Periostin promotes liver fibrogenesis by activating lysyl oxidase in hepatic stellate cells

Liver fibrosis arises from dysregulated wound healing due to persistent inflammatory hepatic injury. Periostin is a nonstructural extracellular matrix protein that promotes organ fibrosis in adults. Here, we sought to identify the molecular mechanisms in periostin-mediated hepatic fibrosis. Hepatic fibrosis in periostin−/− mice was attenuated as evidenced by significantly reduced collagen fibril density and liver stiffness compared with those in WT controls. A single dose of carbon tetrachloride caused similar acute liver injury in periostin−/− and WT littermates, and we did not detect significant differences in transaminases and major fibrosis-related hepatic gene expression between these two genotypes. Activated hepatic stellate cells (HSCs) are the major periostin-producing liver cell type. We found that in primary rat HSCs in vitro, periostin significantly increases the expression levels and activities of lysyl oxidase (LOX) and lysyl oxidase–like (LOXL) isoforms 1–3. Periostin also induced expression of intra- and extracellular collagen type 1 and fibronectin in HSCs. Interestingly, periostin stimulated phosphorylation of SMAD2/3, which was sustained despite short hairpin RNA–mediated knockdown of transforming growth factor β (TGFβ) receptor I and II, indicating that periostin-mediated SMAD2/3 phosphorylation is independent of TGFβ receptors. Moreover, periostin induced the phosphorylation of focal adhesion kinase (FAK) and AKT in HSCs. Notably, siRNA-mediated FAK knockdown failed to block periostin-induced SMAD2/3 phosphorylation. These results suggest that periostin promotes enhanced matrix stiffness in chronic liver disease by activating LOX and LOXL, independently of TGFβ receptors. Hence, targeting periostin may be of therapeutic benefit in combating hepatic fibrosis.

Cirrhosis is a late complication of hepatic fibrosis, which results in significant morbidity and mortality worldwide (1)(2)(3). Hepatic fibrosis is the consequence of a dysregulated woundhealing process triggered by persistent inflammatory injury to the liver (4). Activated hepatic stellate cells (HSCs) 2 and portal myofibroblasts are the primary cellular players involved in the fibrogenic response to chronic liver injury. In health, quiescent HSCs constitute ϳ8 -14% of the total liver cell population (5,6). Following a fibrogenic stimulus, the quiescent HSC transdifferentiates into a myofibroblast-like cell referred to as the "activated HSC" (4). Activated HSCs secrete a wide array of chemokines and extracellular matrix (ECM) proteins, such as fibrillar collagen and fibronectin (4,7). Although key mechanisms of HSC activation have been defined, safe and effective therapeutic targets to reverse dense ECM deposition in the liver, and thus fibrosis resolution, remain elusive.
Periostin is a 90-kDa secretory matricellular protein composed of an N-terminal emilin domain, four repeated fasciclin domains, and a C-terminal heparin binding domain with alternatively spliced variants (8,9). Periostin mediates pleiotropic effects during inflammation, tissue injury, and wound healing (8); importantly, it possesses a highly conserved primary sequence structure related to transforming growth factor (TGF-␤)-induced protein (␤ig-h3) (10). In addition to being an ancient wound-healing molecule, periostin is involved in the embryologic development of the periodontal ligament (11). Periostin is known to activate canonical signaling pathways during wound healing, tissue repair, and matrix remodeling via transmembrane integrins. However, although integrins serve as receptors for protein containing Arg-Gly-Asp (RGD) sequence, periostin lacks this signature (12). Periostin also has the distinct ability to directly interact with key molecules of fibrotic matrix, including fibronectin, type I collagen, elastin, and tenascin, via its emilin domain to promote fibrillogenesis (13). This unique property is thought to be responsible for matrix stiffening via collagen cross-linking that ultimately triggers mechanico-sig-This work was supported by NIDDK, National Institutes of Health, Grants RO1 DK062092, DK111678, and DK113147 (all to F. A. A.). A portion of this work was also supported by funds from Emory University. The authors declare that they have no conflicts of interest with the contents of this article. This article was prepared while Frank A. Anania was employed at Emory University School of Medicine. The opinions expressed in this article are the author's own and do not reflect the views of the Food and Drug Administration, the Department of Health and Human Services, or the United States government. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article contains Tables S1 and S2 and Figs. S1-S4. 1  naling to neighboring HSCs. Additionally, ablation of periostin in vivo prevents cardiac, lung, renal, and hepatic fibrosis (10, 14 -16). Liver stiffness is often associated with excessive accumulation of insoluble and fibrillar collagens in the ECM. Lysyl oxidase (LOX) is a copper-dependent extracellular, matrixembedded amine oxidase protein that plays a critical role in covalent cross-linking of fibrillar collagens and elastin (17,18). There are a total of five LOX isoforms identified to date: LOX and the LOX-like (LOXL) enzymes 1-4, which share a conserved enzymatic domain with divergent N termini (19,20). LOX, LOXL1, and LOXL2 predominantly promote covalent cross-linking of collagen fibers, which is important for collagen stabilization as well as matrix integrity and elasticity. Interestingly, LOX expression and activity significantly up-regulate during HSC activation into myofibroblasts (19).
Epithelial-to-mesenchymal transition (EMT) is a multistep dynamic cellular process considered to be involved in the development of lung cancer after periostin exposure (21,22). However, a role for periostin in the process of EMT promotion of liver fibrosis has not been explored.
In this study, we provide several new molecular mechanisms whereby periostin is pivotal in generating the fibrotic response in liver. Deletion of periostin protects from CCl 4 -induced liver injury in mice. Periostin induced SMAD2/3, independent of TGF phosphorylation by activation of the integrin ␣v␤3-PI3K pathway.

