TLR4 enhances TGF-b signaling and hepatic fibrosisEkihiro Seki1,2, Samuele De Minicis1,2, Christoph H Osterreicher1,2, Johannes Kluwe1, Yosuke Osawa1,David A Brenner2 & Robert F Schwabe1

Hepatic injury is associated with a defective intestinal barrier and increased hepatic exposure to bacterial products. Here we report

that the intestinal bacterial microflora and a functional Toll-like receptor 4 (TLR4), but not TLR2, are required for hepatic

fibrogenesis. Using Tlr4-chimeric mice and in vivo lipopolysaccharide (LPS) challenge, we demonstrate that quiescent hepatic

stellate cells (HSCs), the main precursors for myofibroblasts in the liver, are the predominant target through which TLR4 ligands

promote fibrogenesis. In quiescent HSCs, TLR4 activation not only upregulates chemokine secretion and induces chemotaxis of

Kupffer cells, but also downregulates the transforming growth factor (TGF)-b pseudoreceptor Bambi to sensitize HSCs to TGF-b–induced signals and allow for unrestricted activation by Kupffer cells. LPS-induced Bambi downregulation and sensitization to

TGF-b is mediated by a MyD88–NF-jB–dependent pathway. Accordingly, Myd88-deficient mice have decreased hepatic fibrosis.

Thus, modulation of TGF-b signaling by a TLR4-MyD88–NF-jB axis provides a novel link between proinflammatory and

profibrogenic signals.

Chronic tissue injury leads to fibrosis in many organs, including theliver, lung, kidney and heart. In chronic liver disease, development offibrosis is the first step toward the progression to cirrhosis and itscomplications (such as organ failure, esophageal variceal bleeding andhepatocellular carcinoma), independently of the underlying etiology1.TGF-b and platelet-derived growth factor (PDGF) have been chara-cterized as key cytokines that mediate hepatic fibrogenesis1,2, but thecurrent model of TGF-b– and PDGF-dependent fibrogenesis does notaccount for the role of inflammatory mediators in hepatic fibrogen-esis. Kupffer cells, the resident hepatic macrophages, and hepaticstellate cells (HSCs), the main producers of extracellular matrix inthe fibrotic liver, are key in hepatic fibrogenesis and are both targets ofproinflammatory mediators1,2. Although inflammation is present invirtually all patients with hepatic fibrosis and correlates with fibrosisprogression1, the molecular link between hepatic inflammation andfibrogenesis remains elusive.

TLRs comprise a highly conserved family of receptors that recognizepathogen-associated molecular patterns and allow the host todetect microbial infection. TLRs are not only important in theregulation of innate and adaptive immune responses3, but alsoinvolved in noninfectious inflammatory diseases of the cardiovascularsystem, lung and liver4–9. TLR4 acts as receptor for LPS, a cell-wallcomponent of Gram-negative bacteria that is among the strongestknown inducers of inflammation. After liver injury, LPS levels increasein the portal and systemic circulation owing to changes in theintestinal mucosal permeability and increased bacterial transloca-tion10. Kupffer cells are the best-characterized target of LPS10,11, inthe liver, where they have a crucial role in hepatic fibrogenesis byenhancing HSC activation12,13. We and others have previously shownthat activated HSCs express TLR4 and are highly responsive to low

concentrations of LPS14–16. Thus, TLR4 ligands may modulate hepaticfibrogenesis through two distinct targets, HSCs and Kupffer cells. Herewe report that TLR4 and the intestinal microflora are essential forhepatic fibrogenesis, and that TLR4 downregulates the TGF-b pseu-doreceptor BMP and the activin membrane-bound inhibitor (Bambi)on quiescent HSCs to augment TGF-b–mediated HSC activation andcollagen production.

RESULTS

TLR4 enhances hepatic inflammation and fibrogenesis

To determine whether TLR4 is involved in hepatic fibrogenesis, weperformed bile duct ligation (BDL) on TLR4-mutant mice (C3H/HeJ)and on TLR4–wild-type mice (C3H/HeOuJ). Twenty-one daysafter BDL, TLR4–wild-type mice showed overt hepatic fibrosis,whereas TLR4-mutant mice had a significant reduction in fibrosis asdemonstrated by Sirius red staining, hydroxyproline content andexpression of a–smooth muscle actin (a-SMA, encoded by Acta2)(Fig. 1a–d). In TLR4-mutant mice, a strong suppression of hepaticfibrogenesis was already present 5 d after BDL, as indicated by thereduction in messenger RNA expression of early markers of fibrogen-esis including collagen-a1(I) (encoded by Col1a1), Acta2, TGF-b1(encoded by Tgfb1) and tissue inhibitor of metalloproteinases-1(TIMP1, encoded by Timp1) (Fig. 1e). We observed a similarreduction in BDL-induced hepatic fibrosis, 5 and 21 d after BDL, inmice Tlr4-mutant from another genetic background (data not shown).Liver injury was not different between Tlr4–wild-type andTlr4-mutant mice as assessed by measuring serum ALT levels(Fig. 1f). Next, we examined the recruitment of macrophages, a cellpopulation that is required for HSC activation and hepatic fibrosis12.Tlr4-mutant mice showed significantly reduced hepatic macrophage

Received 19 July; accepted 31 August; published online 21 October 2007; doi:10.1038/nm1663

1Department of Medicine, Columbia University, College of Physicians and Surgeons, New York, New York 10032, USA. 2Department of Medicine, University ofCalifornia, San Diego, School of Medicine, La Jolla, California 92093, USA. Correspondence should be addressed to R.F.S. (rfs2102@columbia.edu).

1324 VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 NATURE MEDICINE

ART ICL ES©

2007

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

emed

icin

e

infiltration in comparison to Tlr4–wild-type mice (Fig. 1g). Addi-tionally, the expression of Ccl2 and Ccl4, two potent mediators ofmacrophage recruitment, was decreased in Tlr4-mutant mice(Fig. 1h). To determine whether TLR4 activation is a universalrequirement for hepatic fibrosis, we assessed fibrogenesis in twoother, chemically-induced models of fibrosis, CCl4 and thioacetamide(TAA). Tlr4-mutant mice showed a significant reduction of hepaticfibrosis after eight doses of CCl4 or 20 weeks of TAA treatment(Supplementary Fig. 1a,b online). In all three models of fibrogenesis,plasma concentrations of the TLR4 ligand LPS were elevated(Supplementary Fig. 1c–f). In contrast with TLR4, the presence ofTLR2, the main receptor for Gram-positive bacterial cell wall compo-nents3, was not required for hepatic fibrosis after BDL (Supplemen-tary Fig. 2 online).

Gut-derived bacterial products mediate hepatic fibrosis

To investigate whether the intestinal bacterial microflora contributesto elevated LPS levels and to hepatic fibrosis, we treated mice witha cocktail of non-absorbable broad-spectrum antibiotics17 andthen performed BDL. This antibiotic cocktail efficiently suppressedthe increase in plasma LPS after BDL (Supplementary Fig. 1f).Mice receiving this cocktail showed a significant reduction inhepatic fibrosis as shown by decreased Sirius red staining, hydroxy-proline content and Acta2 expression (Fig. 2a–e). Moreover, hepaticmacrophage infiltration was suppressed (Fig. 2f,g), suggestingthat intestinally derived TLR ligands mediate macrophage recruitmentto the liver. This is further supported by the observed decrease inTlr4 expression in antibiotic-treated mice (Fig. 2h), as Kupffercells represent the most abundant source of Tlr4 mRNA in theliver. The antibiotic cocktail did not affect Tlr4 expression orthe number of Kupffer cells in the liver before BDL (Fig. 2g,h),and mice treated with this cocktail showed a normal responsetoward exogenous LPS, thus excluding any effects of antibioticson LPS signaling molecules and other downstream mediators(data not shown).

