Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

Abstract

Bmi1 is overexpressed in one-third of hepatocellular carcinoma (HCC) patients and acts as an oncogene in hepatocarcinogenesis. However, the underlying mechanism is unclear. The role of TGFβ signalling in HCC is not well defined as well. Here, we report that TGFβ2 is a target of Bmi1 in HCC and has a tumour-suppressing role. In Bmi1-knockout mouse livers and HCC cell lines, TGFβ2/SMAD cascade proteins were upregulated. TGFβ2 expression was inversely correlated with Bmi1 expression in human and mouse HCC tissues. In vitro, Bmi1 knockdown activated TGFβ2/SMAD signalling and led to cell apoptosis via upregulation of p15 and p21. TGFβ2 inhibition rescued the inhibitory effect of Bmi1 knockdown on HCC cell survival, proliferation, and cell-cycle progression. In vivo, restoration of TGFβ2 expression blocked Bmi1/Ras-driven hepatocarcinogenesis in mice. Chromatin immunoprecipitation and luciferase reporter assays revealed that Bmi1 repressed TGFβ2 expression by binding to its promoter as a co-factor of polycomb repressor complex 1. Our findings elucidate the molecular mechanism underlying hepatic Bmi1-driven carcinogenesis and highlight the importance of TGFβ2 as a tumour suppressor in HCC development.

Full text

Oncogene
https://doi.org/10.1038/s41388-019-1043-8
ARTICLE
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD
signalling axis
Bin Li1●Yuyuan Chen1●Fei Wang2●Jun Guo1●Wen Fu1●Min Li3●Qichang Zheng3●Yong Liu1●Lingling Fan4●
Lei Li1●Chuanrui Xu 1
Received: 13 January 2019 / Revised: 22 September 2019 / Accepted: 24 September 2019
© The Author(s), under exclusive licence to Springer Nature Limited 2019
Abstract
Bmi1 is overexpressed in one-third of hepatocellular carcinoma (HCC) patients and acts as an oncogene in
hepatocarcinogenesis. However, the underlying mechanism is unclear. The role of TGFβsignalling in HCC is not well
defined as well. Here, we report that TGFβ2 is a target of Bmi1 in HCC and has a tumour-suppressing role. In Bmi1-
knockout mouse livers and HCC cell lines, TGFβ2/SMAD cascade proteins were upregulated. TGFβ2 expression was
inversely correlated with Bmi1 expression in human and mouse HCC tissues. In vitro, Bmi1 knockdown activated TGFβ2/
SMAD signalling and led to cell apoptosis via upregulation of p15 and p21. TGFβ2 inhibition rescued the inhibitory effect
of Bmi1 knockdown on HCC cell survival, proliferation, and cell-cycle progression. In vivo, restoration of TGFβ2
expression blocked Bmi1/Ras-driven hepatocarcinogenesis in mice. Chromatin immunoprecipitation and luciferase reporter
assays revealed that Bmi1 repressed TGFβ2 expression by binding to its promoter as a co-factor of polycomb repressor
complex 1. Our findings elucidate the molecular mechanism underlying hepatic Bmi1-driven carcinogenesis and highlight
the importance of TGFβ2 as a tumour suppressor in HCC development.
Introduction
Bmi1 is a member of the mammalian polycomb repressor
complex 1 (PRC1) and is involved in the regulation of
development, stem cell self-renewal, the cell cycle, and
senescence [1–4]. Bmi1 reportedly is overexpressed and
acts as an oncogene in multiple tumour types, including
breast cancer [5], colon carcinoma [6], non-small cell lung
cancer [7,8], glioblastoma [9], ovarian cancer [10], and
bladder cancer [11]. We and other groups have demon-
strated that Bmi1 is overexpressed and functions as an
oncogene in human hepatocellular carcinoma (HCC) [12–
14]. Our previous studies demonstrated that Bmi1 repres-
sion by either virus-encapsulated shRNA or liposome-
delivered siRNA inhibits HCC growth in mice [15,16]. In
accordance herewith, Chiba et al. reported that Bmi1 is
overexpressed in human HCC cell lines, and knockdown
(KD) of Bmi1 diminishes the ‘side population’in these
lines [17]. Together, these studies indicate that Bmi1 plays a
critical role in HCC and is a potential treatment target.
However, its molecular mechanism underlying HCC for-
mation and development remains unclear.
As a major member of the polycomb group protein
family, Bmi1 acts on its targets through repressive epige-
netic modification by ubiquitylating nucleosomal histone
H2A Lys119 [18]. Bmi1 regulates development and stem
cell self-renewal by repressing the INK4a/ARF locus,
which encodes p16INK4a and p14ARF (p19ARF in mice)
These authors contributed equally: Bin Li, Yuyuan Chen
*Chuanrui Xu
xcr@hust.edu.cn
1School of Pharmacy, Tongji Medical College, Huazhong
University of Science and Technology, Wuhan 430030, China
2Department of Biochemistry and Molecular Biology, School of
Basic Medicine, Huazhong University of Science and Technology,
Wuhan 430030, China
3Department of Hepatobiliary Surgery, Union Hospital, Tongji
Medical College, Huazhong University of Science and
Technology, Wuhan 430022, China
4Stem Cell Center, Union Hospital, Tongji Medical College,
Huazhong University of Science and Technology, Wuhan 430022,
China
Supplementary information The online version of this article (https://
doi.org/10.1038/s41388-019-1043-8) contains supplementary
material, which is available to authorized users.
1234567890();,:
1234567890();,:

[3,19,20]. In tumour development, Bmi1 functions in both
INK4a/ARF-dependent and -independent manners. Bmi1
represses INK4a/ARF in non-small cell lung cancer and
prostate cancer [21,22], but does not act on INK4a/ARF in
glioma and Ewing sarcoma [9,23]. In addition, Bmi1
reportedly inhibits either p16 or p19 [2,7]. Our previous
studies showed that Bmi1 drives HCC development inde-
pendent of INK4a/ARF [13,14]. Considering that Bmi1 is a
critical regulator in maintaining HCC growth, we specu-
lated that a hitherto uncovered signalling pathway mediates
hepatic carcinogenesis driven by Bmi1.
TGFβis a homodimer that exists in three isoforms in
mammalian cells: TGFβ1, TGFβ2, and TGFβ3[24,25]. All
TGFβligands transduce signal through the same receptor
signalling systems to activate downstream SMAD or BMP
signalling [25]. The three isoforms have non-redundant
roles in regulating cell growth, differentiation, matrix pro-
duction, and apoptosis [24]. In cancer, TGFβhas dual roles:
it inhibits tumour formation in the early stage and promotes
tumour metastasis in the advanced stage. Among the three
isoforms, TGFβ1 is the most widely investigated in many
cancer types [25]. However, there are few reports on the
role of TGFβ2 in cancer.
In this study, we screened potential targets of Bmi1 in
HCC cell lines and Bmi1-knockout (KO) mouse liver tis-
sues, and we found that TGFβ2 is a target of Bmi1 in HCC.
We confirmed the tumour-suppressor function of TGFβ2by
performing loss-of-function and gain-of-function experi-
ments in cell lines and mouse models. In addition, we
explored the mechanism underlying TGFβ2 repression by
Bmi1 in HCC cells. Our results uncover the mechanism by
which Bmi1 promotes hepatocarcinogenesis and illustrate
the significant role of TGFβ/SMAD signalling in liver
cancer.
Results
TGFβ/SMAD signalling is regulated by Bmi1 in HCC
cells and in mice
To screen for targets of Bmi1 in HCC, we performed cDNA
microarray analyses of Bmi1-KO mouse liver tissues and
Bmi1-KD Hep3B and Huh7 cells. We chose Hep3B and
Huh7 because both cell lines exhibit an early-stage HCC
signature and thus are appropriate for the identification of
alterations in signalling during the onset stage of HCC [26].
We selected genes with a fold-change > 1.2 and a p-value <
0.05 in Bmi1-KO hepatocytes and KD HCC cells as com-
pared with control hepatocytes and HCC cells, respectively.
In Bmi1-KO hepatocytes, 927 genes were significantly up-
or downregulated. In Bmi1-KD Huh7 and Hep3B cells, 493
genes were significantly up- or downregulated. Gene
ontology and pathway enrichment analyses using the
Database for Annotation, Visualization and Integrated
Discovery system (DAVID) revealed that TGFβsignalling
was the only pathway enriched in both Bmi1-KO mouse
hepatocytes and Bmi1-KD Huh7 and Hep3B cells (Fig. 1a, b).
Focusing on genes involved in TGFβ/SMAD signalling, we
found that most genes in this pathway were upregulated in
both the Bmi1-KO hepatocytes and the Bmi1-KD HCC
cells (Fig. 1c, d, and Supplementary Table 1). Therefore,
and considering the fact that TGFβ/SMAD signalling acts as
a‘brake’on the cell cycle whereas Bmi1 promotes cell-
cycle progression in HCC, we speculated that TGFβ/SMAD
signalling might be repressed in Bmi1-induced HCC.
To confirm our speculation and determine which isoform
is the major target of Bmi1, we examined TGFβ1, TGFβ2,
and TGFβ3 levels in Bmi1-KD Hep3B and Huh7 cells.
qRT-PCR analysis showed that only TGFβ2 mRNA was
increased in Bmi1-KD Hep3B and Huh7 cells, consistent
with the microarray data (Fig. 2a). Western blotting
revealed that TGFβ2 protein, but not TGFβ1, was upregu-
lated in the Bmi1-KO cells (Fig. 2b). Increased TGFβ2
expression after Bmi1 KD was confirmed by immuno-
fluorescence staining (Fig. 2c). Accordingly,
TGFβ2 secretion was increased in Bmi1-KD compared with
wild-type Hep3B and Huh7 cell-culture medium (Fig. 2d).
Collectively, these data suggested that Bmi1 regulates
TGFβ/SMAD signalling via repressing TGFβ2 in HCC
cells.
TGFβ2 is inversely correlated with Bmi1 in mouse
and human HCC tissues
We previously reported that Bmi1 is overexpressed in a
large fraction of human HCC and demonstrated its role as
HCC driver gene in a mouse model [13,27]. Given that
TGFβ2 is a target of Bmi1, we speculated that TGFβ2
would be repressed by Bmi1 in mouse and human HCC
tissues. To confirm this hypothesis, we evaluated TGFβ2
and Bmi1 expression in these tissues by immunohis-
tochemistry. In Bmi1/Ras-induced mouse HCC, Bmi1 was
highly expressed in tumour tissues and sporadically
expressed in adjacent tissues, whereas TGFβ2 was expres-
sed mainly in tumour-adjacent tissues, but seldom in tumour
tissues (Fig. 3a). Similarly, in Akt/Ras-induced mouse
HCC, in which Bmi1 is not the driver, but is required for
HCC formation [27], Bmi1 was more strongly expressed in
tumour tissues than in non-tumour tissues, whereas TGFβ2
was expressed at lower levels in tumour tissues than in non-
tumour tissues (Fig. 3b). The inverse correlation between
Bmi1 and TGFβ2 expression in both Bmi1/Ras and Akt/Ras
HCC tissues was confirmed by western blotting (Fig. 3c, d).
Next, we evaluated Bmi1 and TGFβ2 expression in
human HCC samples. In ten fresh human HCC tissues, both
B. Li et al.

