Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: Adipogenesis is one of the key processes during obesity development. Better understanding of this
Received 15 February 2019 process could advance our knowledge on obesity and its treatment. Transcription factors (TFs) are master
Accepted 25 February 2019 regulators during adipogenesis, however, a system-wide analysis of TFs dynamic proteome during adi-
Available online 6 March 2019
pogenesis is lacking. Here, we profiled 472 TFs and systematically elucidated their roles during the first 7
days of adipogenesis of human adipose-derived stem cells (hADSCs) on proteome scale. We identified
Keywords:
two main and four sub-phases during adipogenesis. The commitment phase (0 he8 h) mainly mediated
Obesity
stem cell proliferation, differentiation and chromatin remodeling, in which proteins of SWI/SNF family
Adipogenesis
Transcription factors
are the key centroid nodes. The determination phase (1D-7D) predominately regulated fat cell differ-
Proteomics entiation and response to lipid and oxygen, which could be associated with terminal differentiation of
adipocyte and responsible for maturation. PPARg, CREB1 and MYC are the centroid nodes of this phase.
Remarkably, we identified and verified three TFs (BATF3, MAFF and MXD4) as novel regulators of adi-
pogenesis, whose over-expression could inhibit adipogenesis of hADSCs in vitro. Overall, our study
provided a valuable TFs resource to understand the complex process of adipogenesis.
© 2019 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/j.bbrc.2019.02.134
0006-291X/© 2019 Elsevier Inc. All rights reserved.
856 S. Li et al. / Biochemical and Biophysical Research Communications 511 (2019) 855e861
research of TFs on adipogenesis at proteomics level to date. <1%. As a forward database, Uniprot Human Protein Database was
Here, based on a recently developed approach to enrich TFs by used. A reverse database for the decoy search was generated
Qin Lab [9], we first systematically quantified the dynamic prote- automatically in MaxQuant. Enzyme specificity was set to Trypsin,
ome of endogenous TFs during adipogenesis of hADSCs. Enriching and a minimum number of seven amino acids was required for
TFs by a synthetic DNA made of a concatenated tandem-array of the peptide identification. Variable modification: acetylation (Protein-
consensus response elements (catTFRE) and analyzing by label-free N terminus) and oxidation (methionine); fixed modification: car-
mass spectrometry, we finally quantitatively identified 472 TFs bamidomethylation (cysteine).
during adipogenesis of hADSCs. We found that this process could be
divided into two main and four sub-stages based on the dynamic 2.5. Overexpression of target genes and determination of
TFs proteome. We also found and validated three TFs as novel tar- adipogenesis
gets to regulate adipogenesis, whose over-expression could effi-
ciently inhibit adipogenesis of hADSCs in vitro. Lentiviral vectors, which are over-expressing EGFP only (pLV
[Exp]-EGFP:T2A:Puro-Null), over-expressing MAFF and EGFP
2. Materials and methods (pLV[Exp]-EGFP:T2A:Puro-EF1A > hMAFF[NM_152878.1]), over-
expressing BATF3 and EGFP (pLV[Exp]-EGFP:T2A:Puro-
2.1. Cell culture and differentiation EF1A > hBATF3[NM_018664.2]), over-expressing MXD4 and EGFP
(pLV[Exp]-EGFP:T2A:Puro-EF1A > hMXD4[NM_006454.2] were
hADSCs purchased from ScienCell Research Laboratories were purchased from Cyagen Biosciences Inc (Guangzhou, China) and
seeded at 2 104/cm2 density and proliferated in Mesenchymal were used to infect hADSCs according to manufacturer's in-
Stem Cell Growth Media (Cyagen). Two days after confluence, cells structions, respectively. EGFP expression was observed by fluores-
were induced to differentiate with hADSCs Adipogenesis Basal cence microscopy (Olympus) after infected for 48 h to evaluate the
Medium (Cyagen), and collected at 0 h, 0.5 h, 2 h, 8 h, 1 day, 3 day transfection efficiency. If the efficiency of transfection was more
and 7 day, respectively. Three biological replicates were conducted than 80%, the cells were split equally for subsequent determination
for each time point. All cells were cultured at 37 Cand 5% CO2 in a experiment of adipogenesis. Cells were stained by Oil red O (Sigma)
humidified atmosphere. at 7 day after exposed to differentiation stimuli. Cells were visual-
ized and photographed by digital microscope (Olympus). The area
2.2. catTFRE pull-down and trypsin digestion dyed by Oil red O is calculated by ImageJ (National Institutes of
Health, USA).