Periostin ؊/؊ mice are protected from CCl 4 -induced liver fibrosis
In vivo data demonstrated that periostin Ϫ/Ϫ mice are protected from CCl 4 -induced liver fibrosis (Fig. S1, A-F). Because periostin is a nonstructural ECM protein known to stimulate type I collagen cross-linking and promotes matrix stiffness, we examined the stiffness of the mouse livers by atomic force microscopy (AFM). As shown in Fig. 1A, liver stiffness of the CCl 4 -treated periostin Ϫ/Ϫ mice was significantly less than the WT controls. The density of collagen fibers in CCl 4 -treated periostin Ϫ/Ϫ mice versus WT mice was analyzed by taking transmission electron microscopic (TEM) images of fixed liver tissues as shown in Fig. 1B. The diameter of collagen fibers was measured by ImageJ (Fig. 1C). By this approach, we demonstrated that collagen fibril density in the periostin Ϫ/Ϫ mice was substantially reduced when compared with the livers of CCl 4treated WT mice (Fig. 1C).
We examined whether collagen bundles in the fibrotic matrix were qualitatively different in CCl 4 -treated periostin Ϫ/Ϫ mice compared with WT controls. For instance, the solubility of collagen in the ECM is inversely correlated to the extent of collagen cross-linking. Not surprisingly, the insoluble collagen fraction increased with the development of hepatic fibrosis (Fig.  1D). The ratio of insoluble collagens increased from ϳ5% in untreated groups to ϳ37 and ϳ19% in WT and periostin Ϫ/Ϫ mice treated with CCl 4 , respectively (p Ͻ 0.05; Fig. 1D). Additionally, pepsin-soluble collagen decreased from ϳ90% in untreated controls to ϳ56% in CCl 4 -treated WT mice (Fig. 1D). However, this response to CCl 4 treatment was significantly attenuated in periostin Ϫ/Ϫ mice, which had a reduction in pepsin-soluble collagen to only ϳ76% (Fig. 1D). In sum, the matrix stability assessment in the absence of periostin in vivo is consistent with both the AFM and TEM data.

mRNA expression of fibrosis-related genes is reduced in CCl 4 -treated periostin ؊/؊ mice
In addition to histological and biochemical studies, quantitative RT-PCR analysis confirmed significantly attenuated mRNA expression of fibrosis-related genes in CCl 4 -treated

LOX and LOXL1-2 expression is reduced in CCl 4 -treated periostin ؊/؊ mice
Because in vivo data suggested a pivotal role of periostin in collagen fibrillogenesis during the development of liver fibrosis, we next examined whether mRNA expression of LOX and its four isoforms LOXL1-4 changed in the livers of CCl 4 -treated periostin Ϫ/Ϫ mice. As shown in Fig. 2 (A-C), LOX, LOXL1, and LOXL2 mRNA transcripts, respectively, were significantly reduced in the livers of CCl 4 -treated periostin Ϫ/Ϫ mice compared with CCl 4 -treated WT mice. LOXL3 expression was also induced by CCl 4 treatment, but its expression levels did not differ significantly between WT and periostin Ϫ/Ϫ mice (Fig.  2D). CCl 4 treatment did not change the mRNA expression of LOXL4 in WT and periostin Ϫ/Ϫ mice (Fig. 2E). Protein expression of LOX and LOXL isoforms 1 and 2 was attenuated in periostin Ϫ/Ϫ mice (Fig. 2F). Additionally, we also determined the LOX activity in liver homogenates by fluorometric assay. LOX activity was significantly attenuated in livers from periostin Ϫ/Ϫ mice compared with WT mice (Fig. 2G), both of which were treated with CCl 4 for 6 weeks.

Periostin deficiency does not change the liver injury induced by a single CCl 4 treatment
Because periostin Ϫ/Ϫ mice are protected from chronic liver injury, we explored whether the ablation of periostin had any influence on the acute liver response to a single dose of CCl 4 exposure. Our data indicate that both periostin Ϫ/Ϫ and WT littermate mice develop centrilobular necrosis, which peaked after day 1 following CCl 4 treatment, with recovery by 7 days following CCl 4 exposure (Fig. 3A). We did not observe any significant histological difference between WT and periostin Ϫ/Ϫ mice. Consistent with our histological observations, we also found that alanine transaminase and aspartate transaminase levels in serum had similar peak values in both periostin Ϫ/Ϫ and WT littermate mice at 24 h following the single dose of CCl 4 (Fig. 3, B and C). Next, we examined the expression of several genes related to inflammation, fibrosis, and matrix remodeling by RT-qPCR. The mRNA levels for TGF␤1, CTGF, TIMP1, Col1␣1, and F4/80 were significantly up-regulated after day 1 in both the periostin Ϫ/Ϫ and WT littermate mice (Fig. 3D). Elevated mRNA expression levels persisted through day 3 after injury. Periostin mRNA was significantly up-regulated in WT mice at day 1 and remained up-regulated at day 7. The peak induction in mRNA levels of IL1␤, MMP9, MMP13, and IL6 was observed at day 1 in both periostin Ϫ/Ϫ and WT mice, whereas mRNA of MMP2 was highly up-regulated after day 3 of CCl 4 treatment. We did not observe any significant difference of mRNA expression in either genotype at the indicated time points except for TIMP1 and MMP9, which, after day 1, were reduced in periostin Ϫ/Ϫ mice compared with WT littermates (Fig. 3D). Taken together, these findings suggest that periostin does not appear to play a significant role in the pathogenesis of acute liver injury unlike chronic injury. Importantly, periostin does not appear to play a major role in the acute inflammatory response.

Periostin expression is up-regulated in CCl 4 -induced liver fibrosis and co-localized with ␣-SMA-positive cells in the liver
To investigate periostin expression in major hepatic cell types and the potential source for periostin, we co-stained liver tissue sections of CCl 4 -treated WT mice with periostin and markers of major hepatic cell types: albumin (hepatocytes), ␣-SMA (hepatic stellate cells), CD146 (liver sinusoidal endo-

Periostin promotes liver fibrogenesis
Next, we isolated and cultured four major cell types (hepatocytes, HSCs, Kupffer cells, and LSECs) from healthy mouse livers. Total mRNA was harvested from each of the cell types after 3 days of primary cell culture. As shown in Fig. 4F, periostin mRNA was expressed at markedly higher levels in HSCs than in hepatocytes, Kupffer cells, or LSECs.

Recombinant periostin up-regulates LOX/LOXL expression in HSCs
Previous studies have shown that periostin can regulate LOX, collagen, and fibronectin expression in fibroblasts (25,26). To dissect the role of periostin in LOX and LOX-related isoform expression, we isolated primary HSCs from rats and performed immunoblots of lysates from freshly isolated (quiescent) and activated HSCs as shown in Fig. S3. Culture-activated HSCs highly expressed LOX and its isoforms LOXL1-4 along with periostin, ␣-SMA, collagen type I, and fibronectin (Fig.  S3). We treated cultured HSCs with or without full-length human recombinant periostin (100 ng/ml) for 24 h. Superna-tant and whole-cell lysates were collected for Western blot analysis. In vitro, periostin treatment significantly induced the expression of different LOX isoforms except for LOXL4 (Fig. 5A), and supernatant from cell culture (media) periostin resulted in enhanced secretion of LOXL2 along with fibronectin and collagen type I (Fig. 5B) but not LOXL3 or LOXL4 (data not shown). We confirmed that periostin significantly increased LOX activity in the supernatants of cultured HSCs (Fig. 5C).

Periostin modulates SMAD2/3 phosphorylation in HSCs
TGF-␤ has been reported to induce periostin expression in several cell lines (1,27); however, the relationship between periostin and the TGF␤-SMAD signaling pathway has not been clearly delineated. Furthermore, the prevailing hypothesis is that periostin production depends on TGF␤ activity. To clarify the association between periostin and the TGF␤-SMAD pathway, we stimulated culture-activated HSCs with various concentrations of periostin for 30 min and performed Western blot analysis followed by immunoblotting for phospho-

Silencing SMAD2/3 attenuates periostin-induced LOX expression in HSCs
Because periostin-induced production of LOX is known to correlate with increased collagen I and fibronectin production and can independently activate SMAD2/3 in liver myofibroblast, we assessed whether periostin-mediated SMAD2/3 activation was responsible for increased LOX expression. By using siRNA, we knocked down the SMAD2/3 complex in myofibroblasts and then treated cells with periostin for 24 h. Periostin treatment significantly stimulated the expression of LOX and collagen in scrambled siRNA-transfected HSCs, but periostinmediated LOX expression was abolished in si-SMAD2/3transfected HSCs (Fig. 6F). si-SMAD2/3 also attenuated fibronectin and type I collagen culture-activated HSC protein expression (Fig. 6F).

Integrin ␣v␤3 is critical to periostin-induced LOX expression in HSCs
The above data demonstrate that integrin ␣v␤3 mediates periostin-induced SMAD phosphorylation, and knockdown of SMAD2/3 reduced LOX expression. Therefore, we determined whether integrin ␣v␤3 was critical to periostin-mediated LOX expression by antibody blockade with anti-␣v␤3. As shown in Fig. 7 (A and B), periostin treatment significantly induced LOX expression and activity, whereas anti-integrin ␣v␤3 blocked both periostin-stimulated LOX expression and activity.

siRNA-mediated knockdown of focal adhesion kinase (FAK) fails to block SMAD phosphorylation in HSCs
Because integrins are known to signal through FAK (28), we next determined whether activation of FAK could promote SMAD2/3 phosphorylation by periostin. For this experiment, we employed siRNA knockdown of FAK (si-FAK) and performed Western blot analysis. As shown in Fig. 8A, si-FAK failed to change the status of SMAD2/3 phosphorylation, whereas AKT phosphorylation was attenuated in the presence of periostin (Fig. 8B). Furthermore, blocking antibody against integrin ␣v␤3 attenuated periostin-stimulated phosphorylation of FAK and AKT (Fig. 8B).

Periostin promotes liver fibrogenesis Periostin positively regulates EMT in primary mouse hepatocytes
Recent data indicate that periostin levels are higher in humans with hepatobiliary malignancy (29,30). Because cirrhosis and fibrogenic changes pose a risk for tumor development, we explored whether periostin could promote EMT in murine hepatocytes. After isolating mouse hepatocytes, we treated primary cultures with periostin and examined whether periostin influenced the expression of EMT-associated genes. Periostin treatment significantly induced the expression of mesenchymal markers, including Col1␣1, vimentin, fibronectin, and N-cadherin (Fig. 9, A-D). In contrast, the mRNA expression of epithelial markers, such as E-cadherin and HNF4␣, was effectively suppressed (Fig. 9, E and F). We then isolated and cultured hepatocytes from WT and periostin Ϫ/Ϫ mice and used RT-qPCR to measure EMT-related gene expression. Compared with WT mice, hepatocytes of periostin Ϫ/Ϫ mice expressed significantly higher mRNA levels of epithelial markers and lower mRNA levels of mesenchymal markers (Fig. 9G).

Discussion
In the current study, we have demonstrated several molecular mechanisms associated with periostin-mediated liver fibrosis that have not been previously reported. Some of the data presented corroborate recent reports that periostin enhances matrix stiffness and insoluble fibrillar collagen deposition during liver fibrogenesis (16), and our studies lend credence to several reports (16,31,32) that periostin is in contention to be an important pro-fibrogenic mediator. However, here we have shown that persiostin may be an independent mediator of SMAD activation that does not appear to be TGF␤-dependent. Second, we demonstrated that periostin can activate SMADs via the integrin ␣v␤3 pathway, which depends on PI3K and is specific to this integrin (Fig. 10). Third, we have shown that periostin plays an important role in the EMT of hepatocytes

Periostin promotes liver fibrogenesis
and that in chronic exposure to CCl 4 , the bulk of periostin production is derived from HSCs. Together, our new findings clearly merit additional investigation.
Periostin is highly expressed during embryonic development but expressed at very low levels in adulthood. Recent reports showed that periostin expression was up-regulated in human fibrotic liver tissues compared with healthy controls (16,33). However, periostin Ϫ/Ϫ mice do not suffer significant developmental abnormalities compared with their WT littermates. Recent studies confirm that periostin-deficient mice are protected from fibrotic stimuli in various organs, including liver (14,16,27,34). Our current data indicate fibrotic liver sections of periostin Ϫ/Ϫ mice were less stiff than liver sections of WT mice and also have lower levels of collagen I expression. Liver stiffness is a multifaceted phenomenon that makes outside-inmyofibroblast signal transduction (19,20,27) seminal to the molecular orchestration of deposition and maintenance of dense, fibrillar ECM that ultimately leads to disease pathogenesis. As reported previously (35), we have also observed that periostin can propagate pro-fibrotic signaling via the integrin-PI3K signaling axis, specifically via ␣v␤3 integrins. Earlier stud-

Periostin promotes liver fibrogenesis
ies have shown that collagen cross-linking mediated by the copper-dependent enzyme LOX and its isoforms LOXL1-4 is enhanced by ECM-integrin signaling (36). Mesarwi et al. (37) reported that hepatic LOX mRNA levels were 6.7-fold higher in patients with fibrosis compared with healthy individuals. Furthermore, they correlated serum LOX levels with hepatic mRNA levels (37). It is hypothesized that softer matrices may promote fibrosis resolution and ultimately restore normal matrix density and function by permitting matrix metalloproteinases (MMPs) to ultimately remove dense matrix. Recent studies by Popov et al. (18) and others (38 -40) indicate that the LOX inhibitor ␤-aminopropionitrile may provoke hepatic progenitors to replace the ductular reaction associated with portal fibrosis; hence, there is a strong rationale to investigate additional targets mediating collagen cross-linking in fibrosis injury. These studies are collectively corroborated by recent human studies indicating that children who suffer from biliary atresia have significantly higher circulating levels of periostin (41). Indeed, periostin as a marker for fibrotic injury has long been a surrogate for lung fibrosis (42). Our findings indicate that mRNA expression of LOX and its isoforms LOXL1-3 were significantly lower in periostin Ϫ/Ϫ mice compared with WT mice after CCl 4 gavage, which makes targeting periostin a compelling strategy for development of antifibrotic therapy.
Taking the current findings together with the other data reported thus far, several important questions need to be addressed. First, are activated HSCs and portal myofibroblasts the only source of periostin, or do Kupffer cells, or even populations of infiltrating cells, play a contributory role in periostin production? Because we note that the typical transaminases associated with chronic CCl 4 exposure are significantly reduced in periostin Ϫ/Ϫ mice, we cannot yet exclude this possibility. Not unexpectedly, we identified HSCs as the major cellular source of periostin in CCl 4 -induced liver fibrosis. The expression of periostin in hepatocytes, LSECs, and Kupffer cells in the CCl 4 -treated mice was significantly lower than in HSCs. Together, these observations suggest that HSCs are the major source of hepatic periostin during liver fibrosis, and a definitive source for hepatic periostin production is currently under investigation in our laboratory. A single dose of CCl 4 treatment resulted in a similar acute response in WT and periostin Ϫ/Ϫ mice as demonstrated by serum transaminase levels, histology, and hepatic mRNA levels.
Finally, in the context of these data in periostin Ϫ/Ϫ mice, we set out to determine whether periostin possesses additional functions in the pathogenesis of matrix biology besides its known cross-linking function. Our current data support the notion that periostin plays a critical role in enhanced expression and activity of TGF-␤1 in HSCs (Fig. S4, A and B). We found that periostin significantly induced TGF-␤1 mRNA expression in HSCs and enhanced its activity in culture supernatants. Also, hepatic TGF-␤1 mRNA expression was suppressed in periostin Ϫ/Ϫ mice gavaged with CCl 4 compared with CCl 4 -treated WT mice. TGF-␤ has been reported to induce periostin expression in several cell lines (1,43,44); however, the biological relationship between TGF-␤ and periostin is not clear. Given the interrelationship between TGF-␤1 and periostin, we sought to determine whether periostin could instigate SMAD activation as a novel ligand for the TGF-␤ receptor. By both shRNA-mediated and pharmacological approaches against both TGF-␤ receptor isoforms (RI and RII), we found that periostin-induced SMAD2/3 phosphorylation was sustained despite the knockdown of the TGF-␤ receptor as well as the presence of TGF-␤ receptor inhibitor. Although it may be premature to declare periostin independent of TGF-␤ without studies performed in animal models, our current data strongly suggest that periostin-induced serine phosphorylation of SMAD2/3 functions independently of TGF-␤ receptor(s). Periostin also has the unique ability to interact not only with integrins but has potential to interact with other myofibroblast receptors (e.g. discoidin domain receptor 1) (45). 3 Novel periostin signaling interactions would further shed light on distinct pathways not yet ascribed that would promote production of dense, fibrillar ECM.
Previous studies have demonstrated that periostin contributes to the EMT process in various cancers and fibrosis (21, 46 -48). However, the role of EMT in liver fibrosis has become a highly controversial topic in the past several years. In vitro studies demonstrated that hepatocytes can undergo the process of EMT and express several mesenchymal markers. Furthermore, Zeisberg et al. (49) reported the first in vivo evidence for hepatocyte EMT through lineage-tracing experiments demonstrating that hepatocytes express EMT marker S100A4 (fibroblast-specific protein 1). Similarly, several other research groups have demonstrated that targeting EMT could suppress hepatic fibrogenesis (50 -52). However, a recent lineage-tracing study identified that mouse hepatocytes fail to undergo EMT during CCl 4 -induced liver fibrosis (53). Although this work challenges the existence of hepatocyte EMT in liver fibrosis, lineage-tracing techniques have several notable pitfalls. Extensive studies are required to confirm the role of hepatocyte EMT in liver fibrosis.
In this study, we demonstrated that periostin expression appears to be very low in healthy liver, but following a profibrotic stimulus, hepatic periostin expression is markedly increased. The primary source of injury-driven periostin expression at present is the activated HSC or ␣-SMA-positive myofibroblast. Moreover, periostin Ϫ/Ϫ mice appear to be protected from CCl 4 -induced liver fibrosis and have reduced liver stiffness. These findings were corroborated with reduced insoluble collagen deposition and diminished LOX/LOXL expression and respective enzyme activity. Although there are several novel findings we uncovered here, the results have prompted more questions than we have set out to address. Clearly, additional molecular approaches will shed light on whether periostin would be an effective therapeutic target for the treatment of hepatic fibrosis in the future.

Animals and CCl 4 -induced liver fibrosis in mice
Eight-week old male periostin Ϫ/Ϫ and WT (C57BL/6J) littermate mice were used for animal studies. periostin Ϫ/Ϫ mice were purchased from The Jackson Laboratory (Bar Harbor, ME), stock no. 009067. The mice used in the experiments were the progeny from backcrossed C57BL/6J mice. Mice were cared for in accordance with protocols approved by the animal care and use committee of Emory University. Mice were housed in a temperature-controlled environment with a 12-h/12-h light/ dark cycle. Animals were fed ad libitum with a Purina Laboratory Standard Chow diet (Ralston Purina, St. Louis, MO) and water. In addition to PCR-based genotyping, we also confirmed the absence of periostin in the serum by ELISA (mouse periostin ELISA kit, RayBiotech, Inc., Norcross, GA). WT and periostin Ϫ/Ϫ mice weighing 20 -25 g were gavaged with CCl 4 (2 ml/kg) diluted in olive oil (1:1 ratio) or an equivalent volume of olive oil mixed with saline three times weekly for 6 weeks as described elsewhere (54). Antibodies and primers used in the current study presented in Tables S1 and S2, respectively.

Assessment of liver tissue mechanical properties by AFM
Liver stiffness was measured by AFM. Liver tissue cryosections of 15-m thickness were obtained from periostin Ϫ/Ϫ and WT mice. The AF microscope used in the present study was the MFP-3D (Asylum Research, Santa Barbara, CA) with a Nikon Ti-inverted optical microscope (Nikon, Melville, NY) located at the Georgia Institute of Technology (Atlanta, GA). The detailed procedure for AFM was described elsewhere (55,56).

Assessment of matrix stability
Matrix stability was analyzed by complete collagen fractionation using a biochemical assay via serial extractions from liver tissues followed by hydroxyproline quantitation. The detailed methods of the fibrotic matrix stability assay were described by Popov and co-workers (18,24).

Primary rat HSC isolation and treatments
Primary rat HSCs were isolated from male Sprague-Dawley rats by a two-step in situ liver perfusion protocol as described previously (54). Primary rat HSCs were used between passages 2 and 5. Confluent HSCs were serum-starved and treated with or without recombinant human periostin (Sino Biological Inc., Wayne, PA) as indicated in the figure legends. For inhibitor and blocking antibody experiments, HSCs were pretreated for 30 min, followed by stimulation with periostin along with vehicle, control IgG, inhibitor, and blocking antibody.

Lentivirus-mediated TGF␤ receptor knockdown
Plasmids encoding sh-TGF-␤RI, sh-TGF-␤RII (catalog nos. TR709395 and TR711255), nontargeting shRNA (pRS vector), and packaging plasmids were purchased from Origene Technologies, Inc. (Rockville, MD). Viral particles were generated by the Lentiviral packing kit (Origene) according to the manufacturer's guidelines. Briefly, HEK293T cells were transfected with premix packaging plasmids and pRS-shRNA lentiviral vector using TurboFectin transfection reagent (Origene). After 48 h of transfection, the supernatant containing viral particles was collected and filtered through a 0.45-m polyvinylidene difluoride filter. HSCs were transduced with concentrated and purified lentiviral particles (multiplicity of infection ϭ 5) in the presence of Polybrene (5 g/ml). The infected cells were cul-tured in complete growth medium for 72 h, and stable clones were selected with puromycin dihydrochloride (2 g/ml). Surviving colonies were pooled and expanded in complete cell culture medium containing puromycin (1 g/ml) for further experiments.

Hepatocyte isolation
Hepatocytes were isolated from adult mice using the standard two-step collagenase perfusion method as described previously (23). Briefly, under anesthesia, liver was perfused in situ via the portal vein for 5 min with calcium/magnesium-free HEPES buffer. Liver was digested with magnesium-free buffer containing type IV collagenase (30 mg/ml) and calcium chloride (10 mM) for 6 min. Digested liver was torn with the help of sterile forceps. Cells were filtered through 70-m nylon cell strainers (Thermo Fisher Scientific, Waltham, MA). Cells were centrifuged at 50 ϫ g for 2 min, and supernatant was discarded. Pellet was resuspended in 20 ml of 40% Percoll followed by centrifugation at 200 ϫ g for 5 min. After centrifugation, bottom parts (hepatocytes) were collected. Cells were washed three times with 10 ml of Dulbecco's modified Eagle's medium containing 5% FBS by centrifugation at 50 ϫ g for 2 min.

Statistical analysis
All data are expressed as the mean Ϯ S.E. from 3-4 separate experiments. The differences between groups were analyzed using a two-tailed Student's t test when only two groups were analyzed or analysis of variance when there were more than two groups analyzed. The statistical analyses were conducted using the Excel (Microsoft Office 2015) GraphPad Prism software version 5.04 (GraphPad Software, Inc., La Jolla, CA). A p value Ͻ 0.05 was considered as a statistically significant difference.
Author contributions-P. K. designed the study; collected data; performed all of the experiments, statistical analysis, and interpretation of data; and wrote the manuscript. T. S. maintained the mouse colony and CCl 4 gavage. R. R. and D. M. C. critically revised the manuscript for important intellectual content. H. B. and T. S. acquired AFM data. Y. L. provided assistance with Western blot analyses. F. A. A. was responsible for study concept and design and outlined and revised the manuscript.