anti-mouse IgG microbeads (Miltenyi
Biotec) following the manufacturer’s instruction.
RRBS
Genomic DNA from whole kidney was isolated using the DNeasy Kit (Qiagen). Libraries were generated from the iso-lated DNA using the Premium Reduced Representation
Bisul-fite Sequencing Kit (Diagenode) following the manufacturer’s
instruction. The 2100 Bioanalyzer (Agilent) and High Sensi-tivity DNA Kit (Agilent) was used for quality check. After trimming adapter and low-quality reads using Trim Galore version 0.5.0 (https://github.com/FelixKrueger/TrimGalore)
with the option“–rrbs,”Bismark version 0.19.121was applied
to align reads to the mouse genome (mm10). MethylKit
ver-sion 1.8.122was used to quantify the methylation level of CpG
sites covered by at leastfive reads and to calculate the
meth-ylation difference betweenDnmt3a/3b and control samples.
CpG sites with methylation difference .20% and qvalue
,0.01 were analyzed in edmr version 0.6.4.123to identify
dif-ferentially methylation regions (DMRs) with at least three
CpG sites and methylation difference.20%.
WGBS
Genomic DNA was isolated fromCdh16-positive cells using
the MagAttract HMW DNA Kit (Qiagen) according to the
manufacturer’s instruction. The concentration of DNA was
measured using the Quant-iT PicoGreen dsDNA Assay kit
(Life Technologies) following the manufacturer’s instruction.
DNA quality was checked on agarose gel. After trimming adapter and low-quality reads by Trim Galore, Bismark was used for alignment to the mouse genome (mm10),
dedupli-cation, and quantification of methylation level for each CpG
site. SMART version 2.2.824was used to performde novo
ge-nome segmentation with default thresholds. The fragments
with at least five CpG sites and methylation difference
.20% were identified as DMRs.
RNA Sequencing
Total RNA from whole kidneys were isolated using the RNeasy Mini Kit (Qiagen). RNA quantity and quality was analyzed on the 2100 Bioanalyzer (Agilent) using the RNA 6000 Pico Kit (Agilent). Sequencing reads were aligned to the mouse
ge-nome using STAR version 2.2.125and gene expression was
quantified using RSEM version 1.3.1.26Comprehensive gene
annotation (gencode.vM18.annotation.gtf ) was obtained
from GENCODE.27 Differentially expressed genes (DEGs)
were identified using edgeR version 3.24.3,28with the
thresh-olds of false discovery rate,0.001 and log2 fold change.1.
Functional enrichment of DEGs were performed using DAVID
database version 6.829and the modPhEA database.30
Functional Annotation of DMRs
We downloaded mouse kidney reference epigenomes,
includ-ing WGBS and histone modifications (embryonic day 14.5
[E14.5], E15.5, postnatal day 0 [P0], and 8 weeks old) from the ENCODE website (Y. He, M. Harihran, D. U. Gorkin, D. E.
Dickel, C. Luo, R. G. Castanon,et al., unpublished
observa-tions; D. U. Gorkin, I. Barozzi, Y. Zhang, A. Y. Lee, B. Li,
Y. Zhao,et al., unpublished observations). To define
regula-tory elements, mouse kidney chromatin states (E14.5, E15.5, E16.5, and P0) were downloaded from http://enhancer.sdsc. edu/enhancer_export/ENCODE/chromHMM/replicated/. These chromatin states were estimated by ChromHMM (D. U. Gorkin, I. Barozzi, Y. Zhang, A. Y. Lee, B. Li, Y. Zhao,
et al., unpublished observations) using a 15-state model.
ChromHMM uses a combinatorial pattern of histone modifi
-cations (H3K4me1, H3K4me2, H3K4me3, H3K27ac, H3K27me3, H3K9ac, H3K9me3, and H3K36me3). The
15-state model was further simplified into four states including
promoter, enhancer, transcription, and other. Chromatin states from multiple data sets were merged. BEDTools version
2.27.031was used to intersect chromatin states and DMRs.
Con KspCreDnmt3a/3b
20x
60x
Relative expression
3d 10d 21d 56d Proximal tubule
* *** **
Slc34a1 20
15
10
5
0
3d 10d 21d 56d Loop of Henle
Slc12a1
10 8 6 4 2 0 12
3d 10d 21d 56d Distal tubule
Slc12a3 6
4
2
0
3d 10d 21d 56d Collecting duct
Aqp2
3
1
0 4
2
Control
KspCreDnmt3a/3b
0.04
0.03
0.02
0.01
0.00
P=0.0008
P=0.082
Dnmt3a Dnmt3b
Relative expression
Control
KspCreDnmt3a/3b
0.5
0.4
0.3
0.2
0.1
0.0
Creatinine (mg/dL)
Con
Ksp Cre
Dnmt3a/3b
KspCreDnmt3a/3b+Veh KspCreDnmt3a/3b+Cisplatin
Con+Veh Con+Cisplatin
BUN(mg/dL)
P=0.012 P<0.0001
0 50 100 150 200
0 20 40 60
80 P<0.0001 P=0.037 Cd68
0 10
5 15 20
P=0.027 P=0.0002
Ccl2
Relative expression
P=0.020P=0.065
0 20 10 30 40 50 2500
1500 2000
500 1000
Kim1
P=0.012 P=0.005
0 2000 4000 6000 8000
100 200 300 400 500
Lcn2
Relative expression
Slc12a3
3
1 4
2
0
Relative expression
0.0 0.5 1.0 1.5
Slc12a1 P=0.001
Relative expression
0.0 0.5 1.0 1.5
Slc34a1
P<0.0001
KspCreDnmt3a/3b+Veh
KspCreDnmt3a/3b+Cisplatin Con+Veh
Con+Cisplatin
A
D
E
B
C
F
H
G
Figure 1. Phenotypic characterization ofKspCreDnmt3a/Dnmt3bmice. (A) Breeding scheme for generatingKspCreDnmt3a/3bdouble
knockout mice (KspCreDKO). (B) Transcript levels ofDnmt3aandDnmt3bin kidneys of 3-day-old control andKspCreDnmt3a/3bmice.
Data are represented as mean6SEM;Pvalue was calculated by two-tailedttest. (C) Serum creatinine measurement in 3-week-old
control and KspCreDnmt3a/3b mice. Data are represented as mean6SEM. (D) Representative images of haemotoxylin and
eo-sin–stained kidney sections of control andKspCreDnmt3a/3bmice. Scale bar: upper 20mm; bottom: 10mm. (E) Relative mRNA levels of
kidney segment markersSlc34a1,Slc12a1,Slc12a3, andAqp2in control andKspCreDnmt3a/3bmice on day 3, 10, 21 and 56. Data are
represented as mean6SEM. *P,0.05, **P,0.01, ***P,0.001 by two-way ANOVA withpost hocTukey test. (F) Serum BUN levels in
ChromHMM using the 15-state model using the combinato-rial patterns of chromatin immunoprecipitation followed by
high-throughput sequencing (ChIP-seq) profiles (H3K4me1,
H3K4me3, H3K36me3, H3K27me3, H3K27ac, CTCF, and RNA polymerase II) obtained from the mouse ENCODE
proj-ect.32Similarly, the 15-state model was simplified into four
states such as promoter, enhancer, transcription, and other. Chromatin accessibility of DMRs in kidneys of 8-week-old
male C57BL/6J mice were quantified by deepTools version
3.1.233 using bigwig files obtained from the Mouse
sci-ATAC-seq Atlas.34 Transcription factor motif enrichment
analysis of DMRs was performed usingfindMotifsGenome.pl
of HOMER version 4.10.3.35GREAT version 3.0.036was used
to predict functions ofcis-regulatory regions.
