Stiffness is associated with hepatic stellate cell heterogeneity during liver fibrosis

Abstract

The fibrogenic wound-healing response in liver increases stiffness. Stiffness mechanotransduction, in turn, amplifies fibrogenesis. Here, we aimed to understand the distribution of stiffness in fibrotic liver, how it impacts hepatic stellate cell (HSC) heterogeneity, and identify mechanisms by which stiffness amplifies fibrogenic responses. Magnetic resonance elastography and atomic force microscopy demonstrated a heterogeneous distribution of liver stiffness at macroscopic and microscopic levels, respectively, in a carbon tetrachloride (CCl₄) mouse model of liver fibrosis as compared with controls. High stiffness was mainly attributed to extracellular matrix dense areas. To identify a stiffness-sensitive HSC subpopulation, we performed single-cell RNA sequencing (scRNA-seq) on primary HSCs derived from healthy versus CCl₄-treated mice. A subcluster of HSCs was matrix-associated with the most upregulated pathway in this subpopulation being focal adhesion signaling, including a specific protein termed four and a half LIM domains protein 2 (FHL2). In vitro, FHL2 expression was increased in primary human HSCs cultured on stiff matrix as compared with HSCs on soft matrix. Moreover, FHL2 knockdown inhibited fibronectin and collagen 1 expression, whereas its overexpression promoted matrix production. In summary, we demonstrate stiffness heterogeneity at the whole organ, lobular, and cellular level, which drives an amplification loop of fibrogenesis through specific focal adhesion molecular pathways.

NEW & NOTEWORTHY The fibrogenic wound-healing response in liver increases stiffness. Here, macro and microheterogeneity of liver stiffness correlate with HSC heterogeneity in a hepatic fibrosis mouse model. Fibrogenic HSCs localized in stiff collagen-high areas upregulate the expression of focal adhesion molecule FHL2, which, in turn, promotes extracellular matrix protein expression. These results demonstrate that stiffness heterogeneity at the whole organ, lobular, and cellular level drives an amplification loop of fibrogenesis through specific focal adhesion molecular pathways.

INTRODUCTION

Liver disease accounts for ∼2 million deaths per year worldwide, of which 1 million are due to complications of fibrosis and cirrhosis (1). Hepatic stellate cells (HSCs) are key regulators of liver fibrosis and subsequent cirrhosis. In response to liver injury, HSCs transdifferentiate into myofibroblasts characterized by increased proliferation, migration, and extracellular matrix (ECM) deposition (2, 3). Although the canonical stimuli responsible for myofibroblastic phenotypes, such as tumor growth factor β (TGF-β) and platelet-derived growth factor (PDGF) are well characterized (4, 5), effective therapies are lacking. Until lately, HSCs were considered as a homogenous cell population restricting the discovery of targeted therapies. Some recent studies have described HSCs as a heterogeneous cell population (6, 7). However, pathological pathways and targetable molecular candidates specific to HSC subpopulations during fibrogenesis are yet to be identified.

Hepatic fibrosis and cirrhosis are characterized by an increase in liver stiffness, which can be measured in vivo by magnetic resonance elastography (MRE) (8). This technique is beginning to see widespread clinical use in assessing hepatic fibrosis as a safer, more comfortable, and less expensive alternative to biopsy; it can also reduce biopsy-related complications and sampling errors (9, 10). Another technological method that gives information about the stiffness of a tissue ex vivo is atomic force microscopy (AFM) microindentation. MRE measures the whole organ stiffness, including integrated mechanical response to solid stress and fluid pressure, at a macroscopic level. AFM measures local stiffness, mainly from solid stress, at the micron level (11, 12). It is recognized that increased stiffness amplifies fibrogenesis through mechanotransduction (13–16). Nevertheless, liver stiffness distribution at macroscopic and microscopic levels and its impact on HSC heterogeneity in fibrosis have not been addressed.

In this study, we aim to identify stiffness-specific pathways involved in the amplification of fibrosis. Stiffness heterogeneity was associated with HSC heterogeneity, where HSCs localized in stiff areas were characterized by increased expression of focal adhesion molecules. Disruption of our focal adhesion prototype, four and a half LIM domains protein 2 (FHL2), inhibited ECM production, which is the main feature of liver fibrosis. Together, by identifying stiffness heterogeneity at the whole organ, lobular, and cellular level, we demonstrate that matrix deposition leads to stiff areas in the liver, which enhances the activation of a specific subpopulation of HSCs through focal adhesion pathway to exacerbate fibrosis.

METHODS

In Vivo Experiments

In vivo protocols were approved by the Mayo Clinic Institutional Animal Care and Use Committee. In all in vivo experiments, mice received humane care and were sex- and age-matched. Six-week-old WT C57Bl/6 male and female mice were purchased from Envigo. CCl₄ (1 μL/g of body wt, Sigma Aldrich No. 319961) or olive oil were administered via intraperitoneal injection twice a week for 0–6 wk. Bile duct ligation was performed as a second model of liver fibrosis as previously described (5, 17, 18). Briefly, bile duct was ligated using sterile 3/0 silk ligatures. After 3 wk, mice were euthanized, and the livers harvested for analysis. Hepatic MRE was performed at 0, 3, and 6 wk in all animals in vivo. Livers were collected 2 days after the final injection and analyzed by Sirius Red, Western blotting (WB), immunofluorescence, and AFM. Some of the livers were utilized for scRNA-seq.