catTFRE DNA was obtained from Beijing Proteome Research
Center (Beijing, China). The details of catTFRE pull-down and 2.6. RT-PCT analysis for target gene expression
trypsin digestion could refer to their previous work [10]. Briefly,
nuclear proteins were extracted using NE-PER kit (Thermo Fisher After infected for 48 h, cells were harvested for quantitative
Scientific) according to the manufacturer's instruction. Afterwards, real-time PCR analysis to determine gene expression. Total RNA was
2 mg of biotinylated DNA was immobilized with 40 mL of Dynabeads extracted using TRIzol reagent (Ambion). Reverse transcription was
(Thermo Fisher scientific), and incubated with 100 mg nuclear ex- completed using Reverse Transcription System (promega). TOYOBO
tracts in 4 C for 1.5 h. After incubation, the supernatant was dis- SYBR Green Realtime PCR master mix (Toyobo) was used for
carded, and the NETN buffer (100 mM NaCl, 20 mM Tris-HCl, quantitative real-time polymerase chain reaction. b-actin gene was
0.5 mM EDTA, and 0.5% [vol/vol] Nonidet P-40) and PBS were used used as the internal control.
to wash the Dynabeads twice respectively. Beads were digested hMAFF: F:50 -AGGAGGTGACACGGCTCAAG-3’; R:50 -CCAGCTCC-
overnight with 1.5 mg trypsin at 37 C in 50 mmol NH4HCO3. GACTTCTGCTTC-3’.
hBATF3: F: 50 -AGACCCAGAAGGCTGACAAGC-3’;
2.3. LC-MS/MS using Q Exactive HF R: 50 -GGCACTGGCACAAAGTTCATAG-3’.
hMDX4: F:50 -CATCAAGAAACTGGAGGAGCAGG-30 , R:50 -TCCGT-
Peptides were analyzed on a Q Exactive HF mass spectrometer GGAGACAGCAGAGCC-3’.
(Thermo Fisher Scientific) connected to an Easy-nLC 1000 liquid b-Actin: F: 50 -TGACGTGGACATCCGCAAAG-3’;
chromatography system (Thermo Fisher Scientific). Peptides were R: 50 -CTGGAAGGTGGACAGCGAGG-3’.
loaded to a trap column (100 mm 2 cm, homemade; particle size,
3 mm; pore size, 120 Å; SunChrom, USA) in Solvent A (0.1% formic 2.7. Bioinformatics and statistics
acid), and separated on a home-made 150 mm 30 cm silica
microcolumn (particle size, 1.9 mm; pore size, 120 Å; Dr. Maisch For label-free protein quantification, we used MaxQuant LFQ
GmbH) with a gradient of 5%e35% Solvent B (100% acetonitrile and algorithm to quantitate the MS signals, and the proteins intensities
0.1% formic acid) at a flow rate of 600 nL/min for 150 min. MS were represented in iBAQ. For normalization, protein's intensity in
analysis was performed with one full scan (300e1400 m/z, one sample was first divided by the total intensity of all identified
R ¼ 120,000 at 200 m/z) with an AGC of 3e6, followed by up to 30 proteins of this sample and then multiplied by the median intensity
data-dependent MS/MS scans (R ¼ 15,000 at 200 m/z) with an AGC of summed proteins' intensities of all samples. After normalization,
of 2e4, maximum injection time of 40 ms, isolation window of proteome data was filtered to remove outlier samples until the
1.6 m/z, and normalized collision energy of 27%. Dynamic exclusion standard deviation of protein intensities for triplicate biological
time was enabled at 18 s. replicates was &1.5 for each time point. Proteins identified in at
least two biological replicates were used for subsequent analysis.