western blotting and immunohistochemical staining
demonstrated that Bmi1 was highly expressed in tumour
versus adjacent tissues, whereas TGFβ2 was expressed at
lower levels in HCC than in adjacent tissues (Fig. 4a, b).
Likewise, p-Smad2 and p-Smad3 were expressed at lower
levels in tumour tissues than in adjacent tissues (Fig. 4b).
We analysed the correlation between Bmi1 and TGFβ2
expression using a HCC tissue microarray containing 75
pairs of HCC and adjacent tissues. Bmi1 was expressed in
41% (31/75) of HCC tissues, but in only 15% (11/75) of
adjacent tissues (Fig. 4c). TGFβ2 was expressed in 37%
(28/75) of HCC tissues and in 60% (45/75) of adjacent
tissues. These data confirmed that TGFβ2 and Bmi1
expression is inversely correlated in human HCC tissues,
corroborating a potential causal relation between Bmi1 and
TGFβ2 expression in HCC development.
Bmi1 blocks TGFβ2/SMAD signalling in HCC cells
In the canonical pathway, TGFβbinds to and activates its
receptors, subsequently activating downstream SMAD
proteins through phosphorylation and nuclear translocation
[25]. To confirm that TGFβ2/SMAD signalling is involved
in Bmi1-induced HCC, we examined SMAD protein levels
in Bmi1-KD HCC cells. Silencing of Bmi1 led to increased
TGFβ2 expression in both Hep3B and Huh7 cells (Fig. 5a).
Consequently, Smad2 and Smad3 phosphorylation was
increased. Smad4 showed evident nuclear translocation.
Immunostaining demonstrated increased Smad2 and Smad3
phosphorylation and nuclear translocation, and Smad4
nuclear localization (Fig. 5b). Co-immunoprecipitation
assays revealed that both Samd2 and Smad3 co-
precipitated with anti-Smad4 antibody in Bmi1-KD cells,
Fig. 1 Microarray analysis identifies TGFβsignalling as a downstream
target of Bmi1 in HCC. aSignalling pathways with fold-change > 1.2
and p< 0.05 in Bmi1−/−mice (8-week-old female mice, whole-body
knockout) were enriched by DAVID analysis. bSignalling pathways
enriched in Bmi1-KD Hep3B and Huh7 cells as indicated by DAVID-
based analysis. cExpression of TGFβcascade genes in Bmi1-KO mice
analysed by microarray. dmRNA levels of TGFβcascade genes in
Bmi1-KD Hep3B and Huh7 cells. (red =upregulation, green =
downregulation)
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

indicating that more Smad2 and Smad3 can bind Smad4
upon Bmi1 KD (Fig. 5c). Notably, Smad2/3 complex acti-
vates p21cip1 and p15INK4b in cooperation with C/EBPβand
Miz-1, respectively [25]. In line herewith, phosphorylation
and nuclear translocation of Smad2 and Smad3 led to
increased p21cip1 and p15INK4b expression in Bmi1-KD
Hep3B and Huh7 cells (Fig. 5a). Correspondingly, cyclin
D1 expression was reduced, suggesting that increased
p21cip1 and p15INK4b blocked cell-cycle progression. To
exclude possible off-target effects, we also knocked down
Fig. 2 Knockdown of Bmi1 leads to increased TGFβ2 in Hep3B and
Huh7 cells. Hep3B and Huh7 cells were treated with Bmi1 shRNA1
lentivirus for 3 days and then selected on puromycin for 2 days. Cells
were then collected and lysed for RNA or protein extraction, or used
for an immunofluorescence assay. Medium was collected for the
detection of secreted TGFβ2. aRT-qPCR analysis of TGFβ1, TGFβ2,
and TGFβ3 in Bmi1-KD Hep3B and Huh7 cells. rRNA was used as a
control and data are expressed as the mean ± SD (n=4). **p< 0.01.
bWestern blot for Bmi1 and TGFβ2 expression in Bmi1-KD Hep3B
and Huh7 cells. cImmunofluorescence of Bmi1 and TGFβ2 in Bmi1-
KD Hep3B and Huh7 cells. Nuclei were stained with DAPI. Scale bar
=50 μm. dLevels of secreted TGFβ2 detected by ELISA in Bmi1-KD
and WT cells. Data are expressed as the mean ± SD (n=3). **p< 0.01
B. Li et al.

Bmi1 using a Bmi1-siRNA pool of three siRNAs
(Bmi1 siRNA2) in Hep3B and Huh7 cells. As expected,
similar effects were observed in the two cell lines after
Bmi1 KD (Supplementary Fig. S1a–c). In addition, we
analysed the expression of betaglycan (also known as TGFβ
receptor type III, TGFBR3) in Hep3B and Huh7 cells. Both
Hep3B and Huh7 cells express high levels of betaglycan in
comparison with HeLa or DU145 cells, suggesting that
TGFβ2 may activate SMAD signalling by binding beta-
glycan (Supplementary Fig. S2). Collectively, these data
indicated that Bmi1 repressed TGFβ2 and downstream
SMAD signalling in HCC cells.
TGFβ2/SMAD signalling suppresses
hepatocarcinogenesis by inhibiting cell proliferation
As Bmi1 does not act on INK4a/ARF in HCC cells [13,27],
we next investigated whether Bmi1 promotes HCC cell
proliferation via repression of TGFβ2 using gain- and loss-
of-function experiments. In Hep3B and Huh7 cells, siRNA-
mediated Bmi1 silencing led to cell death, whereas treat-
ment with the TGFβreceptor inhibitor LY2157299
increased the survival of Bmi1-siRNA-treated HCC cells,
indicating that Bmi1 promotes cell growth via repressing
TGFβsignalling in these cells (Fig. 6a). Simultaneous
siRNA-mediated TGFβ2 silencing restored the growth of
Huh7 and Hep3B cells inhibited by Bmi1 KD (Fig. 6b). KD
of both Bmi1 and TGFβ2 was confirmed by western blot-
ting (Fig. 6c). These data corroborated that TGFβ2is
involved in cell-growth inhibition. A BrdU incorporation
assay showed that proliferation was increased by TGFβ2
inhibition in Bmi1-KD HCC cells (Fig. 6d). FACS analysis
revealed that TGFβ2 inhibition suppressed Bmi1-KD-
induced apoptosis (Fig. 6e). Bmi1 is essential for main-
tenance of the tumour-initiating capability of HCC cells
[14]; we determined whether this effect is dependent on
TGFβ2. A colony-formation assay showed that colonies
were reduced in Bmi1-KD Hep3B and Huh7 cells, whereas
TGFβ2 suppression reversed this effect (Fig. 6f). Of note,
KD of TGFβ2 alone only slightly promoted survival,
growth, proliferation, and migration of Hep3B and Huh7
cells, possibly because TGFβ2 is repressed by Bmi1 in
these two cell lines. Again, we confirmed these results using
Bmi1-siRNA/shRNA (Bmi1 siRNA2/shRNA2) and/or
TGFβ2 siRNA/shRNA (TGFβ2 siRNA2/shRNA2) pools of
three siRNA/shRNA sequences to knock down Bmi1 and/or
TGFβ2 in Hep3B and Huh7 cells, respectively (Supple-
mentary Fig. S3a–f). TGFβ2 repression could rescue HCC
cell growth repressed by Bmi1 KD (Supplementary
Fig. S4). Inhibition of TGFβ2 with its inhibitor pirfenidone
rescued the growth Hep3B and Huh7 blocked by the Bmi1
inhibitor PTC-209 (Supplementary Fig. S5). Collectively,
Fig. 3 TGFβ2 protein levels
negatively correlate with Bmi1
expression in mouse HCC.
Expression of TGFβ2 and Bmi1
was determined in mouse
HCC tissues generated by
hydrodynamic injection of
Bmi1/Ras or Akt/Ras together
with transposase ‘sleeping
beauty’(SB) into FVB/N mice
(6-week-old female mice,
n=3 in each group).
aImmunohistochemical analysis
of Bmi1 and TGFβ2 expression
in Bmi1/Ras mouse HCC
tissues. bImmunohistochemical
analysis of Bmi1 and TGFβ2in
Akt/Ras mouse HCC tissues.
Red dotted lines outline HCC
nodules. ‘N’indicates non-
tumour tissue and ‘T’indicates
tumour tissues. c,dWestern blot
for Bmi1 and TGFβ2 expression
in Bmi1/Ras and Akt/Ras mouse
HCC tissues. Samples were
from three mice in each group.
Scale bar =50 μm
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