Six2Binding Sites in NPCs
Six2binding sites (ChIP-seq peaks) in NPCs were downloaded
from GUDMAP database (RID:Q-Y4CY).37Six2peaks
iden-tified in at least two of the three replicates were used for further
analysis. To build a control set, shuffled regions, matching in
size and number, were generated using BEDTools shuffle.
Topologically Associating, Domain-Constrained Map of Enhancer DMR-DEG Associations
We obtained a topologically associating domain-constrained map of enhancer-promoter associations from a reference (D. U. Gorkin, I. Barozzi, Y. Zhang, A. Y. Lee, B. Li, Y. Zhao,
et al., unpublished observations). BEDTools was used for in-tersect analysis. The Spearman correlation test was performed to examine the relationship between enhancer DMR (eDMR) methylation and DEG expression.
Enrichment of Genome-Wide Association Study Single-Nucleotide Polymorphisms in Developmental
DMRs andDnmt3aandDnmt3bDouble Knockout
DMRs
Genome-wide association study (GWAS) single-nucleotide polymorphisms (SNPs) were obtained from the GWAS catalog
(gwas_catalog_v1.0-associations_e96_r2019–06–20.tsv).38
Afterfiltering out SNPs for missing coordinates and signifi
-cance (P,531028), 79,744 SNPs were used for follow-up
analysis. The University of California Santa Cruz (UCSC) Ge-nome Browser LiftOver function was used to lift over the
mouse coordinates into human coordinates,39resulting in a
final set of 36,045 GWAS SNPs. The hypergeometric test was
used to determine the significance of disease trait and
enrich-ment of DMRs.
SNPs showing a significant association with eGFR were
downloaded from recent publications.9,40,41 The significant
(P,531028) SNPs from different studies were combined
(26,637). The human SNP coordinates were converted to mouse genome (mm10) using LiftOver from the UCSC
Ge-nome Browser,39resulting in afinal set of 7923 eGFR SNPs.
Quantification and Statistical Analysis
Statistical analyses were performed using R or GraphPad Prism software (GraphPad Software, La Jolla, CA).
Two-tailed t test or Wilcoxon signed rank sum test was used to
compare two groups. One-way ANOVA was used to compare multiple groups. Spearman rank correlation was used to de-termine the correlation. When needed, multiple testing cor-rection was performed using the false discovery rate.
Data Availability
All sequencing data (WGBS, RRBS, and RNA sequencing) have been deposited in the National Center for Biotechnology
Information’s Gene Expression Omnibus (GEO) and are
ac-cessible through GEO accession number GSE134267.
RESULTS
Deletion ofde NovoMethyltransferasesDnmt3aand
Dnmt3bin Renal Progenitors
To understand the role ofde novomethyltransferases in kidney
development and maturation, we generated double knockout
transgenic mice by crossing the Dnmt3af/f mice and the
Dnmt3bf/f mice with KspCre mice (KspCreDnmt3a/3b or
KspCreDKO) (Figure 1A). KspCre mice express Cre
recombi-nase under the control of the mouse cadherin 16.Chd16or
Ksp-cadherin is expressed from E11.5 in the developing kid-ney (Supplemental Figure 1A). Its expression increases as the
kidney matures (Supplemental Figure 1B). Cell type–specific
expression and open chromatin data indicated that it is ex-pressed in distal tubules, collecting duct, loop of Henle, and proximal tubules in adult kidney (Supplemental Figure 1, C
and D). KspCreDnmt3a/3bmice were born at the expected
Mendelian ratio. The genetic deletion was confirmed by
quan-titative RT-PCR in kidneys of 3-day-old mice (Figure 1B).
Transcript levels of Dnmt3a and Dnmt3b were lower in
KspCreDnmt3a/3bmice (Figure 1B); however, as reported
ear-lier, the expression ofDnmt3bwas around the detection limit
at birth (Figure 1B).
KspCreDnmt3a/3b mice showed no obvious renal pheno-typic alterations at baseline. Serum creatinine level of
KspCreDnmt3a/3b was comparable to littermate controls (Figure 1C). Kidney structural analysis showed no observable abnormalities (Figure 1D). To further understand the role of
one-way ANOVA. (G) Relative mRNA level of AKI markers (Kim1andLcn2) and cytokines (Cd68andCcl2). Data are represented as
mean6SEM;Pvalue was calculated by one-way ANOVA. (H) Relative mRNA level of kidney segment markers such asSlc34a1,Slc12a1,
Dnmt3aandDnmt3bin renal development, we quantified the
expression of renal segment–specific markers (Figure 1E).
Slc34a1 (proximal tubule),Slc12a1(loop of Henle), Slc12a3
(distal tubule), andAqp2(collecting duct) were quantified on
day 3, 10, 21, and 56. Gene expression levels showed minor alterations in the developing and maturing kidneys, however
they were overall similar between adultKspCreDnmt3a/3bmice
and littermate controls (Figure 1E).
Ksp
Cre
Hypo-DMRs
Six2
Cre
Hypo-DMRs
E
Control ControlKspCre DKOSix2Cre
DKO
3-week-old
Kidney
Reduced representation bisulfite sequencing
(RRBS)
Loss of methylation in Dnmt3a/3b
Both n=196
Methylation difference (Six2Cre DKO – Control) 7,184 Six2Cre DKO DMRs
including 20,953 DMCs
Loss 6,777 94.3%
Gain 407 5.7%
Loss 1,039 71.3%
Gain 418 28.7%
Loss 1,546 30.9%
Gain 3,464 69.1% 1,457 KspCre DKO DMRs
including 3,045 DMCs
5,010 Developmental DMRs including 13,194 DMCs
-log10 (q value)
Methylation difference (KspCre DKO – Control)
Methylation difference (P21 – P0) -1.0 -0.5 0 0.5 1.0 -1.0 -0.5 0 0.5 1.0
-1.0 100
75
50
25
0
-0.5 0 0.5 1.0
Fraction of Hypo-DMRs
10%
36%
6%
48%
27% 12% 32% 29% 100%
80%
60%
40%
20%
0%
Hyper DMCs
D
Hypo DMCs No DMCs No CpGs detected Developmental DMC
A
C
Six2CreOnly
Hypo-DMRs n=6,581 KspCreOnly Hypo-DMRs
n=843
#
* *
& *
* *
* *
DNA methylation
0.0 0.25 0.5 0.75 1.0 Both Hypo-DMRs
n=196
Fetal Control (Week 3 )
Ksp
Cre Dnmt3
a/b
Six 2
Cre Dnmt3a/b Control
(Week 3) Fetal
Six2Cre
DKO KspCre
DKO
E14.5 E15.5 E16.5 P0
Mean methylation
1.0
0.5
0.0
1.0
0.5
0.0
1.0
0.5
0.0
B
KspCre Only n=843
Six2Cre
Only n=6,581
Figure 2. Dnmt3aandDnmt3bplay important roles in establishing methylation patterns during kidney development. (A) Schematic
representation of RRBS on whole-kidney lysates of KspCreDnmt3a/3b, SixCreDnmt3a/3b and littermate controls. (B) Volcano plot
showing methylation changes in (DMR or differentially methylated cytosines [DMCs]) in SixCreDnmt3a/3b (left),KspCreDnmt3a/3b
