Effect of Distinct ECM Microenvironments on the Genome-Wide Chromatin Accessibility and Gene Expression Responses of Hepatic Stellate Cells

Abstract

Hepatic stellate cells (HSCs) are one of the primary drivers of liver fibrosis in non-alcoholic fatty liver disease. Although HSC activation in liver disease is associated with changes in extracellular matrix (ECM) deposition and remodeling, it remains unclear how ECM regulates the phenotypic state transitions of HSCs. Using high-throughput cellular microarrays, coupled with genome-wide ATAC and RNA sequencing within engineered ECM microenvironments, we investigated the effect of ECM and substrate stiffness on chromatin accessibility and resulting gene expression in activated primary human HSCs. Cell microarrays demonstrated the cooperative effects of stiffness and ECM composition on H3K4 and H3K9 methylation/acetylation. ATAC sequencing revealed higher chromatin accessibility in HSCs on 1kPa compared to 25kPa substrates for all ECM conditions. Gene set enrichment analysis using RNA sequencing data of HSCs in defined ECM microenvironments demonstrated higher enrichment of NAFLD and fibrosis-related genes in pre-activated HSCs on 1kPa relative to 25kPa. Overall, these findings are indicative of a microenvironmental adaptation response in HSCs, and the acquisition of a persistent activation state. Combined ATAC/RNA sequencing analyses enabled identification of candidate regulatory factors, including HSD11B1 and CEBPb. siRNA-mediated knockdown of HSD11b1 and CEBPb demonstrated microenvironmental controlled reduction in fibrogenic markers in HSCs.

Graphical Abstract

Introduction

Liver disease is the major cause of 2 million deaths every year worldwide, 50% of which are due to complications related to cirrhosis, the advanced stage of Non-Alcoholic Fatty Liver Disease (NAFLD) [1]. Occurrence rates of NAFLD is estimated to be around 25.2 % across the globe, which can further progress to non-alcoholic steatohepatitis (NASH) in 1/4^th of the patients. Current treatments for NASH and NASH-related cirrhosis are typically based on dietary and lifestyle changes in addition to supplements or pharmaceutical treatments such as Vitamin E, pioglitazone and pentoxifylline that have shown limited efficacy [2, 3]. Over the past decade, hepatic stellate cells (HSCs) were identified as one of the primary drivers of fibrosis in an injured liver, despite comprising only ~10% of the organ [4]. HSCs inhabit the perisinusoidal region of the liver sinusoid, a microenvironment which is a rich milieu of cells, ECM, and signaling molecules that is critical to normal liver function. In a healthy liver, HSCs exist in a quiescent, non-proliferative, Vitamin A storing state. During acute injury to the liver, they activate to transdifferentiate to myofibroblast-like cells, secreting ECM which contributes to early fibrotic scarring as a side effect of the repair and regeneration process. During chronic injury, this activation state is prolonged leading to liver fibrosis [5, 6].

Notably, HSC gene expression heterogeneity and plasticity have been described in vivo [7–10], and recently microenvironmental ECM signals have been implicated in modulating HSC phenotypic heterogeneity. Broadly, the ECM provides structural integrity within tissues, and modulates the signaling pathways of resident cells within their local microenvironments. In a healthy liver microenvironment, the ECM is mainly comprised of glycoproteins such as fibronectin, laminin, and collagen IV, as well as proteoglycans including heparan sulfate. In diseased or fibrotic livers, the ECM changes dramatically in composition and quantity. In particular, fibrotic livers exhibit increased levels of ECMs like collagens I, II, III, V and fibronectin and decreased levels of laminin and collagen IV. Relative ECM protein presence was observed to regulate HSC behaviors like proliferation and collagen I expression further implicating the role of the ECM in influencing fibrosis progression. [11–13]. The mechanical stiffness of the liver has also been observed to increase as fibrosis progresses [14, 15]. Based on these reports, we have previously delineated the combinatorial effect of ECM and substrate stiffness on human pre-activated HSC behavior, where single and pairwise combinations of the 10 most common ECMs present in the liver were examined together with 3 distinct substrates spanning the stiffnesses characteristic of healthy and disease states [16].

Epigenetics have been widely studied in the context of differential cellular responses with identical genomic content. Broadly, epigenetic regulation refers to the modifications leading to gene expression changes without a change in the underlying DNA sequence. These modifications such as acetylation/methylation on various histone positions, affect the chromatin structure, which results in differential transcription accessibility of various genes [17].

Furthermore, there have been multiple studies exploring microenvironmental control of epigenetics in tumors and fibrosis. Stowers et al. specifically studied tumor phenotype influenced by matrix stiffness via epigenetic mechanisms [18]. Walker et al. demonstrated stiffness dependent chromatin remodeling in myofibroblasts. Similar mechanisms have also been implicated in various stages of HSC activation [17]. Further, small molecule epigenetic inhibitors such as the DNMT1 (DNA methyltransferase 1) inhibitor 5-azadeoxycytidine (5-AzadC) and the EZH2 (Enhancer of zeste homolog 2) inhibitor 3-deazaneplanocin A (dZNep) potently inhibit HSC activation in vitro and in vivo. However, the specific control of these epigenetic mechanisms as a function of complex composition of extracellular matrix has not been explored [19].

