MCB Accepted Manuscript Posted Online 26 January 2015 Mol. Cell. Biol. doi:10.1128/MCB.01404-14 Copyright © 2015, American Society for Microbiology. All Rights Reserved.
1
(Article, Molecular and Cellular Biology)
2 3
Lysine-specific demethylase LSD2 suppresses lipid influx and
4
metabolism in hepatic cells
5 6
Katsuya Nagaoka1,2, Shinjiro Hino1*, Akihisa Sakamoto1, Kotaro Anan1, Ryuta Takase1,
7
Takashi Umehara3, Shigeyuki Yokoyama3, Yutaka Sasaki2 & Mitsuyoshi Nakao1,4*
8 9 10 11
1
Department of Medical Cell Biology, Institute of Molecular Embryology and Genetics, 2 Department of Gastroenterology and Hepatology, Graduate School of Medical
13
Sciences, Kumamoto University, Kumamoto, 860-0811, Japan, RIKEN Systems and Structural Biology Center, Yokohama, Japan, 4 Core Research for Evolutional Science and Technology (CREST), Japan Science and
14
Technology Agency, Tokyo, Japan.
12
3
15 16 17
*Address correspondence to:
18
Mitsuyoshi Nakao M.D., Ph.D. Shinjiro Hino Ph.D.
19 20 21 22 23 24
Department of Medical Cell Biology, Institute of Molecular Embryology and Genetics, Kumamoto University 2-2-1 Honjo, chuo-ku, Kumamoto 860-0811, Japan. Phone: +81-96-373-6800; Fax: +81-96-373-6804 E-mail: [email protected] or [email protected]
25
1
26
ABSTRACT
27
Cells link environmental fluctuations such as nutrition to metabolic remodeling.
28
Epigenetic factors are thought to be involved in such cellular process, but the molecular
29
basis remains unclear. Here we report that the lysine-specific demethylase LSD2
30
suppresses the flux and metabolism of lipids to maintain the energy balance in hepatic
31
cells. Using transcriptome and chromatin immunoprecipitation-sequencing analyses, we
32
reveal that LSD2 represses the genes involved in lipid influx and metabolism, through
33
demethylation of histone H3K4. Selective recruitment of LSD2 at lipid metabolism
34
gene loci was mediated in part by a stress-responsive transcription factor c-Jun.
35
Intriguingly, LSD2 depletion increased the intracellular levels of many lipid metabolites,
36
which was accompanied by an increased susceptibility to toxic cell damage in response
37
to fatty acid exposure. Our data demonstrate that LSD2 maintains metabolic plasticity
38
under fluctuating environment in hepatocytes, by mediating the crosstalk between the
39
epigenome and metabolism.
40 41
2
42
INTRODUCTION
43
Organisms and cells must adjust their energy strategy to fluctuating nutrient
44
availability and other environmental conditions. Epigenetic mechanisms have been
45
implicated in the phenotypic plasticity in response to environmental changes, as well as
46
in consistent execution of the developmental program (1). It has been shown that
47
nutrients and dietary composition potently influence epigenetic marks including DNA
48
methylation and histone methylation/acetylation in both humans and animal models (2).
49
Because chromatin-modifying enzymes utilize nutrient-derived metabolites as
50
substrates and coenzymes, epigenome formation is, by nature, influenced by nutritional
51
and metabolic conditions (3-6). Lysine-specific demethylase 1 and 2 (LSD1 and LSD2),
52
also respectively known as KDM1A and KDM1B, comprise the flavin-dependent amine
53
oxidase family of histone demethylases (7). These enzymes require flavin adenine
54
dinucleotide (FAD) as a coenzyme for the removal of methyl groups from the lysine
55
residue of histone H3 and other proteins (8, 9). FAD is a vitamin B2-derived metabolite
56
that serves as a redox cofactor in key metabolic processes such as fatty acid oxidation
57
and succinate dehydrogenation in the tricarboxylic acid cycle (10). Thus, the cellular
58
metabolic state may influence the demethylase activity of these proteins. Indeed, we and
59
others have previously demonstrated that LSD1 controls energy metabolism genes in
60
response to extracellular conditions (11, 12), suggesting that FAD-dependent epigenetic
61
factors may link environmental information to metabolic programming. LSD2 was
62
identified as a second flavin-dependent histone demethylase that targets methylated
63
lysines 4 and 9 of histone H3 (H3K4 and H3K9, respectively) (8, 13-15). Although
64
LSD2 has been implicated in the establishment of maternal genomic imprinting in
3
65
oocytes (16), little is known about its biological functions, particularly in relation to
66
metabolic control.
67
In the liver, hepatocytes play a crucial role in the homeostatic control of lipid
68
metabolism (17). Hepatocytes incorporate adipose- and diet-derived fatty acids, which
69
are either stored by themselves as neutral lipids or redistributed to other tissues in the
70
form of very low-density lipoproteins (18). When hepatocytes are exposed to an
71
intolerably high amount of fatty acids, for example due to over-feeding, excessive fatty
72
acids and their toxic metabolites accumulate in the cells, often leading to the lipotoxic
73
liver injury known as nonalcoholic fatty liver disease (NAFLD) (19, 20). Epigenetic
74
alterations in the liver have been linked to insulin resistance and NAFLD in humans
75
(21), and diet-induced steatosis in mice (22). A recent report by Ahrens et al. examined
76
the DNA methylation profiles of liver biopsies from patients with NAFLD and
77
non-alcoholic steatohepatitis (NASH), an advanced form of NAFLD (23). Of particular
78
note, some disease-state dependent methylation patterns could be reversed after
79
improvement of the disease condition by bariatric surgery (23), suggesting that hepatic
80
lipid homeostasis is associated with epigenetic plasticity. However, we still lack
81
knowledge of whether a specific epigenetic factor could be involved in the homeostatic
82
control of hepatic lipid metabolism.