(middle), and development (right). Thexaxis shows mean methylation differences and theyaxis shows statistical significance (as
2log10[qvalue]). (C) The overlap between DMRs identified inSixCreDnmt3a/3b, KspCreDnmt3a/3band during kidney
develop-ment (P0–P21). (D) Overlap of hypo-DMRs identified inKspCreDnmt3a/3bandSixCreDnmt3a/3b. (E) Methylation changes during
mouse kidney development (E14.4, E15.5, E16.5, and P0), followed by methylations inKspCreDnmt3a/3b,SixCreDnmt3a/3band
littermate controls (left). Mean methylation levels from zero (blue) to one (red); mean methylation levels of DMRs in fetal, littermate
controls,KspCreDnmt3a/3bandSixCreDnmt3a/3bkidneys (right). Mean methylation difference between control kidney and fetal,
KspCreDnmt3a/3b and SixCreDnmt3a/3b kidney were compared by Wilcoxon signed rank sum test. #P53.831026, &P50.84,
Mouse phenotype
Hypo-DMRs
Hyper-DMRs
Abn. kidney cortex morphology
Abn. oogenesis
Abn. nerve conduction
Enhanced behavioral response to alcohol
Decreased nerve conduction velocity
Polyphagia
-Log10(FDR)
0 10 20 30
Abn. renal corpuscle morphology
Abn. renal glomerulus morphology
Abn. urinary system development
Abn. kidney development
Chromatin states
Others Transcript Enhancer Promoter
Fraction of DMRs/non-DMRs
overlapped with chromatin states
Hypo-DMRHyper-DMRNon-DMRHypo-DMRHyper-DMRNon-DMR Fetal
Kidney states
Adult Kidney
states
100%
80%
60%
40%
20%
0%
All CpGs Fetal Promoter Adult Promoter Fetal Enhancer Adult Enhancer
Global CpG methylation
0.0 0.2 0.4 0.6 0.8 1.0
E14.5
ESC E15.5 E16.5 P0 Week 3 Week 8
DKO
D
E
G
38%
16% 18% 10% 7% 13%
E14.5
ESC E15.5E16.5P0AdultControlDKO E15.5P0 W8 E15.5P0 W8 E15.5P0 W8 PT_1 PT_2 PT_3 PT_5 PT_S3DCT DCT_CDCD LOH_2LOH_3GlomerularPodocytes
Hypo-DMRs (5,686) overlapped with kidney enhancers
DNA methylation H3K27ac H3K4me1 Accessibility Chromatin accessibility by single cell ATAC-seq
Fetal enhancer
Only (3,321, 58.4%)
Fetal & adult Enhancer
(1,864, 32.8%)
Adult enhancer
Only (501, 8.8%)
DNA
methylation RPKM
(ChIP - Input)
-5 0 5 10 15
-5k DMR
5k -5k DMR
5k -1.5k DMR
1.5k -1.5k DMR
1.5k
0.0 0.25 0.5
ATAC-seq signal (RPKM)
0.0 0.25 0.5
0.75 1.0
F
Control 3-week-old
Cdh16 antibody
MACS column
Whole genome
bisulfite sequencing
(WGBS) Kidney cell
suspension and antibody
incubation
Cdh16+
cell sorting Kidney
KspCre
Dnmt3ab DKO
A
Log2 number of CpGs in segment
10
8
6
4
2
0 Hypo-DMRs 13,276
980,525 fragments (>=5 CpGs) by SMART
Methylation difference in Cdh16+ cells (DKO - Control)
Hyper-DMRs 4,302
-1.0 -0.5 0.0 0.5 1.0
B
CpG methylation DKO Cdh16
+ cells
0.0 0.2 0.4 0.6 0.8 1.0
DMRs after Dnmt3ab DKO (n=17,578)
CpG methylation Control Cdh16+ cells
0.0 0.2 0.4 0.6 0.8 1.0
C
4,302
13,276
Figure 3. Base-resolution methylome analysis of isolated tubule cells in control andKspCreDnmt3a/3bmice. (A) Experimental design.
(B) Volcano plot,xaxis shows differentially methylated regions (20% change in at leastfive CpG of any fragment) andyaxis shows
the number of DMRs. (C) Methylation level of 17,578 DMRs inCdh161cells in control (yaxis) andKspCreDnmt3a/3b(xaxis). Color
Next, we analyzed whether KspCreDnmt3a/3b mice show alterations in response to injury. To model AKI, we treated
control and KspCreDnmt3a/3b mice with cisplatin. Serum
BUN level of cisplatin-treated mice was significantly
in-creased, confirming the injury. Serum BUN level was lower
in KspCreDnmt3a/3b mice when compared with cisplatin-treated controls (Figure 1F). Tubule injury markers such as
kidney injury molecule 1 (Kim-1) and Lipocalin-2 (Lcn2) were
lower in KspCreDnmt3a/3b mice compared with controls
(Figure 1G). The macrophage marker Cd68and cytokine
Ccl2were higher in cisplatin-treated animals but they were
less prominent inKspCreDnmt3a/3bmice (Figure 1G). Kidney
segment–specific marker genes, such as the proximal tubule
marker Slc34a1 and loop of Henle marker Slc12a1 were
decreased after cisplatin treatment and these changes were
comparable inKspCreDnmt3a/3bmice (Figure 1H). We also
analyzed the injury response in the folic acid–induced kidney
fibrosis model. We did not observe significant differences
be-tween control and KspCreDnmt3a/3b mice in the folic
acid–induced kidneyfibrosis model (Supplemental Figure 2).
BecauseChd16expression is segment specific and expresses
at later stages of development, we next genetically deleted
Dnmt3aandDnmt3bin nephron progenitors using theSix2Cre
mice (Six2CreDnmt3a/3b or Six2CreDKO) (Supplemental
Figure 3A).Six2is expressed at E11.5, and it labels the
self-renewing nephron progenitors that give rise to all nephron
epithelia5(Supplemental Figure 1, A, B, and D).Six2CreDnmt3a/
3bmice were born at the expected Mendelian ratio. Kidney
sec-tions of 3-week-oldSix2CreDnmt3a/3bmice showed no obvious
structural abnormalities (Supplemental Figure 3B). Transcript
levels of kidney segment markers (Slc34a1,Slc12a1,Slc12a3, and
Aqp2) in 3-week-oldSix2CreDnmt3a/3bmice were comparable
to littermate controls (Supplemental Figure 3C). Taken together,
Dnmt3aandDnmt3bappeared dispensable in adult kidney tu-bule cells at baseline. However, they seem to confer some
resis-tance to AKI, but not to kidneyfibrosis.
Dnmt3aandDnmt3bPlay Important Roles in
Establishing Methylation Patterns during Kidney Development
To examine the contribution ofDnmt3aandDnmt3bto
kid-ney cytosine methylation changes during development, we
performed RRBS on whole-kidney samples obtained from
3-week-old control andKspCreDnmt3a/3band Six2CreDnmt3a/
3b mice (Figure 2A). On average, RRBS quantified
methyl-ation levels of approximately 2.4-million CpG sites represent-ing approximately 10.8% of CpG sites in the mouse genome.
Using stringent criteria (of qvalue,0.01 and methylation
difference .20%), we identified 7184 DMRs in kidneys of
Six2CreDnmt3a/3b mice and 1457 DMRs in kidneys of
KspCreDnmt3a/3bmice (Figure 2B). As expected, most regions
showed lower methylation levels inDnmt3a/3bknockout mice
(94.3% of the DMRs in Six2CreDnmt3a/3b and 71.3% in
KspCreDnmt3a/3bmice) (Figure 2B), indicating thatDnmt3a
andDnmt3bplay a key role inde novomethylation. Regions
that failed to gain methylation in the Dnmt3ab/3b double
knockout kidneys were mostly fetal kidney enhancers (Supplemental Figure 4).