One of the primary findings of our previous study into HSC phenotypic regulation within cell-ECM microarrays was the relatively increased expression of Lox and Collagen 1 in pre-activated HSCs when cultured on 1kPa compared to 25kPa stiffness substrates [16]. Although this experimental observation was initially counter-intuitive, as 1kPa stiffness is similar to a healthy liver, it suggested that HSCs may exhibit potential adaptations to microenvironmental conditions that could act to influence the responses of HSCs to dynamic changes in microenvironmental signals. Consequently, to specifically examine HSC microenvironmental response mechanisms, we aimed to study epigenetic regulation involved in HSC responses to distinct ECM composition and stiffness microenvironments. We used high-throughput cellular microarrays to quantify overall acetylation and methylation markers at specific histone positions as a function of extracellular matrix and substrate stiffness. Further, 3 specific ECM combinations were down-selected based on the divergent epigenetic regulation responses identified in cell microarray studies, and subsequent paired ATAC-sequencing (ATACseq) and RNA-sequencing (RNAseq) was performed. The ATACseq analysis revealed differential genome-wide chromatin accessibilities in HSCs as a function of the underlying substrate stiffness and specific ECM combination. Additionally, simultaneous RNAseq informed how those chromatin accessibilities resulted in unique gene expression in the HSCs. Lastly, candidate genes HSD11b1 and CEBPb were found to be regulators of characteristic fibrogenic markers on 1kPa substrates, which were investigated further using a transient knockdown for both genes in HSCs. The subsequent alterations in epigenetics and fibrogenic behavior of the HSCs was quantified using the high-throughput cell microarrays.

Materials and Methods

Cell Culture

For all experiments in this study primary human hepatic stellate cells were used between passage 15–17 (Lonza). For passaging, cells were seeded on tissue culture plastic coated with poly-l-lysine (0.02 mg/ml) and subsequently cultured under controlled environmental conditions (37 °C and 5% CO₂). Cells were treated with trypsin-EDTA (0.25% v/v) for 5 min to detach them for sub-culturing. Cell culture media consisted of High-Glucose DMEM with fetal bovine serum (FBS) at 10% v/v, penicillin/streptomycin (P/S) at 1% v/v, and l-glutamine at 1% v/v. The HSD11b1 and CEBPb knockdown and Negative Control (NC) HSC transient cell line was created using lipid-based transfection of HSD11b1 siRNA (Thermofisher Scientific, AM16708, siRNA ID 107742), CEBPb siRNA (Thermofisher Scientific, AM16708, siRNA ID 114494) and nonsense siRNA (Thermofisher Scientific, AM4611). Lipofectamine RNAiMAX (Thermo Fisher Scientific 685 13778075) was used according to the protocol recommended by the manufacturer. Briefly, for 10mL of media, 1000uL of transfection solution was made. 30uL of the RNAiMax was dissolved in 500uL of OptiMEM media and 28uL of 10uM stock of siRNA was dissolved in 5000uL of OptiMEM media. Both the solutions were mixed to get 1000uL of transfection solution and incubated at room temperature for 30 minutes. The transfection solution was added to 10mL of HSC media with cells at 50% confluency after they had been growing on tissue culture plastic for atleast 24 hours after passaging or thawing. They were then transfected for 48 hours before harvesting for microarray studies. For microarray experiments, 30000 cells/1 well of 24-well plate were used. For immunocytochemistry and sequencing experiments cells were left to adhere on arrays overnight (~18 h). After seeding, arrays were washed twice with media and then experiment-specific treatments were delivered. DZNep (EMD Millipore, 252790–2MG) was delivered at 1uM concentration (1mg/mL, 3.8mM stock in DMSO) and DMSO was delivered in equal amounts for the control.

Preparation of polyacrylamide hydrogels

Polyacrylamide (PA) hydrogels were prepared following previous protocols [20–22]. Briefly, 12 mm glass coverslips etched by immersing them 0.2 N NaOH (Sigma-Aldrich 415413–1L) for 1 hour on an orbital shaker and then rinsing with dH₂O. The coverlips were then air-dried and placed on a hot plate at 110°C until dry. For silanization, the cleaned coverslips were immersed in 2% v/v 3-(trimethoxysilyl)propyl methacrylate (Sigma Aldrich 440159–500ML) in ethanol and placed on the shaker for 30 minutes, followed by a wash in ethanol for 5 minutes. The silanized coverslips were air-dried, and again placed on the hot plate at 110°C until dry. For fabrication of hydrogels with specific elastic moduli, prepolymer solution in dH20 with 8% acrylamide (Sigma-Aldrich A3553–100G) and 0.55% bis-acrylamide (Sigma-Aldrich M7279–25G) was prepared to achieve elastic moduli of 25 kPa. The prepolymer solution was then mixed with Irgacure 2959 (BASF, Corp.) solution (20% w/v in methanol) at a final volumetric ratio of 9:1 (prepolymer:Irgacure). This working solution was then deposited onto Rainx (Amazon Rain-X 800002245) coated slides (20uL/coverslip) and covered with silanized coverslips. The sandwiched working solution was transferred to a UV oven and exposed to 365 nm UV A for 10 min (240E3 μJ). The coverslips with the hydrogels attached to it were immersed in dH₂O at room temperature for a day in order to remove excess reagents from the hydrogel substrates. Before microarray fabrication, hydrogel substrates were thoroughly dehydrated on a hot plate for ≥15 minutes at 50°C.