83
In the present study, we provide direct evidence that LSD2 plays an essential
84
role in the homeostatic control of lipid metabolism in hepatocytes. Our integrative
85
investigations
86
immunoprecipitation-sequencing (ChIP-seq) analyses reveal that LSD2 suppresses lipid
87
transport and metabolism by repressing key metabolic genes through the regulation of
88
methylated H3K4 (H3K4me). We further show that LSD2 depletion leads to enhanced
using
the
transcriptome,
4
metabolome
and
chromatin
89
lipotoxic cell damage under fatty acid exposure. We propose an epigenetic mechanism
90
for ensuring metabolic plasticity in response to lipid overload, in which LSD2 maintains
91
the proper expression of lipid metabolism genes in hepatocytes.
92 93 94
MATERIALS and METHODS
95 96
Cell culture
97
HepG2 cells were cultured in high glucose (25 mM D-glucose) Dulbecco’s
98
modified Eagle’s medium (Sigma) supplemented with 10% (v/v) heat-inactivated fetal
99
bovine serum and penicillin/streptomycin. For the knockdown experiments, specific
100
siRNAs were introduced to the cells using RNAiMAX reagent (Invitrogen) when they
101
were approximately 50% confluent. After being cultured for 3 to 4 days, semiconfluent
102
cells were harvested for subsequent analyses. Target sequences used for siRNA design
103
are listed in Table S1. siRNA against firefly luciferase gene was used as a control. For
104
the knockdown of c-Jun gene, siGENOME Human JUN siRNA SMARTpool
105
(Dharmacon) was used, while siGENOME Non-Targeting siRNA SMARTpool #1
106
(Dharmacon) was used as a control.
107 108
Antibodies
109
Anti-LSD2 polyclonal antibody was raised in rabbits by administering
110
recombinant human LSD2 protein (region 26 – 822), which was prepared using a
111
baculovirus expression system. Anti-LSD2 antisera were purified by affinity
112
chromatography of the IgG fraction, and were titrated prior to use. Anti-histone H3 5
113
(ab1791), anti-H3K4me1 (ab8895) and anti-H3K27ac (ab4729) antibodies were
114
purchased from Abcam. The following antibodies were also used: anti-histone
115
H3K4me2 (07-030, Millipore), anti-normal rabbit IgG (sc-2027, Santa Cruz), anti-Flag
116
M2 (F1804, Sigma-Aldrich).
117 118
Plasmid construction
119
To construct pcDNA3-Flag-hLSD2, an LSD2 expression vector, a fragment
120
(from +1 to +2,469) of the human LSD2 gene was PCR-amplified using cDNAs from
121
HepG2 cells, and cloned into EcoRV and XbaI sites of the pcDNA3-Flag-mock vector.
122 123
Gene expression analysis
124
Total RNA from tissues and cells was extracted using Trizol reagent
125
(Invitrogen). cDNAs were produced using a ReverTra Ace qPCR RT-Kit (Toyobo).
126
Quantitative RT-PCR was performed by the SYBR green method using Thunderbird
127
reagents (Toyobo) and an ABI 7300 Sequence Detector (Applied Biosciences). Data are
128
presented as mean ± SD. Statistical analyses were performed using two-tailed Student’s
129
t-test. Primers used in this study are listed in Table S1.
130 131
Microarray analysis
132
Genome-wide expression analysis was performed using a GeneChip Human
133
Genome Array U133 Plus 2.0 in combination with a GeneChip Hybridization, Wash and
134
Stain Kit (Affymetrix). We prepared three HepG2 samples, by introducing each of three
135
different siRNAs against LSD2. We also prepared two control-KD samples. Total RNA
136
from cells was extracted, and the sample integrity was confirmed using a Bioanalyzer 6
137
RNA 6000 Nano Assay (Agilent). Data annotation analysis was performed using
138
GeneSpring GX software (Agilent). GSEA was done using GSEA ver. 2.0 software
139
provided
140
(http://www.broadinstitute.org/gsea/).
by
the
Broad
Institute
of
MIT
and
Harvard
141 142
Chromatin immunoprecipitation
143
In the ChIP experiments for detecting modified histones, cells were
144
cross-linked with 1% formaldehyde. Following cell lysis, isolated nuclei were subjected
145
to sonication for chromatin fragmentation. Chromatin fragments were incubated at 4 °C
146
overnight with appropriate antibodies, followed by a pull-down assay using protein
147
A/G-conjugated agarose beads. Purified DNAs were subjected to quantitative PCR
148
(qPCR) using the primer sets listed in Table S1. To detect LSD2 enrichment on
149
genomic DNA, we employed a protocol for detecting indirect associations between
150
protein and DNA (24). Briefly, enhanced cross-linking of chromatin using
151
formaldehyde
152
3′-dithiobispropionimidate 2HCl (Sigma) was performed to increase the stability of
153
protein-DNA complexes. Chromatin fragmentation was done by sonication in regular
154
RIPA buffer containing 0.1% SDS, followed by immunoprecipitation, as described
155
above.
and
the
protein-protein
chemical
cross-linker
dimethyl
3,
156 157
ChIP-seq analysis
158
For ChIP-seq analysis to detect LSD2 or H3K4me1/DNA interactions, 1×107
159
HepG2 cells were collected for a ChIP experiment. After crosslinking, chromatin DNA
160
was fragmented using a Covaris S220 sonicator (Covaris Inc.), followed by 7
161
immunoprecipitation with either an anti-LSD2 or an anti-H3K4me1 polyclonal antibody.