HSCs, but not Kupffer cells, promote TLR4-dependent fibrosis

Kupffer cells are the main targets of LPS in the liver, and they have apivotal role in HSC activation and fibrogenesis10,12. To explorewhether Kupffer cells mediate TLR4-dependent profibrogenic effects,we generated TLR4-chimeric mice using a combination of clodronate-mediated Kupffer cell depletion, irradiation and bone-marrow trans-plantation (BMT). This protocol achieved full reconstitution ofKupffer cells, whereas HSCs were not replaced by bone marrow–derived cells (Fig. 3a). We confirmed successful BMT in all mice byassessing expression of Il6, the gene encoding interleukin (IL)-6, inLPS-stimulated spleen cells (Fig. 3b). To our surprise, chimeric micecontaining Tlr4-mutant bone marrow showed normal fibrogenesis,whereas chimeric mice containing Tlr4–wild-type bone marrow butTlr4-mutant resident liver cells showed the same reduction in fibrosisas Tlr4-mutant mice transplanted with Tlr4-mutant bone marrow(Fig. 3c–f). These results indicate that bone marrow–derived cells,including Kupffer cells, are not the primary cells that enhance hepaticfibrogenesis in response to TLR4 ligands. To test the hypothesisthat TLR4 ligands directly target HSCs to mediate hepatic fibrogenesis,we first examined whether HSCs are targets of LPS in vivo.After intraperitoneally injecting LPS, we detected nuclear transloca-tion of the NF-kB subunit p65 in Kupffer cells as well as in severalother cell populations, including HSCs and hepatocytes (Fig. 4a).Given that Kupffer cells rapidly release large amounts of TNF-a inresponse to LPS18, the observed activation of NF-kB in HSCsand hepatocytes could not be directly attributed to LPS. To avoidthe confounding effect of TNF-a, livers were depleted of Kupffer cellsby liposomal clodronate and were then injected with LPS. In Kupffercell–depleted mice, we still detected nuclear translocation of p65 in alarge number of HSCs, whereas there was almost no p65 translocationin hepatocytes and in desmin-negative nonparenchymal cells (Fig. 4a).Next, we investigated TLR4 expression in freshly isolated HSCs.HSCs expressed large amounts of TLR4 in the quiescent state, andTLR4 expression in vivo was not further enhanced by activation(Fig. 4b). To further confirm that quiescent HSCs are targets of

Tlr4+ Tlr4+

Tlr4+

Tlr4+

Tlr4+

Tlr4mutTlr4mut

Tlr4mut

Tlr4mut

Tlr4mut

ShamSham Sham Sham

BDL BDL BDLBDL

20

15

10

5

0

Siri

us r

ed–p

ositi

vear

ea (

%)

Sham BDL Sham BDL

*

*

* ***

400

300

200

100

0Hyd

roxy

prol

ine

(µg/

g liv

er)

BDLBDLSham Sham 1 2 31 2 3

α–SMA

β-actin

150

100

50

0Col

1a1

mR

NA

Sham BDL

Sham BDL

Sham

Sham

BDL

BDL

15

10

5

0Act

a2 m

RN

A

****

300

200

100

0Tim

p1 m

RN

A6

4

2

0Tgf

b1 m

RN

A

12

8

4

0Sham BDL

F4/

80-p

ositi

vear

ea (

%)

BDLBDL500400300200100

0 0 5 21Days after BDL

Ser

um A

LT(I

U/li

ter)

200150100

50

0Ccl

2 m

RN

A

Ccl

4 m

RN

A

Sham BDL Sham BDL

200150

100

500

**

a d

f hg

b

c

e

Figure 1 Reduction of hepatic fibrogenesis and macrophage infiltration in TLR4-mutant mice. Tlr4-mutant C3H/HeJ (Tlr4mut; n ¼ 8) and Tlr4–wild-type

C3H/HeOuJ (Tlr4+; n ¼ 8) mice underwent BDL for 21 d. *P ¼ 0.001. (a,b) Collagen deposition was evaluated by Sirius red staining (a and b, left graph)

and hydroxyproline measurement (b, right graph) 21 d after BDL. Scale bar, 50 mm. (c,d) Expression of a-SMA was determined by western blot analysis

(c) and immunohistochemistry (d). Numbers in c indicate individual mice. Scale bar, 50 mm. (e) Hepatic levels of Col1a1, Acta2, TGFb1 and TIMP1 mRNA

were measured by qPCR in Tlr4–wild-type (n ¼ 5) and Tlr4-mutant mice (n ¼ 5) 5 d after BDL. *P ¼ 0.045, **P r 0.02. (f) Hepatic injury as measured

by serum ALT. (g) Macrophage infiltration as determined by immunohistochemical staining of F4/80 21 d after BDL (left) and quantification by counting fiverandomly chosen high-power fields (right). Scale bar, 25 mm. *P ¼ 0.001. (h) Hepatic levels of Ccl2 and Ccl4 mRNA were determined by qPCR 5 d after

BDL. *P r 0.001.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 1325

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

LPS, we measured LPS-induced NF-kB activation in quiescent HSCswith a reporter assay. LPS increased NF-kB–driven luciferase activityin quiescent HSCs at doses as low as 1 ng/ml (Fig. 4c). Moreover,microarray analysis revealed that LPS induced the expression of alarge number of genes in quiescent HSCs, among them genesencoding chemokines, adhesion molecules and Tlr2 (SupplementaryTable 1 online).

TLR4 enhances the interaction between HSC and Kupffer cells

We identified a large number of chemokines that were significantlyupregulated in LPS-stimulated quiescent HSCs (Fig. 4d and Supple-mentary Table 1). Because Tlr4-mutant mice showed decreasedrecruitment of macrophages after BDL (Fig. 1g), we investigatedwhether stimulating HSCs with LPS induces Kupffer cell migration.Conditioned media from LPS-stimulated quiescent HSCs increasedmigration of Kupffer cells 2.5-fold in comparison to conditionedmedia from unstimulated HSCs (Fig. 4e). Kupffer cells fromTlr4-mutant mice induced approximately the same amount ofchemotaxis as Kupffer cells from Tlr4–wild-type mice in this assay,demonstrating that LPS mediates its effects through HSCs and notthrough Kupffer cells. Additionally, we found upregulation ofmRNAs encoding the adhesion molecules Icam1, Vcam1 and Selein LPS-stimulated HSCs and increased adhesion of Tlr4–wild-type aswell as Tlr4-mutant Kupffer cells to LPS-stimulated quiescent HSCs(Fig. 4f,g). To confirm the pivotal role of Kupffer cells in HSCactivation and fibrosis in vivo, we depleted Kupffer cells by liposomalclodronate and then performed BDL. Liposomal clodronate achieved

complete depletion of Kupffer cells (Supplementary Fig. 3a online)and had no direct effects on HSCs as assessed by transcriptionfactor (NF-kB and Smad) reporter assays and an HSC activationassay (data not shown). After BDL, Kupffer cell–depleted mice showedan almost complete suppression of HSC activation and fibrosis(Supplementary Fig. 3b–g).

TLR4 enhances HSC activation induced by TGF-b or Kupffer cells

To explore the mechanisms by which TLR4 contributes to hepaticfibrogenesis, we investigated whether LPS promotes the transdiffer-entiation of quiescent HSCs to extracellular matrix–producing myo-fibroblasts by using a well-characterized reporter mouse that expressesGFP under control of the collagen-a1(I) promoter (Coll-GFP)19. LPStreatment alone did not induce GFP expression or a myofibroblast-likephenotype in HSCs isolated from Coll-GFP mice (Fig. 5a). However,when quiescent HSCs were pretreated with LPS (100 ng/ml for 24 h),

BDLBDL ShamSham0

2

4

6

8

10

Tlr4

mR

NA

ControlAntibiotics

ControlAntibiotics

ControlAntibiotics

**

02468

101214

F4/

80-p

ositi

vear

ea (

%)

AntibioticsControl

BDLBDLBDLBDL

Sham ShamControl Antibiotics

Sham Sham1 2 3 1 2 3α-SMA

β-actin

BDLBDLControl Antibiotics

BDLSham

BDLSham

*

*

0100200300400500

Hyd

roxy

prol

ine

(µg/

g liv

er)

0

5

10

15

20

Siri

us r

ed–p

ositi

vear

ea (

%)

BDL BDL

Sham ShamControl Antibioticsa d

g h

e

f

b

c

Figure 2 Gut sterilization prevents hepatic fibrosis after bile-duct ligation. Tlr4–wild-type C3H/HeOuJ

mice were gut-sterilized for 4 weeks before BDL and 3 weeks after BDL (n ¼ 8) or underwent BDL

without gut sterilization (n ¼ 8). (a,b) Fibrillar collagen deposition was evaluated by Sirius red staining

(a) and determination of the Sirius red–positive area (b) 21 d after BDL. Scale bar, 50 mm. *P o 0.001.