Magnetic Resonance Elastography

All mice were fasted for 4 h before imaging, anesthetized, and maintained with 1.5% inhaled isoflurane during the scan. A 3.0 T whole body MRI scanner (HDx, GE Healthcare, Milwaukee, WI) and a customized 8-channel, 4-cm inner diameter imaging coil were used in this study. All mice were scanned in a supine position, with a silver needle (diameter: 0.26 mm, length: 39 mm, Asahi Medical Instrument Co., Kawaguchi, Saitama, Japan) inserted in the liver tissue from the anterior body wall. The waves were generated by an active driver (Resoundant Co., Rochester, MN) located outside of the scanner room, and transmitted through a long plastic tube, to the passive pneumatic driver, which was connected to the other end of the inserted needle. A free-breathing 4-shot spin-echo-based echo-planar MRE was used to obtain MRE wave images. Acquisition matrix is 96 × 96, TRs 400 ms, TEs 37.5–43.5 ms, section number 8, section thickness 2 mm, and FOV 8 cm × 8 cm. We obtained cylindrically symmetric shear waves throughout the liver at 200 Hz. The means, medians, standard deviations, and quantiles of shear stiffness were calculated at the same time from manually drawn regions of interest (ROIs), which encompass as much liver with significant wave propagation as possible and exclude the vibrating source and adjacent area.

Atomic Force Microscopy

Liver tissues were collected after 0, 3, or 6 wk of olive oil or CCl₄ treatment, embedded in Tissue-Plus O.C.T. Compound (Fisher HealthCare), and frozen in 2-methylbutane cooled in dry ice. Liver sections of 10 µm on poly-l-lysine-coated glass slides (Electron Microscopy Sciences) were obtained using a Leica CM1860 UV cryostat. For measurements, we utilized an AFM (Bioscope Catalyst, Bruker) coupled to an optical phase contrast Olympus microscope. AFM measurements were performed on hydrated tissues in 1× PBS after the experimental protocol previously developed by us and others using a Catalyst Bioscope atomic force microscope (Bruker) and the MIRO 2.0 extension through NanoScope 9.1 software (Bruker) (19, 20). Borosilicate sphere AFM tips with a 2.5-µm radius (NovaScan) were used and had a spring constant k estimated at ∼100 pN/nm by thermal tune method. The force applied on tissue sample was up to 14 nN, an indentation rate at 20 μm/s, and a ramp size of 10 μm. Finally, the elastic modulus (Young’s modulus) was estimated by fitting force curves (NanoScope Analysis 2.0 software, Bruker) using the well-known Hertz contact model:

$$\text{E}=3/4·\left[\right(1-v^2\text{)/(}{R}^{(1/2)}·{\delta }^{(3/2)}\left)\right]·\text{F}$$

with R being the tip radius, ν the Poisson’s ratio assumed at 0.4 for liver tissue (21, 22), and δ the sample indentation. The force curve fitting with Hertz model has been applied only on the first micron of indentation from the contact point (19, 23).

For the livers collected after 3 and 6 wk of CCl₄ treatment, two different types of areas of interest were considered: enriched in extracellular matrix referred to as ECM^S-High and low extracellular matrix areas referred to as ECM^S-Low. For control samples, only one type of area of interest was considered. For each mouse, three different locations from two nonconsecutive liver sections were analyzed per type of area of interest. For each location, 25 different force curves were obtained and analyzed to get the elastic values. Finally, the 25 elastic values were averaged to report one elastic value per location. To study the heterogeneity of liver stiffness, the distribution of elastic values was analyzed. To study the spatial elastic distribution, 169 force curves were performed in a grid pattern to complete the elastic heatmap of 13 × 13 pixels for spatial dimensions of 130 × 130 µm².

Single-Cell RNA-Sequencing

Primary HSCs were isolated utilizing the Accudenz gradient protocol (24) and the nonparenchymal cell isolation protocol (25) from vehicle or CCl₄-injected mice. HSC subpopulations were identified by single-cell RNA sequencing (scRNA-seq). 10× Genomics Cell ranger Single Cell Software Suite (version 3.0.2) was used to generate FASTQ, alignment to mm10 reference genome, filtering, and to quantify barcodes and unique molecular identifiers (UMI). This was followed by determining cell clusters using Seurat package (version 3.2.2) using a shared nearest neighbor (SNN) optimization-based clustering algorithm at resolution 0.1 (26). Genes that are expressed in fewer than three cells and cells that expressed fewer than 200 genes were excluded for downstream analysis in each sample. Custom mitochondrial concentration to UMI counts outlier threshold was determined for CCl₄ and healthy models to filter low-quality cells. In addition, any cell with >40% reads aligning to mitochondrial genes and expressing >300 k genes were excluded. Each data set was log-normalized and scaled for each gene across all cells. Enriched gene markers in each cluster conserved across all samples were identified with log2-fold change larger than 1 and adjusted P value smaller than 0.05. Differentially expressed genes within each cluster between the two conditions were also detected with log2-fold change larger than 1 in either direction and adjusted P value smaller than 0.05. HSCs were extracted from CCl₄ and healthy models and further reclustered independently for the two conditions at resolution 0.1 to output two subclusters. Healthy and CCl₄ HSC^col-high cells were further reclustered at resolution 0.25 to generate three subclusters. All clustering and statistical analyses were performed in R (version 3.6.2). The accession number for these data is https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE175939.

In Vitro Experiments

Primary human hepatic stellate cells (HSCs, ScienCell Research Laboratories No. 5300), liver endothelial cells (LECs, ScienCell Research Laboratories No. 5000), or HepG2 hepatocyte cell line were grown in complete media [DMEM (Life Technologies No. 11995065), 10% FBS, and 1% penicillin-streptomycin] at 37°C, and 5% CO₂. Cells were treated with 10 ng/mL recombinant PDGF (Sigma Aldrich No. P3201) or 5 ng/mL TGF-β (R&D Systems No. 240-B) in basal media (DMEM with 1% penicillin-streptomycin). For stiffness experiments, HSCs were cultured on silicon gels of 0.5 kpa, 32 kpa (Advanced BioMatrix CytoSoft No. 5165 and No. 5144), or on plastic. Cells were collected for further analysis.