these experiments demonstrated that TGFβ2 repressed HCC
cell proliferation and Bmi1 promoted HCC cell proliferation
via repressing TGFβ2.
As Bmi1 drives cell-cycle progression in many cancer
types, including HCC, we examined whether the stimula-
tory effect of Bmi1 on HCC cell-cycle progression depends
on TGFβ2 repression. Bmi1 KD by siRNA resulted in the
accumulation of Hep3B and Huh7 cells in G1 phase
(Supplementary Fig. S6a). Western blotting showed that p-
Smad2 and p-Samd3 as well as cyclins D and E were
reduced following Bmi1 KD (Supplementary Fig. S6b).
Accordingly, cell-cycle inhibitors p15, p21, and c-Myc
were increased. In contrast, inhibition of TGFβ2 by either
inhibitor or siRNA alleviated the G1 arrest in the Bmi1-KD
HCC cells (Supplementary Fig. S6a). p-Smad2 and p-
Smad3 as well as p15, p21, and c-Myc were decreased,
whereas cyclins D and E were increased (Supplementary
Fig. S6b).
TGFβ1 signals via the same downstream proteins and
contributes to the similar cellular mechanisms as TGFβ2.
TGFβ2 siRNA reportedly inhibits TGFβ1 expression to
some extent [28]. To rule out the possibility that TGFβ1
played a role in HCC cell survival and growth after both
Bmi1 and TGFβ2 KD, we evaluated TGFβ1 expression in
HCC cells treated with TGFβ2 siRNA and the effect of
TGFβ1 KD on Bmi1-KD HCC cells. In line with the
observation by Oh et al. [28], TGFβ2 siRNA slightly
reduced the level of TGFβ1 (Supplementary Fig. S7a). In
addition, cell growth in Hep3B and Huh7 cells after Bmi1
KD was promoted strongly by TGFβ2 KD, but not by
TGFβ1 KD (Supplementary Fig. S7b, c). These results
indicated that TGFβ2, not TGFβ1, mediates the inhibitory
role of Bmi1 KD in HCC cells. Together, these data indi-
cated that Bmi1 promotes cell-cycle progression via
repressing TGFβ2/SMAD signalling in HCC cells.
TGFβ2 is a direct target of Bmi1 in HCC cells
We next asked how Bmi1 represses TGFβ2 in HCC cells.
As a core protein of PRC1, Bmi1 generally acts on its target
genes as a transcriptional repressor [29–32]. Therefore, we
investigated whether Bmi1 directly binds to the TGFβ2
promoter using luciferase reporter and chromatin immuno-
precipitation (ChIP) assays. Noma et al. reported that the
Fig. 4 TGFβ2 expression is
inversely correlated with Bmi1
in clinical HCC tissues. aBmi1
and TGFβ2 levels in ten human
HCC tissues determined by
western blotting. ‘P’represents
para-tumour tissue and ‘T’
represents tumour tissue. Note
that 50-kD TGFβ2 bands refer to
its full-length or latent form
(cleaved but still non-covalently
bound). bRepresentative images
of H&E staining and
immunohistochemical staining
of Bmi1 and TGFβ2 in ten
human HCC tissues. Scale bar
=20 μm. ‘N’represents non-
tumour tissue and ‘T’represents
tumour tissue. Staining intensity
was calculated as integrated
optic density in the same area in
different sections using software
Image-Pro Plus and then
normalized to that in para-
tumour sections. #p< 0.05,
**p< 0.01. cRepresentative
images and statistics of
Bmi1 and TGFβ2 protein
levels determined by
immunohistochemical staining
using a human HCC tissue
microarray containing 75 paired
HCC and adjacent tissues
B. Li et al.

promoter sequence of TGFβ2 stretches from −1729 to +63,
with a core promoter sequence spanning from −508 to +63
that has transcriptional activity equivalent to that of the full
promoter [33]. To ensure luciferase expression, we cloned
both the full (−2000 to +100, promoter 1) and the core
(−508 to +63, promoter 2) TGFβ2 promoters into the
luciferase reporter pGL3 (Fig. 7a). Both promoter sequen-
ces effectively drove luciferase expression, consistent with
the observation of Noma et al. (Fig. 7b). Notably, KD of
Bmi1 in Huh7 cells significantly increased luciferase
expression under both TGFβ2 promoters, whereas over-
expression of Bmi1 had the opposite effect (Fig. 7c). These
results suggested that Bmi1 binds to the TGFβ2 core
promoter region (promoter 2). To confirm this, we per-
formed a ChIP assay using antibodies against Bmi1 and
Ring1B. We used Ring1B because this is an another core
protein of PRC1 [30–32]. We designed two primer pairs to
detect eluted TGFβ2 promoter DNA (Fig. 7d). As expected,
TGFβ2 promoter DNA was precipitated by antibodies
against both Bmi1 and Ring1B (Fig. 7e, f). Overexpression
of Bmi1 led to significantly higher precipitation of TGFβ2
promoter DNA. These results confirmed that Bmi1 directly
binds to the TGFβ2 promoter together with Ring1B.
Together with the luciferase reporter assays, these experi-
ments indicated that Bmi1 protein binds to the TGFβ2
promoter via PRC1 and thus inhibits TGFβ2 transcription.
Fig. 5 Canonical TGFβ2/SMAD signalling is activated by silencing
Bmi1 in Hep3B and Huh7 cells. aWestern blot of TGFβ2/SMAD
pathway proteins in Hep3B and Huh7 cells after Bmi1 knockdown
with siRNA1. Proteins were extracted from Hep3B or Huh7 cells after
Bmi1 siRNA1 transfection for 48 h. Scramble siRNA (SC) was used as
a control. bImmunofluorescence of p-Smad2, p-Smad3, and Smad4 in
Hep3B and Huh7 cells. p-Smad2, p-Smad3, and Smad4 were stained
red, and cell nuclei were counterstained with DAPI (blue). Scale
bar =20 μm. cWestern blot detection of Smad2 and Smad3 pre-
cipitated with Smad4 antibody. After 3 days of infection with
Bmi1 shRNA1 lentivirus and 2 days of puromycin selection, Hep3B
and Huh7 cells were collected and lysed. Cell lysates were incubated
with anti-Smad4 antibody for 2 h and then with protein A/G agarose at
4 °C overnight. Levels of Smad2, Smad3, and Smad4 were analyzed
by western blotting
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