(c) Hepatic hydroxyproline content was measured 21 d after BDL. *P ¼ 0.035. (d,e) Expression of

a-SMA was determined by immunohistochemistry (d) and western blotting (e) 21 d after BDL. Numbers

in e indicate individual mice. Scale bar, 50 mm. (f,g) Macrophage infiltration as determined by immuno-

histochemical staining of F4/80 (f) and quantification by counting five randomly chosen high-power fields (g). IU, international units. *P ¼ 0.016. Scale bar,

25 mm. (h) Tlr4 mRNA was quantified by qPCR in sham- or BDL-operated antibiotic-treated mice and sham- or BDL-operated control mice.*P ¼ 0.048.

a b

c d

e f

BDLSham

BDLSham

0

100

200

300

Hyd

roxy

prol

ine

(µg/

g liv

er)

**

*

**

*

***

0

4

8

12

16

Siri

us r

ed–p

ositi

vear

ea (

%)

++++LPS0

2040

6080

100

II6 m

RN

A *Tlr4mut BM→ Tlr4mut

Tlr4+ BM→ Tlr4+ Tlr4+ BM→ Tlr4mut

Tlr4mut BM→ Tlr4mutTlr4mut BM→ Tlr4+

Tlr4+ BM→ Tlr4+

Tlr4mut BM→ Tlr4+ Tlr4mut BM→ Tlr4mut

Tlr4+ BM→ Tlr4mut

Tlr4mut BM→ Tlr +Tlr + BM→ Tlr4mutTlr + BM→ Tlr +Green:GFP

Red:DesminGreen:GFPRed:F4/80

Figure 3 Bone marrow–derived cells do not mediate TLR4-dependent

profibrogenic effects. (a) Bone marrow from transgenic mice expressing GFP

from the b-actin promoter was transplanted into clodronate-treated,

irradiated wild-type mice. Kupffer cells and HSCs were detected by

immunofluorescent staining of desmin (left) and F4/80 (right), respectively.

Scale bar, 10 mm. (b–f) Chimeric mice consisting of transplanted Tlr4-

mutant bone marrow into irradiated and clodronate-treated Tlr4–wild-type

mice (n ¼ 8) or Tlr4-mutant mice (n ¼ 8) and vice versa (n ¼ 8 eachgroup) were examined. Successful BMT occurred, as measured by Il6 mRNA

in splenocytes after stimulation with 100 ng/ml LPS (b) Hepatic fibrogenesis

in chimeric mice was assessed by Sirius red staining (c) and determination

of the Sirius red–positive area (d), a-SMA immunohistochemistry (e) and

hydroxyproline measurement (f). Scale bar, 50 mm. *P r 0.0001,

**P r 0.002, ***P ¼ 0.003.

ART ICL ES

1326 VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

the response to the profibrogenic cytokine TGF-b1 was stronglyenhanced, as shown by threefold higher GFP expression (Fig. 5a).The enhanced response to TGF-b1 in LPS-pretreated HSCs wasconfirmed by a reporter assay in which luciferase expression is drivenby Smad-binding CAGA elements (Fig. 5b), and by quantitative real-time PCR (qPCR) amplification of Col1a1 mRNA (data not shown).We also investigated whether LPS promotes fibrosis by sensitizingHSCs to platelet-derived growth factor (PDGF), another key player inHSC activation and fibrogenesis. However, we found neitherLPS-mediated sensitization to PDGF-induced phosphorylation ofthe kinases Akt and Erk, nor increased PDGF-induced [3H]thymidineuptake in LPS-treated HSCs (Supplementary Fig. 4a,b online). AsKupffer cells are an important source of TGF-b in the fibrotic liver, weinvestigated whether LPS pretreatment primes HSCs to Kupffer cell–induced activation. To exclude the effects of LPS on Kupffer cellsin these experiments, we cultured HSCs from Coll-GFP mice withTlr4–wild-type as well as Tlr4-mutant Kupffer cells. Under bothconditions, LPS pretreatment strongly enhanced GFP expression,confirming HSCs as LPS targets in this assay (Fig. 5c). Kupffercell–induced GFP expression was completely blocked by solubletype II TGF-b receptor in LPS- and vehicle-treated HSCs, confirmingthat LPS-mediated enhancement of HSC activation requiresTGF-b (Fig. 5c).

TLR4-mediated Bambi downregulation sensitizes HSCs to TGF-bThe TGF-b pseudoreceptor Bambi20 was the only TGF-b–related geneamong 121 LPS-regulated genes that we identified by microarray

analysis in quiescent HSCs (Supplementary Table 1). Microarrayanalysis, qPCR and immunoblotting showed a strong downregulationof Bambi mRNA and Bambi protein levels in HSCs after LPS stimula-tion (Supplementary Table 1 and Fig. 5d). Notably, we also observed asignificant downregulation of Bambi in HSCs isolated from Tlr4–wild-type mice after BDL (Fig. 5e) or CCl4 treatment (data not shown), butnot in those from Tlr4-mutant mice, suggesting that TLR4-dependentdownregulation of Bambi is an integral part of the HSC activationprocess in vivo. LPS also decreased the activity of a Bambi-drivenluciferase reporter (data not shown). The levels of several other TGF-bsignaling molecules, including Smad2, Smad3, Smad4, Smad7, Sar1a,Skil, Ski, Tgfbr1, Tgfbr2 and Tgfbr3, were not significantly altered byLPS treatment of quiescent HSCs, as assessed by microarray experi-ments (data not shown) and qPCR (Supplementary Fig. 4c). More-over, LPS was at least as efficient in sensitizing HSCs to active TGF-b asin sensitizing them to latent TGF-b, a result that excludes any effects ofLPS on TGF-b activation (Supplementary Fig. 4d). To explore thepotential role of Bambi in mediating the TGF-b–sensitizing effects ofLPS, we infected quiescent HSCs with adenoviruses expressing domi-nant-negative Bambi (AddnBambi), full-length Bambi (AdBambi) orGFP (AdGFP), and measured Smad-driven luciferase activity afterstimulation with LPS and TGF-b1. In uninfected and AdGFP-infectedHSCs, LPS pretreatment enhanced TGF-b–induced Smad reporteractivity 2.5-fold (Fig. 5f). AddnBambi-infected cells showed a signifi-cant increase in Smad-driven luciferase activity, which was similar tothat of LPS-pretreated, AdGFP-infected HSCs (Fig. 5f). Notably,TGF-b1–induced luciferase activity could not be further enhanced byLPS in AddnBambi-infected HSCs, suggesting that Bambi downregula-tion is the major mechanism by which LPS sensitizes HSCs to TGF-b–induced signals. Conversely, Smad reporter activity induced byTGF-b1 or TGF-b1 plus LPS was blunted in AdBambi-infectedHSCs. In addition, we investigated whether Bambi altered TGF-b–induced HSC activation by using collagen-driven GFP expressionas a marker of activation. AddnBambi- or AdBambi-infected HSCsshowed increased or reduced GFP expression, respectively, in compar-ison to HSCs infected with AdShuttle control virus (Fig. 5g). We nextwanted to assess the effects of modulating Bambi activity in vivo withan adenoviral approach. To exclude any confounding effects of over-expressing dominant-negative Bambi in hepatocytes and Kupffer cells,