Next we examined global methylation patterns at birth (P0) and at 3 weeks of age (P21) (Figure 2B), which is the time when nephron development ceases in mice. As reported
ear-lier, there was a significant gain in CpG sites methylation
(69.1%) during this period (P0–P21). This is consistent with
the terminal differentiation of cells during this period.
Develop-mental DMRs failed to gain methylation inDnmt3a/3b
knock-out mice (chi-squared testP51.05310233; Figure 2C). This
effect was more pronounced in theSix2CreDnmt3a/3banimals
compared with KspCreDnmt3a/3b mice. There was a limited
overlap between hypo-DMRs observed inSix2CreDnmt3a/3b
and KspCreDnmt3a/3b mice (Figure 2, D and E), indicating
that the targets ofDnmt3aandDnmt3bare spatially and
tem-porally different. In summary, cytosine methylation was severely
reduced in kidneys of mice lackingde novomethyltransferases
such asSix2CreDnmt3a/3bandKspCreDnmt3a/3b.
Dnmt3aandDnmt3bAre Necessary forde Novo
Methylation of Fetal-Specific Enhancers
For accurate cell type–specific, base-resolution methylome
analysis, we performed WGBS on sortedCdh16
(Ksp-cadherin)-positive cells (Figure 3A). Cells were isolated from kidneys of
3-week-oldKspCreDnmt3a/3band control mice. The methylation
patterns of Cdh16-positive cells showed strong concordance
with whole-kidney methylation (Supplemental Figure 5A). To
identify DMRs, wefirst segmented the genome into 980,525
KspCreDnmt3a/3bbased on GREAT for functional annotation. (E) The overlap between hypo-DMRs and hyper-DMRs inKspCreDnmt3a/
3bmice and different chromatin states in fetal and adult kidneys. Chromatin states from fetal and adult kidneys were used to classify
the genome location of each segment by WGBS. (F) Functional annotation of hypo-DMRs (5686, lost methylation inDnmt3a/3b
knockout mice). First panel showed the number and fraction of hypo-DMRs overlapped with fetal and/or adult enhancers. Second
panel shows the methylation levels of hypo-DMRs during kidney development, control, andKspCreDnmt3a/3b. The data were ordered
according to the methylation level in control kidneys. Third panel shows density of chromatin mark (associability, H3K27ac, and
H3K4me1) in each hypo-DMR and itsflanking regions (5 kb) in fetal kidney (E15.5, P0) and adult kidney (8 weeks old). The last panel
shows the chromatin accessibility by single cell assay for transposase-accessible chromatin using sequencing single cell (ATAC-seq) in adult kidney (8-week-old) cell types. (G) Global CpG methylation levels in kidney enhancers and promoters at different stages of
development and inDnmt3a/3bknockout mice. Abn., abnormal; CD, collecting duct; DCT, distal tubule; DKO, double knockout; ESC,
Fetal
A
Motif Hoxc9 Six2 PBX2 Six1 Dlx3 NF1 HNF1b EWS:ERG PU.1 Etv2Gain methylation in adult (23,640)
Lose methylation in adult (23,293) Type Homeobox 1.00E-287 1.00E-188 1.00E-136 1.00E-132 1.00E-131 1.00E-181 1.00E-102 1.00E-43 1.00E-26 1.00E-24 Homeobox Homeobox Homeobox Homeobox CTF ETS ETS ETS Homeobox P Consensus Developmental DMRs Adult Hoxc9 Six2 Dlx3 Lhx1 CDX4 PBX2 Six1 Lhx3 Lhx2 Pdx1 Homeobox Homeobox Homeobox Homeobox Homeobox Homeobox Homeobox Homeobox Homeobox Homeobox
Motif Type Consensus P
DMRs gaining methylation in adult kidney and losing methylation in
DKO kidney cells (n=1,951)
1.00E-30 1.00E-25 1.00E-23 1.00E-22 1.00E-20 1.00E-19 1.00E-17 1.00E-15 1.00E-13 1.00E-13
Fetal Adult DKO
Hoxc9 Six2 PBX2 Six1 Dlx3 Gata6 EWS:ERG ETS PU.1 ELF3 Etv2
Gain methylation in DKO (4,302) Lose methylation in DKO (13,276)
Homeobox Homeobox Homeobox Homeobox Homeobox ETS Zf ETS ETS
Motif Type Consensus P
DKO DMRs 1.00E-63 1.00E-44 1.00E-34 1.00E-24 1.00E-16 1.00E-16 1.00E-10 1.00E-09 1.00E-06 1.00E-05 Control DKO
B
C
p < 2.2e-160.3 0.6 0.0 -0.3 -0.6 Methylation difference
(adult kidney – fetal kidney)
p < 2.2e-16
NPC Six2 peak Shuffled regions NPC Six2 peaks Shuffled regions 0.6 Methylation difference
(DKO - Control)
0.3
0.0
-0.3
-0.6
Methylation difference (adult kidney – fetal kidney) NPC Six2 peaks Shuffled regions
Percentage of segments overlapped with enhancer 100% 80% 60% 40% 20% 0% Fetal enhancer only Fetal & adult enhancer Adult enhancer only Not kidney enhancer
D
0.6 0.3 0.0 -0.3NPC Six2 peaks
Methylation difference
(DKO - Control) 56.1%
-0.6
Shuffled regions
26.9%
-0.6 -0.3 0.0 3.0 6.0 -0.6 -0.3 0.0 3.0 6.0
F
E
G
NPC Six2 peaks Shuffled regions
ESC
E14.5 E15.5 E16.5 Week 3 Week 8
ESC
E14.5 E15.5 E16.5 Week 3 Week 8
P0 P0 1.00 0.75 0.50 0.25 0.00
Global CpG methylation
>100kb
Enhancer
2,946 bp 122,489 bp
Hypo-DMR CpG island Gene WGBS H3K27ac Six2 E15.5 P0 Adult Control DKO E15.5 P0 Adult NPC NPC NPC Promoter
NPC Six2 peaks Shuffled regions
Fetal enhancer only Adult enhancer only Fetal & adult enhancer 1.0 0.5 -0.5 -1.0 0.0
Correlation between Six2
expression and
CpG methylation
8.9e-16 1.1e-09 0.073
Figure 4. KspCreDnmt3a/3b DMRs are enriched for developmental transcription factor binding. (A) Transcription factor motif
en-richment (HOMER) of DMRs, during kidney development (P0–P21), in control versusKspCreDnmt3a/3bkidneys. Fragments that gained
methylation during development but failed to gain methylation inDnmt3a/3bknockout mice. (B) The degree of overlap between
DMRs that are in enhancer regions and on Six2 binding sites in nephron progenitors (NPC) from ChIP-seq (cyan) compared with
background (gray). (C) Methylation differences in control versusKspCreDnmt3a/3bkidneys (left) or adult versus fetal kidneys (right) that
overlap withSix2binding versus background shuffled region. (D) Scatter plot of methylation changes ofSix2binding sites in NPCs,
fragments, each of them containing at leastfive CpG sites, as
established in the SMART method (Supplemental Figure 5B).24
The differential methylation analysis identified 17,578 fragments
(DMRs) showing at least 20% change in methylation between
KspCreDnmt3a/3bmice and controls (Figure 3B, Supplemental
Table 2). Consistently,.75% (13,276) of DMRs showed a lower
methylation level (hypo-DMRs) in kidneys ofKspCreDnmt3a/3b
mice. Loci with intermediate (40%–80%) methylation level were
mostly affected in theKspCreDnmt3a/3bmice (Figure 3C).