FHL2 Knockdown and Overexpression

For FHL2 knockdown, human FHL2 shRNA was purchased from Sigma Millipore. Lentivirus generation in one 100-mm dish was done by cotransfecting 293 T cells with 2 µg shRNA, 1.5 µg p8.91, 0.5 µg pMD.G/pVSV.G plasmids, 300 µL buffer EC, 32 µL enhancer, and 40 µL Effectene (QIAGEN No. 301425). 293 T cells were then cultured in complete media and their conditioned media was collected after 24 h. For infection with FHL2 shRNA lentivirus, primary human HSCs were incubated for 24 h with the conditioned media containing the virus followed by 48 h of complete media. Infected cells were selected by puromycin treatment for 48 h. Cells were washed with 1× PBS and used for further experiments.

For FHL2 overexpression, FHL2 plasmid was purchased from OriGene (No. RC200398). Primary human HSCs were washed and incubated with 1 ng plasmid in 100 μL OptiMEM media (Life Technologies No. 31985070) supplemented with 2 µL Turbofect (Thermo Scientific No. R0531) for 48 h. Cells were washed with 1X PBS and used for further experiments.

Gene Expression

The mRNA from isolated cells, cultured cells, or liver tissues was extracted using RNeasy Mini kit (QIAGEN No. 74104). Equal amounts of total RNA (500 ng) were reverse-transcribed into cDNA using SuperScript III kit (Invitrogen No. 18080-051). Then, qPCR was performed using SYBR Green supermix (Bio-Rad No. 1725120). The manufacturers’ protocols were followed in all the experiments. The primers are listed as follows: mouse FHL2 forward: TGACCTGCTTCTGTGACTTG; mouse FHL2 reverse: AGAGATGTACTTTGTGCCACC; human FHL2 forward: ACTTTGCCTACTGCCTGAAC; human FHL2 reverse: AGTCGTTATGCCACTGCC; human collagen I forward: CCCCTGGAAAGAATGGAGATG; human collagen I reverse: TCCAAACCACTGAAACCTCTG; human fibronectin forward: ACTGTACATGCTTCGGTCAG; human fibronectin reverse: AGTCTCTGAATCCTGGCATTG; human α smooth muscle actin (αSMA) forward: AATGCAGAAGGAGATCACGG; and human αSMA reverse: TCCTGTTTGCTGATCCACATC.

Western Blot

EVs, liver tissues, or whole cell pellets were lysed using RIPA lysis buffer (Cell signaling No. 9806S) with protease inhibitor cocktail (Roche No. 4693159001). Equal amounts of protein for cell lysates (25 µg) or liver lysates (70 µg) were loaded onto SDS-PAGE gel, resolved, and transferred to nitrocellulose membrane. The membrane was blocked in 5% of BSA or 5% of milk and then incubated overnight with primary antibodies. Primary antibodies to detect αSMA (Abcam No. ab5694), collagen I (Southern Biotech No. 1310-01), FHL2 (Abcam No. ab202586), Fibronectin (BD BioSciences No. 610077), PDGFRα (Santa Cruz No. sc-398206), and HSC70 (Santa Cruz No. sc7298) were utilized. The blots were developed and detected using chemiluminescence (Millipore No. WBLUR0100 or Santa Cruz No. sc-2048) and autoradiography. The blots were quantified using ImageJ and HSC70 was used as internal control.

Imaging

A Zeiss Definite Focus .2 microscope with Axiocam 702 mono camera was used.

Statistics

GraphPad Prism 9 was used for statistical analysis. Normal distribution was examined utilizing Shapiro–Wilk test. One-way ANOVA and Kruskal–Wallis with multiple comparison tests, Mann–Whitney, and t tests were used to analyze the data. The difference was considered significant for P value lower than 0.05. Results are presented as means ± SE. Samples have been randomly assigned to processing orders. The investigators have been blinded to the sample group allocation during the analysis of the experimental outcome.

RESULTS

MRE-Assessed Stiffness Is Heterogeneously Increased in a CCl₄ Mouse Model of Liver Fibrosis

Liver stiffness is an indicator of liver disease, and especially of liver fibrosis (8). Nevertheless, the spatial distribution of stiff areas and how this affects the development of fibrosis has not been addressed, yet. Therefore, we aimed to understand the spatial distribution of liver stiffness at the macroscopic level, referred to as macroheterogeneity, in a mouse model of liver fibrosis. Mice were administered either with olive oil or carbon tetrachloride (CCl₄) for 6 wk to induce fibrosis. The establishment of liver fibrosis after 6 wk of CCl₄ treatment was confirmed by increased Sirius Red staining (Supplemental Fig. S1; see https://doi.org/10.6084/m9.figshare.15093954.v1) as well as collagen I and αSMA protein levels Supplemental Fig. S2). Macroheterogeneity was examined utilizing MRE-assessed shear stiffness, which represents the resistance of a material to deformation in response to a force applied at a given frequency or rate. In vivo, tissue is biphasic and contains both hard extracellular matrix and softer interstitial fluid pressures. In our study, liver shear stiffness was measured following a high-frequency force (200 Hz) in live animals at 0, 3, and 6 wk of treatment for each mouse. Although shear stiffness in olive oil-treated mice did not change throughout the duration of the MRE study, in CCl₄-treated mice it significantly increased from 1 kPa at 0 wk to 1.3 kPa at 6 wk (Fig. 1A) and it correlated to matrix deposition (Supplemental Fig. S1). To understand the distribution of shear stiffness values in the liver, the range of shear stiffness was examined. The range of liver stiffness in olive oil-treated group was 0.6–1.4 kPa and did not change from 0 wk to 6 wk, as represented by the superimposed curves (Fig. 1B). However, in CCl₄-treated group, the range of shear stiffness values was higher as represented by a flatter curve, indirectly suggesting a heterogeneous distribution of stiff areas throughout the liver. In addition, the mean peak of shear stiffness shifted toward stiffer values at 6 wk compared with 0 wk, indicating an increase in liver stiffness (Fig. 1C). Stiffness heterogeneity was also observed when comparing olive oil to CCl₄ groups at 6 wk of treatment (Fig. 1D).