+–+–

Tlr4+ KCs Tlr4mut KCs

Tlr4+ KCs Tlr4mut KCs

LPS

+–+–LPS

+–LPS+–LPS+–LPS

0

150

300

450

Adh

esio

n

*

***

*

*** *

**

*NS

010203040506070

Sel

e m

RN

A

0

2

4

6

8

0

2

4

6

8

Vca

m1

mR

NA

Icam

1 m

RN

A

LPS – + LPS – + LPS – +

LPS – + LPS – + LPS – +

010

20

30

40

Ccl

2 m

RN

AC

xcl1

mR

NA

Cxc

l2 m

RN

A

Cxc

l10

mR

NA

Ccl

3 m

RN

A

Ccl

4 m

RN

A

80

60

40

20

0 020406080

1000

1

2

3

Mig

ratio

n

* **NS

25201510

50

25201510

50

18

12

6

0

100101LPS (ng/ml)0

0

12345

NF

-κB

luci

fera

se

104103102101100

200

100

0

TLR4 fluorescenceintensity (log scale)

TLR4 fluorescenceintensity (log scale)

Num

ber

of c

ells

Quiescent HSCsIn vivo–activated

HSCs

NPCNPCHepato-cytes Desmin-neg Desmin-pos

+++

++++

++++

+ –––

––––

––––

–ClodrLPS

0

20

40

60

80

100

p65

nucl

ear

tran

sloc

atio

n(%

)

+LPS

–LPS

Control Clodronatea

b

d

f

g

e

c200

100

0104103102101100

Figure 4 Quiescent HSCs are a target of LPS in vitro and in vivo. (a) Kupffer

cell–depleted and untreated mice were injected with LPS (0.1 mg/g

intraperitoneally) and killed 2 h later. NF-kB nuclear translocation in HSCs

was determined by staining for NF-kB p65 (in green) and desmin (in red),

which was visualized by confocal microscopy. Graph shows the pecrcentage

of cells counted in ten high-power fields that showed NF-kB nuclear

translocation. Scale bar, 10 mm. Clodr, clodronate. (b) TLR4 expression was

determined in quiescent HSCs (6 h after isolation) and BDL-activated HSCs

by flow-cytometric analysis. (c) Quiescent HSCs were infected with an

adenovirus expressing NF-kB–driven luciferase 12 h after isolation and then

treated with LPS for 6 h. NF-kB activation is expressed as fold induction

after normalization to b-galactosidase activity. (d) Expression of Cxcl1,

Cxcl2, Cxcl10, Ccl2, Ccl3 and Ccl4 mRNA was measured by qPCR in

quiescent HSCs (plated for 6 h) after stimulation with LPS (100 ng/ml) or

vehicle for 6 h. *P r 0.0001, **P ¼ 0.0002. (e) Conditioned media fromquiescent HSCs treated with LPS (100 ng/ml for 8 h) was placed in the

lower chamber and Kupffer cells were placed in the upper chamber of the

migration chamber. Migration of Kupffer cells toward HSCs was determined

by fluorescent microscopy. *P ¼ 0.001, **P ¼ 0.006. KCs, Kupffer cells.

(f) Expression of Icam1, Vcam1 and Sele mRNA levels as determined by

qPCR in quiescent HSCs treated with LPS (100 ng/ml) or vehicle for 6 h.

(g) Adhesion of Kupffer cells to LPS- or vehicle-treated quiescent HSCs as

determined as described in Methods. All figures represent at least three

independent experiments. *P o 0.0001 for f,g.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 1327

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

two hepatic cell types that are readily infected by adenoviruses, we firstmeasured Bambi expression in these cells and found no significantexpression (Supplementary Fig. 5a). We then injected mice withAddnBambi and performed BDL. AddnBambi-treated mice showedincreased expression of fibrogenic markers such as Col1a1 and Acta2five days after BDL (Supplementary Fig. 5b). Notably, there was alsoan increase in fibrogenesis in sham-operated mice infected withAddnBambi, further supporting our hypothesis that Bambi acts anegative regulator of HSC activation.

TLR4 downregulates Bambi through MyD88 and NF-jB

Finally, we investigated the TLR4 signaling pathways through whichLPS regulates Bambi expression. First, we assessed the potentialinvolvement of NF-kB activation by using an adenovirus thatexpresses IkB super-repressor (IkBsr), a potent and specific inhibitorof NF-kB activation21. Whereas Bambi expression fell by 70% incontrol virus–infected HSCs after LPS stimulation, AdIkBsr-infectedHSCs were completely resistant to LPS-mediated Bambi downregula-tion (Supplementary Fig. 6a online). In AdIkBsr-infected HSCs, wealso observed a decrease in TGF-b–induced collagen production afterLPS treatment similar to the decrease observed in AdBambi-infectedHSCs (Supplementary Fig. 6b), suggesting an important role for theNF-kB–Bambi pathway in TLR4-mediated fibrogenic responses. Twoadaptor molecules, MyD88 and Trif, are essential mediators of TLR4-induced NF-kB signaling. Therefore, we next investigated whetherMyD88 and Trif regulate Bambi expression. Bambi expression wasdownregulated by LPS stimulation in wild-type and Trif-deficientHSCs, whereas Myd88-deficient HSCs were completely resistant to

LPS-mediated Bambi downregulation (Fig. 6a). We then analyzed thecontribution of MyD88 and Trif to fibrogenic responses of the liver.Myd88-deficient mice showed a significant reduction in the expressionof Col1a1, Acta2 and Tgfb, which encode profibrogenic markers, andof proinflammatory cytokine genes Tnf-a and Il6 5 d after BDL(Fig. 6b). In contrast, Trif-deficient mice had normal fibrogenicresponses 5 d after BDL (Fig. 6c). Evaluation 3 weeks after BDLconfirmed a significant reduction in fibrosis in Myd88-deficient mice(Fig. 6d). We also observed a reduction in fibrogenesis in Trif-deficientmice 3 weeks after BDL (Fig. 6e), suggesting that Trif acts as aregulator of fibrosis in later stages of fibrogenesis in a Bambi-independent manner. However, in comparison to Myd88-deficientmice, the reduction of fibrogenesis in bile duct–ligated, Trif-deficientmice was less pronounced.

DISCUSSION

Chronic hepatic inflammation is tightly linked to fibrosis in virtuallyall individuals with liver disease and in experimental models offibrogenesis. Whereas chronic activation of inflammatory pathwayshas been shown to promote hepatocarcinogenesis22,23, the molecularlink between inflammation and hepatic fibrogenesis remains elusive.Here we demonstrate that TLR4 drives myofibroblast activation andfibrogenesis in the liver and that TLR4-dependent modulation ofTGF-b signaling provides a link between proinflammatory andprofibrogenic signals. Tlr4-mutant mice show markedly decreasedfibrogenesis in three different models of experimental hepatic fibrosis,and LPS sensitizes HSCs to TGF-b–induced signals through aMyD88–NF-kB pathway.

Control dn-Bambi Bambi+++

++++

++++

+ –––

––––

––––

–LPSLPS TGF-β

0

10

20

30

40

50

Col

lage

n-G

FP

+

cells

(%

)

Col

lage

n-G

FP

+ c

ells

(%

)

Col

lage

n-G

FP

+

cells

(%

)

*

*NS

NS

+LPS+LPS+LPS+LPS+LPS+LPS

–LPS–LPS–LPS–LPS–LPS–LPS

–TGF-β –TGF-β –TGF-β+TGF-β +TGF-β +TGF-βAdBambiAddnBambiControl

Bambidn-BambiGFP+++

++++

++++

+–– –

–––

–– –

– ––TGF-β

TGF-β

0

5

10

15

20

CA

GA

-luci

fera

se

*** ***

NS

BDL

*

LPS

++

+++

+++

– +0

0.2

0.4

0.6

0.8

1

1.2

Bam

bi m

RN

A

Tlr4mut KCs

+Tlr4mut KCs

Tlr4+ KCs

+Tlr4+ KCs

––

–– –

– ––

sTRIILPS

0

20

40

60 * ********

+LPS +sTRII+LPS +sTRII

+LPS +LPS

+LPS+LPS

–LPS–LPS

–LPS –LPSβ-actin

β-actin

BambiLPS

LPSTGF-β

+TGF-β–TGF-β

LPS–– –

–––

––

– +

Tlr4mutTlr4+

Tlr4mutTlr4+

– + – +0

0.20.40.60.8

11.21.41.6

Bam

bi m

RN

A

BambiBDL – – ++

**

**

+++ +

++ +

+

0

5

10

15

CA

GA

-luci

fera

se

40

30

20

10

0

a c

g

d

f

b

e

Figure 5 LPS downregulates Bambi to enhance TGF-b signaling and HSC activation. (a) Quiescent HSCs from Coll-GFP mice were treated with LPS