Func-tional annotation of these DMRs indicated that demethylation events tended to be close to genes associated with morphogen-esis and kidney development (Figure 3D).
To define regions that are specifically altered byDnmt3a
-and Dnmt3b-mediated differential methylation, we mapped
DMRs to functional regulatory elements. Kidney-specific
functional regulatory elements such as enhancers and pro-moters were annotated by integrating multiple histone ChIP
data obtained from fetal and adult mouse kidney samples.42
We found thatDnmt3a- andDnmt3b-mediated DMRs were
enriched on enhancers, but hardly ever observed on promoter regions (Figure 3E, Supplemental Figure 5C). When we com-pared fetal and adult kidneys, we found that loci with lower
methylation (hypo-DMRs) inDnmt3a/3bknockout mice were
strongly enriched on fetal-enhancer regions (38% versus 13%,
chi-squared testP,2.2310216), and to lesser degree on
re-gions identified as transcribed regions in fetal kidneys
(Figure 3E).
Next, we wanted to understand the fate of the fetal en-hancers and the role of methylation. Upon analyzing cytosine methylation on a genome-wide scale during kidney develop-ment, we found that adult promoters and enhancers showed a mild, gradual decline in global methylation level, indicating their openness was determined earlier (Figure 3G). On the
other hand, fetal enhancers and promoters gained significant
methylation during the P0–P21 time period. Kidneys of
Dnmt3a/3bknockout mice failed to gain methylation,
consis-tent with their role asde novomethyltransferases (Figure 3G).
To further explore the function ofDnmt3a/3bin kidney
development, we identified 3334 DMRs that showed
methyl-ation changes during development and also showed changes inDnmt3a/3bknockout mice. Most (58.5%) of these shared
DMRs underwentde novo methylation in development but
failed to increase their methylation inDnmt3a/3bknockout
mice, and were enriched in fetal-enhancer regions (Supplemental Figures 5, C and D, and 6). When focused our analysis on the hypo-DMRs that overlapped with kidney
enhancers, we found that 58% were fetal specific whereas only
a small fraction (8.8%) of hypo-DMRs were annotated as en-hancers only in the adult mouse kidney (Figure 3F, Supplemental Figure 5C). Hypo-DMRs gained methylation, losing enhancer marks and chromatin accessibility in the adult kidney (Figure 3F, Supplemental Figure 5, E and F). These
results revealed Dnmt3aand Dnmt3b were necessary forde
novo methylation and decommissioning of fetal-specific
enhancers.
Dnmt3aandDnmt3bAre Required for
Decommissioning of Fetal Enhancers Bound by Developmental Transcription Factors
Because DMRs were enriched on developmental-enhancer re-gions, we were interested to understand whether we could identify critical transcription factors associated with these sites. We used transcription factor motif analysis established
by HOMER35to analyze DMRs from the WGBS data set. Upon
comparing regions that gained methylation during develop-ment, we found a measurable enrichment for homeobox
tran-scription factors, includingSix2which is known to play a key
role in kidney development (Figure 4A). Through examining
DMRs identified in kidneys of KspCreDnmt3a/3b mice, we
again found enrichment for homeobox transcription factors,
such asSix2binding sites (Figure 4A). The overlap between
the developmental DMRs and Dnmt3a/3bknockout DMRs
showed significant enrichment for kidney developmental
transcription factor binding sites including Hoxc9, Six2,
Six1, andDlx3(Figure 4A).Six2, a kidney developmental
tran-scription factor, is required for nephron progenitor
mainte-nance.7,43These results indicate that a good portion of fetal
enhancers that are decommissioned byDnmt3a/3bare bound
bySix2.
Because motif analysis cannot distinguish among closely
related transcription factors, we next specifically examined
Six2 binding sites identified in NPCs by ChIP-seq.37 The
Six2peaks in NPCs were affected by methylation because
they were enriched for fetal hypo-DMRs (Figure 4B)
com-pared with a shuffled background. More than half (56%) of
the fragments that overlapped with NPCSix2 peaks gained
methylation in development but failed to be methylated in
KspCreDnmt3a/3b mice (Figure 4, C and D, Supplemental Figure 7A). These fragments were localized to genes that are
known to play roles in kidney development. For example,Six2
peaks in nephron progenitors overlapped with the distal
en-hancer of Pax2 which is a critical regulator of kidney
methylation in adult kidneys, but not inKspCreDnmt3a/3bmice. (E) IGV genome browser view of thePax2region (chromosome 19:
44729363–44851852). DNA methylation changes in hypo-DMRs (lower methylation inKspCreDnmt3a/3b), note the methylation pattern
in the developing mouse kidney (E15.5, P0), wild-type and Dnmt3a/3b knockout (DKO) mice, H3K27ac enhancer mark, andSix2
binding. Note the failure to increase in methylation of a fetal-enhancer region. (F) Global CpG methylation in NPCSix2 peaks and
shuffled regions at different stages of kidney development. (G) Correlation between gene expression ofSix2and CpG methylation of
segments overlapped with NPC Six2 peaks and shuffled regions. Wilcoxon signed rank sum test was carried out to calculate the
significance of difference between NPCSix2peaks and shuffled regions, andPvalues was provided for each comparison. ESC,
B
D
A
C
F
E
1815
12
-log10(FDR)
9
6
3
0
-10 -5
Log2 transformed fold change
0 5 10
Down-regulated (27) Up-regulated (122)
3-week-old Kidney
RNA sequencing
(RNA-seq)
eDMR-DEG associations
8
6
4
2
0
2
-4
-0.4 0.4
Pdia2
Tiam1
Spearman’s rho
Elf3
Methylation difference (DKO - Control)
Expression difference Log2 (DKD/Wild type)
0
31 eDMR-DEG associations (31 eDMRs ~ 21 DEGs) Spearman’s rho = -0.4, p = 0.02
0.8
0.6
0.4
0.2 24
16
8
0
Tiam1 expression
(TPM)
CpG Methylation
E14.5 E15.5 E16.5
W
eek 3 Control Week 3 DKO
P0
Spearman’s rho = -0.84 p-value = 0.0027 DKO Hypo-DMR
Gene
E15.5
WGBS
RNA-seq Chromatin state
P0
P0 Adult
Adult Control
Control
Control DKO
DKO
DKO
chr16:89705992-90036540 chr16:89943409-89945895
E15.5
Biological processes
Mouse phenotype
Enhancer
Promoter DMR
Positive
Negative
0.2 0.4 0.6 0.8
Tiam1 chr16:89944403-89944892 Epithelial cell differentiation
Programmed cell death
Collecting duct development Kidney development Wound healing Alpha-amino acid catabolic process Cellular amino acid catabolic process Reponse to wounding Anion transport
Epithelium development Up
Down
-Log10(p value)
0 2 4 6
Renal/urinary system phenotype Abnormal renal/urinary system physiology Abnormal urine osmolality Abnormal renal/urinary system morphology Abnormal kidney physiology Abnormal thrombosis Heart left ventricle hypertrophy Decreased t cell proliferation Homeostasis/metabolism phenotype Abnormal blood homeostasis
Up
Down
TAD-constrained map of enhancer-promoter associations Control
KspCre DKO
Figure 5. Dnmt3a- andDnmt3b-mediated methylation represses developmental genes in late development stage. (A) Schematics of
the experiments. (B) Gene expression changes inKspCreDnmt3a/3bmice. Volcano plot;xaxis shows fold-change difference, andyaxis
shows statistical difference (2log false discovery rate [2logFDR]). Red dots represent genes with higher expression inKspCreDnmt3a/
3b, whereas blue dots represent genes with lower expression. (C) Function enrichment (biologic processes and mouse phenotype)
analysis of differentially regulated genes. Red colors represent genes with increased expression, whereas blue represent decreased
expression. (D) Enhancer and promoter associations obtained from topologically associating domain (TAD)–constrained maps. eDMRs
represent enhancers that also showed differential methylation. The association (Spearman rank correlation coefficient) between eDMRs
development.44This enhancer showed low methylation levels
in the fetal kidney and its methylation level increased in the adult kidney (Figure 4E, Supplemental Figure 7A). In contrast to the wild-type mice, this enhancer failed to gain methylation
in kidneys of theKspCreDnmt3a/3bmice and the regions
re-mained similar to those in fetal kidneys. Furthermore, chro-matin conformation capture contact matrices revealed
interactions between this enhancer and the Pax2 locus
(Supplemental Figure 7B). Integrative analysis revealed that
Six2binding sites had lower methylation levels in fetal kidneys
and gained methylation after birth (Figure 4F), indicating an
interaction between Six2 binding and methylation changes.