2.4. MS data processing Proteins in each time point was compared to proteins at 0 h. Pro-
teins with fold change j2j and p value < 0.05 were defined as
Raw MS files were managed by MaxQuant software (version significant proteins. Biological function annotation was conducted
1.6.0.16). MS/MS-based peptide identification was carried out with by GO enrichment through String platform or Ingenuity Pathway
the Andromeda search engine in MaxQuant. Andromeda uses a Analysis platform (QIAGEN Redwood City). Protein-protein inter-
target-decoy approach to identify peptides and proteins at an FDR action network was conducted in String platform and high
S. Li et al. / Biochemical and Biophysical Research Communications 511 (2019) 855e861 857
confidence (0.7) was set as requited interaction score. Pearson proteomic profiles. First, supervised PLS-DA analysis was per-
correlation, PLS-DA, unsupervised hierarchical clustering, and formed to group samples according to their quantitative protein
bubble plot analysis were handled in R package. A two-sample, profiles. Our results revealed two main and four sub-phases during
two-side student t-test was applied to compare the difference be- adipogenesis, in which 0 he8 h was grouped as one main phase,
tween the mean values of two groups. and 1D to 7D belong to the other phase. We further defined 0 h as
the pre-early stage and 7D as the post-late stage, which are the
3. Results demarcation points for sub-phases (Fig. 2A). The number of iden-
tified proteins also provided similar stratified features, in which the
3.1. Overview of identification and quantification of TFs proteome number of identified proteins jumped at 0.5 h, and persisted to 8 h,
while straightly dropped down from 1D to 7D (Fig. 2B). The pattern
To systematically depict the dynamic features of TFs during adi- of significant proteins showed that 8 h could be the critical point to
pogenesis of hADSCs, we performed quantitative proteomics of switch the stage of adipogenesis, in which the number of down-
endogenous TFs across a high-density temporal course (0, 0.5 h, 2 h, regulated proteins started to increase, and reached the most at
8 h, 1D, 3D, and 7D), which covered both early and later stages of 7D (Fig. 2C). All these results indicated that proteomic profiles of
adipogenesis [5,11]. A synthetic DNA containing a concatenated TFs could well characterize and provide more clues about the dy-
tandem array of the consensus TFREs (catTFRE) was used as an af- namic process of adipogenesis.
finity reagent to permit efficient enrichment of TFs (Fig. 1A).
Consequently, aided by annotation in TcoF-DB v2 database (https:// 3.3. Hierarchical clustering and network analysis revealed featured
tools.sschmeier.com/tcof/home/) [12], we could assign 472 of iden- biological responses during different stages of adipogenesis
tified proteins as TFs, 285 out of which (60.38%) were found matched
TF family information in DBD database (http://dbd.mrclmb.cam.ac. Since dynamic proteomic profiles of TFs revealed different
uk/DBD) and almost all TFs families were recovered [13]. TF fam- stages during adipogenesis, we next investigated what biological
ilies of ZF-C2H2, Homeobox, and HLH accounted for the most of responses were involved. Unsupervised hierarchical clustering was
identified TFs (Fig. 1B). Our results showed good quality of triplicate conducted on significant proteins in at least one time points to
replicates with a correlation value of R ¼ 0.93 on average (Fig. 1C) depict biological clusters. Consequently, two main clusters and four
and an average of 71% overlap of protein identification in each group sub-clusters were obtained (Fig. 3A). Proteins in cluster 1 pre-
(Fig. 1D). Notably, the degree of Pearson correlation and overlap of dominantly kept unchanged at 0.5 h and 2 h, while increased from
TFs identification were decreased during the later process of adipo- 1D to 7D. Proteins in cluster 2 highly expressed at 0.5 and 2 h, while
genesis, indicating some special features underlying. markedly decreased from 1D to 7D. Distinct patterns of 8 h in either
cluster 1 or cluster 2 further revealed that it could be the turning
3.2. TF's proteome characterized different stages of adipogenesis point that switch the process of adipogenesis to the next stage.
Cluster 1 mainly involved in process of response to lipid and fat cell
We first examined the temporal features of adipogenesis via TFs differentiation (Fig. 3B). Network analysis of proteins in cluster 1
Fig. 1. Overview of TFs proteome during adipogenesis of hADSCs. (A) Illustration of hADSCs adipogenesis and workflow of catTFRE enrichment, MS-based quantitative proteomic
and bioinformatic analyses. (B) Family information for identified TFs. (C) Pearson correlation analysis on quality of biological replicates. (C) Overlap of identified TFs of triplicate
biological replicates.