(100 ng/ml) or vehicle for 24 h, and this was followed by TGF-b1 (100 pg/ml) or vehicle treatment for an additional 48 h. GFP expression was visualized by

fluorescent microscopy (left) and quantified in six high-power fields (right). Scale bar, 25 mm. *P ¼ 0.0002. (b) Quiescent HSCs were transduced with

CAGA-luciferase and then treated with LPS (100 ng/ml) for 24 h and TGF-b1 (100 pg/ml) for 8 h. CAGA-driven luciferase activity is expressed as fold

induction after normalization to b-galactosidase activity. *P ¼ 0.003. (c) Quiescent HSCs from Coll-GFP mice were treated with LPS (100 ng/ml) or vehiclefor 24 h and were then co-cultured with Kupffer cells in the presence or absence of soluble TGF-b receptor type II (sTRII) for an additional 48 h. Scale bar,

25 mm. *P ¼ 0.003, **P ¼ 0.002, ***P r 0.02. (d–e) Bambi protein expression (top) and mRNA expression (bottom) were measured by western blotting

and qPCR in untreated and LPS-treated (100 ng/ml for 6 h) quiescent HSCs (d) and in HSCs isolated from TLR4–wild-type and TLR4-mutant mice that had

undergone BDL or sham operation for 2 weeks (e). *P ¼ 0.03, **P ¼ 0.005. (f) Quiescent HSCs were infected with CAGA-luciferase (multiplicity of

infection (MOI) of 100), b-galactosidase (MOI of 100) and either AddnBambi, AdBambi or AdGFP (both with an MOI of 300), and this was followed by

TGF-b1 treatment (100 pg/ml). CAGA-driven luciferase activity is expressed as fold induction after normalization to b-galactosidase activity. *P ¼ 0.007,

**P ¼ 0.002, ***P ¼ 0.001. (g) Quiescent HSCs from Coll-GFP mice were infected with AddnBambi or AdBambi (MOI of 300), pretreated with LPS

(100 ng/ml) for 24 h and then treated with TGF-b1 (100 pg/ml) for an additional 48 h. Scale bar, 25 mm. *P o 0.001. All figures are representative of at

least three independent experiments. dn, dominant negative.

ART ICL ES

1328 VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

Although Kupffer cells are considered to be the main targets of TLRligands in the liver, our study provides several lines of evidence thatHSCs, and not Kupffer cells, are the primary targets that drivefibrogenesis in response to TLR4 ligands. First, chimeric mice contain-ing Tlr4-mutant Kupffer cells and Tlr4–wild-type HSCs show normalfibrogenesis, whereas chimeric mice containing Tlr4-mutant HSCsand Tlr4–wild-type Kupffer cells show a similar reduction in fibrogen-esis as Tlr4-mutant mice transplanted with Tlr4-mutant bone marrow.Second, quiescent HSCs express high levels of TLR4 and are highlyresponsive to LPS in vitro and in vivo. Lastly, LPS sensitizes HSCsto TGF-b– and Kupffer cell–mediated activation. Other cell popula-tions, including hepatocytes, biliary epithelial cells and endothelialcells, are known to express TLR4 (ref. 10) but do not show a strongactivation of NF-kB after LPS injection in vivo, which is consistentwith previous reports24.

The liver is a main target of intestinally-derived bacterial products,and the rate of bacterial translocation increases in various models ofhepatic disease, rendering LPS a likely candidate mediator of TLR4-dependent profibrogenic effects. Accordingly, we found increasedamounts of LPS in all three models of fibrogenesis. Moreover, LPSlevels and fibrogenesis were strongly reduced in mice that were treatedwith a combination of three nonabsorbable oral antibiotics andmetronidazole, suggesting that the intestinal flora is the main sourceof LPS and that intestinally derived LPS drives fibrogenesis. Recently, ithas been suggested that TLR4 is also activated by endogenous ligandssuch as HMGB1 and hyaluronan4,25–28. Although hepatic HMGB1and hyaluronan were upregulated after BDL (data not shown), wefound almost the same reduction of hepatic fibrosis in gut-sterilizedmice as in Tlr4-mutant mice. Moreover, a blocking HMGB1 antibodydid not reduce hepatic fibrogenesis after BDL (R.F.S., unpublishedobservations), suggesting that intestinally derived bacterial productsare the predominant profibrogenic TLR4 agonists in hepatic fibrosis.

However, we cannot completely exclude a contribution of endogenousTLR4 ligands to hepatic fibrogenesis, for example at other time pointsor in different models of liver injury. Our hypothesis that gut-derivedLPS is an important mediator of hepatic fibrogenesis is furthersupported by a recent study that has demonstrated a reduction inhepatic fibrogenesis in CD14-deficient and LPS-binding protein(LBP)–deficient mice29. In contrast to our study, the authors concludethat Kupffer cells are the LPS targets that enhance hepatic fibrogenesis.In their study, the suppression of fibrogenesis was not as pronouncedas the one we observed in Tlr4-mutant mice. This difference is mostprobably due to CD14- and LBP-independent effects of LPS. Wepropose that the increase in intestinal permeability and the rise ofportal and systemic LPS levels after liver injury act as danger signals toprime HSCs for activation and to prepare the liver for wound-healingresponses. Although these signals may be important for the responseto acute injury, our results suggest that LPS-mediated TLR4 activationcontributes to the excessive activation of wound-healing responsesthat is characteristic of chronic liver injury and that leads to thedevelopment of liver cirrhosis and its serious consequences in humans.Further investigations in germ-free and mono-associated animals arerequired to reveal how individual components of the intestinalmicrobiota contribute to hepatic fibrogenesis.

Our study defines HSC-mediated Kupffer-cell chemotaxis andsensitization to TGF-b–induced signals as two independent, yetcomplementary, mechanisms by which TLR4 enhances HSC activationand hepatic fibrosis. TGF-b is considered the most powerful mediatorof HSC activation in vitro and in vivo1,2. Kupffer cells are a mainsource of TGF-b in the liver and promote HSC activation andfibrogenesis12,13,30. Thus, TLR4 mediates HSC activation throughtwo TGF-b–dependent mechanisms: increased exposure to Kupffercell–derived TGF-b and enhanced sensitivity to TGF-b. The impor-tance of HSC-mediated Kupffer cell chemotaxis is emphasized by our

1.2

Veh30 8 10

8

6

4

2

0

10

8

6

4

2

0

6

4

2

0

20

10

0

3

40

30

20

10

0

100

50

0

2

1

0

Sham BDL

Sham BDL

Sham BDL

Sham

WT WT

BDL

Sham BDL Sham BDL

WT12

9

6

3

0

Sham BDL

Sham BDL Sham BDL

Sham BDL

Sham BDL Sham BDL

Sham BDL Sham BDL

*

**

* *

*** *

Col

1a1

mR

NA

Tgf

b1 m

RN

A

Col

1a1

mR

NA

Tim

p1 m

RN

A

Tim

p1 m

RN

A

Act

a2 m

RN

A

Tnf

mR

NA

Il6 m

RN

A

Il6 m

RN

A

10

8

6

4

2

0

2

Tgf

b1 m

RN

A

0

1

5120

25

20

15

10

5

0

90

60

30

0

4

3

10

8

6

4

2

0

Act

a2 m

RN

A

Tnf

mR

NA

LPS

1.0

0.8

0.6

Bam

bi m

RN

A

0.4

0.2

0.0

WT

Myd

88−/−

WTMyd88−/−

Myd88 −/−Myd88−/−

WTTrif lps2/lps2

Trif lps2/lps2WT

* **

Trif lps2/lps2

10

8

6

4

Siri

us r

ed−p

ositi

vear

ea (

%)