Microarray fabrication

Microarrays were fabricated as described previously [23–25]. ECMs (Supplemental Table 1) for arraying were diluted in 2×ECM printing buffer. to a final concentration of 250 μg/mL and loaded in a 384-well V-bottom microplate. To prepare 2× ECM protein printing buffer, 164 mg of sodium acetate and 37.2 mg of ethylenediaminetetraacetic acid (EDTA) was added to 6 mL dH2O. After solubilization, 50 μL of pre-warmed Triton X-100 and 4 mL of glycerol was added. 40 – 80 μL of glacial acetic acid was added, titrating to adjust the pH to 4.8.A robotic benchtop microarrayer (OmniGrid Micro, Digilab) loaded with SMPC Stealth microarray pins (ArrayIt) was used to microprint ECM combinations from the 384 microwell plate to polyacrylamide hydrogel substrate, resulting in ~600 μm diameter arrayed domains. Fabricated arrays were stored at room temperature and 65% RH overnight and left to dry under ambient conditions in the dark. For printing on the 24-well matrigen plates (Matrigen, Softwell24, Easy Coat, 1kPa, 2kpa, 25kPa), the plate was dehydrated at 55C for 15 minutes prior to printing.

Sequencing Sample Collection

For ATACseq, live cells were collected from the microarrays at 72 hours after the overnight seeding. Briefly, 100uL of trypsin (Fisher Scientific: SH30042.02) was added directly to the 24-well plate wells after washing with PBS (without calcium and magnesium) and the cells were incubated at 37 degrees for 5 minutes. After removal of trypsin, 100uL of 5mg/mL solution of Collagenase D (Sigma Aldrich: 11088858001) in HBSS buffer which was added to each well uniformly and incubated at 37°C for 20 minutes. Collagenase was deactivated by adding warm media and cells were centrifuged, resuspended in fresh media and transferred to sequencing facility. Briefly, the libraries were prepared using the ATAC-Seq Kit from Active Motif. The libraries were pooled, quantitated by qPCR and sequenced on two SP lanes for 101 cycles from both ends of the fragments on a NovaSeq 6000. Fastq files were generated and demultiplexed with with the bcl2fastq v2.20 Conversion Software (Illumina). For RNAseq, RNA sample was collected simultaneously using RNeasy Plus Kit (Qiagen, QIA74134). Briefly, RLT buffer was added directly to microarrays, incubated for 1 min at followed by collection of the sample in a tube and vortexed for homogenization. The RNA was extracted using the manufacturers protocol from there on and transferred to the sequencing facility. Briefly, the RNAseq libraries were prepared with Illumina′s′TruSeq Stranded mRNAseq Sample Prep kit (Illumina). The libraries were pooled, quantitated by qPCR and sequenced on one SP lane for 101 cycles from one end of the fragments on a NovaSeq 6000. Fastq files were generated and demultiplexed with the bcl2fastq v2.20 Conversion Software (Illumina). The quality control files for both the RNAseq and ATACseq fastq files can be found in Supplemental File 1.

Sequencing Data Analysis

For ATACseq the fastq files were processed using the ENCODE pipeline: https://www.encodeproject.org/atac-seq/. BigBed files of the peaks found in every sample was obtained from the pipeline. The counts for every peak/region was obtained using bedtools coverage function. For RNAseq data analysis, the fastaq files were processed using the ENCODE pipeline: https://www.encodeproject.org/pipelines/ENCPL002LSE/. The rsem RESULTS output file obtained from the pipeline was used to extract counts for each transcript using tximport package in R. For both the sequencing, the counts were normalized and log transformed. Surrogate variable analysis was done using RUVseq package in R, and removeBatchEffect function was used to remove the surrogate variables found. Differential accessibility analysis was done using lmFit function of package limma in R. Specific comparisons (Stiffness/ECM) were statistically tested in the model using contrasts.fit function. Adjusted P-value and fold change was calculated for each region and comparison.

For subsequent analysis with the ATACseq data, the regions were annotated according to proximity to a gene using annotatePeak function in ChIPseeker package in R. The peaks were visualized using the desktop application of Integrative Genome Viewer using the bigwig files obtained from the ENCODE pipeline. For motif analysis, region of interest for every specific comparison and p-value filtration was converted to bed files and fasta files subsequently using getfasta tools in bedtools. The fasta files were submitted in the online MEME-Chip tool: https://meme-suite.org/meme/tools/meme-chip. The classic mode was used, with the HOCOMOCO Human (v11 FULL) database selected for alignment with known motifs.

A common gene list for a specific comparison between ATACseq and RNAseq was made for enrichment analysis on EnrichR website. For GSEA analysis, the java desktop application was used, with instructions from their Wiki page was used.