162
The protein-bound chromatin fraction was collected using Dynabeads Protein A/G (Life
163
Technologies), and the DNA was purified. A DNA fragment library for sequencing was
164
constructed using an Ion Fragment Library Kit (Life Technologies). Adopter-ligated
165
DNA fragments were purified using Agencourt AMPure XP (Beckman Coulter Inc.).
166
High throughput sequencing was performed using Ion PGM and Ion Proton
167
semiconductor sequencers (Life Technologies), according to the manufacturer’s
168
instruction. All sequence data obtained from Ion PGM and Proton were merged before
169
alignment onto the human reference genome hg19 using the BWA algorithm (CLC
170
Genomics Workbench Software). With the same software, the duplicate reads and low
171
read/mapping quality reads were trimmed out. For LSD2 ChIP-seq, final numbers of
172
mapped reads were 36,409,704 reads for LSD2, and 27,141,872 reads for Input. For
173
H3K4me1 ChIP-seq, final numbers of mapped reads were 33,087,078 reads for the
174
control siRNA sample, and 53,725,212 reads for input of the control siRNA sample;
175
49,410,448 reads for the LSD2-KD sample, and 58,829,288 reads for input of the
176
LSD2-KD sample.
177
Peak detection was done using the MACS algorithm in Avadis NGS software.
178
LSD2 binding sites were detected based on the LSD2 peaks significantly enriched over
179
input peaks at a cutoff value of p=10-5. H3K4me1 peaks enhanced by LSD2-KD were
180
detected based on the H3K4me1 peaks in LSD2-KD samples significantly enriched over
181
H3K4me1 peaks in control siRNA sample peaks at a cutoff value of p=10-5.
182
For identification of the direct regulatory target genes of LSD2, we first
183
detected 414,095 LSD2 peaks using the MACS algorithm (25) and selected 15,532
184
robust peaks that met the criteria of >25 reads and >5-fold enrichment. Based on this 8
185
peak detection, 6,079 neighboring genes were identified within 5,000 bases of the LSD2
186
peaks using Avadis NGS software with the gene annotation provided by Ensembl.
187
Among these genes, we identified 226 genes with >1.5-fold expression change in
188
response to LSD2-KD.
189
For identifying genes with both increased H3K4me1 and expression change in
190
response to LSD2-KD, we first detected 180,505 exclusive H3K4me1 peaks under
191
LSD2-KD using the MACS algorithm, and selected 12,735 robust peaks that met the
192
criterion of >30 reads and >5-fold enrichment. Based on this peak detection, 5,768
193
neighboring genes were identified within 5,000 bases of the H3K4me1 peaks using
194
Avadis NGS with the gene annotation provided by Ensembl. Of these genes, we
195
identified 207 with >1.5-fold expression change in response to LSD2-KD.
196
Visualization of ChIP-Seq data with smoothing (smoothing window size = 2
197
bp) was done using Avadis NGS. ChIP-seq data for chromatin modifications in HepG2
198
cells were obtained from the ENCODE/Broad Institute via the UCSC Genome Browser
199
website (http://genome.ucsc.edu/). The accession numbers of the files are as follows:
200
wgEncodeEH000082
201
(GSM733743) for H3K27ac. ChIP-seq data for p300 enrichment in HepG2 cells was
202
from ENCODE/Stanford/Yale/USC/Harvard (wgEncodeEH001862), FAIRE data was
203
from ENCODE/OpenChrom (UNC Chapel Hill) (wgEncodeEH000546 (GSM864354)),
204
c-Jun data was from ENCODE/Stanford/Yale/USC/Harvard (wgEncodeEH001794) and
205
c-Myc data was from ENCODE/Open Chrom (UT Austin) (wgEncodeEH000542
206
(GSM822286)).
(GSM733693)
for
207 208
Correlation analyses of ChIP-seq peaks 9
H3K4me2
and
wgEncodeEH000094
209
Correlations of LSD2 peaks with histone modifications were analyzed using
210
GenomeInspector
(Genomatix)
according
to
the
manufacturer’s
instructions
211
(https://www.genomatix.de/online_help/help_regionminer/GenomeInspector.html).
212
reference data sets, histone modification data of HepG2 cells were obtained from the
213
ENCODE/Broad Institute via the UCSC Genome Browser website. The accession
214
numbers of the files used in the analyses are as follows: wgEncodeEH001749
215
(GSM798321) for H3K4me1, wgEncodeEH001023 (GSM733754) for H3K27me3.