858 S. Li et al. / Biochemical and Biophysical Research Communications 511 (2019) 855e861
Fig. 2. TF's proteome characterized different stages of adipogenesis. (A) Partial least squares discrimination analysis (PLS-DA) of temporal proteomic data. (B) Distribution of
identified TFs in each time points. (C) Distribution of significant proteins in each time point. Each time point was compared with 0 h.
Fig. 3. Hierarchical clustering and network analysis TF's proteome. (A) Hierarchical clustering analysis of significant proteins. 221 proteins identified across all time points were
used and 160 of which were significant proteins in at least one time points (compared with 0 h). (B) GO analysis of proteins in cluster 1. (C) Protein-protein interaction network
analysis of proteins in cluster 1. (D) GO analysis of proteins in cluster 2. (E) Network analysis of proteins in cluster 2. (F) Fold changes of protein of component of SWI/SNF complex.
(G) Number of unique proteins. (H) GO analysis of unique proteins on 7D.
showed three major sub-networks, in which PPARg, CREB1 and and differentiation, and chromatin assembly or disassembly,
MYC were the centroid nodes, regulating lipid response, fat cell accompanying the acquisition of lineage-committed cellular
proliferation and Notch signaling pathway (Fig. 3C). Proteins in phenotype [11,15]. 1D to7D matched the terminal differentiation,
Cluster 2 mainly regulate stem cell proliferation and differentiation which response to lipid and regulate fat cell differentiation, facili-
(Fig. 3D). Network analysis showed that proteins, including tating to get the characteristics of mature adipocyte.
SMARCA1-5, SMARCC1-2, SMARCD1-3, and SMARCE1, which are
components of SWI/SNF chromatin remodeling complexes, were 3.4. Time restricted biological response revealed by TFs proteome
the major nodes (Fig. 3E and F). These proteins regulate chromatin
remodeling, and mediate transition from proliferating stem/pro- To examine the time restricted biological responses during
genitor cells to adult state. Previous evidences pointed out that adipogenesis, we first investigated the distribution of unique TFs
adipogenesis was composed of commitment phase and terminal numbers. Proteins that were not identified in all triple replicates or
differentiation phase. During commitment phase, stem cells lose only identified in one of triple replicates, while identified in at least
pluripotency and become preadipocytes, and in terminal differen- two replicates of all other time points were defined as unique
tiation phase, preadipocytes acquire some traits of mature adipo- proteins of this time point. Our results showed that 7D had the
cytes [11,14,15]. Hence, we speculated that 0 he8 h matched the most heterogeneity with the greatest number of unique proteins,
commitment phase, which is responsible for stem cell proliferation indicating its specific role during adipogenesis (Fig. 3G). GO
S. Li et al. / Biochemical and Biophysical Research Communications 511 (2019) 855e861 859
enrichment showed that these unique TFs was mainly focused on technology of catTFRE [10], which could provide high-quality of TFs
response to oxygen and cell death (Fig. 3H), which is necessary for data to build more precise models of TFs network. Our data could
terminal differentiation of adipogenesis and adipocyte maturation also serve as a valuable public resource for adipogenesis study.
[16,17], indicating that 7D may be the critical point to enter Increasing evidences have showed that differentiation may be a
adipocyte maturation. stochastic process, in which an increase in cell-cell variability have
been observed in many mammalian systems [20]. Notably, some
3.5. Three TFs as novel regulators responsible for adipogenesis scholars believe that this stochastic process is caused by the
random response of TFs to external signals [21]. Our results also
We also aimed to discover new potential regulators of adipo- observed similar phenomena that Pearson correlation and over-
genesis. As TF's function is strongly dose-dependent, it could be lapped number of proteins among three biological replicates
considered that drastic changes in expression are often accompa- decreased in later stages of adipogenesis, indicating the increased
nied by drastic functional changes [18]. Hence, we selected three variability of TFs network and suggesting that changes of TFs
TFs, that is, MAFF, BATF3 and MXD4, which presented declining network during adipogenesis is also a stochastic process.