Figure 1.

MRE reveals spatial macroheterogeneity of stiffness in liver fibrosis. C57Bl/6 wild-type male and female mice were treated with either olive oil or CCl₄ for 6 wk. Livers were analyzed by MRE at 0, 3, or 6 wk. The color stiffness maps (A), distribution of shear stiffness values in the olive oil group from 0 to 6 wk (B), in the CCl₄ group from 0 to 6 wk (C), and the comparison between olive oil and CCl₄ groups at 6 wk (D) are shown. C57Bl/6 wild-type male and female mice underwent sham or bile duct ligation (BDL) surgeries. Mice were kept for 3 wk. Livers were analyzed by MRE at 3 wk. The color stiffness maps (E) and the comparison between sham and BDL groups at 3 wk (F) are shown. Two-way ANOVA test was used to obtain the P value. *P < 0.05, n = 4 or 5 animals/group. CCl₄, carbon tetrachloride; MRE, magnetic resonance elastography.

The shear stiffness and its heterogeneity were also examined in bile duct ligation (BDL)-mediated liver injury. The establishment of liver fibrosis was confirmed by Sirius Red staining (Fig. 1C). Sham or BDL surgeries were performed, and mice were kept for 3 wk to obtain liver fibrosis. Similar to the CCl₄ model, BDL surgery increased the shear stiffness in the liver from 1 kPa to 1.4 kPa (Fig. 1E). The range and skewness of stiffness were higher in the BDL group, 0.8–2.8 kPa, compared with the sham group, 0.6–1.4 kPa (Fig. 1F). Altogether, these results suggest that stiffness macroheterogeneity is increased in fibrotic livers compared with olive oil group.

AFM Elastic Modulus Is Increased in Matrix-Rich Areas of the Liver

Next, we performed AFM to understand at the microscale whether liver stiffness changes are broadly distributed across the tissue or focused in particular fibrotic remodeled regions of tissue. Stiffness distribution was next examined at a microscopic level, referred to as stiffness microheterogeneity, utilizing AFM microindentation-assessed elastic modulus. The elastic modulus represents the resistance to deformation under normal loading. A fibrotic tissue has a higher elastic modulus as compared with a nonfibrotic one. In our study, we examined the elastic modulus on liver sections of untreated mice (control) and mice treated for 3 or 6 wk either with olive oil or CCl₄. Microscopically, we utilized the phase-contrast feature of IX73, Olympus microscope coupled to the AFM to select regions of interest, distinguishing macroscopically normal from remodeled regions of the liver (Fig. 2A). Livers of untreated and olive oil-treated mice had a homogenous appearance under phase contrast, thus random areas were selected to measure the elastic modulus. However, livers from CCl₄-treated mice showed the establishment of extracellular matrix (ECM)-rich areas, referred to as ECM^S-High, as opposed to ECM^S-Low areas as seen under light microscopy (Fig. 2A). Therefore, to understand the distribution of stiffness in the CCl₄ group, ECM^S-High and ECM^S-Low areas were selected (Fig. 2A). In each selected area, the elastic modulus was measured at multiple locations, and regions of 100 μm were analyzed as shown in Fig. 2B, where each pixel represents one measurement. Our results on the mean elastic modulus demonstrate that livers of control (0 wk, untreated) and olive oil (3 and 6 wk) mice had similar stiffness (Fig. 2C). At 3 wk of CCl₄ administration, ECM^S-High areas in the liver were more than twice stiffer than the livers from olive oil-treated mice. Surprisingly, ECM^S-Low areas were also stiffer than the olive oil livers (Fig. 2C), indirectly suggesting that these areas might also contribute to the overall liver stiffness. At 6 wk of CCl₄ administration, ECM^S-High areas presented an elastic modulus more than three times higher than the livers from olive oil-treated mice. Moreover, ECM^S-Low areas displayed a slightly higher stiffness than the livers from olive oil-treated mice (Fig. 2C), but this difference was not significant. Thus, our AFM analysis demonstrates that the most profound changes in tissue stiffness are focused on fibrotic areas characterized by ECM deposition rather than cellular components. In line with these results and the MRE-based measurements, the distribution of stiffness values was similar between control and olive oil conditions, for all the time points (Fig. 2D). However, the administration of CCl₄ increased the range of stiffness values and shifted the peak toward stiffer values (Fig. 2E), suggesting an increasing stiffness microheterogeneity. Moreover, at 6 wk of treatment, livers from CCl₄-treated mice presented stiffer values with higher heterogeneity than the livers from olive oil-treated mice (Fig. 2F).

Figure 2.