216
LSD2 peaks were selected using the criteria of >28 reads and >5-fold enrichment from
217
the peaks detected using the MACS algorithm. Peaks for H3K4me1 and H3K27me3
218
were detected using NGS analyzer (Genomatix) or the MACS algorithm in Avadis
219
NGS.
As
220 221
Metabolomic analysis
222
Metabolomic analyses were performed using CE-TOFMS and LC-TOFMS at
223
Human Metabolome Technologies (HMT, Japan, http://humanmetabolome.com).
224
Triplicate samples of control and LSD2-KD2 cells were subjected to both CE-TOFMS
225
and LC-TOFMS, while LSD2-KD1 was analyzed by LC-TOFMS. Cells were washed
226
with PBS and collected by trypsinization, then washed twice with 5% (w/w) mannitol
227
solution at room temperature. Cells were re-dispersed in LC/MS grade methanol
228
(Wako) for CE-TOFMS or LC/MS grade ethanol (Wako) for LC-TOFMS, both
229
containing HMT’s Internal Standard Solution 1. Peaks were extracted as previously
230
described (26). After quantification, metabolite concentrations were normalized by cell
231
number. Of ~900 (for CE-TOFMS) and ~500 (for LC-MS) metabolites from the HMT
232
databases, 275 metabolites were detected above the signal-to noise threshold (listed in 10
233
Table S4). The databases include glycolysis and TCA cycle intermediates, amino acids,
234
nucleic acids, nucleotides, nucleosides, Coenzyme A, organic acids, nicotinamide
235
coenzymes, fatty acids, bile acids, lipids, steroid derivatives and polyphenols.
236
Clustering of metabolites according to changes in concentration was
237
performed using Cluster 3.0, which was obtained from the Laboratory of DNA
238
Information Analysis of the Human Genome Center in the Institute of Medical Science,
239
The
240
(http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm#ctv/).
241
clustering was calculated using Pearson correlation (centered correlation) and complete
242
Linkage. Clustering results were visualized using Java TreeView software (27).
University
of
Tokyo Hierarchical
243 244
Fatty acid uptake assay
245
Cellular fatty acid uptake was tested using QBT Fatty Acid Uptake Assay Kit
246
(Molecular Devices), which uses dodecanoic acid conjugated with BODIPY, a
247
fluorescent dye (BODIPY-DA). Fatty acid uptake assay stock solutions were dissolved
248
completely by adding 10 ml of 1× HBSS buffer (1× Hank’s balanced salt solution with
249
20 mM HEPES and 0.2% fatty acid-free bovine serum albumin). HepG2 cells were
250
cultured in the medium containing the assay reagent for 10 min at 37 °C. Cells were
251
either subjected to microscopic analysis, or trypsinized for fluorescence-activated cell
252
sorting analysis, using a FACS Canto cytometer (Becton Dickinson).
253 254
Lipotoxicity analysis
255
To analyze cellular lipotoxicity, siRNA-introduced HepG2 cells were treated
256
with bovine serum albumin (BSA)-conjugated oleic acid for 48 hours. Following 11
257
trypsinization, cells were counted using an automatic cell imaging counter, Cytorecon
258
(GE Healthcare). For oleic acid conjugation to BSA, a sodium oleate (≥99% (capillary
259
GC), Sigma) solution in 150 mM NaCl was added to a fatty acid-free BSA solution
260
(Wako). The final molar ratio of oleic acid to BSA was 6:1.
261 262
Animal studies
263
Animal experiments were conducted in accordance with the guidelines of the
264
Animal Care and Use Committee of Kumamoto University. For NAFLD induction,
265
7-week-old male C57BL/6J mice were fed a high fat diet (HFD) containing 21% kcal
266
fat (A02082003BP: Research Diets Inc.) or a methionine/choline-deficient diet (MCD,
267
A02082002BG: Research Diets Inc.), which also contains 21% kcal fat for four weeks.
268
Body weight was monitored weekly throughout the test period. After 16-hour fasting,
269
liver tissues were dissected, and the sections were either snap-frozen in liquid nitrogen
270
for RNA analyses or fixed with formalin for histological analyses.
271 272 273 274
Statistical analyses All statistical analyses between two groups were done by Student’s t-test unless otherwise stated. Equality of variance was examined using F-test.
275 276
Accession Number
277
The accession number of microarray and ChIP-seq data in GEO is GSE59695.
278 279 280
RESULTS 12
281 282
LSD2 selectively represses lipid metabolism genes in hepatocytes
283
Because LSD1 regulates the metabolic genes under diverse cellular contexts
284
(11, 12, 28), we tested, in this study, whether another FAD-dependent demethylase
285
LSD2 could be involved in the metabolic programming. During our initial examinations
286
in mouse tissues, we found that LSD2 was expressed more highly in the liver than in
287
other metabolic tissues (Fig. 1A). To gain insight into the role of LSD2 function in
288
hepatocytes, we depleted LSD2 in HepG2 human hepatic cells using three different
289
siRNAs (Fig. 1B), then carried out an expression microarray experiment. We detected
290
1,362 probe sets with more than 1.5-fold difference between the control and
291
LSD2-knockdown (KD) cells (Fig. 1C). Of these, 906 probe sets were up-regulated,
292
while 456 were down-regulated. Using gene set enrichment analysis (GSEA) (29, 30),
293
we established that genes associated with “metabolism of lipids and lipoproteins” were
294
significantly enriched in the probe sets that were up-regulated by LSD2-KD (statistical
295
threshold, p-values
1
(Article, Molecular and Cellular Biology)
2 3
Lysine-specific demethylase LSD2 suppresses lipid influx and
4
metabolism in hepatic cells
5 6
Katsuya Nagaoka1,2, Shinjiro Hino1*, Akihisa Sakamoto1, Kotaro Anan1, Ryuta Takase1,
7
Takashi Umehara3, Shigeyuki Yokoyama3, Yutaka Sasaki2 & Mitsuyoshi Nakao1,4*
8 9 10 11
1
Department of Medical Cell Biology, Institute of Molecular Embryology and Genetics, 2 Department of Gastroenterology and Hepatology, Graduate School of Medical
13
Sciences, Kumamoto University, Kumamoto, 860-0811, Japan, RIKEN Systems and Structural Biology Center, Yokohama, Japan, 4 Core Research for Evolutional Science and Technology (CREST), Japan Science and
14
Technology Agency, Tokyo, Japan.