Immunostaining

Array samples were fixed in 4% w/v paraformaldehyde, 1 × PBS for 15 min. Fixed samples were then permeabilized with a 0.25% v/v Triton X-100, 1 × PBS for 10 min and incubated in 1% w/v BSA and 0.25% v/v Triton X-100, 1 × PBS for one hour at room temperature. After blocking, samples were incubated for overnight at 4 degrees with primary antibodies in blocking buffer. Samples were subsequently washed with 1x PBS thrice and then incubated for one hour at room temperature with secondary antibodies diluted in blocking buffer. Finally, samples were mounted in Fluoromount G with DAPI (Southern Biotech, 0100–20). The concentrations of the primary antibodies used are as followed: Anti-Histone H3 (acetyl K4) (Abcam, ab232931, 1:200), Anti-Acetyl Histone H3 (Lys9) (C5B11) (Cell Signaling Technologies, 9649S, 1:200), Anti-Di/tri Methyl Histone H3 (Lys9) (6F12) (Cell Signaling Technologies, 5327S, 1:200), Anti-Tri Methyl Histone H3 (Lys4) (C42D8) (Cell Signaling Technologies, 9751S, 1:200), Anti-Collagen 1, (R&D Systems, AF6220, 1:100), Anti-LOX antibody [EPR4025] (Abcam, ab174316, 1:250).

Microscopy and Image analysis

The microarrays for the immunostaining were imaged using Axioscan.Z1 Slide Scanner and 10X objective. A wide tile region was defined for the whole array region which was then stitched offline using Zen and exported into TIFF Images for each individual channel. Images of entire arrays were converted to individual 8-bit TIFF files per channel (i.e., red, green, blue) by Fiji (ImageJ version 1.52p) [26]. The images were cropped in MATLAB (version R2018b) to separate each array in a single image. Positional information for each array was automatically calculated using their relative position from the positional dextran-rhodamine markers. CellProfiler (version 4.0.0) [27] was used get per cell measurement for each channel. Nuclei were identified using the DAPI channel image using IdentifyPrimaryObject module and other stains were associated with a specific nuclie was identified by looking at the red/green stain around these nuclei using IdentifySecondaryObject module. The MeasureObjectIntensity module was used to quantify single-cell intensity. The data were exported to CSV files that were then imported in RStudio for data visualization.

Data analysis

All microarray experiments consisted of at least three biological replicates, with 15 technical replicates, or islands, per biological replicate per combination of gene knockdown, treatment, and readout. For comparison between conditions in this study, Wilcoxon tests were performed using the wilcox.test function in R. P values of <0.05 were considered significant.

Data Availability Statement

All the raw data is provided in the box folder: https://uofi.box.com/s/fkmgmpu17w4u5mxgkgljuljzoccybccg. The sequencing data is uploaded on GEO repository with the accession codes GSE210966 and GSE210967.

Results

Methylation and acetylation at H3K9 and K3K4 quantification higher on 1kPa and regulated by underlying ECM

An initial investigation of the epigenetic landscape of HSCs as a function of substrate stiffness was performed by evaluating gene expression of certain histone modifiers that have been implicated in HSC activation (Supplemental Figure 1). Our results demonstrated higher expression of EZH2, a histone methyl transferase, on the 25kPa substrates compared to the 2kPa substrates. Based on our previous studies examining HSC phenotypic regulation within different ECM microenvironments [16] and our previously determined variations in ECM composition in during liver fibrosis progression [28], we selected 8 unique two-factor ECM conditions, in combination with defined stiffnesses, for investigation of HSC epigenetic response mechanisms. Specifically, cell microarrays were fabricated on polyacrylamide hydrogels on 12mm coverslips of stiffness 1, 6 and 25kPa [16]. We aimed to evaluate the response of HSCs as function of stiffness of a healthy tissue (1kPa), a fibrotic tissue (25kPa), and an intermediate stiffness of 6kPa to capture transitionary information. The ECM combinations were chosen based on the representative phenotypic clusters of HSCs found in our previous analyses [16]. Namely, Collagen 1, Fibronectin, 6kPa; Collagen 4, Collagen 5, 1kPa; Collagen 4, Fibronectin, 6kPa; and Collagen 4, Lumican, 1kPa; represented the HSC phenotype cluster promoting elevated levels of proliferation and decreased levels of Collagen 1 production. Further, Collagen 3, Collagen 4, 25kPa; and Fibronectin, Laminin, 6kPa represented the HSC phenotype cluster exhibiting decreased levels of proliferation and Lox expression. Lastly, Collagen 4, Tenascin C, 25kPa and Collagen 1, Hyaluronic Acid, 1kPa represented the HSC phenotype cluster demonstrating decreased levels of proliferation but elevated collagen 1 and Lox expression. Furthermore, we had observed overall stiffness trends where 1kPa substrates resulted in higher Lox and Collagen 1 production and significantly decreased proliferation rates were quantified in HSCs on 25kPa compared to 6kPa substrates. The fabricated cellular microarrays were cultured in individual wells of a 24-well plate allowing for an efficient use of cells leading to high-throughput quantification of various epigenetic measures (Fig. 1a). The representative images of cellular microarrays via DAPI nuclear stain on the 3 stiffness and 6 ECM combinations are shown in Supplementary Figure 2a–c.