tendency and dramatically changed during adipogenesis in our Our TFs proteome results revealed two-main phases during the
proteomics results (Fig. 4A). Due to the robustness and redundancy first 7 days of adipogenesis. The first phase (commitment phase,
of TFs network, inactivation of any one of target genes probably 0 h-8h) could be characterized by stem cell proliferation and dif-
would not affect the function of TFs network, while over-expression ferentiation, in which proteins of components of SWI/SNF chro-
could be an alternative way to examine their intrinsic functions matin remodeling complexes, that dramatically increased at this
[19]. We respectively over-expressed MAFF, BATF3 and MXD4 in period, could play key roles in this process (Fig. 3F). Chromatin-
hADSCs. EGFP fluorescence from target genes showed good trans- remodeling complexes is an important step in cell differentiation
fection efficiency (Fig. 4B). RT-PCR showed 30e10000-fold of over- to facilitate the access of multiple transcription factors. SWI/SNF
expression for target genes (Fig. 4C). Each overrepresented hADSCs complex is involved in the differentiation of a variety of mamma-
were then induced to differentiate. Since 7D could be the point that lian tissues, including adipocytes [22]. It has been proven to be
adipocytes begin to enter maturation, we examined the efficiency necessary for transcription of PPARg2, which is the most important
of adipogenesis of transfected cells on 7D. We found significantly regulator of adipogenesis [23]. The first phase can be further
inhibited Oil Red O accumulation, especially in MXD4 over- divided into two sub-phases (0 h; 0.5e8 h). The number of identi-
expressed hADSCs (Fig. 4DeF). These results showed that over- fied TFs rapidly reached maximum at 0.5 h, and about 10% of the TFs
expression of MAFF, BATF3 and MXD4 could inhibit adipogenesis changed dramatically, suggesting that TFs network at this time
of hADSCs and MXD4 showed the most efficiency. point responded to differentiation stimulus rapidly and dramati-
cally. The second phase (determination phase, 1De7D)was char-
4. Discussion acterized by response to lipid and fat cell differentiation. In this
phase, PPARg, CREB1 and MYC are the key nodes. PPARg is the most
Our study presented the first dynamic TFs profiles about adi- important regulator in adipogenesis. CREB1 and MYC has also been
pogenesis of hADSCs at proteome level, which covered high- proven to promote adipogenesis in 3T3-L1 and ADSCs, respectively
density temporal courses, ranging from stem cell stage (0 h) to [24,25]. The second phase could also be further divided into two
later stage of adipogenesis(7D). This systematic TFs proteomics sub-phases (1D-3D; 7D) and 7D is the special time, in which the
study benefited from the recently well-developed enrichment most unique TFs were found, and these time-restricted proteins
Fig. 4. MAFF, BATF3 or MXD4 over-expression inhibits adipogenesis of hADSCs. (A) Time course expression of MAFF, BATF3and MXD4 in proteomics data. (B) EGFP expression in
hADSCs infected for 48 h. Scale bar, 50 mm. (C) mRNA over-expression of targeted genes in hADSCs infected for 48 h. (DeE) Oil Red-O staining and microscopic images. Scale bar,
50 mm.
860 S. Li et al. / Biochemical and Biophysical Research Communications 511 (2019) 855e861
H. Takatori, T. Tamachi, K. Hirose, A. Onodera, J. Suzuki, O. Ohara, exacerbate autoimmune diabetes, J. Clin. Invest. 128 (2018) 3779e3793.
M. Yamashita, T. Nakayama, H. Nakajima, Sox5 and c-Maf cooperatively [30] Tsuchiya, Suppression of MafA mRNA with siRNA prevents adipose cell dif-
induce Th17 cell differentiation via RORgammat induction as downstream ferentiation in 3T3-L1 cells, Int. J. Mol. Med. 23 (2009).
targets of Stat3, J. Exp. Med. 211 (2014) 1857e1874. [31] S.S. Dong, Y.J. Zhang, Y.X. Chen, S. Yao, R.H. Hao, Y. Rong, H.M. Niu, J.B. Chen,
[29] C.Y. Hsu, L.T. Yeh, S.H. Fu, M.W. Chien, Y.W. Liu, S.C. Miaw, D.M. Chang, Y. Guo, T.L. Yang, Comprehensive review and annotation of susceptibility SNPs
H.K. Sytwu, SUMO-defective c-Maf preferentially transactivates Il21 to associated with obesity-related traits, Obes. Rev. 19 (2018) 917e930.