12
3
15 16 17
*Address correspondence to:
18
Mitsuyoshi Nakao M.D., Ph.D. Shinjiro Hino Ph.D.
19 20 21 22 23 24
Department of Medical Cell Biology, Institute of Molecular Embryology and Genetics, Kumamoto University 2-2-1 Honjo, chuo-ku, Kumamoto 860-0811, Japan. Phone: +81-96-373-6800; Fax: +81-96-373-6804 E-mail: [email protected] or [email protected]
25
1
26
ABSTRACT
27
Cells link environmental fluctuations such as nutrition to metabolic remodeling.
28
Epigenetic factors are thought to be involved in such cellular process, but the molecular
29
basis remains unclear. Here we report that the lysine-specific demethylase LSD2
30
suppresses the flux and metabolism of lipids to maintain the energy balance in hepatic
31
cells. Using transcriptome and chromatin immunoprecipitation-sequencing analyses, we
32
reveal that LSD2 represses the genes involved in lipid influx and metabolism, through
33
demethylation of histone H3K4. Selective recruitment of LSD2 at lipid metabolism
34
gene loci was mediated in part by a stress-responsive transcription factor c-Jun.
35
Intriguingly, LSD2 depletion increased the intracellular levels of many lipid metabolites,
36
which was accompanied by an increased susceptibility to toxic cell damage in response
37
to fatty acid exposure. Our data demonstrate that LSD2 maintains metabolic plasticity
38
under fluctuating environment in hepatocytes, by mediating the crosstalk between the
39
epigenome and metabolism.
40 41
2
42
INTRODUCTION
43
Organisms and cells must adjust their energy strategy to fluctuating nutrient
44
availability and other environmental conditions. Epigenetic mechanisms have been
45
implicated in the phenotypic plasticity in response to environmental changes, as well as
46
in consistent execution of the developmental program (1). It has been shown that
47
nutrients and dietary composition potently influence epigenetic marks including DNA
48
methylation and histone methylation/acetylation in both humans and animal models (2).
49
Because chromatin-modifying enzymes utilize nutrient-derived metabolites as
50
substrates and coenzymes, epigenome formation is, by nature, influenced by nutritional
51
and metabolic conditions (3-6). Lysine-specific demethylase 1 and 2 (LSD1 and LSD2),
52
also respectively known as KDM1A and KDM1B, comprise the flavin-dependent amine
53
oxidase family of histone demethylases (7). These enzymes require flavin adenine
54
dinucleotide (FAD) as a coenzyme for the removal of methyl groups from the lysine
55
residue of histone H3 and other proteins (8, 9). FAD is a vitamin B2-derived metabolite
56
that serves as a redox cofactor in key metabolic processes such as fatty acid oxidation
57
and succinate dehydrogenation in the tricarboxylic acid cycle (10). Thus, the cellular
58
metabolic state may influence the demethylase activity of these proteins. Indeed, we and
59
others have previously demonstrated that LSD1 controls energy metabolism genes in
60
response to extracellular conditions (11, 12), suggesting that FAD-dependent epigenetic
61
factors may link environmental information to metabolic programming. LSD2 was
62
identified as a second flavin-dependent histone demethylase that targets methylated
63
lysines 4 and 9 of histone H3 (H3K4 and H3K9, respectively) (8, 13-15). Although
64
LSD2 has been implicated in the establishment of maternal genomic imprinting in
3
65
oocytes (16), little is known about its biological functions, particularly in relation to
66
metabolic control.
67
In the liver, hepatocytes play a crucial role in the homeostatic control of lipid
68
metabolism (17). Hepatocytes incorporate adipose- and diet-derived fatty acids, which
69
are either stored by themselves as neutral lipids or redistributed to other tissues in the
70
form of very low-density lipoproteins (18). When hepatocytes are exposed to an
71
intolerably high amount of fatty acids, for example due to over-feeding, excessive fatty
72
acids and their toxic metabolites accumulate in the cells, often leading to the lipotoxic
73
liver injury known as nonalcoholic fatty liver disease (NAFLD) (19, 20). Epigenetic
74
alterations in the liver have been linked to insulin resistance and NAFLD in humans
75
(21), and diet-induced steatosis in mice (22). A recent report by Ahrens et al. examined
76
the DNA methylation profiles of liver biopsies from patients with NAFLD and
77
non-alcoholic steatohepatitis (NASH), an advanced form of NAFLD (23). Of particular
78
note, some disease-state dependent methylation patterns could be reversed after
79
improvement of the disease condition by bariatric surgery (23), suggesting that hepatic
80
lipid homeostasis is associated with epigenetic plasticity. However, we still lack
81
knowledge of whether a specific epigenetic factor could be involved in the homeostatic
82
control of hepatic lipid metabolism.