Siri

us r

ed−p

ositi

vear

ea (

%)

2

0

a b

d e

Trif lps

2/lps

2

c

Figure 6 LPS downregulates Bambi through Myd88-dependent signals. (a) Quiescent HSCs isolated from wild-type, Myd88-deficient and Trif lps2/lps2 mice

were cultured for 6 h and then stimulated with LPS (100 ng/ml) for 6 h. Bambi expression was analyzed by qPCR. (b) Myd88-deficient and wild-type

littermates underwent BDL or sham operation for 5 d. Levels of Col1a1, Acta2, Tnf, TGFb1, TIMP1 and Il6 mRNA were determined by qPCR. *P r 0.03,

**P ¼ 0.003, ***P ¼ 0.047. (c) Trif lps2/lps2 and C57BL/6 control mice underwent BDL or sham operation for 5 d. Levels of Col1a1, Acta2, Tnf, TGFb1,

TIMP1 and Il6 mRNA were determined by qPCR. *P ¼ 0.007. (d,e) Myd88-deficient and wild-type littermates underwent BDL or sham operation. On the

eighteenth day after BDL vs sham, tissue from Myd88-deficient and wild-type littermates (d) and Trif lps2/lps2 and C57BL/6 control mice (e) were tested for

fibrillar collagen by Sirius red staining (left) and then quantified (right). Scale bar, 50 mm. WT, wild-type. *P o 0.001, **P ¼ 0.014.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 1329

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

finding that HSC activation and fibrogenesis were almost completelysuppressed in Kupffer cell–depleted mice and by our observation thatCcr1-, Ccr2- and Ccr5-deficient mice show a strong suppression ofhepatic Kupffer cell recruitment and a significant reduction of hepaticfibrogenesis (E.S. and R.F.S., unpublished data). Because the transdif-ferentiation of quiescent HSCs to activated myofibroblast-like HSCs isa key event in hepatic fibrogenesis1,2, it is most likely that thesensitization of quiescent HSCs to TGF-b– and Kupffer cell–inducedactivation constitutes a main mechanism by which TLR4 promotesfibrogenesis. However, we cannot exclude the possibility that TLR4enhances hepatic fibrogenesis by additional mechanisms—perhaps byinducing proinflammatory pathways in activated HSCs, for example,14

or by upregulating antiapoptotic mediators such as Birc2 and Birc3 inHSCs (Supplementary Table 1). This notion is further supported byour recent finding that the presence of LPS modulates the unphysio-logical gene-expression pattern of culture-activated HSCs toward thepattern that is observed in in vivo-activated HSCs31.

Microarray analysis revealed that the TGF-b pseudoreceptor Bambiis a target of LPS in quiescent HSCs. Bambi is a TGF-b family type Ireceptor that lacks an intracellular kinase domain and blocks signaltransduction after stimulation with ligands of the TGF-b superfamilysuch as TGF-b, BMPs and activin20,32,33. Bambi is upregulated inhepatocellular carcinoma and inhibits TGF-b signaling in cancer, butits role in non-transformed hepatic cell populations is not known32,34.The greatly reduced abundance of Bambi in both LPS-treated quies-cent HSCs and in vivo–activated HSCs from Tlr4–wild-type, but notTlr4-mutant, mice strongly suggests that TLR4-mediated downregula-tion of Bambi is an integral part of HSC activation. TGF-b1– andKupffer cell–mediated HSC activation is enhanced by LPS pretreat-ment or by overexpression of dominant-negative Bambi in quiescentHSCs, suggesting that downregulation of Bambi allows Kupffer cells tofully activate HSCs in a TGF-b–dependent manner. Thus, the down-regulation of Bambi and the induction of Kupffer cell–attractingchemokines by LPS complement each other to enhance TGF-b–dependent HSC activation by Kupffer cells. Adenoviral overexpressionof dominant-negative Bambi enhanced hepatic fibrosis, but becausehepatocytes and Kupffer cells do not express Bambi, the effects ofadenoviral dominant-negative Bambi on Bambi–TGF-b signaling inthese cell populations are largely excluded. The moderate increase inhepatic fibrosis is explained by the strong downregulation of Bambiduring HSC activation in vivo and the resulting restriction ofdominant negative Bambi effects to early stages of fibrogenesis.Notably, we even observed an upregulation of HSC activation inAddnBambi-infected mice in the absence of BDL, suggesting thatBambi may act as a gatekeeper of HSC activation. Previous studieshave shown that Bambi expression is regulated by Smad andb-catenin34,35, but these pathways are not directly activated byTLR4. We now present data demonstrating that Bambi expression isregulated through a MyD88–NF-kB–dependent pathway in HSCs.Our results suggest a role for the NF-kB pathway in regulatingfibrogenic responses in the liver. This finding is consistent with arecent study revealing decreased hepatic fibrogenesis after NF-kBinhibition36. Although Trif-deficient mice show no alterations inLPS-mediated Bambi regulation in HSCs and have normal fibro-genesis after five days of BDL, they have less severe histologic fibrosisthree weeks after BDL. Thus, it seems that TLR4 affects fibrosisthrough several mechanisms and that Trif-dependent signals areimportant only in later stages of fibrogenesis, independently of Bambi.

TLRs fulfill a variety of functions in the liver, and inhibition ofTLR4 signaling may alter not only fibrogenesis but also bio-logical processes not directly related to fibrogenesis10. Whereas

TLR4 promotes disease progression in alcoholic and nonalcoholicsteatohepatitis8,10,37,38, TLR4 activation may be beneficial in chronichepatitis B virus infection, owing to TLR4-mediated reduction ofhepatitis B virus replication24. In chronic hepatitis C virus infection,TLR4 is upregulated, but the determination of its precise role requiresfurther investigation10. We and others have shown that liver regenera-tion is not altered in TLR4-deficient mice9,39, suggesting that adecrease of functional liver mass by TLR4 antagonism is unlikely.There are several TLR4 single-nucleotide polymorphisms (SNPs) thatrender humans less responsive to LPS. Notably, a recent study hasidentified the TLR4 SNP that results in a T399I change as a factor thatconfers a significantly reduced risk for fibrosis progression in patientswith chronic hepatitis C virus infection40. Thus, TLR4 inhibitionappears to be a promising strategy for the prevention of hepaticfibrosis in patients with chronic liver disease. Preventing LPS releasefrom the intestinal flora may represent an additional strategy toinhibit hepatic fibrogenesis. Moreover, it is likely that the TLR4-Bambi–TGF-b signaling pathway plays a role in other cell types andaffects wound-healing responses in organs such as the lung, kidneyand heart.

METHODSMice and fibrosis induction. Specific pathogen-free 8-week-old male TLR4–

wild-type (C3H/HeOuJ), TLR4-mutant (C3H/HeJ), Tlr2-deficient, Trif Lps2/Lps2

(ref. 41) and C57BL/6 mice (all from Jackson Laboratories) and MyD88-

deficient mice42 underwent BDL, or were given CCl4 or TAA (see Supplemen-

tary Methods online). We anesthetized mice with ketamine and xylazine. After

midline laparotomy, we ligated the common bile duct twice with 6-0 silk

sutures and closed the abdomen. We performed the sham operation similarly,

except that the bile duct was not ligated. We killed the mice 5 or 21 d after BDL.

We gave the mice humane care according to US National Institutes of Health

recommendations outlined in the ‘‘Guide for the Care and Use of Laboratory

Animals’’. All animal experiments were approved by the Columbia University

Institutional Animal Care and Use Committee.

Gut sterilization. Mice were treated with ampicillin (1 g/l; Sigma), neomycin

sulfate (1 g/l; Sigma), metronidazole (1 g/l; Sigma) and vancomycin (500 mg/l;

Abbott Labs) in drinking water for 4 weeks. This was followed by BDL and

continuation of antibiotic treatment until mice were killed17.