B. Li et al.

Ectopic expression of TGFβ2 inhibits Bmi1/Ras-
induced HCC in mice
The fact that Bmi1 together with Ras can induce HCC in
mice is direct evidence of the oncogenic role of Bmi1 in
hepatocarcinogenesis [13,27]. Thus, we next asked whether
repression of TGFβ2 is responsible for Bmi1-induced HCC
formation and progression. To address this question, we
examined whether restoration of TGFβ2 could block HCC
formation induced by Bmi1/Ras in mice. We cloned TGFβ2
into pT3-EF1a/Bmi1 downstream of Bmi1, with an inde-
pendent internal ribosome entry site (IRES) (Fig. 8a). We
injected FVB/N mice with the sleeping beauty transposase
together with Bmi1/Ras or Bmi1/TGFβ2/Ras and monitored
tumour formation (Fig. 8b). In the second week post
injection, western blotting showed that Bmi1 and Ras pro-
tein levels in Bmi1/TGFβ2/Ras mice were comparable to
those in Bmi1/Ras mice, whereas TGFβ2 was expressed in
Bmi1/TGFβ2/Ras-, but not in Bmi1/Ras-injected liver tis-
sues (Supplementary Fig. S8). After 20 weeks, Bmi1/Ras
injection resulted in multiple HCC nodules spread over the
liver surface, whereas Bmi1/TGFβ2/Ras-injected mice had
only sporadic HCC tumours on the liver surface (Fig. 8c).
Bmi1/TGFβ2/Ras-injected mice had similar body weights
as Bmi1/Ras mice, but their livers weighed less, indicating
reduced tumour burden in Bmi1/TGFβ2/Ras mice (Fig. 8d).
Quantitative analysis showed that each mouse in the Bmi1/
Ras group (n=8) had at least two tumours on the liver,
with an average tumour number of six and a largest tumour
volume of 1560 mm3(Fig. 8e). In contrast, only two mice in
the Bmi1/TGFβ2/Ras group (n=10) had tumours (four and
one, respectively), and the largest tumour volume was
17 mm3. In Bmi1/Ras-induced HCC, TGFβ2 and p-Smad2/
p-Smad3 were repressed, whereas cyclins D1 were robustly
expressed (Fig. 8f). In Bmi1/TGFβ2/Ras HCC, TGFβ2 and
p-Smad2/p-Smad3 levels were evidently higher and cyclin
D1 levels were lower than those in Bmi1/Ras tumours.
Slower tumour growth was confirmed by attenuated
Ki67 staining in Bmi1/TGFβ2/Ras HCC tissues. These
results suggested that the restoration of TGFβ2 blocked
hepatocarcinogenesis induced by Bmi1 and Ras. To confirm
that the effects of restoration of TGFβ2 were attributable to
Bmi1, we examined the effect of TGFβ2 silencing in Bmi1-
KO tumours using a xenograft tumour model. In sub-
cutaneous tumours established using Hep3B cells, shRNA-
mediated Bmi1 silencing significantly inhibited tumour
growth, whereas KD of TGFβ2 alone had a slightly sti-
mulatory effect (Supplementary Fig. S9a). As expected, co-
infection with TGFβ2 shRNA lentivirus partially reversed
the inhibitory effect of Bmi1 repression on tumour growth.
Examination of tumour volumes and weights indicated that
TGFβ2 KD reversed the effect of Bmi1 KO on tumour
growth (Supplementary Fig. S9b, c). Western blotting
confirmed that both Bmi1 and TGFβ2 were repressed by
shRNA in those tumours (Supplementary Fig. S9d). These
results corroborated that increased TGFβ2 is responsible for
the inhibitory effect of Bmi1 KD on HCC growth and that
Bmi1 drives HCC by repressing TGFβ2.
Discussion
The current study demonstrated that elevated Bmi1
repressed TGFβ2 expression in HCC cells by directly
binding to its promoter, thus blocking its transcription.
Repression of TGFβ2 inhibited SMAD signalling and
resulted in cell-cycle progression and the propagation of
HCC cells. In mice, overexpression of TGFβ2 in the liver
significantly suppressed hepatic carcinogenesis driven by
Bmi1. Therefore, our findings provide robust evidence to
support that TGFβ2 is a direct target of Bmi1 in HCC and
that TGFβ/SMAD signalling functions as a ‘brake’on
Bmi1-driven hepatic carcinogenesis (Supplementary Fig.
S10).
In this study, we focused on TGFβ2 and SMAD sig-
nalling for several reasons. First, we found that SMAD
signalling was elevated in both Bmi1-KO mouse hepato-
cytes and Bmi1-KD human HCC cell lines. Second, SMAD
signalling blocks cell-cycle progression, exactly opposite to
the role of Bmi1 in HCC. Third, Bmi1 negatively regulates
gene expression at the transcriptional level, whereas SMAD
proteins are activated mainly by phosphorylation. There-
fore, we speculated that Bmi1 represses the SMAD pathway
via transcriptional inhibition of TGFβs. Finally, only
TGFβ2 among the TGFβisoforms was repressed by Bmi1
(Fig. 2). Accordingly, the role of TGFβ2 in Bmi1-driven
HCC was examined in this study.
The first significance of this study is that we uncovered
an important target of Bmi1 in HCC. A number of studies
revealed that Bmi1 acts as an oncogene independently of its
ability to repress INK4a/ARF in some cancers, including
Fig. 6 Blocking TGFβ2 reverses the growth inhibition caused by
Bmi1 suppression. aCell viabilities of Hep3B and Huh7 cells after
transfection with Bmi1 siRNA1 and/or treatment with TGFβ2 inhibitor
LY2157299 for 48 h. bCell viabilities of Hep3B and Huh7 cells after
transfection with Bmi1 siRNA1 and/or TGFβ2 siRNA1 for 48 h.
cProtein levels of Bmi1 and TGFβ2 detected by western blotting in
Hep3B and Huh7 cells treated with Bmi1 and/or TGFβ2 siRNA1 for
48 h. dResults of BrdU assay to determine the proliferation of Hep3B
and Huh7 cells treated with Bmi1 siRNA1 and/or TGFβ2 siRNA1 for
48 h. eFACS analysis of apoptosis in Hep3B and Huh7 cells treated
with Bmi1 siRNA1 and/or TGFβ2 siRNA1 for 48 h. fColonies formed
by Hep3B and Huh7 cells infected with Bmi1 shRNA1 and/or
TGFβ2 shRNA1 lentivirus. Note that colonies formed by Bmi1-KD
and TGFβ2-KD cells were smaller than those formed by Bmi1 WT and
TGFβ2-KD cells, possibly because of the larger amount of virus used
for double knockdown in the former. Data are expressed as the mean ±
SD (n=4). * or #p< 0.05, **p< 0.01
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

HCC [9,13,14,17,23,34–37]. Therefore, there are other
targets involved in Bmi1-driven tumorigenesis that
remained to be identified. Our results from both Bmi1-KO
mice and Bmi1-KD HCC cell lines indicated that the levels
of TGFβ2 and downstream effectors were increased upon
Bmi1 suppression. Importantly, TGFβ2 repression restored
Fig. 7 Bmi1 inhibits TGFβ2
transcription in vitro by binding
to its promoter. aSchematic
illustration of the two luciferase
reporter constructs used.
bLuciferase activities in Huh7
cells transfected with luciferase
reporter constructs containing
entire or core promoter for 48 h.
cLuciferase activities in Huh7
cells transfected with luciferase
reporter plasmids and Bmi1-
siRNA or Bmi1-expression
plasmid for 48 h. dLocation of
the two primer pairs used for
ChIP on the TGFβ2 promoter.
e,fChIP assay results using
respectively Ring1B and Bmi1
antibodies to pull down TGFβ2
promoter DNA. Chromatin was
extracted from Huh7 cells
transfected with pT3-Bmi1 or
pT3. IgG and pT3 were used as
control antibody and empty
vector, respectively. Data are
expressed as mean ± SD (n=4).
*or#p< 0.05. ** or ##p< 0.01
B. Li et al.

HCC cell growth inhibited by Bmi1 silencing, whereas
repression of TGFβ2 alone did not affect HCC cell growth,
indicating that TGFβ2/SMAD signalling mediates the reg-
ulatory function of Bmi1 on HCC cell proliferation. Hence,
these data demonstrated that TGFβ2/SMAD proteins are
critical targets of Bmi1 in HCC.
Furthermore, our findings explain how Bmi1 regulates
cell-cycle progression in HCC independent of INK4a/ARF.
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