For example, pioneering transcription factors not only play roles in opening closed chromatin sites during development, but their binding footprint can be observed even after the
transcription factor is no longer expressed.45To explore this
hypothesis, we calculated the correlation of fragment
methyl-ation andSix2 expression during kidney development. The
Six2-bound fragment methylation showed negative
correla-tion with Six2expression, which was particularly obvious
for fetal enhancers (Figure 4G). For example, methylation of
a locus on chromosome 7 was significantly negatively
corre-lated (Spearmanr520.97,Pvalue50.0002) with the
expres-sion ofSix2 (Supplemental Figure 8). This region included
Six2binding sites that failed to gain methylation inDnmt3a/
3bknockout mice. These results indicated that DNA
methyl-ation, mediated by Dnmt3a and Dnmt3b, preferentially
affected enhancer regions bound bySix2during kidney
devel-opment. The methylation of these sites correlated withSix2
expression.
Transcriptional Changes Observed inKspCreDnmt3a/
3bMice
Next, we performed unbiased gene expression analysis by
RNA sequencing of kidneys of control andKspCreDnmt3a/3b
mice (Figure 5A). We identified 149 genes that passed the
significance threshold for differential expression (DEGs).
Consistent with the demethylation events, most (82%, 122/ 149) DEGs showed an increase in their expression in
KspCreDnmt3a/3bmice, including several kidney
developmen-tal genes such as Wnt4and Wnt9b (Figure 5B). Functional
annotation indicated that DEGs were enriched for kidney de-velopment and epithelial differentiation functions (Figure 5C).
Next, we tested whether enhancer methylation correlates with gene expression changes. To identify targets of enhancer DMR (eDMR), we obtained topologically associating, domain-constrained maps for accurate enhancer-promoter
associations (D. U. Gorkin, I. Barozzi, Y. Zhang, A. Y. Lee,
B. Li, Y. Zhao,et al., unpublished observations). We
identi-fied 31 associations between 31 eDMRs and 21 DEGs
(eDMR~DEG). The methylation changes in these eDMRs were associated with changes in their target gene expression (Figure 5D, Supplemental Figure 9A, Supplemental Table 3). Most of the eDMR-DEG associations (84%) were direction consistent, such as lower methylation was associated with higher expression (Figure 5D). For example, an eDMR on the Tiam1 locus showed lower methylation in
KspCreDnmt3a/3bmice and an increase in transcript
expres-sion ofTiam1(Figure 5, E and F, Supplemental Figure 9B).
Tiam1 was reported to play a role in Wnt signaling and
epithelial-mesenchymal transition,46,47 and it is mostly
si-lenced in adult kidney tubules. Overall, the effect of
Dnmt3a- and Dnmt3b-mediated methylation changes on gene expression modulation was modest, and mostly af-fected genes involved in kidney development.
Dnmt3a- andDnmt3b-Mediated Methylation Changes
Are Enriched for Kidney Disease Genetic Risk Loci
GWASs have identified nucleotide variations that are
en-riched in patients with CKD. Previously, we showed that a good portion of such loci are enriched on enhancer regions
in the adult kidney,10however, more than half of GWAS loci
remain unannotated. Here, we hypothesized that kidney
disease risk loci might be specific to the developmental
stage, i.e., might be active in the fetal tissue, explaining
the lack of regulatory annotation in the adult human kid-ney. As we showed earlier, such fetal enhancers are
specif-ically methylated and decommissioned by Dnmt3a/3b in
adult kidney.
As a first step, we overlapped the entire human GWAS
catalog and DMRs identified during development. We found
variants associated with kidney function (Figure 6A) and
other kidney-associated traits were specifically localized to
kidney developmental DMRs. Next, we narrowed the DMRs
only those were found to be methylated by Dnmt3aand
Dnmt3b. These DMRs showed a strong enrichment for
kid-ney function–associated traits (Figure 6A). Functional
an-notation revealed that 68% of DMRs that overlapped with
kidney function–associated SNPs were fetal kidney–specific
enhancers, which was significantly higher than expected by
chance (chi-squared test P51.0831021 2; Figure 6B,
Supplemental Figure 10A, Supplemental Table 4). Cross-species comparison indicated these fetal kidney enhancers were conserved between mouse and human, both in se-quence and in their methylation patterns (Supplemental
in kidney development and inKspCreDnmt3a/3bmice. Transcripts per million (TPM) was used for quantification of RNA expression.