83
In the present study, we provide direct evidence that LSD2 plays an essential
84
role in the homeostatic control of lipid metabolism in hepatocytes. Our integrative
85
investigations
86
immunoprecipitation-sequencing (ChIP-seq) analyses reveal that LSD2 suppresses lipid
87
transport and metabolism by repressing key metabolic genes through the regulation of
88
methylated H3K4 (H3K4me). We further show that LSD2 depletion leads to enhanced
using
the
transcriptome,
4
metabolome
and
chromatin
89
lipotoxic cell damage under fatty acid exposure. We propose an epigenetic mechanism
90
for ensuring metabolic plasticity in response to lipid overload, in which LSD2 maintains
91
the proper expression of lipid metabolism genes in hepatocytes.
92 93 94
MATERIALS and METHODS
95 96
Cell culture
97
HepG2 cells were cultured in high glucose (25 mM D-glucose) Dulbecco’s
98
modified Eagle’s medium (Sigma) supplemented with 10% (v/v) heat-inactivated fetal
99
bovine serum and penicillin/streptomycin. For the knockdown experiments, specific
100
siRNAs were introduced to the cells using RNAiMAX reagent (Invitrogen) when they
101
were approximately 50% confluent. After being cultured for 3 to 4 days, semiconfluent
102
cells were harvested for subsequent analyses. Target sequences used for siRNA design
103
are listed in Table S1. siRNA against firefly luciferase gene was used as a control. For
104
the knockdown of c-Jun gene, siGENOME Human JUN siRNA SMARTpool
105
(Dharmacon) was used, while siGENOME Non-Targeting siRNA SMARTpool #1
106
(Dharmacon) was used as a control.
107 108
Antibodies
109
Anti-LSD2 polyclonal antibody was raised in rabbits by administering
110
recombinant human LSD2 protein (region 26 – 822), which was prepared using a
111
baculovirus expression system. Anti-LSD2 antisera were purified by affinity
112
chromatography of the IgG fraction, and were titrated prior to use. Anti-histone H3 5
113
(ab1791), anti-H3K4me1 (ab8895) and anti-H3K27ac (ab4729) antibodies were
114
purchased from Abcam. The following antibodies were also used: anti-histone
115
H3K4me2 (07-030, Millipore), anti-normal rabbit IgG (sc-2027, Santa Cruz), anti-Flag
116
M2 (F1804, Sigma-Aldrich).
117 118
Plasmid construction
119
To construct pcDNA3-Flag-hLSD2, an LSD2 expression vector, a fragment
120
(from +1 to +2,469) of the human LSD2 gene was PCR-amplified using cDNAs from
121
HepG2 cells, and cloned into EcoRV and XbaI sites of the pcDNA3-Flag-mock vector.
122 123
Gene expression analysis
124
Total RNA from tissues and cells was extracted using Trizol reagent
125
(Invitrogen). cDNAs were produced using a ReverTra Ace qPCR RT-Kit (Toyobo).
126
Quantitative RT-PCR was performed by the SYBR green method using Thunderbird
127
reagents (Toyobo) and an ABI 7300 Sequence Detector (Applied Biosciences). Data are
128
presented as mean ± SD. Statistical analyses were performed using two-tailed Student’s
129
t-test. Primers used in this study are listed in Table S1.
130 131
Microarray analysis
132
Genome-wide expression analysis was performed using a GeneChip Human
133
Genome Array U133 Plus 2.0 in combination with a GeneChip Hybridization, Wash and
134
Stain Kit (Affymetrix). We prepared three HepG2 samples, by introducing each of three
135
different siRNAs against LSD2. We also prepared two control-KD samples. Total RNA
136
from cells was extracted, and the sample integrity was confirmed using a Bioanalyzer 6
137
RNA 6000 Nano Assay (Agilent). Data annotation analysis was performed using
138
GeneSpring GX software (Agilent). GSEA was done using GSEA ver. 2.0 software
139
provided
140
(http://www.broadinstitute.org/gsea/).
by
the
Broad
Institute
of
MIT
and
Harvard
141 142
Chromatin immunoprecipitation
143
In the ChIP experiments for detecting modified histones, cells were
144
cross-linked with 1% formaldehyde. Following cell lysis, isolated nuclei were subjected
145
to sonication for chromatin fragmentation. Chromatin fragments were incubated at 4 °C
146
overnight with appropriate antibodies, followed by a pull-down assay using protein
147
A/G-conjugated agarose beads. Purified DNAs were subjected to quantitative PCR
148
(qPCR) using the primer sets listed in Table S1. To detect LSD2 enrichment on
149
genomic DNA, we employed a protocol for detecting indirect associations between
150
protein and DNA (24). Briefly, enhanced cross-linking of chromatin using
151
formaldehyde
152
3′-dithiobispropionimidate 2HCl (Sigma) was performed to increase the stability of
153
protein-DNA complexes. Chromatin fragmentation was done by sonication in regular
154
RIPA buffer containing 0.1% SDS, followed by immunoprecipitation, as described
155
above.
and
the
protein-protein
chemical
cross-linker
dimethyl
3,
156 157
ChIP-seq analysis
158
For ChIP-seq analysis to detect LSD2 or H3K4me1/DNA interactions, 1×107
159
HepG2 cells were collected for a ChIP experiment. After crosslinking, chromatin DNA
160
was fragmented using a Covaris S220 sonicator (Covaris Inc.), followed by 7
161
immunoprecipitation with either an anti-LSD2 or an anti-H3K4me1 polyclonal antibody.