AFM identifies spatial microheterogeneity of stiffness in highly fibrotic regions in the liver. C57Bl/6 wild-type male and female mice were treated with either olive oil or CCl₄ for 0, 3, or 6 wk. Liver sections were analyzed by AFM. A: two zones per slide were analyzed by selecting ECM^S-High and ECM^S-Low areas. B: representative color stiffness maps of the selected areas. C: elastic modulus measured by AFM. D: distribution of elastic modulus of livers from olive oil group at 0, 3, and 6 wk of treatment. E: distribution of elastic modulus of livers from CCl₄ group at 0, 3, and 6 wk of treatment. F: comparison of the distribution of elastic modulus of livers between olive oil and CCl₄ groups at 6 wk. C57Bl/6 wild-type male and female mice underwent bile duct ligation (BDL) surgeries. Mice were kept for 3 wk. Liver sections were analyzed by AFM. G: two zones per slide were analyzed by selecting ECM^S-High and ECM^S-Low areas. H: elastic modulus measured by AFM. I: comparison of the distribution of elastic modulus of livers between control and BDL groups at 3 wk. Kruskal–Wallis test was used to obtain the P value. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, n = 3–5 animals/group, AFM, atomic force microscopy; CCl₄, carbon tetrachloride; ECM, extracellular matrix; ns, nonsignificant.

The elastic modulus and its heterogeneity were also examined in a bile duct ligation (BDL)-mediated liver injury model. To understand the distribution of stiffness in this model, areas around the vessels where the fibrosis was developed (ECM^S-High) and areas far from the vessels (ECM^S-Low) were selected (Fig. 2G). Three weeks after BDL surgeries, ECM^S-High areas were more than five times stiffer than the livers from control mice (Fig. 2H). Surprisingly, ECM^S-Low areas were also stiffer than the control livers (Fig. 2H), indirectly suggesting that these areas might also contribute to the overall liver stiffness. The range and skewness of stiffness were higher in the BDL group, 0.3–1.2 kPa, compared with the control group, 1.2 to more than 2.5 kPa (Fig. 2I).

Taken together, these results demonstrate that in liver fibrosis, stiffness is mainly attributed to ECM^S-High areas and its distribution is more heterogeneous in fibrotic livers compared with their healthy controls.

Single-Cell Transcriptomics Identifies Heterogeneity among HSCs

As demonstrated above, ECM deposition is accompanied by an increased liver stiffness in CCl₄ injured livers. Stiff environments can promote HSC activation (13). To understand how stiffness spatial microheterogeneity promotes cell heterogeneity among HSCs by activating stiffness-specific pathways, we performed scRNA-seq from healthy versus 6-wk CCl₄-treated mice (2 mice per group). To maximize the purification of HSC subpopulations, HSCs were isolated utilizing two methods: the Accudenz gradient protocol (24) and the nonparenchymal cell isolation protocol (25) (Supplemental Fig. S2). Living isolated primary cells were then barcoded, droplet-based captured, and sequenced following the 10X Chromium protocol (Supplemental Fig. S2A). In total, 2009 cells from healthy and 1349 cells from CCl₄-treated mice were sequenced. Seurat clustering using a resolution of 0.1 separated the non-parenchymal cells into eight clusters (Fig. 3A, Supplemental Fig. S2B). HSC cluster was identified based on the expression of conserved Decorin (Dcn) gene and HSC markers such as platelet-derived growth factor receptor α (Pdgfra), Pdgfrb, desmin (Des), collagen 1 α1 (Col1a1), and Col1a2 (Fig. 3B, Supplemental Fig. S3). The HSC cluster included 170 HSCs from healthy mice and 360 HSCs from CCl₄-treated mice.

Figure 3.

Single-cell RNA sequencing identifies two main HSC subpopulations. Livers from healthy or CCl₄-treated C57Bl/6 wild-type mice were utilized. A: tSNE plot of nonparenchymal cells were based on conserved genes (top) and split by condition (bottom). B: feature plots of the expression of HSC markers Dcn, Lrat, Pdgfrb, and Des. C: heatmap of conserved genes in each cluster. Col1a1, Adamtsl2, and Alcam are between the genes upregulated in subcluster 0 as compared with subcluster 1. Gpm6a, Fabp1, and Bhmt are between the genes upregulated in subcluster 1 as compared with subcluster 0. D: tSNE plot of HSCs from healthy and CCl₄ conditions, clustered in HSC^col-high and HSC^col-low. Adamtsl2, a disintegrin and metalloproteinase with thrombospondin repeats-like 2; CCl₄, carbon tetrachloride; Fabp1, fatty acid binding protein 1; Gpm6a, glycoprotein M6A; HSC, hepatic stellate cell; tSNE, t-distributed stochastic neighbor embedding.

To understand HSC heterogeneity and their relation to stiffness, we next extracted all HSCs based on their markers, subclustered them, and examined their expression profile. Although less than 500 HSCs were obtained, at a resolution of 0.1, two subclusters were identified as defined by their respective conserved genes [Log2(fold change)>1 and adjusted P value < 0.05] in a nonsupervised manner (Supplemental Tables S1 and S2). Conserved genes are those abundantly expressed by one cluster compared with the other clusters, disregarding the experimental condition. Thus, the conserved genes are markers of a specific cluster. The first cluster was characterized by a high expression of Col1α1, a disintegrin, and metalloproteinase with thrombospondin motifs-like 2 (Adamtsl2) and activated leukocyte cell adhesion molecule (Alcam) (Fig. 3C). Actin α2 (Acta2), another canonical marker of HSCs, was upregulated in this first cluster by 1.44-fold. As previously demonstrated, Adamtsl2⁺ HSCs reside in areas rich in collagen (27), which present a higher stiffness than areas low in collagen (Fig. 2). Therefore, this first cluster will be referred to as HSC^col-high. HSC^col-high were also present in the healthy liver, where HSC^col-high areas are absent (Figs. 1 and 2). However, Col1α1 levels were lower compared with the fibrotic condition (Fig. 3C, Supplemental Tables S1 and S2). On the opposite, the second cluster of HSCs expressed low levels of Col1α1 and Adamtsl2 and high levels of fatty acid-binding protein 1 (Fabp1) and betaine-homocysteine S-methyltransferase (Bhmt) (Fig. 3C), two molecules that are usually upregulated in resting HSCs (28, 29). This second cluster will be referred to as HSC^col-low.