Bone marrow transplantation. We performed BMT experiments as previously

described with slight modifications7,43. Because only 30% of Kupffer cells are

reconstituted by donor-derived bone marrow cells 6 months after BMT44, we

gave mice an intravenous injection of liposomal clodronate (200 ml intrave-

nously) before irradiation to deplete Kupffer cells and accelerate macrophage

turnover45. We then flushed the tibias and femurs of donor mice to obtain bone

marrow. We washed bone marrow cells twice in HBSS and injected 1 � 107

bone marrow cells into the tail veins of lethally irradiated (11 Gy) recipient

mice. We performed BDL 12 weeks after BMT and checked the efficiency of

reconstitution by transplanting wild-type mice with bone marrow from mice

expressing transgenic GFP from the b-actin promoter43, as described above .

To determine successful BMT in TLR4-mutant and TLR4–wild-type mice,

spleen cells were isolated from bile duct–ligated chimeric mice, then stimulated

with LPS (100 ng/ml) and analyzed by qPCR to measureme induction of Il6

mRNA expression.

HSC and Kupffer cell isolation and culture. We isolated quiescent HSCs by a

two-step collagenase-pronase perfusion of mouse livers followed by

8.2% Nycodenz (Accurate Chemical and Scientific Corporation) two-layer

discontinuous density gradient centrifugation as previously described46,47. To

ensure the highest possible purity, we depleted HSCs of Kupffer cells and

macrophages by magnetic antibody cell sorting (MACS, Miltenyi Biotec) with

antibody to F4/80 antigen (eBioscience) and CD11b-conjugated microbeads

(Miltenyi Biotec). We used the same procedure to isolate in vivo–activated

HSCs from mice that had undergone BDL for 2 weeks. To isolate Kupffer cells,

ART ICL ES

1330 VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

we performed collagenase-pronase perfusion followed by 15% Nycodenz

gradient centrifugation and subsequent positive selection of F4/80–expressing

cells by MACS11. The described procedures resulted in 99% purity of HSCs and

95% purity of Kupffer cells, as judged by retinoid autofluorescence and flow-

cytometric analysis for F4/80 expression, respectively. We cultured HSCs on

uncoated plastic tissue culture dishes in DMEM containing 1% or 10% FBS

used them as non-passaged primary cultures only. Murine Kupffer cells were

not passaged and were cultured in DMEM containing 1% FBS.

Immunohistochemistry. For immunohistochemical analysis, we fixed liver

specimens in 10% buffered formalin and incubated them with monoclonal

antibody to a-SMA (clone 1A4, DakoCytomation) or monoclonal antibody to

F4/80 (clone BM8; eBioscience) using the MOM kit (Vector Laboratories).

For immunofluorescent staining, we fixed liver specimens with 4% parafor-

maldehyde, incubated them in PBS containing 30% sucrose and froze them

at –80 1C. We then incubated them with antibody to F4/80, antibody to desmin

and antibody to p65 (Santa Cruz Biotechnology) and imaged them with

confocal microscopy.

Western blot. We performed electrophoresis of protein extracts and subsequent

blotting as described9. We incubated blots with mouse antibody to a-SMA

(Sigma), antibodies to mouse Bambi (R&D Systems), phospho-Erk, total Erk,

phospho-Akt (on Ser473), total Erk (all from Cell Signaling Technologies, Inc.)

or HMGB1 (R&D Systems) at a dilution of 1:500–1:5,000 and visualized them

by the enhanced chemiluminescence light method (Amersham Biosciences).

Blots were reprobed with mouse antibody to b-actin (Sigma).

Measurement of collagen-driven green fluorescent protein expression in

hepatic stellate cells. To measure collagen-promoter activity, we isolated HSCs

(1 � 105 cells/well) from collagen promoter–driven GFP transgenic mice19

(pCol9GFP-HS4,5 transgene) and incubated them with LPS (100 ng/ml) or

vehicle for 24 h, and this was followed by treatment with TGF-b1 (300 pg/ml)

for an additional 24 h. We determined the number of GFP-positive cells by

counting GFP-positive cells and total cells in ten randomly chosen high-power

fields. For co-culture experiments, we isolated HSCs (1 � 105 cells/well) from

Coll-GFP–transgenic mice, cultured them in 24-well plates in the presence or

absence of LPS (100 ng/ml for 24 h) and then co-cultured them with

Kupffer cells (5 � 105 cells/well) from either Tlr4–wild-type or Tlr4-mutant

mice for 48 h in the presence or absence of 20 ng/ml soluble type II TGF-breceptor (R&D Systems).

Statistical analysis. All data are expressed as mean ± s.d. Differences between

experimental and control groups were assessed by two-tailed unpaired

Student’s t-tests using SPSS 13.0 (SPSS, Inc.). P-values less than 0.05 were

considered statistically significant.

Additional methods. Detailed methodology is described in Supplementary

Methods.

Note: Supplementary information is available on the Nature Medicine website.

ACKNOWLEDGMENTSThe authors thank S. Akira (Osaka University, Osaka, Japan) for providingMyD88-deficient mice, J. Massague (Memorial Sloan-Kettering Cancer Center,New York, NY, USA) for providing dominant-negative Bambi plasmid, T. Akiyama(University of Tokyo, Tokyo, Japan) for providing Bambi-reporter plasmid,Y. Zhang for assistance with microarray analysis and A. Mencin (ColumbiaUniversity, New York, New York, USA) for assistance with mouse studies. Thisstudy was supported by a Research Scholar Award from the American Gastro-enterological Association (to R.F.S.), grants from the Uehara Memorial Foundationand the American Liver Foundation (to E.S.), an Alimenti e Salute grant from theUniversity of Ancona (to S.D.M.), US National Institutes of Health grantR01GM041804 (to D.A.B.) and the Austrian Science Foundation (to C.H.O.).

AUTHOR CONTRIBUTIONSE.S. performed animal surgery and fibrosis evaluation, cell isolations, cell-culturestudies, reporter assays and real-time PCR, generated recombinant adenoviruses,created figures and contributed to the design of the study and the writing of themanuscript. S.D.M. performed cell isolations, real-time PCR and microarraystudies. Y.O. generated recombinant adenoviruses. C.H.O. generated recombinant

adenoviruses. J.K. performed animal surgeries and cell isolations, performedwestern blots and assisted in fibrosis evaluation and in manuscript preparation.D.A.B. contributed to the design of the study, experimental design and thewriting of the manuscript. R.F.S. (the principal investigator) designed andcoordinated the study, contributed to experimental setup, data analysis andinterpretation, and drafted and edited the manuscript.

Published online at http://www.nature.com/naturemedicine

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions

1. Bataller, R. & Brenner, D.A. Liver fibrosis. J. Clin. Invest. 115, 209–218 (2005).2. Friedman, S.L. Molecular regulation of hepatic fibrosis, an integrated cellular response

to tissue injury. J. Biol. Chem. 275, 2247–2250 (2000).3. Akira, S. & Takeda, K. Toll-like receptor signalling. Nat. Rev. Immunol. 4, 499–511

(2004).4. Jiang, D. et al. Regulation of lung injury and repair by Toll-like receptors and

hyaluronan. Nat. Med. 11, 1173–1179 (2005).5. Bjorkbacka, H. et al. Reduced atherosclerosis in MyD88-null mice links elevated serum

cholesterol levels to activation of innate immunity signaling pathways. Nat. Med. 10,416–421 (2004).

6. Zhai, Y. et al. Cutting edge: TLR4 activation mediates liver ischemia/reperfusioninflammatory response via IFN regulatory factor 3–dependent MyD88-independentpathway. J. Immunol. 173, 7115–7119 (2004).

7. Tsung, A. et al. Hepatic ischemia/reperfusion injury involves functional TLR4 signalingin nonparenchymal cells. J. Immunol. 175, 7661–7668 (2005).

8. Uesugi, T., Froh, M., Arteel, G.E., Bradford, B.U. & Thurman, R.G. Toll-like receptor 4is involved in the mechanism of early alcohol-induced liver injury in mice. Hepatology34, 101–108 (2001).