Spearman rank correlation coefficient and significance was calculated and showed. (F) IGV genome browser view of theTiam1locus
(chromosome 16: 89944403–89944892), including WGBS tracks during kidney development E15.5–P0, chromatin state and gene
expression by RNA sequencing in adult andKspCreDnmt3a/3bmice. The right panel is a zoom-in region of the eDMR. DKD, diabetic
SNPs associated with eGFR (eGFR-SNPs)
Wuttke
13,327
553
9 17 5 4 83 65 1,418
3,510 3,124 4,934
13
311Morris Combined 26,637
eGFR-SNPs
7,923 (29.7%) eGFR-SNPs conserved in mouse
4,886 (61.7%) eGFR-SNPs overlapped with 2,901 WGBS segments Lift over to Mouse (mm10)
Overlap with WGBS segments Hellwege
Fetal enhancer
Development DMR
2,901 WGBS segments
DKO DMR
Kidney function-related traits based on GWAS
catalog
DKO DMRsDevelopment DMRs
Blood urea nitrogen levels Glomerular filtration rate (creatinine)
Glomerular filtration rate in non diabetics (creatinine) Hemoglobin
Hemoglobin levels
Idiopathic membranous nephropathy Renal function-related traits (eGRFcrea) Renal function-related traits (sCR) Serum alkaline phosphatase levels Thyroid stimulating hormone levels Vitamin D levels
-log10(p-value)
3
0 6
DMRs overlapped with GWAS SNPs associated with kidney function-related traits
Spearman’s rho = -0.58, p = 0.0018
Fetal enhancer only Fetal & adult enhancer Adult enhancer only Non enhancer
Methylation difference
in human chronic kidney disease
(DKD - Control)
Methylation difference in mouse kidney development
(Adult - Fetal) HNF1A
BCAR3 A1CF
GNAS
BCAS3
UNCX 0.2
0.1
0.0
-0.1
-0.2
-0.6 -0.3 0.0 0.3 0.6
Kidney development
Speraman’s rho = –1.0 p-value = 0.0028
0.8
0.6
0.4
0.2
0
Uncx
chr7:24389472-24389973
Six2 expression
(TPM)
CpG methylation
120
100
80
60
40
20
0 140
E14.5 E15.5 E16.5 P0
Week 3 Week 8
Fraction of enhancers
overlapped with eGFR-SNPs
900
750
600
450
300
150
0 Neither K4me1 K27ac K27ac&K4me1
49%
23%
E15.5 Week 8
Fraction of DMRs
overlapped with eGFR-SNPs
200
150
100
50
0 Other Transcript Enhancer Promoter
51% 34%
Fetal Adult
Fraction of enhancers overlapped with GWAS SNPs associated with kidney function-related traits
50
40
30
20
10
0 Neither K4me1 K27ac K27ac&K4me1
59%
28%
E15.5 Week 8
Fraction of DMRs overlapped with GWAS SNPs associated
with kidney function-related traits
Other Transcript Enhancer Promoter
39% 68%
Fetal Adult 25 30
15
10
5
0 20
B
G
H
C
E
F
A
D
Figure 6. Dnmt3a/3b-methylated regions harbor kidney disease risk loci. (A) Enrichment of human disease risk loci (obtained from
GWAS catalog) in developmental and Dnmt3a/3b double knockout DMRs. Only the significantly enriched human kidney
dis-ease–related traits were shown, and the significance was represented by color from white to dark blue. (B) Genomic location of DMRs
overlapping with kidney disease risk variants. Chromatin states from fetal and adult kidneys were used to classify the genome location
of each DMR. (C) Histone-modification transition from fetal mouse kidney to adult mouse kidney in enhancers overlapping with kidney
Figure 10, B and C). Specifically, we integrated the different enhancer marks (H3K27ac and H3K4me1) in fetal (E15.5)
and adult (week 8) kidneys, and identified 51 GWAS loci that
overlapped with kidney enhancers (Supplemental Table 5). Most of these enhancers (59%) were positive for both en-hancer marks (H3K27ac and H3K4me1) in the fetal stage, but only 28% of them remained positive for both marks in
the adult kidney (chi-squared test P50.0052) (Figure 6C,
Supplemental Figure 10A). To confirm the GWAS
cata-log–based finding (which only reports the top associated
SNPs), we combined a comprehensive list of 26,637 SNPs
that were significantly associated with eGFR (eGFR-SNPs)
in the most recent GWAS studies9,40,41 (Figure 6D). More
than a half (51%) of DMRs overlapped with eGFR-SNPs were localized to fetal enhancers, and 49% of enhancers that overlapped with eGFR-SNPs were enriched for enhancer marks (H3K27ac and K3K4me1) in fetal kidneys (Figure 6, E and F). These results raise the possibility that genetic variants
in fetal enhancers decommissioned byDnmt3aandDnmt3b
contribute to human kidney disease development.
To further understand the clinical significance of ourfi
nd-ings, we analyzed the methylation of DMRs that overlapped
with GWAS SNPs associated with kidney function–related
traits in microdissected kidney tubule samples obtained from healthy subjects and patients with diabetic kidney
dis-ease.14Methylation changes during kidney development were
significantly negatively correlated with methylation changes
observed in diabetic kidney disease (Figure 6, G and H), sug-gesting the methylation patterns established during develop-ment were either reversed in diabetic kidney disease or failed to establish during development.
For example, GWASs have revealed that nucleotide
vari-ants nearbyUNCXwere significantly associated with kidney
functions (Supplemental Figure 10D). The methylation pat-tern of this locus in healthy human kidneys was similar to the mouse kidney, including a large area of lowly methylated
region (Figure 7, A and B). Kidney function–associated
GWAS variants were localized to a fetal enhancer which showed the active enhancer marks H3K27ac and H3k4me1 in fetal kidneys. Although the region remained minimally H3K4me1 positive in the adult kidney, this region was no longer positive for H3K27ac, indicating that it was not an active enhancer in adult kidneys. This region showed an in-crease in methylation level during development and failed to
gain methylation in absence ofDnmt3aandDnmt3b,
indi-cating the key role ofDnmt3aandDnmt3bin establishing
the methylation of this GWAS region. Consistent with the
notion that this region is a fetal-specific active enhancer,
Uncx/UNCXwas expressed in fetal but not in adult kidney (Figures 6H and 7A, Supplemental Figure 10E), indicating
the important role of Dnmt3aand Dnmt3bin
decommis-sioning fetal enhancers. Finally, when compared with healthy kidney samples, kidney tubules from patients with diabetic kidney disease showed strong similarities to kidneys of Dnmt3a/3bknockout mice, such as the loss of cytosine methylation of this region (Figure 7B). In addition to the
UNCX/Uncx locus, we also examined the Hoxd/HOXD
locus. Again, we found a similar pattern, such as
conserva-tion between the human and mouse locus, Dnmt3a/3b
-mediated methylation, and decommissioning of fetal enhancers (Figure 7, C and D, Supplemental Figure 10F).
In summar y, our results indicate that Dnmt3a- and
Dnmt3b-mediated methylation of fetal enhancers are en-riched on kidney disease risk loci.
DISCUSSION
Here,viaintegrating mouse genetic studies and genome-wide
methylome and expression profiling, we elucidated the role of
Dnmt3aand Dnmt3bin renal tubule epithelium in develop-ment, maturation, adult, and diseased mouse kidneys. Using
base-resolution temporal profiling, we described dynamic
changes of DNA methylation during kidney development. Globally, we observed the largest decline in global methylation level between embryonic stem cells and renal progenitors, whereas the greatest increase in methylation was observed
during postnatal maturation (P0–P21). In mice, during the
first 3 weeks, new nephrons are formed, epithelial cells
pro-liferate, and the kidney enlarges drastically.48Regions that act
as enhancers in the adult kidney, show very small and gradual changes in their methylation level during development. Changes in methylation, on the other hand, are associated with alterations in transcript expression that again occur in the postnatal stage. These results indicate that the epigenetic state of these regions is likely established early during kidney development and their postnatal expression is mostly tran-scriptionally controlled.
Our results indicated that kidney-specific fetal enhancers
underwent important changes during postnatal kidney devel-opment. We observed a substantial increase in enhancer
methylation after birth (P0–P21). Histone-modification data
(D) Integration of eGFR-associated SNPs (eGFR-SNPs) from recently published GWAS.9,40,41The combined eGFR-SNPs after the lift
over to mouse genome (mm10) were overlapped with WGBS segments. (E) Genomic location of DMRs overlapped with eGFR-SNPs.