These data suggest that at our resolution, there are two subclusters of HSCs in both healthy and CCl₄ conditions, activated or prone-to-be-activated HSC^col-high, and HSC^col-low with a more quiescent profile (Fig. 3D).

Stiffness-Associated HSCs Upregulate the Expression of Mechanical Force-Related Molecules

To identify differences in HSC^col-high signatures that might reflect the stiff fibrotic liver versus the normal liver, we examined the differentially expressed genes in CCl₄ HSC subclusters compared with healthy HSC subclusters [Log2(fold change)>1 and adjusted P value < 0.05, excluding ribosomal proteins] (Fig. 4A, Supplemental Tables S3 and S4). Compared with healthy HSCs and HSC^col-low, there were 122 differentially expressed genes in HSC^col-high, where 60 genes were upregulated and 63 genes were downregulated (Supplemental Tables S3 and S4). From these 122 genes, 54% coded for secreted proteins and 46% coded for intracellular molecules (Fig. 4B). Given our focus on stiffness-dependent intracellular signaling and mechanosensing proteins, we performed an EnrichR analysis on the genes coding for intracellular proteins (https://maayanlab.cloud/Enrichr/). GO cellular components 2018 revealed that the most upregulated pathway for these intracellular protein-coding genes was focal adhesion (Fig. 4B), recognized to be triggered by mechanical forces (30, 31). The most upregulated genes of this focal adhesion pathway in HSC^col-high were four and a half LIM domains protein 2 (Fhl2), Alcam, Integrin α8 (Itga8), vimentin (Vim), heat shock factor binding protein 1 (Hspb1), and palladin (Palld) (Fig. 4C). Even though HSC^col-high subpopulation appeared also under healthy conditions (Fig. 3D), these focal adhesion genes were a signature of only HSC^col-high in the fibrotic liver (Fig. 4C).

Figure 4.

A HSC^col-high subpopulation upregulates the expression of mechanical force-related molecules. A: heatmap of differentially expressed genes in each cluster. Compared with healthy group, 122 genes were differentially regulated in subcluster 0 from CCl4 group. B: Enrichr analysis of the upregulated genes that code for intracellular proteins, with focal adhesion being the most upregulated pathway. C: violin plots for upregulated genes from focal adhesion pathway. Each dot represents one cell. D: tSNE plot of HSC^col-high reclustered into three subpopulations. E: feature plot of Fhl2 expression across the three subpopulations of HSC^col-high. C57Bl/6 wild-type mice were treated with either olive oil or CCl₄ for 6 wk; (4–6 animals/group). Whole liver lysates were analyzed by qPCR (left), WB for FHL2 expression (F), and immunofluorescence (G, Scale bar = 50 μm). In F, a Mann–Whitney test was used to obtain the P value. H: Fhl2 mRNA expression in primary mouse HSCs from olive oil or CCl₄-treated mice. A t test was used to obtain the P value; (n = 3), *P < 0.05, **P < 0.01. I: C57Bl/6 wild-type male and female mice underwent sham or bile duct ligation (BDL) surgeries. Mice were kept for 3 wk. Livers were analyzed by immunofluorescence (Scale bar = 50 μm, 3 animals/group). CCl₄, carbon tetrachloride; FHL2, four and a half LIM domains protein 2; HSC, hepatic stellate cell; tSNE, t-distributed stochastic neighbor embedding; WB, Western blot.

Next, we aimed to understand whether HSC^col-high within the stiff areas is heterogenous by differentially expressing focal adhesion molecules. For this, HSC^col-high were extracted and further subclustered utilizing a resolution of 0.25. Three clusters of HSC^col-high were obtained (Fig. 4D). Focal adhesion molecules including Fhl2, Alcam, and Hspb1 were conserved in cluster 1 as compared with clusters 0 and 2 (Supplemental Tables S5–S7). Col1α1 and Col1α2 also were conserved in cluster 1 as compared with clusters 0 and 2 (Supplemental Tables S5–S7). In addition, CCl₄ administration induced a higher upregulation of Fhl2 in cluster 1 than in clusters 0 and 2 (Fig. 4E, Supplemental Tables S8–S10). These data suggest that focal adhesion molecules were associated with ECM deposition and expressed by a subset of HSC^col-high localized in stiff areas.

As a following step, we examined how focal adhesion molecule expression correlates with fibrosis and cirrhosis. We took as example one of the most upregulated focal adhesion genes, Fhl2, coding for a protein localized at focal adhesions and stress fibers and participating in cell migration (32, 33). Fhl2 protein was more than four times upregulated in whole liver lysates from CCl₄-injured mice as compared with olive oil-treated group (Fig. 4F). Moreover, our scRNA-seq results show that Fhl2 was mainly expressed by HSCs as compared with other non-parenchymal cells (Supplemental Fig. S4A). In patients, FHL2 transcript was more than 5 times upregulated in cirrhotic livers as compared with healthy livers (GSE25097 published database) (Supplemental Fig. S4B), and is mostly expressed by mesenchymal cells (Supplemental Fig. S4C) (7). We confirmed these data by immunofluorescence where Fhl2 was upregulated in the CCl₄ condition compared with olive oil group. Moreover, Fhl2 was expressed only in ECM^S-High areas in livers from CCl₄-injured mice (Fig. 4G), which correspond to the stiffest area as demonstrated by AFM (Fig. 2). Finally, Fhl2 mRNA expression was enhanced in primary murine HSCs isolated from CCl₄-treated mice as compared with HSCs isolated from control olive oil-treated mice (Fig. 4H). In addition, Fhl2 protein expression was examined in a BDL-mediated model of liver fibrosis. Similar to the CCl₄ model, in BDL livers, Fhl2 was expressed in collagen-rich areas and was undetected in sham liver sections (Fig. 4I). Taken together, these results suggest that in fibrotic livers only a subset of HSC^col-high localized in stiff ECM^S-High areas significantly upregulates the expression of focal adhesion molecules, including FHL2, and this subset is associated with ECM deposition.