162
The protein-bound chromatin fraction was collected using Dynabeads Protein A/G (Life
163
Technologies), and the DNA was purified. A DNA fragment library for sequencing was
164
constructed using an Ion Fragment Library Kit (Life Technologies). Adopter-ligated
165
DNA fragments were purified using Agencourt AMPure XP (Beckman Coulter Inc.).
166
High throughput sequencing was performed using Ion PGM and Ion Proton
167
semiconductor sequencers (Life Technologies), according to the manufacturer’s
168
instruction. All sequence data obtained from Ion PGM and Proton were merged before
169
alignment onto the human reference genome hg19 using the BWA algorithm (CLC
170
Genomics Workbench Software). With the same software, the duplicate reads and low
171
read/mapping quality reads were trimmed out. For LSD2 ChIP-seq, final numbers of
172
mapped reads were 36,409,704 reads for LSD2, and 27,141,872 reads for Input. For
173
H3K4me1 ChIP-seq, final numbers of mapped reads were 33,087,078 reads for the
174
control siRNA sample, and 53,725,212 reads for input of the control siRNA sample;
175
49,410,448 reads for the LSD2-KD sample, and 58,829,288 reads for input of the
176
LSD2-KD sample.
177
Peak detection was done using the MACS algorithm in Avadis NGS software.
178
LSD2 binding sites were detected based on the LSD2 peaks significantly enriched over
179
input peaks at a cutoff value of p=10-5. H3K4me1 peaks enhanced by LSD2-KD were
180
detected based on the H3K4me1 peaks in LSD2-KD samples significantly enriched over
181
H3K4me1 peaks in control siRNA sample peaks at a cutoff value of p=10-5.
182
For identification of the direct regulatory target genes of LSD2, we first
183
detected 414,095 LSD2 peaks using the MACS algorithm (25) and selected 15,532
184
robust peaks that met the criteria of >25 reads and >5-fold enrichment. Based on this 8
185
peak detection, 6,079 neighboring genes were identified within 5,000 bases of the LSD2
186
peaks using Avadis NGS software with the gene annotation provided by Ensembl.
187
Among these genes, we identified 226 genes with >1.5-fold expression change in
188
response to LSD2-KD.
189
For identifying genes with both increased H3K4me1 and expression change in
190
response to LSD2-KD, we first detected 180,505 exclusive H3K4me1 peaks under
191
LSD2-KD using the MACS algorithm, and selected 12,735 robust peaks that met the
192
criterion of >30 reads and >5-fold enrichment. Based on this peak detection, 5,768
193
neighboring genes were identified within 5,000 bases of the H3K4me1 peaks using
194
Avadis NGS with the gene annotation provided by Ensembl. Of these genes, we
195
identified 207 with >1.5-fold expression change in response to LSD2-KD.
196
Visualization of ChIP-Seq data with smoothing (smoothing window size = 2
197
bp) was done using Avadis NGS. ChIP-seq data for chromatin modifications in HepG2
198
cells were obtained from the ENCODE/Broad Institute via the UCSC Genome Browser
199
website (http://genome.ucsc.edu/). The accession numbers of the files are as follows:
200
wgEncodeEH000082
201
(GSM733743) for H3K27ac. ChIP-seq data for p300 enrichment in HepG2 cells was
202
from ENCODE/Stanford/Yale/USC/Harvard (wgEncodeEH001862), FAIRE data was
203
from ENCODE/OpenChrom (UNC Chapel Hill) (wgEncodeEH000546 (GSM864354)),
204
c-Jun data was from ENCODE/Stanford/Yale/USC/Harvard (wgEncodeEH001794) and
205
c-Myc data was from ENCODE/Open Chrom (UT Austin) (wgEncodeEH000542
206
(GSM822286)).
(GSM733693)
for
207 208
Correlation analyses of ChIP-seq peaks 9
H3K4me2
and
wgEncodeEH000094
209
Correlations of LSD2 peaks with histone modifications were analyzed using
210
GenomeInspector
(Genomatix)
according
to
the
manufacturer’s
instructions
211
(https://www.genomatix.de/online_help/help_regionminer/GenomeInspector.html).
212
reference data sets, histone modification data of HepG2 cells were obtained from the
213
ENCODE/Broad Institute via the UCSC Genome Browser website. The accession
214
numbers of the files used in the analyses are as follows: wgEncodeEH001749
215
(GSM798321) for H3K4me1, wgEncodeEH001023 (GSM733754) for H3K27me3.
216
LSD2 peaks were selected using the criteria of >28 reads and >5-fold enrichment from
217
the peaks detected using the MACS algorithm. Peaks for H3K4me1 and H3K27me3
218
were detected using NGS analyzer (Genomatix) or the MACS algorithm in Avadis
219
NGS.