Figure 1: Experimental design of high-throughput quantification of epigenetic markers and principal component analysis of quantification of H3K9Me and H3K9Ac in preactivated-HSCs.

a) Schematic of the cellular microarray pipeline for high-through quantification of epigenetic markers as a function of substrate stiffness and underlying extracellular matrix combination b) Correlation plot of measured methylation and acetylation on H3K4 and H3K9 to PC1 and PC2 respectively. c) Principal component plane for 4 epigenetic markers as a function of the specific microenvironment. Analysis done in R, n>= 3 biological replicates (independent experiments) and n>= 10 technical replicates (individual islands).

This multiwell array system was then used to quantify methylation and acetylation at H3K9 and H3K4 position using immunostaining of these markers. Methylation levels at the H3K9 position has been inversely correlated with the expression of the PPARy gene, that aids in maintaining quiescent state of HSCs [29]. Di/Trimethylation at H3K4 has been found in the promoter regions of pro-fibrogenic genes such as Col1a, a-SMA and TIMP-1 having a positive effect in their expression [30, 31]. Further, global acetylation levels have been linked to fibrogenicity in HSCs in various contexts [32]. However, acetylation at specific histone positions has not been evaluated in the context of stellate cell activation. Consequently, single-cell quantification of the methylation and acetylation of both H3K9 (H3K9Me, H3K9Ac) and H3K4 (H3K4Me, H3K4Ac) were performed for the cell microarray-based HSC cultures. Principal component analysis (PCA) of the 4 quantitative immunostaining measures of the specific histone modifications for the HSCs cultured on 8 different ECMs across 3 different substrate stiffness allowed us to classify the microenvironmental conditions based on the epigenetic measures. The overall variation in the quantified histone modifications could be well defined by the first 2 dimensions, in which Dimension 1 captured 84.6% of the variance and Dimension 2 captured additional 10% of the variance. For PCA Dimension 1, all 4 epigenetic metrics correlated with each other. However, the 10% of the variance captured by PCA Dimension 2 highlighted the variability between these different metrics (Fig 1b). From this analysis, it was observed that stiffness was the factor that contributed to the separation of the epigenetic phenotype along PCA Dimension 1, whereas ECM separated the data on PCA Dimension 2 (Fig. 1c). Specifically, we observed higher H3K9Ac and H3K4Ac expression in HSCs cultured on 1kPa, with a decreasing trend with increasing substrate stiffness (Fig. 2). Notably, for the HSCs cultured on the 25kPa substrates, an increased presence of H3K9Ac value was determined for cells cultured on Fibronectin, Laminin compared to Collagen 1, Fibronectin. Furthermore, culture of HSCs on 6kPa substrates together with Collagen 1, Fibronectin resulted in a 50% increase in anti-H3K9Ac immunofluorescence labeling intensity compared to HSCs cultured on 6kPa, Collagen 4, Lumican, whereas HSCs cultured on 25kPa, Collagen 1, Hyaluronic Acid resulted in 80% increase in the H3K9Ac intensity compared to HSCs cultured on 25kPa, Collagen1, Fibronectin. Lastly, HSCs cultured on Fibronectin, Laminin also exhibited elevated H3K9Ac presence compared to Collagen 4, Lumican on 6kPa substrates. An increased expression of both H3K4Me and H3K4Ac was additionally observed for HSCs cultured on 1kPa compared to 6kpa and 25kPa (Supplemental Figure 3). In particular, a 40% increase in H3K9Me was observed for culture on 1kPa substrates relative to 25kPa. Further, within each stiffness, ECM significantly affected these quantification values. HSCs cultured on Collagen 1, Fibronectin had 28% more H3K9Me intensity compared Collagen 4, Tenascin C on 1kPa substrates.

Figure 2: Quantification of H3K9Ac and H3K4Ac in preactivated-HSCs.

a,d) Box-jitter plot of H3K9Ac and H3K4Ac fluorescence intensity as a function of stiffness. Every data point represents an individual microarray island, where the color of each datapoint is a specific ECM combination. b,e) Fluorescence Images of HSCs microarray islands. Scale Bar 100 microns. c,f) Box plots for H3K9Ac and H3K4Ac fluorescence intensity as a function of specific ECM combination on 1kPa, 6kPa and 25kPa. Boxplots: ‘*’: p-value <0.05; ‘****’: p-value <0.0001, calculated using Wilcox test in R, n>= 3 biological replicates (independent experiments) and n>= 10 technical replicates (individual islands).