Bmi1 has a dual role in inducing tumorigenesis: it promotes
cell-cycle progression and inhibits apoptosis [4]. In the
canonical pathway, Bmi1 promotes cell-cycle progression
by repressing p16INK4A and blocks apoptosis by inhibiting
p14ARF [38]. However, how Bmi1 regulates cell-cycle
progression independent of INK4a/ARF in cancer remains
largely unknown. Our results showed that Bmi1 KD led to
cell-cycle arrest in G1, whereas repression of TGFβ2 could
relieve the blockade. Interestingly, TGFβs reportedly inhibit
cell proliferation by causing growth arrest in the G1 phase
[39]. TGFβ-induced G1 arrest has been attributed to the
inhibitory effect of TGFβon both the levels and activities of
G1 cyclins and various cyclin-dependent protein kinases
[40], as well as on cyclin-dependent kinase inhibitors,
including p21WAF1, p27KiPl, p16, and pl5INK4B [39,41,42].
The present study highlights the important role of TGFβ2in
controlling cell-cycle progression and helps understand how
Bmi1 regulates the cell cycle via TGFβ2 in HCC cells.
Based on these previous findings and our observations, we
conclude that Bmi1 represses TGFβ2 and subsequent
SMAD proteins and hence, promotes cell-cycle progression
via multiple cell-cycle-regulatory pathways.
Interestingly, several recent reports revealed that Bmi1
acts on TGFβ/SMAD signalling in other cancer types.
Gargiulo et al. showed that Bmi1 controls glioblastoma
multiforme development by regulating TGFβ/BMP signal-
ling [43]. Kim et al. reported that Bmi1 suppresses senes-
cence and prolongs life span by inhibiting TGFβsignalling
in normal human oral keratinocytes [44]. In general, our
data are consistent with these studies in that Bmi1 drives
tumour formation by regulating TGFβsignalling. However,
we provided more robust evidence of the role of TGFβ2in
Bmi1-induced HCC. The HCC mouse model experiments
demonstrated that repression of TGFβ2 is essential for Bmi1
to promote cell-cycle progression and HCC cell prolifera-
tion. When TGFβ2 was restored, Bmi1 failed to induce
HCC formation. In addition, our study clarified the reg-
ulatory action of Bmi1 on TGFβ2. Using ChIP and luci-
ferase assays, we demonstrated that Bmi1 directly binds to
the promoter of TGFβ2 to represses its transcription.
Our data showed that Bmi1 KD led to increased secre-
tion of TGFβ2, which may act in an autocrine and/or
paracrine manner. Of note, secreted TGFβs form homo-
dimers after cleavage and then bind to receptors either
directly or with the assistance of the accessory co-receptor
betaglycan [45,46]. Betaglycan is particular critical for
TGFβ2, which acts in both betaglycan-dependent and
-independent pathways. In the canonical pathway, TGFβ2
first binds to betaglycan, then to its receptor Tβ-R2, which
recruits Tβ-R1. In the non-canonical pathway, TGFβ2 binds
to Tβ-R2/Tβ-R1 pair directly and activates SMAD signal-
ling [47,48]. One study reported that binding of TGFβ2to
Tβ-R2 requires co-expression of Tβ-RI or Tβ-RIII [49]. Our
results indicate that betaglycan is highly expressed in
Hep3B and Huh7 cells, suggesting that TGFβ2 may act in a
betaglycan-dependent manner in HCC cells. However, we
did not examine the impact of betaglycan deletion on
TGFβ2/SMAD signalling transduction. Whether TGFβ2/
SMAD signalling transduction depends on betaglycan in
HCC remains to be investigated in future.
Another important finding of our study is that the TGFβ/
SMAD axis functions as a tumour suppressor during HCC
development, which is contrary to the conclusion reached
by Dropmann et al. [50]. TGFβplays contradictory roles in
tumorigenesis: on the one hand, it suppresses tumour for-
mation; on the other hand, it promotes metastasis. In normal
and premalignant cells, TGFβenforces homoeostasis and
suppresses tumour formation through cell-autonomous
tumour-suppressive effects (cytostasis, differentiation, and
apoptosis) or effects from the micro-environment (sup-
pression of inflammation and stroma-derived mitogens)
[25]. After tumour formation, cancer cells may lose TGFβ
tumour-suppressive responses and use TGFβto their
advantage to advance immune evasion, growth factor pro-
duction, transformation into an invasive phenotype, and
metastatic dissemination [25]. Dropmann et al. reported that
TGFβ2 was highly expressed in many HCC patients, and its
levels were reversely correlated with survival, demonstrat-
ing an oncogenic role of TGFβ2[50]. We observed that
TGFβ2 repression by Bmi1 promoted HCC formation,
whereas restoration of TGFβ2 expression blocked tumour
formation induced by Bmi1 and Ras. Conversely, KD of
TGFβ2 restored tumour growth inhibited by Bmi1 repres-
sion in HCC xenografts. Therefore, our study confirms that
TGFβ2 is a tumour suppressor in HCC, at least at the early
stages.
Fig. 8 TGFβ2 inhibits tumour formation induced by Bmi1 and Ras in
FVB/N mice. aSchematic illustration of the four constructs used for
hydrodynamic injections. bIllustration of the establishment of trans-
genic Bmi1/Ras and Bmi1/TGFβ2/Ras HCC mice via hydrodynamic
injection. In both groups, Bmi1, Bmi1-TGFβ2, and Ras plasmids are
injected at the same amount into each FVB/N mouse (female, 6 weeks
old, n=8 for the Bmi1/Ras group and n=10 for the Bmi1/TGFβ2/
Ras group). cGross morphological images of livers from Bmi1/Ras
and Bmi1/TGFβ2/Ras mice. dBody/liver weights of Bmi1/Ras and
Bmi1/TGFβ2/Ras mice. eTumour numbers and maximal tumour
volumes in Bmi1/Ras and Bmi1/TGFβ2/Ras mice. Data are expressed
as the mean ± SD (n=8 for the Bmi1/Ras group and n=10 for the
Bmi1/TGFβ2/Ras group). *p< 0.05, **p< 0.01. fHaematoxylin and
eosin and immunohistochemical staining of liver tissues in Bmi1/Ras
and Bmi1/TGFβ2/Ras mice. Statistical analysis was performed based
on the ratio of positively stained nuclei versus total nuclei or relative
staining intensity in the tumour tissues. Relative staining intensity was
calculated by determining the integrated optic density with Image-Pro
Plus in Bmi1/TGFβ2/Ras HCC tissues and then normalized to that in
Bmi1/Ras tissues. Data are the mean ± SD (n=3 mouse sections for
both Bmi1/Ras and Bmi1/TGFβ2/Ras groups). *p< 0.05, **p< 0.01.
Scale bar =20 μm. ‘N’indicates non-tumour tissues and ‘T’indicates
tumour tissues
B. Li et al.

Our study provides several useful hints for targeted
therapy against either Bmi1 or TGFβin HCC. First, our data
indicate that TGFβ2 functions as a tumour suppressor in
early-stage HCC, especially in Bmi1-driven HCC. Both
cell-culture and mouse studies supported its anti-tumour
role in HCC formation or growth. Therefore, monotherapy
targeting TGFβmay be not applicable to HCC patients in
early stages or with high Bmi1 levels, and might even
worsen rather than cure the cancer. Second, therapeutic
intervention blocking Bmi1 in HCC may enhance TGFβ2
expression and subsequent SMAD signalling activation. As
we have demonstrated, activated TGFβ2/SMAD signalling
blocks HCC cell growth. However, whether it would pro-
mote HCC cell metastasis later on remains unknown. Given
that TGFβ/SMAD also promotes metastasis, sequential
treatment with Bmi1 inhibitors followed by TGFβ2 inhibi-
tors would be worthy of exploring. Last, but not least, our
tissue microarray data showed that some HCC tissues
exhibit high TGFβ2 expression, in line with findings in a
previous report [50]. However, the role of elevated TGFβ2
has not been clearly demonstrated in experimental or
mechanistic studies. Considering the dual roles of TGFβin
different contexts, whether therapies targeting TGFβ2 are
applicable in these patients requires more pre-clinical
studies.
Regretfully, we did not examine the long-term effects of
TGFβexpression in HCC progression and potential
metastasis, and thus failed to evaluate its role in advanced
HCC. We plan to evaluate the final outcome of over-
expressing TGFβin HCC development and progression in a
follow-up study. In addition, we noticed that in some clin-
ical tissues, increased Bmi1 was not accompanied with
decreased TGFβ2. It is possible that TGFβ2 is also regu-
lated by factors other than Bmi1 considering that there are
plentiful known driver oncogenes and signalling molecules
in HCC. Conversely, Bmi1 may have other targets other
than TGFβ2 in HCC. To elucidate the roles and mechan-
isms of both Bmi1 and TGFβ2 in hepatocarcinogenesis
further, more clinical samples and mouse models will be
utilized in our future studies, and we also hope to uncover
novel functions of Bmi1 and TGFβsignalling in other liver
cancer types.
Materials and methods
Cell culture and lentiviral infection
Human Hep3B and Huh7 HCC cell lines were purchased
from China Center for Type Culture Collection at Wuhan
University, China. Both cell lines were authenticated using
STR profiling and treated with plasmocin (25 μg/mL) for
2 weeks before used in this study. The cells were cultured in
DMEM supplemented with 10% foetal bovine serum at 37 °C.
Lentivirus expressing Bmi1 shRNA was transfected into
HCC cells to knock down Bmi1 expression. Three days
post infection, the cells were expanded and selected with
1μg/mL puromycin for 2 days, and then harvested for
protein and RNA extraction.
Constructs, siRNA, and transfection
The Bmi1-targeting shRNA1 construct Bmi1/pLKO.1
(TRCN0000020155, NM_005180.5-693s1c1) used to
knock down Bmi1 expression was obtained from Open-
Biosystems (Thermo Fisher, Waltham, MA, USA). Control
SC/pLKO.1 (with a scrambled sequence) plasmids were
obtained from Addgene. TGFβ1 shRNA (sc-270322-SH),
Bmi1 siRNA2 (sc-29814) and shRNA2 (sc-29814-v), and
TGFβ2 siRNA2 (sc-39802), shRNA1 (sc-39802-SH), and
shRNA2 (sc-39802-v) lentivirus pools were from Santa
Cruz Biotechnology (Santa Cruz, CA, USA). All shRNA
constructs were packaged into lentivirus and used for
transfection. The following siRNAs were used in this study:
Bmi1 siRNA1 sense: 5′-CCAGACCACUACUGAAUA
UAA-3′, anti-sense: 5′-UUAUAUUCAGUAGUGGUCUG
GUU-3′; TGFβ2 siRNA1 sense: 5′-CACUCGAUAUGGA
CCAGUUTT-3′, anti-sense: 5′-AACUGGUCCAUAUC
GAGUG-3′. SiRNA oligos were transfected into Huh7 and
Hep3B cells using Lipofectamine 2000 (Invitrogen, Carls-
bad, CA, USA), according to the manufacturer’s instruc-
tions. The hyperactive sleeping beauty construct (pCMV/
SB), Bmi1 construct pT3-EF1α-Bmi1-V5, N-Ras construct
pCaggs-NRasV12, and Akt construct pT3-myr-Akt-HA
were provided by Dr Xin Chen of the University of Cali-
fornia, San Francisco. The pT3-EF1αvector containing
duplicated inverted repeats for sleeping beauty-mediated
integration and the EF1αpromoter (pT3-EF1α) used for
hydrodynamic injection were previously described [51].
Construction of pT3-EF1α-Bmi1-V5, pCaggs-NRasV12,
and pT3-myrAkt-HA has been described elsewhere
[13,52]. Human TGFβ2 and an IRES sequences were
cloned into pT3-EF1α-Bmi1-V5 to generate pT3-EF1α-
Bmi1-Flag-IRES-TGFβ2-V5. All plasmids were purified
using the Endotoxin-free Maxi Prep Kit (Omega Bio-Tek,
Norcross, GA, USA) before injection into mice.
Cell viability, proliferation, growth, colony-
formation, and cell-cycle analyses
Hep3B or Huh7 cells were transfected with siRNA or
shRNA or treated with drugs for 48 h. ShRNA-transfected
cells were treated with puromycin (1 µg/mL) for 48 h. For
cell viability assays, cells were seeded into 96-well plates at
5000 cells/well and then transfected with siRNA or treated
with drugs. Then, cell viability was determined using an
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