As
220 221
Metabolomic analysis
222
Metabolomic analyses were performed using CE-TOFMS and LC-TOFMS at
223
Human Metabolome Technologies (HMT, Japan, http://humanmetabolome.com).
224
Triplicate samples of control and LSD2-KD2 cells were subjected to both CE-TOFMS
225
and LC-TOFMS, while LSD2-KD1 was analyzed by LC-TOFMS. Cells were washed
226
with PBS and collected by trypsinization, then washed twice with 5% (w/w) mannitol
227
solution at room temperature. Cells were re-dispersed in LC/MS grade methanol
228
(Wako) for CE-TOFMS or LC/MS grade ethanol (Wako) for LC-TOFMS, both
229
containing HMT’s Internal Standard Solution 1. Peaks were extracted as previously
230
described (26). After quantification, metabolite concentrations were normalized by cell
231
number. Of ~900 (for CE-TOFMS) and ~500 (for LC-MS) metabolites from the HMT
232
databases, 275 metabolites were detected above the signal-to noise threshold (listed in 10
233
Table S4). The databases include glycolysis and TCA cycle intermediates, amino acids,
234
nucleic acids, nucleotides, nucleosides, Coenzyme A, organic acids, nicotinamide
235
coenzymes, fatty acids, bile acids, lipids, steroid derivatives and polyphenols.
236
Clustering of metabolites according to changes in concentration was
237
performed using Cluster 3.0, which was obtained from the Laboratory of DNA
238
Information Analysis of the Human Genome Center in the Institute of Medical Science,
239
The
240
(http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm#ctv/).
241
clustering was calculated using Pearson correlation (centered correlation) and complete
242
Linkage. Clustering results were visualized using Java TreeView software (27).
University
of
Tokyo Hierarchical
243 244
Fatty acid uptake assay
245
Cellular fatty acid uptake was tested using QBT Fatty Acid Uptake Assay Kit
246
(Molecular Devices), which uses dodecanoic acid conjugated with BODIPY, a
247
fluorescent dye (BODIPY-DA). Fatty acid uptake assay stock solutions were dissolved
248
completely by adding 10 ml of 1× HBSS buffer (1× Hank’s balanced salt solution with
249
20 mM HEPES and 0.2% fatty acid-free bovine serum albumin). HepG2 cells were
250
cultured in the medium containing the assay reagent for 10 min at 37 °C. Cells were
251
either subjected to microscopic analysis, or trypsinized for fluorescence-activated cell
252
sorting analysis, using a FACS Canto cytometer (Becton Dickinson).
253 254
Lipotoxicity analysis
255
To analyze cellular lipotoxicity, siRNA-introduced HepG2 cells were treated
256
with bovine serum albumin (BSA)-conjugated oleic acid for 48 hours. Following 11
257
trypsinization, cells were counted using an automatic cell imaging counter, Cytorecon
258
(GE Healthcare). For oleic acid conjugation to BSA, a sodium oleate (≥99% (capillary
259
GC), Sigma) solution in 150 mM NaCl was added to a fatty acid-free BSA solution
260
(Wako). The final molar ratio of oleic acid to BSA was 6:1.
261 262
Animal studies
263
Animal experiments were conducted in accordance with the guidelines of the
264
Animal Care and Use Committee of Kumamoto University. For NAFLD induction,
265
7-week-old male C57BL/6J mice were fed a high fat diet (HFD) containing 21% kcal
266
fat (A02082003BP: Research Diets Inc.) or a methionine/choline-deficient diet (MCD,
267
A02082002BG: Research Diets Inc.), which also contains 21% kcal fat for four weeks.
268
Body weight was monitored weekly throughout the test period. After 16-hour fasting,
269
liver tissues were dissected, and the sections were either snap-frozen in liquid nitrogen
270
for RNA analyses or fixed with formalin for histological analyses.
271 272 273 274
Statistical analyses All statistical analyses between two groups were done by Student’s t-test unless otherwise stated. Equality of variance was examined using F-test.
275 276
Accession Number
277
The accession number of microarray and ChIP-seq data in GEO is GSE59695.
278 279 280
RESULTS 12
281 282
LSD2 selectively represses lipid metabolism genes in hepatocytes
283
Because LSD1 regulates the metabolic genes under diverse cellular contexts
284
(11, 12, 28), we tested, in this study, whether another FAD-dependent demethylase
285
LSD2 could be involved in the metabolic programming. During our initial examinations
286
in mouse tissues, we found that LSD2 was expressed more highly in the liver than in
287
other metabolic tissues (Fig. 1A). To gain insight into the role of LSD2 function in
288
hepatocytes, we depleted LSD2 in HepG2 human hepatic cells using three different
289
siRNAs (Fig. 1B), then carried out an expression microarray experiment. We detected
290
1,362 probe sets with more than 1.5-fold difference between the control and
291
LSD2-knockdown (KD) cells (Fig. 1C). Of these, 906 probe sets were up-regulated,
292
while 456 were down-regulated. Using gene set enrichment analysis (GSEA) (29, 30),
293
we established that genes associated with “metabolism of lipids and lipoproteins” were
294
significantly enriched in the probe sets that were up-regulated by LSD2-KD (statistical
295
threshold, p-values