We additionally measured 3D nuclear morphological features of HSCs, such as nuclear volume and sphericity, on the microarray platform (Supplemental Fig. 4). Overall, HSCs cultured on 1kPa exhibited more spherical nuclei, and had increased nuclear volume compared to HSCs cultured on 6 and 25kPa. Furthermore, specific ECM combinations regulated both the sphericity and nuclear volume on each stiffness. Subsequently, we quantified microenvironmental changes in cellular morphology of HSCs as function of substrate stiffness and ECM combination. Mean HSC cellular area was significantly decreased for HSCs cultured on 25kPa compared to 1kPa. Furthermore, cellular eccentricity (1= perfect ellipse, 0= perfect circle) was highest in HSCs cultured on 1kPa compared to both 6kPa and 25kPa, indicating that the HSCs were more elliptical on the 1kPa substrates (Supplemental Fig. 4).

Elevated chromatin accessibility on relatively soft substrates that is cooperatively regulated by ECM composition

To gain further insights into the microenvironmental regulation of HSC epigenetics, we sought to perform ATACseq with HSCs cultured on well-defined microenvironments and systematically evaluate genome-wide chromatin accessibilities. Since the cell microarray studies indicated that several epigenetic markers demonstrated a decreasing trend with increasing substrate stiffness, we chose 1 kPa and 25kPa for further epigenetic evaluation in subsequent experiments. Additionally, 3 ECM combinations that spanned different regions on the principal component plane were chosen; Collagen 1, Fibronectin; Collagen 4, Tenascin C and Fibronectin, Laminin (Fig. 1). For ATACseq studies, an adapted cell microarray platform was integrated, as it offered the ability to control microenvironmental context, incorporate additional replicates, and reduce material usage compared to bulk culture. In particular, the protocol for synthesizing the microarrays was optimized for a 24-well plate format, using commercially available defined stiffness surfaces (Matrigen) that cover the bottom surface of each well completely. Following array fabrication, each well contained a single arrayed ECM combination, which facilitated cell collection and lysis from each condition. In addition, we aimed to correlate the chromatin accessibility information obtained from ATACseq data to gene expression, therefore, cultured cell RNA was collected simultaneously from each well for performing RNAseq. Both ATACseq and RNAseq were performed on HSC samples collected after 72 hours of culture on the multiwell microarrays (Fig. 3a). The 72-hour timepoint was chosen based on the phenotypic and epigenetic differences observed in HSCs as a function of ECM and stiffness in the microarray experiments described above and our previous determination of phenotypic response [16].

Figure 3: Schematic of the sequencing experiments and ATACseq data quantifications.

a) Schematic of the cellular microarray pipeline for bulk ATACseq and RNAseq quantifications as a function of controlled microenvironment b,c) Representation and box plot quantification of FRiP score for a specific ECM and stiffness combination d) Heatmap of log(counts) for region differentially accessible between each ECM combination on specific stiffness e,f) ATACseq derived peaks visualized in specific regions of genes using Integrative Genomics Viewer (IGV), blue: 1kPa samples, red: 25kPa samples, green: C4TC, purple: C1FN.

Evaluation of the ATACseq data demonstrated an overall effect of stiffness and ECM composition on the fraction of reads in peaks (FRiP) score. This score informs the fraction of the open DNA fragments that lie in the statistically identified peaks. In particular, a higher FRiP score signifies higher number of DNA fragments in localized regions/peaks on the genome and hence decreased chromatin accessibility (Fig. 3b). For the HSCs cultured on 1kPa, a significantly lower FRiP score was observed regardless of the underlying ECM, signifying higher chromatin accessibility (Fig. 3c, Supp. Fig. 5e; p-value <0.01 when only stiffness is compared). Differential chromatin accessibility analysis was then performed that revealed more than 44000 distinct regions on the genome that were significantly differentially accessible as a function of the both the substrate stiffness and ECM. The heatmap of regions significantly (p-value < 0.01) differentially accessible compared between the ECM combination on each stiffness is shown in Fig. 3d. These regions were then mapped to a gene based on the genome based on its proximity to a gene to infer how accessibility could be related to phenotype. We also characterized the regions in terms of their occurrence on/near a promoter, intron, exon or distal intergenic position for a gene (Supplemental Fig. 4). A fragment on the promoter region of the gene HSD11B1 and CXCL8 in the HSCs on 1kPa showed a 6-fold increase in counts, signifying 6-fold higher accessibility, when compared to the 25kPa whereas intron regions corresponding to SLC1A7 and NRXN3 were amongst the region with highest fold-increase in HSCs cultured on the 25kPa substrates compared to the 1kPa (Fig. 3e). Furthermore, an intronic region associated with COL4A1 was significantly more accessible on C4TC on 1kPa substrates compared to C1FN on 1kPa. Similarly, an intronic region associated with GLUD1 was significantly more accessible on C4TC on 25kPa substrates compared to C1FN on 25kPa (Fig. 3f).