MTT detection kit (Beyotime, Beijing, China). To assay cell
growth, 5000 cells/well were seeded in 12-well plates and
cells were recounted at indicated time points. Cell pro-
liferation was assayed using BrdU labelling as described
previously [53]. For colony-formation assays, 5 × 103viable
transfected cells/well were seeded in 6-well plates and
maintained in complete medium for 1 week. Colonies were
fixed with methanol and stained with 0.1% crystal violet in
20% methanol. Colonies were counted with Image J (NIH,
Bethesda, MD, USA). Triplicate wells were prepared for
each transfectant and all experiments were repeated twice.
The cell cycle was analysed by flow cytometry (Cyto-
micsTM FC 500; Beckman Coulter, Brea, CA, USA) after
propidium iodide staining, and the results were analysed
using FlowJo7.6 (FlowJo, Ashland, OR, USA).
Quantitative reverse-transcription (RT-q)PCR,
western blotting, and co-immunoprecipitation
Total RNA was extracted from cells or frozen liver tissues
using TRIzol (Invitrogen) and digested with DNase I. After
reverse transcription, SYBR green-based qPCR was carried
out to detect mRNA levels of genes of interest. The primer
pairs used are listed in Supplementary Table 2.
For western blotting, liver or HCC tissues/cells were
lysed in M-PER Mammalian Protein Extraction Buffer
(Thermo Fisher, Rockford, IL, USA) plus Proteinase Inhi-
bitor Cocktail (Roche, Indianapolis, IN, USA) and Halt
Phosphatase Inhibitor Cocktail (Thermo Fisher). Proteins
were quantified using a BCA protein assay kit (Thermo
Fisher).
For co-immunoprecipitation, cells were harvested and
incubated with IP-lysis buffer (Beyotime). Cell lysates were
incubated with rabbit anti-Smad4 monoclonal antibody
(1:50, Cell Signalling) at 4 °C for 2 h, followed by incu-
bation with 20 μL of protein A/G agarose (Santa Cruz
Biotechnology, CA, USA) at 4 °C overnight. Immunopre-
cipitates were washed four times with lysis buffer and
analysed by western blotting using anti-Smad2 and anti-
Smad3 monoclonal antibodies (1:1000, Cell Signalling).
Antibodies used for western blotting and co-
immunoprecipitation are listed in Supplementary Table 3.
Enzyme-linked immunosorbent assay (ELISA) for
TGFβ2
Hep3B and Huh7 cells were infected with Bmi1 shRNA
lentivirus for 48 h and selected on puromycin for 48 h.
Then, Bmi1 KD and control (infected with scramble
shRNA) cells were seeded into 6-well plates at 4 × 105
cells per well. After 24 h, the culture supernatant was
collected. TGFβ2 concentrations were measured using an
ELISA kit for human TGFβ2(CusabioBiotechCo.,
Houston, TX, USA) according to the manufacturer’s
instruction. In brief, purified human TGFβ2 standards and
samples were incubated in a high-binding 96-well
microtiter plate pre-coated with TGFβ2 antibody at 37 °C
for 2 h. Subsequently, liquid was removed and diluted
biotin-conjugated antibody was added to each well. After
1 h of incubation at 37 °C and washing, the plate was
incubated with horseradish peroxidase-avidin working
solution at 37 °C for 1 h. After final washing, the plates
were incubated with 3,3′,5,5′-tetramethylbenzidine at
37 °C in the dark for 30 min. Finally, the optical density at
450 nm was determined with a Multilabel Reader Spec-
trophotometer (Victor X5, Perkin Elmer). The con-
centration of TGFβ2 in each well was calculated based on
the standard curve.
Histology, immunohistochemistry, and
immunofluorescence
Mice were euthanized and the livers were removed and
rinsed in PBS. Samples collected from the livers were either
frozen in dry ice for RNA and protein extraction or fixed
overnight in fresh, cold 4% paraformaldehyde. Fixed tissue
samples were embedded in paraffin and sectioned to 5 μm
for haematoxylin and eosin (H&E) or immunohistochemical
staining. For immunofluorescence staining, cells were fixed
with 4% paraformaldehyde, permeabilized with 0.1% Triton
X-100, and pre-treated with 0.5% goat serum. The cells
were incubated with primary antibodies for 1.5 h and sub-
sequently with Alexa Flour 596-conjugated AffiniPure goat
anti-rabbit IgG (Proteintech, Rosemont, IL, USA) at a 1:100
dilution for 45 min. Nuclei were counterstained with DAPI.
Antibodies used for immunohistochemistry and immuno-
fluorescence are listed in Supplementary Table 3. Staining
intensity was quantified using Image-Pro Plus (Media
Cybernetics, Rockville, MD, USA, version 7.0) to calculate
the integrated optic density.
Primary and xenograft HCC mouse models
Wild-type FVB/N mice and BALB/c nude mice were
obtained from Huaukang Technology Corporation (Beijing,
China) and were fed standard rodent chow. All mice were
housed, fed, and treated in accordance with protocols
approved by the Animal Experiments Ethical Committee at
Huazhong University of Science and Technology. All the
mice used in this study were randomized to different groups
using an online randomization tool Randomizer (https://www.
randomizer.org/), but the animal experiments were not blin-
ded. Hydrodynamic transfection was performed as described
previously [51]. To generate Bmi1/Ras HCC in mice, plas-
mids (pT3-EF1α-Bmi1-V5:pCaggs-NRasV12:pCMV-SB =
25:25:4 μgorpT3-EF1α-Bmi1-FLAG-TGFβ2-V5:pCaggs-
B. Li et al.

NRasV12:pCMV-SB =25:25:4 μg per 20 g of mouse weight)
were diluted in 2 mL of saline, filtered through a 0.22-μm
filter, and injected into the tail vein of 6-week-old female
FVB/N mice (18–20 g) in 5–7 s. The mice were monitored
weekly and sacrificed at 20 weeks post injection for growth
evaluation and histological examination. To generate Akt/Ras
HCC in mice, plasmids (pT3-myr-Akt-HA:pCaggs-
NRasV12:pCMV-SB =25:25:4 μg per 20 g of mouse weight)
were injected into the tail vein of 6-week-old female FVB/N
mice (18–20 g) in 5–7s.
For the xenograft tumour model, subcutaneous tumours
were established by inoculating 5 × 106Hep3B cells in the
front armpit of 6-week-old female BALB/c nude mice
(18–20 g). Prior to inoculation, the Hep3B cells were
infected with Bmi1 and/or TGFβ2 shRNA lentivirus and
selected with puromycin (1 µg/mL). Mice inoculated with
Hep3B cells transfected with sc/pLKo.1 lentivirus served as
a control. Tumour sizes and mouse weights were measured
simultaneously. Tumour volume (V) was monitored by
measuring the length (L) and width (W) with a vernier
caliper and calculated as V=(L×W2) × 0.5. At the end of
the experiment, mice were sacrificed and tumours were
collected and photographed.
Microarray analysis
Liver tissues of Bmi1−/−mice (whole-body knockout) were
kindly provided by Dr Xin Chen of the University of
California, San Francisco. Total RNA was extracted from
HCC cells and 8-week-old female wild type and Bmi1−/−
mouse liver tissues. Human (HuGene 1.0 ST; Affymetrix,
Santa Clara, CA, USA) and mouse (MoGene 1.0 ST;
Affymetrix) GeneChip arrays were used for gene expression
profiling. Hybridization, washing, staining, scanning, and
data analysis were performed by Phalanx Biotech Group
(Palo Alto, CA, USA). Expression levels were analysed
using Microarray Analysis Suite 5.0 (Affymetrix). The
microarray data were deposited in the NCBI Gene
Expression Omnibus public database (http://www.ncbi.nlm.
nih.gov/geo/, accession number GSE97172).
Patients and tissue samples
Ten fresh HCC tissue samples were collected at Union
Hospital affiliated to the Tongji Medical College of Huaz-
hong University of Science and Technology (Wuhan,
China) in 2016. The collection of HCC tissues was
approved by Medical Ethics Committees of Huazhong
University of Science and Technology and written informed
consent was obtained from all patients before surgery. The
HCC tissue microarray (Cat Number: LivH150CS03; Lot
Number.: XT15-037) was from Shanghai Outdo Biotech
Company (Shanghai, China).
Luciferase reporter and ChIP assays
The regions −2000 to +100 and −508 to +63 of the human
TGFβ2 promoter were PCR-amplified and cloned into the
pGL3 luciferase vector. To analyse TGFβ2 transcriptional
activity, Huh7 cells were transfected with the pGL3 plasmid
and an internal control reporter plasmid, pRL-TK (Promega,
Madison, WI, USA), along with Bmi1-siRNA or pT3-Bmi1
plasmid. Forty-eight hours after transfection, luciferase
activity was measured with the Dual-Glo luciferase assay
system (Promega) according to the manufacturer’s
instructions.
ChIP assays were essentially performed as previously
described [54,55]. ChIP DNA was analysed by qPCR with
SYBR Green (Bio-Rad Laboratories, Berkeley, California,
USA) in an ABI-7500 instrument (Thermo Fisher) using the
following primers: primer 1F: 5′-TTGGGAGGCTGTGA
CTGA-3′, primer 1R: 5′-GCTGTGGGTAAGGGAGGA-3′;
primer 2F: 5′-TAATACAGGAGGGAAGCC-3′, primer
2R: 5′-TGCCAGCAGATAACATCA-3′. Antibodies used
were as follows: rabbit anti-Ring1B (Abcam, Cambridge,
MA, USA) and normal rabbit IgG (Abcam).
Statistical analysis
Data are reported as the mean ± standard deviation (SD).
Means of two groups were compared using Student’sttest
(SPSS Software, Chicago, IL, USA). Means of multiple
groups were compared by one-way ANOVA and Dun-
nett’s posttests. Variances between multiple groups were
statistically compared. Values of p<0.05 and p<0.01
were considered significant and highly significant,
respectively. Sample sizes of animal experiments were
calculated using the ‘resource equation’:E=total number
of animals−total number of groups, in which Eshould lie
between 10 and 20 [56].
Acknowledgements This study was supported by the National Science
Foundation of China (81572723 and 81872253) and the Innovation
Foundation of Higher Education of China (2016JCTD109).
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflicts of
interest.
Publisher’s note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
References
1. Jacobs JJ, Kieboom K, Marino S, DePinho RA, van Lohuizen M.
The oncogene and Polycomb-group gene bmi-1 regulates cell
proliferation and senescence through the ink4a locus. Nature.
1999a;397:164–8.
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis

2. Molofsky AV, Pardal R, Iwashita T, Park IK, Clarke MF, Mor-
rison SJ. Bmi-1 dependence distinguishes neural stem cell self-
renewal from progenitor proliferation. Nature. 2003;425:962–7.
3. Park IK, Qian D, Kiel M, Becker MW, Pihalja M, Weissman IL,
et al. Bmi-1 is required for maintenance of adult self-renewing
haematopoietic stem cells. Nature. 2003;423:302–5.
4. Park IK, Morrison SJ, Clarke MF. Bmi1, stem cells, and senes-
cence regulation. J Clin Investig. 2004;113:175–9.
5. Pietersen AM, Horlings HM, Hauptmann M, Langerod A,
Ajouaou A, Cornelissen-Steijger P, et al. EZH2 and BMI1
inversely correlate with prognosis and TP53 mutation in breast
cancer. Breast Cancer Res. 2008;10:R109.
6. Dhawan S, Tschen SI, Bhushan A. Bmi-1 regulates the Ink4a/Arf
locus to control pancreatic beta-cell proliferation. Genes Dev.
2009;23:906–11.
7. Dovey JS, Zacharek SJ, Kim CF, Lees JA. Bmi1 is critical for
lung tumorigenesis and bronchioalveolar stem cell expansion.
Proc Natl Acad Sci USA. 2008;105:11857–62.
8. Becker M, Korn C, Sienerth AR, Voswinckel R, Luetkenhaus K,
Ceteci F, et al. Polycomb group protein Bmi1 is required for
growth of RAFdriven non-small-cell lung cancer. PLoS ONE.
2009;4:e4230.
9. Bruggeman SW, Hulsman D, Tanger E, Buckle T, Blom M,
Zevenhoven J, et al. Bmi1 controls tumor development in an
Ink4a/Arf-independent manner in a mouse model for glioma.
Cancer Cell. 2007;12:328–41.
10. Jacobs JJ, Scheijen B, Voncken JW, Kieboom K, Berns A, van
Lohuizen M. Bmi-1 collaborates with c-Myc in tumorigenesis by
inhibiting c-Myc-induced apoptosis via INK4a/ARF. Genes Dev.
1999b;13:2678–90.
11. Kalra N, Kumar V. c-Fos is a mediator of the c-myc-induced
apoptotic signaling in serum-deprived hepatoma cells via the p38
mitogen-activated protein kinase pathway. J Biol Chem.
2004;279:25313–9.
12. Levy L, Renard CA, Wei Y, Buendia MA. Genetic alterations and
oncogenic pathways in hepatocellular carcinoma. Ann N Y Acad
Sci. 2002;963:21–36.
13. Xu CR, Lee S, Ho C, Bommi P, Huang SA, Cheung ST, et al.
Bmi1 functions as an oncogene independent of Ink4A/Arf
repression in hepatic carcinogenesis. Mol Cancer Res.
2009;7:1937–45.
14. Chiba T, Miyagi S, Saraya A, Aoki R, Seki A, Morita Y, et al. The
polycomb gene product BMI1 contributes to the maintenance of
tumor-initiating side population cells in hepatocellular carcinoma.
Cancer Res. 2008;68:7742–9.
15. Qi S, Li B, Yang T, Liu Y, Cao S, He X, et al. Validation of Bmi1
as a therapeutic target of hepatocellular carcinoma in mice. Int J
Mol Sci. 2014;15:20004–21.
16. Yang T, Li B, Qi S, Liu Y, Gai Y, Ye P, et al. Co-delivery of
doxorubicin and Bmi1 siRNA by folate receptor targeted lipo-
somes exhibits enhanced anti-tumor effects in vitro and in vivo.
Theranostics. 2014;4:1096–111.
17. Chiba T, Seki A, Aoki R, Ichikawa H, Negishi M, Miyagi S, et al.
Bmi1 promotes hepatic stem cell expansion and tumorigenicity in
both Ink4a/Arf-dependent and -independent manners in mice.
Hepatology. 2010;52:1111–23.
18. Goel HL, Chang C, Pursell B, Leav I, Lyle S, Xi HS, et al. VEGF/
neuropilin-2 regulation of Bmi-1 and consequent repression of
IGF-IR define a novel mechanism of aggressive prostate cancer.
Cancer Discov. 2012;2:906–21.
19. Molofsky AV, He S, Bydon M, Morrison SJ, Pardal R. Bmi-1
promotes neural stem cell self-renewal and neural development
but not mouse growth and survival by repressing the p16Ink4a
and p19Arf senescence pathways. Genes Dev. 2005;19:1432–7.
20. Sangiorgi E, Capecchi MR. Bmi1 is expressed in vivo in intestinal
stem cells. Nat Genet. 2008;40:915–20.
21. Fan C, He L, Kapoor A, Gillis A, Rybak AP, Cutz JC, et al. Bmi1
promotes prostate tumorigenesis via inhibitingp16(INK4A)
and p14(ARF) expression. Biochim Biophys Acta. 2008;
1782:642–8.
22. Vonlanthen S, Heighway J, Altermatt HJ, Gugger M, Kappeler A,
Borner MM, et al. The bmi-1 oncoprotein is differentially
expressed in non-small cell lung cancer and correlates with
INK4A-ARF locus expression. Br J Cancer. 2001;84:1372–6.
23. Douglas D, Hsu JH, Hung L, Cooper A, Abdueva D, van Door-
ninck J, et al. BMI-1 promotes ewing sarcoma tumorigenicity
independent of CDKN2A repression. Cancer Res. 2008;
68:6507–15.
24. Heldin CH, Miyazono K, ten Dijke P. TGF-beta signalling from
cell membrane to nucleus through SMAD proteins. Nature.
1997;390:465–71.
25. Massague J. TGFbeta in Cancer. Cell. 2008;134:215–30.
26. Coulouarn C, Factor VM, Thorgeirsson SS. Transforming growth
factor-beta gene expression signature in mouse hepatocytes pre-
dicts clinical outcome in human cancer. Hepatology. 2008;
47:2059–67.
27. Fan L, Xu C, Wang C, Tao J, Ho C, Jiang L, et al. Bmi1 is
required for hepatic progenitor cell expansion and liver tumor
development. PLoS ONE. 2012;7:e46472.
28. Oh S, Kim E, Kang D, Kim M, Kim JH, Song JJ. Transforming
growth factor-beta gene silencing using adenovirus expressing
TGF-beta1 or TGF-beta2 shRNA. Cancer Gene Ther.
2013;20:94–100.
29. Kim JH, Yoon SY, Kim CN, Joo JH, Moon SK, Choe IS, et al.
The Bmi-1 oncoprotein is overexpressed in human colorectal
cancer and correlates with the reduced p16INK4a/p14ARF pro-
teins. Cancer Lett. 2004;203:217–24.
30. Bentley ML, Corn JE, Dong KC, Phung Q, Cheung TK, Cochran
AG. Recognition of UbcH5c and the nucleosome by the Bmi1/
Ring1b ubiquitin ligase complex. EMBO J. 2011;30:3285–97.
31. McGinty RK, Henrici RC, Tan S. Crystal structure of the PRC1
ubiquitylation module bound to the nucleosome. Nature.
2014;514:591–6.
32. Blackledge NP, Farcas AM, Kondo T, King HW, McGouran JF,
Hanssen LL, et al. Variant PRC1 complex-dependent H2A ubi-
quitylation drives PRC2 recruitment and polycomb domain for-
mation. Cell. 2014;157:1445–59.
33. Noma T, Glick AB, Geiser AG, O’Reilly MA, Miller J, Roberts
AB, et al. Molecular cloning and structure of the human trans-
forming growth factor-beta 2 gene promoter. Growth Factors.
1991;4:247–55.
34. Datta S, Hoenerhoff MJ, Bommi P, Sainger R, Guo WJ, Dimri M,
et al. Bmi-1 cooperates with H-Ras to transform human mammary
epithelial cells via dysregulation of multiple growth-regulatory
pathways. Cancer Res. 2007;67:10286–95.
35. Hoenerhoff MJ, Chu I, Barkan D, Liu ZY, Datta S, Dimri GP,
et al. BMI1 cooperates with H-RAS to induce an aggressive breast
cancer phenotype with brain metastases. Oncogene.
2009;28:3022–32.
36. Jagani Z, Wiederschain D, Loo A, He D, Mosher R, Fordjour P,
et al. The Polycomb group protein Bmi-1 is essential for the
growth of multiple myeloma cells. Cancer Res. 2010;70:5528–38.
37. Silva J, Garcia JM, Pena C, Garcia V, Dominguez G, Suarez D,
et al. Implication of polycomb members Bmi-1, Mel-18, and Hpc-
2 in the regulation of p16INK4a, p14ARF, h-TERT, and c-Myc
expression in primary breast carcinomas. Clin Cancer Res.
2006;12:6929–36.
38. Kreso A, van Galen P, Pedley NM, Lima-Fernandes E, Frelin C,
Davis T, et al. Self-renewal as a therapeutic target in human
colorectal cancer. Nat Med. 2014;20:29–36.
39. Datto MB, Li Y, Panus JF, Howe DJ, Xiong Y, Wang XF.
Transforming growth factor beta induces the cyclin-dependent
B. Li et al.

kinase inhibitor p21 through a p53-independent mechanism. Proc
Natl Acad Sci USA. 1995;92:5545–9.
40. Reynisdottir I, Polyak K, Iavarone A, Massague J. Kip/Cip and
Ink4 Cdk inhibitors cooperate to induce cell cycle arrest in
response to TGF-beta. Genes Dev. 1995;9:1831–45.
41. Li CY, Suardet L, Little JB. Potential role of WAF1/Cip1/p21 as a
mediator of TGF-beta cytoinhibitory effect. J Biol Chem.
1995;270:4971–4.
42. Sandhu C, Garbe J, Bhattacharya N, Daksis J, Pan CH, Yaswen P,
et al. Transforming growth factor beta stabilizes p15INK4B pro-
tein, increases p15INK4B-cdk4 complexes, and inhibits cyclin
D1-cdk4 association in human mammary epithelial cells. Mol Cell
Biol. 1997;17:2458–67.
43. Gargiulo G, Cesaroni M, Serresi M, de Vries N, Hulsman D,
Bruggeman SW, et al. In vivo RNAi screen for BMI1 targets
identifies TGF-beta/BMP-ER stress pathways as key regulators of
neural- and malignant glioma-stem cell homeostasis. Cancer Cell.
2013;23:660–76.
44. Kim RH, Lieberman MB, Lee R, Shin KH, Mehrazarin S, Oh JE,
et al. Bmi-1 extends the life span of normal human oral kerati-
nocytes by inhibiting the TGF-beta signaling. Exp Cell Res.
2010;316:2600–8.
45. Batlle E, Massague J. Transforming growth factor-beta signaling
in immunity and cancer. Immunity. 2019;50:924–40.
46. Mitchell EJ, Lee K, O’Connor-McCourt MD. Characterization of
transforming growth factor-beta (TGF-beta) receptors on BeWo
choriocarcinoma cells including the identification of a novel 38-
kDa TGF-beta binding glycoprotein. Mol Biol Cell.
1992;3:1295–307.
47. del Re E, Babitt JL, Pirani A, Schneyer AL, Lin HY. In the
absence of type III receptor, the transforming growth factor
(TGF)-beta type II-B receptor requires the type I receptor to bind
TGF-beta2. J Biol Chem. 2004;279:22765–72.
48. Rotzer D, Roth M, Lutz M, Lindemann D, Sebald W, Knaus P.
Type III TGF-beta receptor-independent signalling of TGF-beta2
via TbetaRII-B, an alternatively spliced TGF-beta type II receptor.
EMBO J. 2001;20:480–90.
49. Rodriguez C, Chen F, Weinberg RA, Lodish HF. Cooperative
binding of transforming growth factor (TGF)-beta 2 to the types I
and II TGF-beta receptors. J Biol Chem. 1995;270:15919–22.
50. Dropmann A, Dediulia T, Breitkopf-Heinlein K, Korhonen H,
Janicot M, Weber SN, et al. TGF-beta1 and TGF-beta2 abundance
in liver diseases of mice and men. Oncotarget. 2016;7:19499–518.
51. Tward AD, Jones KD, Yant S, Cheung ST, Fan ST, Chen X, et al.
Distinct pathways of genomic progression to benign and malig-
nant tumors of the liver. Proc Natl Acad Sci USA.
2007;104:14771–6.
52. HoC,WangC,MattuS,DestefanisG,LaduS,DeloguS,etal.AKT
(v-akt murine thymoma viral oncogene homolog 1) and N-Ras
(neuroblastoma ras viral oncogene homolog) coactivation in the
mouse liver promotes rapid carcinogenesis by way of mTOR
(mammalian target of rapamycin complex 1), FOXM1 (forkhead box
M1)/SKP2, and c-Myc pathways. Hepatology. 2012;55:833–45.
53. Wojtowicz JM, Kee N. BrdU assay for neurogenesis in rodents.
Nat Protoc. 2006;1:1399–405.
54. Jiang Y, Yan B, Lai W, Shi Y, Xiao D, Jia J, et al. Repression of
Hox genes by LMP1 in nasopharyngeal carcinoma and modula-
tion of glycolytic pathway genes by HoxC8. Oncogene.
2015;34:6079–91.
55. Shi Y, Tao Y, Jiang Y, Xu Y, Yan B, Chen X, et al. Nuclear
epidermal growth factor receptor interacts with transcriptional
intermediary factor 2 to activate cyclin D1 gene expression trig-
gered by the oncoprotein latent membrane protein 1. Carcino-
genesis. 2012;33:1468–78.
56. Charan J, Kantharia ND. How to calculate sample size in animal
studies? J Pharmacol Pharmacother. 2013;4:303–6.
Bmi1 drives hepatocarcinogenesis by repressing the TGFβ2/SMAD signalling axis