FHL2 Is Upregulated by Stiffness and Promotes HSC Activation

Given the significant upregulation of Fhl2 in HSC^col-high and liver fibrosis, we next examined whether its expression is directly due to a stiff environment and if this exacerbates ECM production. In humans, healthy liver stiffness is less than 6 kPa, while the stiffness of a cirrhotic liver can reach 75 kPa (34). Therefore, in our experiments with primary human HSCs, we chose 0.5 kPa as the healthy condition and 32 kPa as the fibrotic/cirrhotic condition, which correspond to healthy individuals and patients with advanced stage of liver fibrosis, respectively. HSCs cultured on 32 kPa spread more and presented more stress fibers than HSCs cultured on 0.5 kPa plates, as shown by bright-field images and phalloidin staining, respectively (Supplemental Fig. S5A), similar to human liver endothelial cells or HepG2 hepatocyte cell line (Supplemental Fig. S5B). These results confirmed a previous study where HSCs cultured on stiff gels spread more, expressed higher levels of the activation marker αSMA, and had a different transcriptome than HSCs cultured on soft gels (13). At the protein level, FHL2 protein was upregulated in HSCs cultured on stiff 32 kPa plates and on plastic as compared with HSC cultured on soft 0.5 kPa plates (Fig. 5A), confirming the mRNA results of a published database (Supplemental Fig. S5C) (13). Interestingly, FHL2 expression in human primary HSCs was not affected by TGF-β, nor by PDGF-BB, the two most potent factors driving HSC activation and matrix deposition (Supplemental Fig. S5D). This suggests that FHL2 is selectively sensitive to stiffness.

Figure 5.

FHL2 is upregulated by stiffness and promotes a fibrogenic amplification loop in HSCs. A: primary human HSCs were cultured on 0.5 kPa, 32 kPa, or plastic plates for 24 h and were analyzed by WB; (n = 3). Kruskal–Wallis nonparametric test was used to measure the P value. Primary human HSCs were cultured on 32-kPa silicon plates and infected with shControl or shFHL2 for 48 h to obtain FHL2 downregulation. Cell lysates were analyzed by WB (B) and qPCR (C) and Kruskal–Wallis nonparametric test was used to measure the P value; (n = 4). Primary human HSCs were cultured on plastic plates and were transfected with empty vector (EV) or FHL2 plasmid for 48 h to obtain FHL2 overexpression. Cell lysates were analyzed by WB (D) and qPCR (E) and Mann–Whitney nonparametric test was used to measure the P value; (n = 5) *P < 0.05, **P < 0.01. FHL2, four and a half LIM domains protein 2; HSCs, hepatic stellate cells; WB, Western blot.

For our following steps, we challenged FHL2 expression by knockdown or overexpression and examined the HSC fibrogenic program such as collagen, fibronectin, and αSMA expression. FHL2 shRNA (shFHL2) reduced FHL2 expression at mRNA and protein levels in HSC cultured on 32 kPa matrix (Fig. 5B). Moreover, this FHL2 knockdown abrogated protein levels of HSC fibrotic markers, such as fibronectin and collagen I, without affecting αSMA (Fig. 5B). However, fibronectin and collagen I downregulation were not significant at the mRNA level (Fig. 5C). In line with these data, transient FHL2 overexpression in HSCs cultured on plastic not only upregulated FHL2 expression at protein and mRNA levels but also promoted fibronectin and collagen I expression (Fig. 5D). However, this upregulation of fibrotic markers was not observed at the mRNA level (Fig. 5E), indirectly suggesting that it might be a posttranslational regulation of these fibrogenic molecules.

In summary, stiffness due to ECM mediates the upregulation of focal adhesion molecules such as FHL2, which drives a profibrotic program in HSCs by upregulating fibronectin and collagen I expression leading to an amplification of fibrogenesis (Fig. 6).

Figure 6.

Schema of the proposed study. HSCs localized in stiff ECM^S-High areas increase the expression of focal adhesion molecules including FHL2. This leads to increased expression of ECM molecules such as collagen I and fibronectin leading to more matrix production, a stiffer environment, and a fibrosis amplification loop. HSCs localized in soft ECM^S-Low areas have decreased expression of focal adhesion molecules and abrogated matrix protein production. Created with BioRender.com. ECM, extracellular matrix; FHL2, four and a half LIM domains protein 2; HSCs, hepatic stellate cells.

DISCUSSION

Liver cirrhosis is associated with increased liver stiffness, which is currently utilized for noninvasive disease detection and staging (9). Nevertheless, in the current era of describing liver heterogeneity and zonation, how liver stiffness heterogeneity affects cell subpopulations and fibrosis progression remains elusive. In developing fibrosis, we found that stiffness macro- and microheterogeneity are mainly due to matrix deposition leading to ECM^S-High and ECM^S-Low areas. We identified an activated HSC subpopulation in both healthy and fibrotic livers. This HSC subpopulation upregulated focal adhesion pathway in stiff ECM^S-High areas in the fibrotic liver, a pathway that was triggered exclusively by stiffness. Enhancement of focal adhesion molecule expression in HSC^col-high subpopulation led to the amplification of the fibrogenic program through increasing ECM deposition, the main feature of liver fibrosis progression (Fig. 6). This study attests to the existence of a new stiffness-mediated fibrosis amplification loop, advancing our understanding of disease progression.

MRE-measured liver stiffness is being increasingly utilized to stage fibrosis for patients and preclinical studies in mice (8, 35–37). However, the mean value of stiffness does not reflect its spatial heterogeneity within the liver. Understanding stiffness heterogeneity is of interest, especially when the gold standard for staging liver fibrosis relies on a small liver biopsy that might not accurately reflect the overall burden of fibrosis (36). Here, we report that the MRE-measured spatial macroheterogeneity of stiffness attested to the presence of highly fibrotic regions, which were not reflected by the modest increase in mean shear stiffness, but which confirmed previously published data (38). Microscopically, the AFM-measured highly fibrotic regions presented a stiffness microheterogeneity with stiff ECM^S-High and soft ECM^S-Low areas. Given the role of stiffness in regulating cell behavior (13, 39), our data indirectly suggest that stiffness heterogeneity might drive liver cell heterogeneity during fibrogenesis.

Recently, liver cell heterogeneity has emerged as an important concept to assess cell zonation in patients and mouse models of liver diseases (40). In this regard, single-cell RNA sequencing (scRNA-seq) studies have revealed rare cell types, cell subtypes, disease-specific cell types, and cell-to-cell interactions via ligand-receptor expression analysis (40, 41). Indeed, zonation between hepatic periportal and pericentral regions following spatial analysis has identified several subpopulations of hepatocytes, endothelial, immune, and HSCs in healthy and injured livers (7, 27). Recently, HSCs have been clustered in two main subpopulations, ADAMTSL2⁺ collagen-producing HSCs with a pericentral location, and nerve growth factor receptor (NGFR)⁺ HSCs localized in the periportal area (27). However, scRNA-seq studies have been mainly focused on characterizing HSC and myofibroblast subpopulations rather than identifying pathological pathways and molecular candidates. In addition, how stiffness is related to and mediates the heterogeneity of HSCs within the stiff ECM^S-High areas remains unknown. In our study, we identify the HSC^col-high subpopulation in both healthy and fibrotic livers. In addition to the fibrogenic HSC subpopulation previously described (27), our results underscore the existence of a subset of HSCs in healthy livers, which express Col1α1 and Adamtsl2 and might be prone to activate in case of injury. However, Col1α1 and Adamtsl2 expression levels in the healthy HSC subset are lower than in HSC^col-high from fibrotic livers. In fibrotic livers, HSC^col-high is heterogeneous and only a fraction of these cells within the stiff area upregulates the focal adhesion pathway. These data suggest that during fibrogenesis, stiff areas present heterogeneity of HSCs which differentially regulate specific signaling pathways.

Matrix stiffness is a central regulator of myofibroblast function in several organs (13, 39). Moreover, increased stiffness is shown to participate in disease progression (13–16). In hepatocytes, stiffness decreases albumin production and the expression of functional genes within the HNF4 regulatory network (42). Stiffness has been described to induce migratory contractile phenotype in HSCs through focal adhesion kinase (43). However, the role of focal adhesion molecules in matrix regulation in hepatic stellate cells is elusive. Here we report that in the liver, HSC^col-high in stiff fibrotic livers in vivo as well as HSCs cultured on stiff matrix in vitro significantly enhance the expression of focal adhesion molecules. It has been shown that deletion of one of the focal adhesion molecules FHL2 selectively in fibroblasts attenuates renal fibrosis (44). Moreover, FHL2 has been reported to be involved in scar formation in cardiac ischemia (45). Nevertheless, the role of HSC-specific FHL2 in the liver has not been assessed. Here, we demonstrate that this focal adhesion molecule is selectively upregulated by stiffness and drives subsequent matrix deposition, highlighting the fact that focal adhesions not only participate in canonical cell migration but also in scar formation and stiffness-mediated progression of liver fibrosis.

In conclusion, spatial stiffness heterogeneity leads to HSC functional diversity where focal adhesion is a stiffness-specific pathway driving the amplification loop of fibrogenesis and might require special attention for the development of new specific inhibitors.

SUPPLEMENTAL DATA

Supplemental Tables S1– S10 and Supplemental Figs. S1–S5: https://doi.org/10.6084/m9.figshare.15093954.v1.

GRANTS

This study was supported by the American Association for the Study of Liver Diseases Pinnacle Research Award (to E. Kostallari), Mayo Clinic Center for Cell Signaling in Gastroenterology Pilot/Feasibility Award P30DK084567 (to E. Kostallari), and the National Institutes of Health Grants UH2/3 AA026887 (to M. Yin and V. H. Shah), R01 EB017197 (to M. Yin), R01 HL133320-1 (to D. J. Tschumperlin and V. H. Shah), R01 HL092961 (to D. J. Tschumperlin), R37 AA021171-06, and R01 DK59615-16 (to V. H. Shah).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

E.K. conceived and designed research; E.K., B.W., D.S., J.L., and J.G. performed experiments; E.K., B.W., D.S., J.L., S.A.C., M.D., Y.L., and M.Y. analyzed data; E.K., B.W., D.S., J.L., and S.C. interpreted results of experiments; E.K., D.S., J.L., and S.A.C. prepared figures; E.K. and B.W. drafted manuscript; E.K., D.J.T., and V.H.S. edited and revised manuscript; E.K. and V.H.S. approved final version of manuscript.