RNA sequencing analysis reveals increased expression and enrichment of liver fibrosis-related genes in pre-activated HSCs cultured on relatively soft microenvironments

RNAseq data of HSCs cultured on C1FN, C4TC and FNLN on 1kPa and 25kPa revealed gene expression differences based on the microenvironment condition. A heatmap of genes exhibiting significant (p-value < 0.05) differential expression when compared across ECM/stiffness conditions is shown in Fig. 4a. Overall, the representative regions for differential chromatin accessibility that emerged from the analyses displayed in Fig 3, showed similar trends in their associated gene expression differences. For example, HSD11B1 and CXCL8 were shown to have a 30-fold and 14 fold higher expression in HSCs on 1kPa compared to 25kPa whereas SLC1A7 and NRXN3 had a 6-fold and 10 fold higher expression in HSCs on 25kPa compared to 1kPa (Fig 4b,c). For both COL4A1 and GLUD1 genes, ECM dependent regulation of differential chromatin accessibility was found resulting in an ECM dependent differential gene expression was also observed (Fig. 4d). To further analyze the genome-wide expression data for each microenvironmental condition, we decided to perform Gene Set Enrichment Analysis (GSEA) [33, 34]. A custom gene set was prepared using the data previously reported in [16], consisting of ‘matrisome’-related proteins upregulated in liver biopsies from human patients at different stages of liver fibrosis compared to healthy human patients. This list of upregulated genes was used to calculate enrichment in our RNAseq dataset with various comparisons. First, when stiffness was compared, it was determined that this gene set was enriched significantly more in HSCs on 1kPa compared to 25kPa on all ECMs. Secondly, for the 1kPa substrate conditions, significantly higher enrichment of this gene set was found on C1FN substrates compared to FNLN, implicating a important cooperative role of ECM combinations on the global phenotype of the HSCs (FDR < 0.25, considered significant for GSEA analysis). Similar comparisons were performed using other gene sets relating to NAFLD from MolSigDB [35] and fibrosis using human liver biopsy data reported in the literature [36]. For both the previously established gene sets, higher enrichment for found for HSCs on 1kPa compared to 25kPa on all ECMs (Fig 4e,f, Supp Fig 5). In our previous work, we have reported higher expression of Collagen1 and Lox in pre-activated HSCs following transition to 1kPa substrates, and these new findings represent a genome-wide confirmation of this phenotypic response exhibited by tissue culture plastic-activated HSCs on 1kPa compared to 25kPa. Additionally, as part of the RNAseq analyses, we assessed YAP signaling pathway gene expression in HSCs as function of the substrate stiffness (Supplementary Figure 6). RNAseq revealed higher expression of YAP on 1kPa substrates, with higher expression of GLI2 (downstream YAP target gene in HSCs) and TEAD2 (co-transcription factor) also on 1kPa compared to 25kPa. Simultaneously, TEAD4 (YAP co-transcription factor) and AXL (YAP target gene in HSC activation) were found to be significantly higher on 25kPa. Collectively, these observations point towards a complex mechanism of YAP signaling as a function of both the stiffness and ECM composition in HSC activation.

Figure 4: RNAseq data analysis and Gene Set Enrichment Analysis.

a) Heatmap of log(counts per million) for genes differentially expressed between each ECM combination on specific stiffness b,c,d) Box plot quantification of log(counts per million) for specific genes as function of stiffness and/or ECM. e,f) GSEA enrichment plot liver fibrosis and matrisome related genes when compared between activated HSCs on 1kPa and 25kPa.

Combined analysis of ATACseq and RNAseq data

To gain further insights into the relative presence of specific transcription factors regulating accessibility/expression of multiple downstream genes, and to evaluate the collective implications of the ATACseq and RNAseq data, we performed motif analysis. Specifically, motif analysis was first run on selective regions from the ATACseq data using the MEME suite tool [37]. Subsequently, the resulting motifs were matched to an existing database of known motifs for 700 transcription factors in humans [38], followed by the validation of expression trends using the acquired RNAseq data. Specifically, the top 1% of regions that were highly accessible on 1kPa compared to 25kPa were analyzed on motif enrichment. The top 3 motifs found on these regions were matched to known motif sequences for proteins CEBPb, TEAD3 and ZNF341 respectively. The gene expression data of CEBPb especially demonstrated higher expression 1kPa compared to 25kPa (Fig 5a,b). This gene expression trend was a secondary checkpoint in validating the motif analysis results. Similarly, the top 1% regions that were highly accessible on 25kPa compared to 1kPa were analyzed for motif enrichment. The top 3 motifs found on these regions were matched to known motif sequences for proteins WT1, ZNF18 and BCL6 respectively. The gene expression of WT1 was also found to be higher 25kPa compared to 1kPa (Fig. 5e,f). Motif analysis for deciphering ECM regulation was performed using all of the specific genetic regions shown to have significantly different (p-value <0.05) chromatin accessibility, for every ECM combination comparison on each stiffness. The top 3 motifs found on these regions were matched to known motif sequences for proteins CPEB1, ZNF770 and PITX2. Further, CPEB1 and ZNF770 showed strong ECM based gene expression differences in the HSCs (Fig 5i,k).