9. Seki, E. et al. Contribution of Toll-like receptor/myeloid differentiation factor 88signaling to murine liver regeneration. Hepatology 41, 443–450 (2005).

10. Schwabe, R.F., Seki, E. & Brenner, D.A. Toll-like receptor signaling in the liver.Gastroenterology 130, 1886–1900 (2006).

11. Seki, E. et al. Lipopolysaccharide-induced IL-18 secretion from murine Kupffer cellsindependently of myeloid differentiation factor 88 that is critically involved ininduction of production of IL-12 and IL-1b. J. Immunol. 166, 2651–2657 (2001).

12. Duffield, J.S. et al. Selective depletion of macrophages reveals distinct, opposing rolesduring liver injury and repair. J. Clin. Invest. 115, 56–65 (2005).

13. Rivera, C.A. et al. Attenuation of CCl4-induced hepatic fibrosis by GdCl3 treatmentor dietary glycine. Am. J. Physiol. Gastrointest. Liver Physiol. 281, G200–G207(2001).

14. Paik, Y.H. et al. Toll-like receptor 4 mediates inflammatory signaling by bacteriallipopolysaccharide in human hepatic stellate cells. Hepatology 37, 1043–1055(2003).

15. Brun, P., Castagliuolo, I., Pinzani, M., Palu, G. & Martines, D. Exposure to bacterial cellwall products triggers an inflammatory phenotype in hepatic stellate cells. Am. J.Physiol. Gastrointest. Liver Physiol. 289, G571–G578 (2005).

16. Muhlbauer, M. et al. LPS-mediated NF-kB activation varies between activated humanhepatic stellate cells from different donors. Biochem. Biophys. Res. Commun. 325,191–197 (2004).

17. Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S. & Medzhitov, R.Recognition of commensal microflora by Toll-like receptors is required for intestinalhomeostasis. Cell 118, 229–241 (2004).

18. Magilavy, D.B. & Rothstein, J.L. Spontaneous production of tumor necrosis factoralpha by Kupffer cells of MRL/lpr mice. J. Exp. Med. 168, 789–794 (1988).

19. Yata, Y. et al. DNase I-hypersensitive sites enhance alpha1(I) collagen gene expressionin hepatic stellate cells. Hepatology 37, 267–276 (2003).

20. Onichtchouk, D. et al. Silencing of TGF-b signalling by the pseudoreceptor Bambi.Nature 401, 480–485 (1999).

21. Iimuro, Y. et al. NF-kB prevents apoptosis and liver dysfunction during liver regenera-tion. J. Clin. Invest. 101, 802–811 (1998).

22. Maeda, S., Kamata, H., Luo, J.L., Leffert, H. & Karin, M. IKKbeta couples hepatocytedeath to cytokine-driven compensatory proliferation that promotes chemical hepato-carcinogenesis. Cell 121, 977–990 (2005).

23. Pikarsky, E. et al. NF-kB functions as a tumour promoter in inflammation-associatedcancer. Nature 431, 461–466 (2004).

24. Isogawa, M., Robek, M.D., Furuichi, Y. & Chisari, F.V. Toll-like receptor signalinginhibits hepatitis B virus replication in vivo. J. Virol. 79, 7269–7272 (2005).

25. Izuishi, K. et al. Cutting edge: high-mobility group box 1 preconditioning protectsagainst liver ischemia-reperfusion injury. J. Immunol. 176, 7154–7158 (2006).

26. Lotze, M.T. & Tracey, K.J. High-mobility group box 1 protein (HMGB1): nuclear weaponin the immune arsenal. Nat. Rev. Immunol. 5, 331–342 (2005).

27. Termeer, C. et al. Oligosaccharides of hyaluronan activate dendritic cells via Toll-likereceptor 4. J. Exp. Med. 195, 99–111 (2002).

28. Tsung, A. et al. The nuclear factor HMGB1 mediates hepatic injury after murine liverischemia-reperfusion. J. Exp. Med. 201, 1135–1143 (2005).

29. Isayama, F. et al. LPS signaling enhances hepatic fibrogenesis caused by experimentalcholestasis in mice. Am. J. Physiol. Gastrointest. Liver Physiol. 290, G1318–G1328(2006).

30. Friedman, S.L. & Arthur, M.J. Activation of cultured rat hepatic lipocytes by Kupffercell conditioned medium. Direct enhancement of matrix synthesis and stimulation of

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 1331

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

cell proliferation via induction of platelet-derived growth factor receptors. J. Clin.Invest. 84, 1780–1785 (1989).

31. De Minicis, S. et al. Gene expression profiles during hepatic stellate cell activation inculture and in vivo. Gastroenterology 132, 1937–1946 (2007).

32. Sasaki, T., Sasahira, T., Shimura, H., Ikeda, S. & Kuniyasu, H. Effect of Nma ongrowth inhibition by TGF-b in human gastric carcinoma cell lines. Oncol. Rep. 11,1219–1223 (2004).

33. Tsang, M. et al. Zebrafish nma is involved in TGF-b family signaling. Genesis 28,47–57 (2000).

34. Sekiya, T. et al. Identification of BMP and activin membrane-bound inhibitor (Bambi),an inhibitor of transforming growth factor-b signaling, as a target of the b-cateninpathway in colorectal tumor cells. J. Biol. Chem. 279, 6840–6846 (2004).

35. Sekiya, T., Oda, T., Matsuura, K. & Akiyama, T. Transcriptional regulation of the TGF-bpseudoreceptor Bambi by TGF-b signaling. Biochem. Biophys. Res. Commun. 320,680–684 (2004).

36. Oakley, F. et al. Inhibition of inhibitor of kB kinases stimulates hepatic stellatecell apoptosis and accelerated recovery from rat liver fibrosis. Gastroenterology 128,108–120 (2005).

37. Thurman, R.G. et al. The role of gut-derived bacterial toxins and free radicals inalcohol-induced liver injury. J. Gastroenterol. Hepatol. 13 Suppl, S39–S50 (1998).

38. Diehl, A.M., Li, Z.P., Lin, H.Z. & Yang, S.Q. Cytokines and the pathogenesis of non-alcoholic steatohepatitis. Gut 54, 303–306 (2005).

39. Campbell, J.S. et al. Proinflammatory cytokine production in liver regeneration isMyd88-dependent, but independent of Cd14, Tlr2, and Tlr4. J. Immunol. 176, 2522–2528 (2006).

40. Huang, H. et al. A 7 gene signature identifies the risk of developing cirrhosis in patientswith chronic hepatitis C. Hepatology, 46, 297–306 (2007).

41. Hoebe, K. et al. Identification of Lps2 as a key transducer of MyD88-independent TIRsignalling. Nature 424, 743–748 (2003).

42. Adachi, O. et al. Targeted disruption of the MyD88 gene results in loss of IL-1- andIL-18-mediated function. Immunity 9, 143–150 (1998).

43. Kisseleva, T. et al. Bone marrow-derived fibrocytes participate in pathogenesis of liverfibrosis. J. Hepatol. 45, 429–438 (2006).

44. Kennedy, D.W. & Abkowitz, J.L. Kinetics of central nervous system microglial andmacrophage engraftment: analysis using a transgenic bone marrow transplantationmodel. Blood 90, 986–993 (1997).

45. Van Rooijen, N. & Sanders, A. Liposome mediated depletion of macrophages:mechanism of action, preparation of liposomes and applications. J. Immunol. Methods174, 83–93 (1994).

46. Bataller, R. et al. NADPH oxidase signal transduces angiotensin II in hepatic stellatecells and is critical in hepatic fibrosis. J. Clin. Invest. 112, 1383–1394 (2003).

47. Siegmund, S.V., Uchinami, H., Osawa, Y., Brenner, D.A. & Schwabe, R.F. Anandamideinduces necrosis in primary hepatic stellate cells. Hepatology 41, 1085–1095(2005).

ART ICL ES

1332 VOLUME 13 [ NUMBER 11 [ NOVEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine