FSTL1 Secreted by Activated Fibroblasts Promotes Hepatocellular Carcinoma Metastasis and Stemness

Globally, hepatocellular carcinoma (HCC) is the sixth most common cancer and third deadliest cancer, causing more than 800,000 new cases and approximately 800,000 deaths in 2018 (1). The presence of liver fibrosis, a consequence of continuous repair and damage, is a prominent risk factor of HCC development with 80% to 90% of all cases observed to be accompanied by a fibrotic or cirrhotic liver (2). HCC can be illustrated as a multistep disease in which a normal liver develops cirrhosis, followed by the manifestation of dysplastic features, ultimately neoplastic transformation into HCC. Depending on the demographical and etiologic factors such as hepatitis B and C viral infection, 5% to 30% of the patients with cirrhotic liver develop HCC in the next 5 years (3). Accordingly, patients with more advanced cirrhosis have greater risk of developing HCC. Together, a fibrotic liver that is regularly undergoing repair may offer a permissive premalignant microenvironment and tumor microenvironment (TME) for hepatocyte regeneration and cancer cell progression, respectively (4). Myofibroblast, an activated form of fibroblast, plays a vital role in tissue repair and fibrosis, partly by producing growth factors crucial for growth (4). A fibrotic or cirrhotic liver is enriched for proinflammatory factors secreted from activated fibroblasts and immune cells to promote growth upon injury (4, 5), thereby providing the opportunity for cancer cells to exploit this recovery machinery for progression.

To date, merely two categories of FDA-approved drugs are in clinical practice for advanced HCC, namely multikinase and immune checkpoint inhibitors. Among the multikinase inhibitors, sorafenib is commonly administered as first-line therapy (6). However, sorafenib can only extend the median survival for approximately 3 months (7), thereby highlighting the urgent need to develop novel treatment for advanced HCC. Tumor-initiating cells (TIC), alternatively named cancer stem cells (CSC), are categorized by their enhanced self-renewal properties, which can confer resistance to therapy by regeneration of the depleted tumor (8). TICs rely heavily on the TME to maintain its stemness (9). Cancer-associated fibroblast (CAF), a component of the stromal fraction in the TME, can contribute to the stemness of TICs by secreting growth factors, such as chemokine (C-C motif) ligand 2 (CCL2) and hepatocyte growth factor (HGF) to upregulate Notch and Wnt stemness signaling (10–12). CAFs have been shown to promote chemoresistance and maintain cancer stemness by the secretion of IL6 (13). CAF-derived factors are also shown to promote cancer cell invasion and migration (14–16). However, the mechanism in which a highly inflammatory premalignant microenvironment during liver fibrosis translates to a TME that contributes to malignant properties of HCC remains largely unknown.

FSTL1 is a proinflammatory factor, which is found to be upregulated during inflammation in tissues such as the joint, lung, and heart (17–19). FSTL1 has been reported to promote malignant properties in various cancers, such as breast, esophageal, and colon cancers (20–22). Considering the known role of FSTL1 in inflammation and the fact that most HCCs develop in a cirrhosis background, the role of FSTL1 in mediating HCC warrants further investigation. Here, we identified Fstl1 to be upregulated in regenerative, fibrotic, and disease states using different mouse models of the liver. Fstl1 lineage cells gave rise to myofibroblasts in a CCL₄-induced hepatic fibrosis mouse model. We further demonstrated that FSTL1 is predominantly secreted from the stromal cells rather than liver cancer cells. In vitro and in vivo functional assays showed FSTL1 augments the stemness, proliferation, metastasis, and therapy resistance properties of liver cancer cells. Importantly, neutralizing antibody attenuates FSTL1-mediated malignancy. Mechanistically, FSTL1 binds to TLR4 receptor on HCC cells to activate oncogenic effects through a deregulated AKT/mTOR/4EBP1 signaling cascade. Preclinical immunocompetent mouse model demonstrated the treatment of HCC tumors with a specific FSTL1 neutralizing mAb (nAb) sensitized it to sorafenib, prolonged survival, and eradicated the liver tumor-initiating cell subset. Together, we unraveled FSTL1's potential as a novel prognostic biomarker and therapeutic target in HCC.

Hepatic stellate cell (HSC) line human telomerase reverse transcriptase (hTERT)-HSC was obtained as a gift from David Brenner of University of California San Diego, La Jolla, CA (23). CHO-K1 cell line and HCC cell lines Hep3B, HepG2 were purchased from ATCC. HCC cell line PLC8024 was obtained from the Institute of Virology of the Chinese Academy of Sciences. HCC cell line Huh7 was purchased from the JCRB Cell Bank. HCC cell line SNU878 was purchased from the Korean Cell Line Bank. HCC cell lines MHCC97L and MHCC97H were obtained from the Liver Cancer Institute, Fudan University. MIHA was provided by Dr. JR Chowdhury of Albert Einstein College of Medicine, Bronx, NY. Normal liver fibroblast line was purchased from iBiologics. Cell lines used in this study were regularly authenticated by morphologic observation and AuthentiFiler STR (Invitrogen) and tested for absence of Mycoplasma contamination. Experiments were performed within 30 passages after cell thawing.

For patient-derived organoid cultures, cells were isolated and cultured as previously described (24–26). HCC tissue used for organoid establishment of HCC patient-derived organoid HK-HCC P1 was obtained from a patient with HCC undergoing surgery at Queen Mary Hospital with written informed consent obtained and protocol approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster. Patient did not receive any previous local or systemic treatment prior to operation.

Frozen primary human HCC and adjacent nontumor liver tissue samples were obtained from patients with HCC undergoing hepatectomy at Queen Mary Hospital with written informed consent obtained from all patients and protocol approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster. Samples were collected from patients who had not received any previous local or systemic treatment prior to operation.

CAF score was computed as described in a previous study (27). Liver cancer data was downloaded from https://xena.ucsc.edu/. Liver CAF score signature was the mean of genes specific to liver CAF, as determined by previous study (28). The correlation between FSTL1 and CAF score were analyzed using GraphPad Prism 6.0.

Animal research ethics was approved by and performed in accordance with the Committee of the Use of Live Animals in Teaching and Research at The University of Hong Kong and the Animals (Control of Experiments) Ordinance of Hong Kong or the Animal Care and Use Committee at Nankai University.

Surgery was performed as previously described (29). Briefly, C57BL/6 mice were anesthetized with 80 mg/kg ketamine and 10 mg/kg xylazine (i.p.). The left lobe and the median lobe were ligated separately and resected to achieve two-thirds hepatectomy. Animals were sacrificed on days 1.5, 3, or 7 after surgery.

Mice were treated with diethylnitrosamine (DEN; i.p., 1 mg/kg) at the age of 14 days. Starting at 8 weeks of age, CCL₄ (i.p., 0.2 ml/kg) was administered twice weekly for 12 to 16 weeks (30). Tumors started to grow at the age of approximately 21 weeks, and the humane endpoint of this model was at the age of 24 weeks. Animals were sacrificed at weeks 8, 16, 18, 20, 22, 23, and 24.

Six to 8 weeks old male wild-type C57BL/6 mice were used and the procedure was performed as previously described (31, 32). In brief, 20 μg of plasmids encoding human AKT1 (myristylated AKT1) and human neuroblastoma Ras viral oncogene homolog (N-RasV12) along with sleeping beauty (SB) transposase in a ratio of 25:1 were diluted in 2 mL saline (0.9% NaCl), filtered through 0.22 μm filter and injected into the lateral tail vein of C57BL/6 mice in 5 to 7 seconds. The constructs used in this study showed long term expression of genes via hydrodynamic injection (32). Three weeks post hydrodynamic tail vein injection (HTVI) of proto-oncogenes and SB transposase, mice were separated into 4 groups and administered with (i) DMSO + IgG, or (ii) sorafenib + IgG, or (iii) DMSO + FSTL1 nAb, or (iv) sorafenib + FSTL1 nAb. Sorafenib was administered at 30 mg/kg/day. FSTL1 nAb was administered at a concentration of 2.5 mg/kg every 3 days.

Fstl1-CreERT2 knock in mice were generated in Model Animal Research Center of Nanjing University. Rosa-mT-mG (B6.129(Cg)-Gt(ROSA)26Sortm4 (ACTBtdTomato-EGFP)LuoJ mice were purchased from Model Animal Research Center of Nanjing University (stock number: J007676). Fstl1-CreERT2; Rosa26-mT-mG mouse line was obtained by crossing Fstl1-CreERT2 mice with lineage reporter Rosa26-mT-mG mice. The genotypes of the mice were determined by PCR as previously described (33). A tamoxifen stock solution of 10 mg/mL was produced by dissolving tamoxifen powder in corn oil. To examine recombination in the liver, mice received five doses of tamoxifen by intraperitoneal injection before or after CCL₄ treatment and were sacrificed at indicated time for sectioning and analysis.

Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X (Sigma-Aldrich), blocked with 5% BSA and incubated with FSTL1 (1:200, Abcam, ab71548), αSMA (1:400, Abcam, ab21027), FAP (1:100, Abcam, ab28244), CD31 (1:100, Invitrogen, MA3100), pan-cytokeratin (1:500, Abcam, ab7753), followed by Alexa-Fluor conjugated secondary antibody (1:500, Life Technologies). Immunofluorescence (IF) on Fstl1 lineage model was performed as previously reported (33). Antibodies used were GFP (1:1000, Abcam, ab13970) and αSMA (1:500, Santa Cruz, sc-32251). For colocalization of TLR4 and FSTL1, MHCC97L cells were first treated with 10 ng/mL of recombinant human FSTL1 (R&D Systems) and incubated for 8 hours. Cells were then fixed, permeabilized, and blocked as indicated above, and incubated with FSTL1 (1:100, Abcam, ab71548) and TLR4 (1:100, Novus Biologicals, NB100–56566), followed by Alexa-Fluor conjugated secondary antibody (1:500, Life Technologies). Cells were counterstained with anti-fade DAPI (Invitrogen), visualized by fluorescent confocal microscope (Carl Zeiss LSM 780) and analyzed using ZEN 2 (blue edition) software.

After dewaxing and rehydration of the paraffin-embedded sections, the antigen retrieval was conducted using Envision Flex Target Retrieval Solution of high pH (DAKO). For FSTL1 and αSMA staining, the sections were blocked with Opal Antibody Diluent/Block (AKOYA), followed by incubation with FSTL1 (1:100, Abcam, ab71548) for 1 hour at room temperature. Signal was detected by incubating Opal Polymer horseradish peroxidase (HRP), Mouse plus Rabbit (AKOYA) for 30 minutes, followed by Opal 570 Reagent (AKOYA). Another round of antigen retrieval and antibody stripping were conducted before incubating with αSMA (1:100, Abcam, ab5694) for 1 hour at room temperature. Signal was detected by incubating Opal Polymer HRP, Mouse plus Rabbit (AKOYA) for 30 minutes, followed by Opal 520 Reagent (AKOYA). DAPI (AKOYA) was used to counterstain the nuclei. pAKT and pan-CK staining was done as described above, with pAKT (Ser473) (1:200, Cell Signaling Technology, 4060) and pan-CK (1:200, Abcam, ab7753) stained with Opal 570 and Opal 520 Reagent (AKOYA), respectively. The sections were imaged using Vectra Polaris imaging system (Perkin Elmer). Analysis to segment and quantify the cells based on their immunostaining were performed using inForm versions 2.2 (Akoya Biosciences).

Total RNA was extracted using RNA-IsoPlus (Takara) and cDNA was synthesized by PrimeScript RT Master Mix (Takara). qPCR was performed with EvaGreen qPCR Master Mix (ABM) and primers listed in Supplementary Table S1 on a LightCycler 480 II analyzer (Roche) with data analyzed using the LightCycler 480 II software (Roche). Relative expression differences were calculated using the 2^–ΔΔC_t method.

CD14-specific shRNA expression vectors (NM_000591), DIP2A-specific shRNA expression vectors (NM_015151), and scrambled shRNA nontarget control (NTC) were purchased from Sigma-Aldrich. TLR4-specific shRNA expression vectors (NM_138554) were manufactured by Guangzhou IGE Biotechnology Ltd. Sequences were packaged using MISSION Lentiviral Packaging Mix (Sigma-Aldrich) and transfected into 293FT cells. Virus-containing supernatants were collected for subsequent transduction to establish cells with CD14, DIP2A, or TLR4 stably repressed. Puromycin was used for cell selection. Short hairpin RNA (shRNA) sequences used are listed in Supplementary Table S2.

siRNA sequences used are listed in Supplementary Table S3. Please see Supplemental File for methods details.

Atypical fibrotic response is widely associated with increased aggressiveness and poor prognosis in cancers. In HCC, around 90% of cases are associated with fibrosis or cirrhosis, a disease that is characterized by an enrichment of activated fibroblasts due to chronic inflammation. We initiated the study by looking at fibroblasts in a liver regenerating mouse model induced by 70% partial hepatectomy where hepatocyte proliferation and enhanced expression of activated fibroblasts are commonly observed. We found Fstl1 to be significantly upregulated during active liver regeneration on day 3 with its expression slightly decreasing towards the end of liver regeneration when the liver is mostly restored on day 7. Notably, expressions of Fstl1 tightly correlated with Acta2 mRNA expression, which encodes for αSMA, a major marker of myofibroblasts (Fig. 1A). Next, to understand the regulatory mechanism of fibroblasts in HCC, we examined Fstl1 and Acta2 expression in 2 HCC mouse models induced either by HTVI of oncogenic plasmids (NRAS+AKT) and SB transposase or administration of DEN+CCL₄. Consistently, we found Fstl1 expression to be greatly enhanced at advanced stages of tumor development (approximately 4–5 weeks for HTVI-induced HCC and approximately 22 weeks for DEN+CCL₄–induced HCC) with its expression closely correlated with Acta2 levels (Fig. 1B and C). Expression of Fstl1 and αSMA proteins is also colocalized in the tumor tissue sections of the DEN+CCL₄ mouse model, as demonstrated by IF (Supplementary Fig. S1). Lineage tracing of Fstl1 in a CCL₄-induced liver fibrosis model in Fstl1-CreERT2;Rosa26-mT-mG mice found de novo (model 1) but not resident (model 2) Fstl1 lineage cells to be labeled with GFP and that de novo Fstl1 cells overlap with αSMA fibroblasts, suggesting that Fstl1 expressing myofibroblasts is induced in CCL₄ mediated liver fibrosis (Fig. 1D).

To understand the role of CAFs in hepatocarcinogenesis, we obtained freshly resected nontumor liver and HCC clinical samples from patients who underwent surgery for HCC removal and isolated nontumor fibroblasts (NF) and CAF pairs (n = 3). All isolated fibroblasts displayed a homogenous spindle-shaped fibroblastic morphology that stained positive for both αSMA and fibroblast activation protein (FAP; Supplementary Fig. S2). To ensure the established fibroblasts were free of contaminating HCC, endothelial, and epithelial cells, we also examined for expression of the respective markers, alpha-fetoprotein (AFP), CD31, and pan-cytokeratin, which all stained negative (Supplementary Fig. S2). We then examined the secretory expression levels of FSTL1 in normal liver fibroblasts, hTERT- HSC, NFs, CAFs, and a panel of HCC cell lines. All immortalized normal liver and HCC cells (MIHA, MHCC97L, MHCC97H, SNU878, HepG2, PLC8024, Huh7, and Hep3B) produced no or negligible levels of FSTL1, while normal fibroblasts, hTERT-HSC, NFs, and CAFs produced significantly higher amounts (Fig. 1E). Notably, normal fibroblasts secreted some levels of FSTL1, but all 3 CAF lines secreted significantly more than their paired NF and normal fibroblast, suggesting that fibroblast and in particular CAFs in the tumor stroma represent a major source of FSTL1. hTERT-HSC also secreted more FSTL1 compared with normal fibroblast (Fig. 1E). We also found FSTL1 to be significantly associated with an HCC CAF signature as defined by a 12-marker panel comprising expressions of FGF5, CXCL5, IGFL2, MMP1, ADAM32, ADAM18, IGFL1, FGF8, FGF17, FGF19, FGF4, and FGF23 (Fig. 1F; ref. 28). Consistent with the findings in our mouse models, we found the majority of FSTL1 expressing cells to overlap with those of αSMA in HCC (n = 4) clinical samples and that FSTL1 is secreted in more abundance in HCC than as compared with its adjacent nontumor liver (Fig. 1G and H). Analysis of our data collected in-house as well as in the The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) dataset found FSTL1 expression to positively correlate with ACTA2 and FAP, respectively, suggesting FSTL1 to be primarily derived from myofibroblasts/CAFs (Fig. 1I). To explore the clinical relevance of FSTL1 expression, we further examined the TCGA LIHC dataset and found high FSTL1 in the FAP⁺ CAF subset to be significantly correlated with advanced tumor stages in HCC tumor tissues as compared with low FSTL1 in FAP⁺ CAFs, suggesting the clinical implication of high FSTL1 in CAFs of HCC (Fig. 1J).

To investigate the causative relationship between enhanced FSTL1 secretion by CAFs in hepatocarcinogenesis, we overexpressed empty vector control or FSTL1 in CHO-K1 cells, chosen for its high yield and reduced nonspecific production of recombinant proteins, and collected empty vector conditioned medium (EV CM) or FSTL1-overexpressing conditioned medium (FSTL1 OE CM), respectively. For subsequent functional studies, MHCC97L cell line was chosen because it does not express any endogenous FSTL1 and for its ability to grow when injected orthotopically in the liver for the study of metastasis. Note that secretory FSTL1 level in CHO-K1 cells was expressed at physiologic level that is comparable with that of CAFs (Fig. 1E). FSTL1 OE CM enhanced the abilities of MHCC97L HCC cells to proliferate (Fig. 2A), migrate (Fig. 2B), and invade (Fig. 2C). Stem cell–like properties of cancer cells can contribute to cancer relapse due to their ability to self-renew and resist standard therapy. Thus, we also evaluated whether FSTL1 OE CM can alter these properties. Coculture of FSTL1 OE CM enhanced ability of HCC cells to resist molecular targeted therapy sorafenib as demonstrated by a decrease in number of dead cells measured by Annexin V propidium iodide (PI) flow cytometry analysis (Fig. 2D) and also to self renew with the latter demonstrated by an increase in the frequency of TICs in a limiting dilution spheroid formation assay (Fig. 2E). Consistently, administration of recombinant FSTL1 (rhFSTL1) at a level comparable with that secreted by CAFs (Fig. 1E) also resulted in similar protumorigenic functional effects (Fig. 2A–E, right). We also extended our study to a more physiologic setting, utilizing organotypic ex vivo culture of HCC tumor tissues where pathophysiology of the original tumor is better preserved than as compared with cell lines. Note the HCC patient-derived organoid used here is thoroughly characterized at both molecular and phenotypic levels, with comparison made against the original tissue samples. Consistently, similar results could be attained when HCC patient-derived organoids were treated with FSTL1 OE CM or rhFSTL1 (Supplementary Fig. S3A–S3E).

In line with our findings in vitro and ex vivo, treatment of MHCC97L cells with FSTL1 OE CM also promoted tumor growth potential in vivo in immunocompromised nude mice, as demonstrated by increased proliferative and incidence rates (Fig. 2F). In a separate experiment, we also implanted MHCC97L HCC cells at limiting dilutions subcutaneously into immunocompromised NOD-SCID mice and administered either EV CM or FSTL1 OE CM to the tumor injection site every 3 days. HCC cells treated with FSTL1 OE CM exhibited a significantly worse tumor-free survival compared with HCC cells treated with EV CM (Fig. 2G) as well as increased tumor incidence, expedited tumor latency, and close to a five-fold higher frequency of TIC (Supplementary Table S4). At end point, HCC tumor cells were harvested from the liver for ex vivo limiting dilution spheroid formation assay, where it was demonstrated that the frequency of TIC capable of forming spheres also increased when HCC cells were treated with FSTL1 OE CM (Fig. 2H). An orthotopic liver metastasis model was also carried out to examine ability of FSTL1 to alter ability of HCC cells to metastasize in vivo. And indeed, MHCC97L HCC cells treated with FSTL1 OE CM displayed a superior ability to metastasize to the lung than as compared with HCC cells treated with EV CM, as supported by stronger bioluminescence signal, greater liver to body weight ratio (Fig. 2I) as well as number of metastatic nodules detected in the lung (Fig. 2J). Collectively, these findings support a key role of secretory FSTL1 in promoting tumorigenicity, self-renewal and metastasis in HCC.

In light of the above findings, we reasoned that blocking FSTL1 signaling with a nAb may be therapeutic for HCC. Our coauthor W. Ning has previously generated a FSTL1 nAb (clone 22B6) with binding affinity, kinetics, and neutralization function fully characterized (19). Due to its high similarity (>95%) in gene expression with activated human HSC, hTERT-HSC was used as a model to study the paracrine effect of the stromal fraction on HCC cells (23). Treatment of MHCC97L HCC cells with CM collected from the hepatic stellate cell line hTERT-HSC (23) neutralized by FSTL1 nAb resulted in diminished ability of cells to proliferate (Fig. 3A), migrate (Fig. 3B), and invade (Fig. 3C). Treatment of FSTL1 OE CM with nAb also reversed sensitivity of HCC cells to sorafenib (Fig. 3D) and reverted TIC frequency of MHCC97L cells to levels comparable with the EV CM + IgG control (Fig. 3E). Similar effects could also be observed when treatment was done in HCC patient-derived 3D organoids (Supplementary Fig. S4A–S4D). Consistently, we were able to see comparable findings when we used the FSTL1 nAb against FSTL1 OE CM, where the neutralizing antibody reverted the functional phenotypes back to the level of EV CM (Fig. 3A–E, right).

Complementary in vivo studies were also performed. Proliferative growth of tumors formed with MHCC97L cells treated with FSTL1 OE CM was significantly attenuated in the presence of FSTL1 nAb, with tumor volume and weight returning to a similar level than of tumors formed with MHCC97L cells treated with EV CM (Fig. 4A). When the same cells were injected at limiting dilutions in immunodeficient NOD-SCID mice, FSTL1 nAb delivery clearly reduced tumor incidence, delayed time needed for successful engraftment, and reverted the TIC frequency to a similar level than that of the EV CM + IgG control group in primary implantation. When liver tumors from these primary implantations were harvested for serial transplantation into secondary mouse recipients, a further decrease in tumor incidence and TIC frequency could be observed in the nAb treated group (500 cells – 2/10; 1000 cells – 2/9; TIC frequency 1:2674) as compared with the control (500 cells – 6/10; 1000 cells – 4/9; TIC frequency 1:1000), suggesting FSTL1 nAb could decrease self-renewal ability and the functional cancer stemness subset of the HCC tumor (Supplementary Table S5). FSTL1 nAb could also prolong tumor-free survival in both primary and secondary implantations of the mouse model (Fig. 4B); and led to further inhibition of tumor growth in secondary transplantations (Fig. 4C). Ex vivo limiting dilution spheroid formation assay with liver tumor cells harvested from primary implantations also yielded similar results as those of in vivo secondary implantations (Fig. 4D) further confirming our observations. Similar inhibitory effects could also be extended in the orthotopic liver metastasis model, where treatment of MHCC97L cells with FSTL1 nAb attenuated abilities of cells not only led to reduced bioluminescence signal and liver to body-weight ratio (Fig. 4E), but also a marked decrease in number of metastatic nodules detected in the lung (Fig. 4F).

Previous studies have found FSTL1 to bind to TLR4, CD14, follistatin (FST), or disco-interaction protein 2 homology A (DIP2A; refs. 34–36). Using HCC data from the TCGA-LIHC dataset, we found FSTL1 expression to be more closely correlated with TLR4 (Fig. 5A; r = 0.46) than as compared with DIP2A (r = 0.22), and that CD14 and FST were not associated with FSTL1 (Supplementary Fig. S5). We then confirmed colocalization and binding of FSTL1 and TLR4 in MHCC97L cells by dual color IF and coimmunoprecipitation assays, respectively (Fig. 5B and C). To functionally characterize the importance of TLR4 in mediating the cancer properties directed by FSTL1, we treated MHCC97L cells with rhFSTL1, in the absence or presence of the specific TLR4 inhibitor, TAK242. Inhibition of TLR4 by TAK242 abrogated rhFSTL1-induced proliferation, migration, invasion, sorafenib resistance, and self-renewal capacities (Fig. 5D–H), suggesting FSTL1 to regulate cancer stemness properties in HCC via the TLR4 axis. Consistently, stable knockdown of TLR4 by lentiviral-based shRNA also resulted in similar findings, both in vitro and in vivo mouse models (Supplementary Figs. S6A–S6E and S7A–S7H). We also investigated on other known receptors of FSTL1 including CD14 and DIP2A, but stable knockdown of CD14 nor DIP2A did not result in any reduced FSTL1-mediated oncogenic effects, suggesting that FSTL1 seems to work primarily through TLR4 in HCC (Supplementary Fig. S8A–S8J).

To better understand how CAFs-derived FSTL1 contributes to maintain aggressive cancer properties in HCC, the transcriptome profiles of CAFs, as marked by FAP expression, was analyzed in the HCC tumors of the TCGA LIHC dataset. In the FAP+ CAF subpopulation, genes coexpressed with FSTL1 were subjected for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis in DAVID. By this means, we found FSTL1 to be highly associated with PI3K/AKT signaling in patients with HCC enriched with CAFs (Fig. 6A). Consistently, critical molecular players in the AKT/mTOR/4EBP1 pathway were further confirmed by Western blot where pAKT, p-mTOR, p-4EBP1, and c-myc were all found enhanced when MHCC97L HCC cells and HCC patient-derived organoids were treated with either FSTL1 OE CM as compared with EV CM or rhFSTL1 as compared with PBS. Conversely, administration of FSTL1 nAb to the HCC cells and organoids treated with hTERT-HSC CM or FSTL1 OE CM would attenuate the activated signaling (Fig. 6B). To further validate the significance of AKT driven mTOR/c-myc activation in FSTL1-induced HCC, we analyzed the impact of introducing MK-2206, an inhibitor of AKT, into HCC cells and HCC patient-derived organoids cultured in the presence of rhFSTL1. Successful suppression of AKT/mTOR/4EBP1/c-myc pathway following addition of MK-2206 was confirmed by Western blot (Fig. 6B). Functionally, MK-2206 also suppressed the ability of FSTL1-driven HCC cells to proliferate (Fig. 6C; Supplementary Fig. S9A), migrate (Fig. 6D; Supplementary Fig. S9B), invade (Fig. 6E; Supplementary Fig. S9B), resist sorafenib (Fig. 6F; Supplementary Fig. S9C), and self-renew (Fig. 6G; Supplementary Fig. S9D). In line with these findings, inhibition of the more upstream TLR4 by TAK242 also attenuated activation of AKT signaling (Fig. 6I).

Since we identified FSTL1′s ability to confer resistance to sorafenib, we then explored the in vivo benefit of targeting FSTL1 in combination with sorafenib in HCC using the NRAS+AKT HTVI immunocompetent HCC mouse model where we found Fstl1 to be overexpressed (Fig. 1B). FSTL1 nAb and/or sorafenib were administered following HCC formation at 3 weeks and monitored for survival (Fig. 7A). While sorafenib treatment alone did lead to a decrease in liver to body-weight ratio (Fig. 7B), it did not lead to any survival benefit (Fig. 7C). But importantly, FSTL1 nAb and sorafenib combination treatment led to a clear decrease not only in liver to body-weight ratio, but also enhanced survival as compared with mice treated with control DMSO+IgG or just sorafenib or FSTL1 nAb alone (Fig. 7B and C). Subsequent ex vivo limiting dilution spheroid formation assays using cells isolated from the liver of this HCC mouse model demonstrated FSTL1 nAb and sorafenib combination to result in the most significant decrease in self-renewal ability of the cells (Fig. 7D), suggesting FSTL1 nAb to sensitize HCC cells to sorafenib and preferentially eradicated the tumor-initiating subset of the tumor. IHC analysis for pAKT (Ser473) and proliferation marker PCNA expressions in the resected livers of the 4 treatment groups showed a marked decrease in both pAKT and PCNA in the FSTL1 nAb and combination treatment groups, suggesting FSTL1 nAb did effectively impair AKT signaling and proliferative capacity of the tumor (Fig. 7E). Dual color IF for the epithelial marker pan-cytokeratin and pAKT (Ser473) confirmed abrogated AKT activation specifically in HCC cells in the 2 groups treated with FSTL1 nAb (Fig. 7E). Consistently, we observed a similar activation of pAKT signaling in tumors driven by FSTL1 overexpressing condition medium; while FSTL1 nAb of these tumors would lead to mitigation of this signaling (Fig. 7F). Collectively, this data demonstrates the potential of utilizing FSTL1 neutralizing antibody to enhance the efficiency and efficacy of sorafenib.

To understand how fibroblasts maintain FSTL1 expression, we focused on TGFβ1 as it is reported that FSTL1 expression is governed by TGFβ1 signaling (37). First, we examined the secretory level of TGFβ1 in a panel of immortalized normal liver cells, HCC cells, as well as fibroblasts (Supplementary Fig. S10A). Of note, the level between HCC cells and fibroblasts are comparable. Furthermore, the secretory level of TGFβ1 and FSTL1 in fibroblasts is highly correlated (Supplementary Fig. S10B), suggesting that FSTL1 level may be maintained by autocrine signaling of TGFβ1. Using hTERT-HSC and human CAFs as models, we showed that TGFβ1 knockdown in fibroblast resulted in a significant reduction of FSTL1 expression at the mRNA and protein level (Supplementary Fig. S10C–S10H). Consistently, TGFβ1 nAb diminished the production of secretory FSTL1 (Supplementary Fig. S10I–S10J). Computational analysis by JASPAR identified 3 binding sites of the major downstream effector of TGFβ, SMAD2/3/4, on the promoter of FSTL1 (Supplementary Fig. S10K). Further analysis by LASAGNA showed SMAD4 to be a top hit as a transcription factor regulating FSTL1 expression. Further experiments whereby SMAD4 was knocked down by siRNA approach found FSTL1 expression to be decreased (Supplementary Fig. S10L), which further confirms that fibroblast-derived TGFβ1/SMAD4 signaling plays a role in the control of FSTL1 through an autocrine manner.

In lights of the secretory TGFβ1 level in HCC cells, we also reasoned that HCC cells can mediate cross-talk with fibroblast via TGFβ1 paracrine signaling. To this end, we cultured fibroblasts with conditioned media collected from HCC cells with TGFβ1 suppressed (Supplementary Fig. S11A). Compared with the siNTC control, conditioned media of TGFβ1 knockdown HCC cells showed diminished ability to promote FSTL1 production in the fibroblasts (Supplementary Fig. S11B and S11C). In accordance, HCC cells conditioned media treated with TGFβ1 nAb demonstrated reduced ability to stimulate FSTL1 (Supplementary Fig. S11D and S11E). Thus, our results depicted HCC to fibroblast cross-talk via TGFβ1 signaling can enhance production of FSTL1 from fibroblasts, in turn, potentiates malignant properties of HCC cells.

The limited options and modest efficacy of drugs for advanced HCC urge for the development of novel treatment. Given most HCC develops in a fibrosis or cirrhosis liver, how a highly inflammatory TME contributes hepatocarcinogenesis warrants further investigation. Here, we demonstrated Fstl1 expression is upregulated in cirrhosis, regenerative and disease state, as evident in liver regeneration model, DEN+CCL₄ fibrotic HCC model, and HTVI HCC model. RNA sequencing (RNA-seq) of DEN+CCL₄ fibrotic model depicted several CAF-secreted genes to be elevated in tumor compared with vehicle control. Of interest, among the other reported known activated fibroblast-derived factors like Ctgf, Igf2, Tnf, Ccl5, Pdgfc (38, 39), we find levels of Fstl1 to be expressed at high levels in HCC versus vehicle control. A high correlation between Acta2 and Fstl1 gene expression was observed in all mouse models. Additionally, FSTL1 secretory level is substantially greater in HCC patient-derived fibroblast lines compared with HCC cell lines while IF depicted colocalization ACTA2 and FSTL1. Together, we reasoned that FSTL1 is preferentially secreted from the fibroblasts. Using tamoxifen-inducible FSTL1 reporter Fstl1-CreERT2;Rosa26-mT-mG (33), we showed FSTL1 lineage cells to give rise to αSMA⁺ fibroblasts in CCL₄-induced liver fibrosis, indicating FSTL1 as an important growth factor involved in promoting growth upon injury in the liver. Another study found FSTL1 to be elevated in HCC tissue compared with the adjacent nontumor liver tissue and high FSTL1 was associated with poor prognosis (40). As such, we demonstrated that FSTL1 protein level was higher in CAF CM compared with NF CM, alluding to the possibility of cancer cells being modulated by FSTL1 expression to accelerate their growth (41). FSTL1 may be used as a potential prognostic marker as more advanced tumor–node–metastasis (TNM) staging in patients with HCC was associated with high FSTL1 in FAP⁺ fibroblasts. It is worthy to note that FSTL1 has been shown in one study to be elevated in HCC cancer cells (40). This is contrary to our study where we found FSTL1 to be predominantly expressed from activated fibroblasts and that FSTL1 is only secreted at very low levels, if at all, in HCC cancer cells. We suspect that authors in the study led by Yang (40) did not explore into other cell types other than HCC cells; so while it is possible that HCC cells may secrete some expression of FSTL1, majority of the source is not from there.

The cross-talk between CAF and TICs is pivotal for cancer progression (42). In this study, we revealed FSTL1, which is elevated during fibrosis, promotes stemness of liver cancer cells. FSTL1 OE CM and rhFSTL1 both promote proliferation, stem cell frequency, migration and invasion, and therapy resistance in MHCC97L HCC cell line and HCC patient-derived organoid. Complementary in vivo studies further confirmed the tumorigenic and metastatic functions of FSTL1 in HCC. The malignant properties driven by FSTL1 in HCC are consistent with other studies done on esophageal, liver, colon, and breast cancers (20–22, 40, 43). However, FSTL1 was reported to function as tumor suppressor in renal cell carcinoma and lung cancer (44, 45), illustrating FSTL1 facilitates as an oncogene or tumor-suppressor gene in a tissue-dependent manner. Here, we demonstrated FSTL1 mediates its oncogenic influence via TLR4, a binding receptor that has been shown to be involved in HCC progression (46, 47).

A systematic analysis of the transcriptomes of patients with HCC enriched with CAF found the PI3K/AKT signaling pathway to be highly associated with FSTL1. AKT/mTORC1 regulates protein synthesis predominantly through two independent effectors, 4EBP and S6K1. Upon activation, mTORC1 phosphorylates 4EBP, triggering the release of eukaryotic translation initiation factor 4E (eIF4E), which is essential for translation initiation. Studies have also demonstrated that mTORC1 signaling is essential for c-myc–driven HCC progression (48), with 4EBP1/eIF4E regulating c-myc translational levels (49). There is also ample literature to show the crucial role of c-myc in hepatocarcinogenesis (50–52). In regard to the upstream regulator of mTOR, AKT is known to control mTOR signaling (53), while FSLT1 has also been shown to function through AKT (40, 54). Accordingly, we explored if FSTL1 regulates mTOR/c-myc pathway through AKT activation in HCC. We showed that inhibiting AKT using MK-2206 resulted in the ablation of FSTL1-mediated malignancy and reduced activation of mTOR signaling as analyzed using Western blot, hence providing robust evidence that FSTL1 functions via the AKT/mTORC1/4EBP1/c-myc axis in HCC cells. Our study postulates an interesting link between elevated FSTL1 in fibrosis state and FSTL1-driven c-myc, alluding to a possible link of how fibrosis may fuel hepatocyte transformation by elevating AKT/mTOR/c-myc signaling.

Ever since the discovery of TICs, numerous strategies have been explored to target TICs (55). However, TME, which functions as the TIC niche can restore the eradicated TIC population. Moreover, TME-derived autocrine and paracrine factors are essential for the maintenance of TIC phenotype. Therefore, targeting oncogenic growth factors in the TME presents an intriguing option to circumvent TIC phenotype. We explore the potential of attenuating CAF-derived FSTL1-driven malignancy by using FSTL1 neutralizing antibody (19). Similarly observed in our previous study, which demonstrated targeting FSTL1 with neutralizing antibody in esophageal cancer can circumvent the oncogenic effect (21), we demonstrated treating FSTL1 OE CM with neutralizing antibody can hinder the oncogenic properties and restore the HCC cells to the level comparable with EV CM. Further in vivo functional assay investigating growth, stemness, and metastasis recapitulated the in vitro results. To evaluate the translational value of inhibiting FSTL1, we tested the efficacy of FSTL1 nAb in combination with sorafenib in an immunocompetent preclinical model. Strikingly, our results indicated administration of sorafenib with FSTL1 nAb markedly extended the survival of the mice, possibly by suppressing the self-renewal and proliferative ability conferred by FSTL1. In sum, our findings showed targeting FSTL1 presents clinical relevance in treating HCC. In conjunction with other studies like for instance targeting CAF-secreted Netrin-1 in lung and colon cancers (27), we provide evidence that targeting CAF-secreted factors may hold therapeutic value.

In conclusion, FSTL1, a growth factor secreted by activated fibroblasts, augments malignancy and holds potential clinical value as a prognostic marker and therapeutic target, particularly in patients with high FAP and FSTL1 expression. Our findings illuminate FSTL1 confers HCC progression by enhancing TIC features of the HCC via TLR4-mediated AKT/mTORC/4EBP1/c-myc signaling axis.

No disclosures were reported.

J.J. Loh: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. T.W. Li: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology. L. Zhou: Formal analysis, investigation, methodology. T.-L. Wong: Formal analysis. X. Liu: Formal analysis, investigation, visualization, methodology. V.W.S. Ma: Formal analysis, investigation, methodology. C.M. Lo: Resources. K. Man: Resources. T.K. Lee: Resources. W. Ning: Resources, formal analysis, investigation, visualization, methodology. M. Tong: Resources. S. Ma: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

This project is supported in part by grants from the Research Grants Council of Hong Kong – Theme Based Research Scheme (T12–704/16-R) and Collaborative Research Fund (C7026–18G), Health and Medical Research Fund from Food and Health Bureau of the Hong Kong Government (06172546), as well as the “Laboratory for Synthetic Chemistry and Chemical Biology” under the Health@InnoHK Program launched by Innovation and Technology Commission, The Government of Hong Kong Special Administration Region of the People's Republic of China. The authors thank the Centre for PanorOmic Sciences (The University of Hong Kong) for providing and maintaining the equipment and technical support needed for flow cytometry, animal imaging, and confocal microscopy studies. They thank the Centre for Comparative Medicine Research (The University of Hong Kong) for supporting our animal work studies.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.