(F) Histone-modification transition from fetal mouse kidney to adult mouse kidney in enhancers overlapping with eGFR-SNPs. (G)
DNA-methylation changes during mouse kidney development and human diabetic kidney disease (DKD). Spearman rank correlation
co-efficient and significance was calculated and showed. (H) Gene expression ofUncxand CpG methylation of eDMR overlapped with
kidney function–associated SNPs during kidney development from E14.5 to 8 weeks after birth. sCR, serum creatinine; TPM, transcripts
WGBS Six2 E15.5 P0 Adult Control DKO H3K27ac chr5:139515562-139580379 (mm10) E15.5 P0 Adult E15.5 P0 Adult E14.5 E15.5 E16.5 P0 Adult RNA-seq H3K4me1
A
Gene CpG islandRenal disease SNPs
eGFR-SNPs
4930500l23Rik
Renal_function-related_traits_(BUN)
rs1123164 [0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 20]
[0 - 20]
[0 - 20]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
rs76210971 rs6951209 rs4724805 rs7785293 rs4724817 rs73670555
Hemoglobin Uncx
CpG:_69 CpG:_124 CpG:_26
chr7:1239604-1308516 (hg19) Diabetic kidney disease Normal Kidney WGBS
B
WGBS −log10(p value) 60 50 40 30 20 10 0 Phastcons Gene CpG island Renal disease SNPs eGFR-SNPsRenal_function-related_traits_(BUN)
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
UNCX
CpG:_26 CpG:_82
rs1123164 rs6951209 rs7785293 rs4724828
rs62435145 r2 0.8 0.6 0.4 0.2 CpG:_923
C
chr2:74593010-74824038 (mm10)WGBS Six2 E15.5 P0 Adult Control DKO Gene CpG island
Renal disease SNPs
eGFR-SNPs H3K27ac E15.5 P0 Adult E15.5 P0 Adult E14.5 E15.5 E16.5 P0 Adult RNA-seq H3K4me1 rs72919076
rs72914763 rs72923424 rs72923454 rs72927167 rs72916126 rs72916158
Evx2
Estimated_glomerular_filtration_rate
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 20]
[0 - 20]
[0 - 20]
[1.000 - 20]
[1.000 - 20]
[1.000 - 20]
[0 - 20]
[0 - 20]
[0 - 20]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
[0 - 10.00]
Hoxd12 Hoxd9 Hoxd3 Haglr
CpG:_84 CpG:_27 CpG:_99 CpG:_35 CpG:_84
chr2:176872325-177132115 (hg19)
D
−log10(p value) 20 15 10 5 Diabetic kidney disease Normal Kidney WGBS WGBS 0 Phastcons Gene CpG island Renal disease SNPs eGFR-SNPs[0 - 1.00] [0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
[0 - 1.00]
6253455 rs72919076 rs72923454 rs72927180 rs57225327
Estimated_glomerular_filtration_rate_(eGFR)_ Evx2
rs187355703
HOXD10 HOXD3
CpG:_77 CpG:_63 CpG:_25 CpG:_141
r2 0.8 0.6 0.4 0.2
Figure 7. Genetic and epigenetic features ofUNCXandHOXDhuman kidney disease associated loci. (A) IGV genome browser of the
mouseUncxlocus (chromosome 5 [chr5]: 139515562–139580379). The top panel shows GWAS and kidney function (eGFR)–associated
variants followed by methylation patterns (WGBS) in the developing kidney andDnmt3a/3bknockout kidney,Six2binding, H3K27ac
and H3K4me1 (enhancer marks), and gene expression by RNA sequencing (RNA-seq). (Note the failure of methylation of the enhancer
region in absence ofDnmt3a/3b). (B) IGV genome browser view of the humanUNCXlocus (chr7: 1239604–1308516). Lift-over from the
indicated that the change in methylation (at birth) was
asso-ciated with a loss of H3K27ac, a histone mark that defines
active enhancers.Dnmt3aandDnmt3bplay a critical role in
methylation of these fetal enhancers and we identified
thou-sands of enhancer regions whose methylation was not
estab-lished in absence ofDnmt3aand Dnmt3b. These enhancers
show intermediate methylation levels at the fetal stage. Al-though their methylation increases to the adult stage, they do not seem to gain full methylation in the normal adult mouse kidney. Previously, these loci have also been
called“vestigial enhancers.”49Here we showed thatDnmt3a
andDnmt3b played key roles in methylation of these fetal
enhancers. Fetal enhancers were the most significantly
en-riched group among the DMRs.
Six2expression shows a strong correlation with the
open-ness and methylation of fetal enhancers. Fetal enhancers are
enriched forSix2binding. Furthermore, the increase in
meth-ylation ofSix2-bound regions is strongly correlated with the
decrease inSix2 expression. Dnmt3a andDnmt3b play key
roles in methylation ofSix2-bound fetal-enhancer regions. It
seems thatSix2-bound enhancers do not achieve full
methyl-ation in the adult kidney, indicating thatSix2might act as a
pioneering factor, which will need to be tested in future experiments.
Although we observed a failure of full silencing of
devel-opmental genes in theDnmt3aandDnmt3bknockout mice,
it was highly unexpected to observe minimal phenotypic changes at baseline and after injury in these animals. This is a key contrast to the observed methylation changes and to
the key roleDnmt3aandDnmt3bin other progenitor
com-partments.Dnmt3ais essential for hematopoietic stem cell
(HSC) differentiation becauseDnmt3a-null HSCs show a
marked decline in differentiation capacity over serial trans-plantation, resulting in accumulation of undifferentiated
HSCs in the bone marrow.50 Moreover, mutations in
DNMT3Aare prevalent in myeloid malignancies51–53 and
lymphoid leukemias,54consistent with its important
func-tion in hematopoiesis. The role ofDnmt3a and Dnmt3b
seems to be more pronounced in rapidly proliferating stem cells such as HSCs, where methylation loss over time is associated with leukemia development. However, it is worth noting that not all animals develop malignancy and even those that develop disease will do so relatively later in
life. These results indicate that cells exhibit significant
plas-ticity in their enhancer methylation level. In addition, it seems that enhancer methylation is not critical for cell-fate stabilization. This notion will need further
experimen-tal evidence. Furthermore, althoughDnmt3a/3bknockout
mice did not show alterations in a CKD model, it showed
increased resistance to AKI, which requires rapid prolifer-ation and cell differentiprolifer-ation.
Here we show that developmental enhancers, whosede
novomethylation is specifically mediated byDnmt3aand
Dnmt3b, are enriched for kidney disease genetic risk loci. These regions are annotated as active enhancers in the fetal
kidney, many bound bySix2, a critical kidney
developmen-tal transcription factor. However, these regions are no longer annotated as active enhancers in the adult stage. Fetal enhancer methylation level increases during
develop-ment and Dnmt3a and Dnmt3b are responsible for the
methylation of these regions. These genetic variants have not been functionally annotated in the past because these regions are no longer active enhancers in adult kidneys. Furthermore, methylation of these regions shows strong correlation with gene expression. However, because these regions are methylated in the adult kidney, the target gene expression is limited to the fetal stage. It will be important to study how these regions contribute to kidney disease development.
Here we propose a locus-specific convergence of genetic
and epigenetic factors in kidney disease development. The in-terplay of sequence and post-translational variations could explain how genetic and environmental factors could
contrib-ute to common disease development.55Furthermore, because
environmental and nutrient availability are critical in
estab-lishing the epigenome, it is possible that Dnmt3a- and
Dnmt3b-mediated methylation changes play roles in kidney disease development. We found striking similarities when we
compared methylation of theUNCXandHOXD9regions in
healthy human and mouse kidneys. However, methylation of
diseased human kidney samples was more similar toDntm3a
andDnmt3bknockout kidneys, raising the interesting
possi-bility that human kidney disease–specific epigenetic changes
are already established during fetal development. This
hypoth-esis was raised by Barkeret al.56in the past, however it has
never been conclusively proven that these changes are medi-ated by epigenetic factors.
In summary, we established the critical role ofDnmt3aand
Dnmt3bin mouse kidney development.Dnmt3aandDnmt3b
play critical roles inde novomethylation and
decommission-ing of fetal-enhancer regions. Interestdecommission-ingly, most of their effect in mice is observed in the postnatal period when the most
significant change in methylation occurs and is associated
with the decline inSix2expression. Fetal enhancers
methyl-ated by Dnmt3a andDnmt3b appear to harbor key kidney
disease risk loci, potentially indicating their key roles in kidney
disease development and the locus-specific convergence of
ge-netic and epigege-netic factors.
association between SNPs and kidney function. PhastCons conservation scores among 46 vertebrate species. Whole-genome bisulfite
methylation patterns were shown in control and diabetic kidney disease (DKD) samples (note the lower methylation in DKD samples). (C