Figure 5: Motif Analysis on highly accessible regions and combinatorial analysis of ATACseq and RNAseq data.

a,e) Motifs found on highly accessible region in 1kPa compared to 25kPa and vice versa using MEME-ChiP. b,f) mRNA quantification of the transcription factor, whose known motif sequence aligned best to motifs found in Fig. 5a,e. c,g) Venn diagram of regions/genes highly accessible and highly expressed in activated-HSCs on 1kPa compared 25kPa and vice versa. d,h) Enrichment of the gene list obtained in Fig. 5c,h to the KEGG Pathway database. i) Motifs found on accessible regions regulated by ECM in activated-HSCs using MEME-ChiP. j) mRNA quantification of the transcription factor, whose known motif sequence aligned best to motifs found in Fig. 5i. k) Venn diagram of regions/genes regulated by the underlying ECM composition in activated-HSCs. d,h) Enrichment of the gene list obtained in Fig. 5l to the KEGG Pathway database.

For further combined analysis of the ATACseq and RNAseq data, a common gene list was prepared by selecting the intersection of highly/differentially accessible gene regions with highly/differentially expressed genes for specific comparisons. 2352 and 2643 genes were found to have a significant expression increase, while also exhibiting a significant relative increase in chromatin accessibility in HSCs on 1kPa compared to 25kPa substrates, and vice-versa, respectively (Fig. 5c,g). These gene sets were then used for enrichment analysis in comparison with multiple databases provided within the publicly available EnrichR tool [39–41]. The top 3 KEGG pathways for genes/regions more expressed/accessible on 1kPa conditions were pathways in cancer, TNF signaling pathway, and the MAPK signaling pathway. This is in contrast to ribosome-related, cell cycle-related, and tight junctions for genes/regions that more expressed/accessible on 25kPa (Fig. 5d,h). This was a strong indication towards the differential genome-wide effect of substrate stiffness on TCP-activated HSCs. The underlying specific ECM combination of HSCs regulated 83 genes/regions that were differentially expressed/accessible that enriched to focal adhesion, oxytocin signaling pathway and Hippo signaling pathway in KEGG database (Fig 5k,l). Additionally, an increased representation of extracellular matrix and structure organization-related GO Biological Processes was found for the highly expressed/accessible genes/regions in the 1kPa conditions. Furthermore, these gene/regions were also enriched for alterations associated with diseases such as hepatocellular carcinoma and diabetes, evaluated using the ClinVar Database. It was also noted that these 3 different gene sets for HSCs (high on 1kPa, high on 25kPa, regulated by ECM) obtained by comparing the underlying stiffness and regulation by ECM enriched to different diseases in the ClinVar database, signifying the importance of the microenvironment in modelling different phenotypes and stages of human diseases (Supplemental Figure 7).

CEBPb and HSD11B1 knockdown resulted in decreased fibrogenic phenotype under specific microenvironmental conditions

The substantial enrichment of NALFD and fibrosis-related genes in GSEA, as well as the correlations with disease relevant alterations catalogued collectively pointed the importance of ECM composition and stiffness in controlling HSC phenotype and further elucidation of the cellular response mechanisms guiding these phenotypic changes. Next, we sought to examine the role of 2 genes identified from the genome-wide analyses: HSD11b1 and CEBPb. HSD11b1 (11β-Hydroxysteroid dehydrogenase type 1) is an enzyme that converts inactive glucocorticoid cortisone to active cortisol. It has been implicated in lipid metabolism in the liver and glucose sensitivity [42]. It has also been a target for clinical trials in NAFLD patients [43]. Further, the promoter region of HSD11b1 was determined to be significantly more accessible on 1kPa compared to 25kPa, which also translated to more than 30-fold higher expression of the gene. CEBPb belongs to the CCAAT/enhancer-binding protein (C/EBP) family of transcription factors that are pivotal regulators of various liver functions such as nutrient metabolism, hormone response and liver regeneration [44]. The motif sequence for CEBPb was identified within multiple regions that were in the top 1 % regions highly accessible on 1kPa compared to 25kPa including the promoter region for HSD11b1. Based on literature evidence and our experimental findings, we performed two independent transient knockdown of HSD11b1 and CEBPb in the HSCs using lipid based transfection of siRNA and quantified various fibrogenic and epigenetic markers (Fig. 6a). Additionally, in order to delineate the effect these gene expression perturbations within the context of defined microenvironmental conditions, we quantified phenotypic and epigenetic marker expression as a function of varying substrate stiffness, ECM combination and soluble factors. In addition, we explored the effect of the drug 3-Deazaneplanocin A (DZNep), which is an inhibitor of the methyl transferase EZH2, which we found to be more highly expression in 25kPa conditions. The potential therapeutic efficacy of DZNep as modifier of HSC phenotype, has been previously explored using in vitro and in vivo models of HSC activation [45–47]. The multiwell cell microarray platform facilitated the simultaneous investigation of combinations of 3 ECM (C1FN, C4TC, FNLN), 2 substrate stiffness (1kPa, 25kPa), 2 gene knockdowns and a drug treatment (Fig. 6a). Microenvironmental regulated reduction in morphological parameters such as cellular eccentricity and nuclear area was observed with both the CEBPb KD and HSD KD HSCs compared to the NC (Fig. 6b–e, Supplementary Figure 8). An increase in average HSCs cell number/island was observed with CEBPb KO on both 1kPa and 25kPa stiffness compared to the negative and a similar increase of HSD KO but only on 1kPa substrates (Supplementary Figure 8).

Figure 6: Downstream investigation of HSD11b1 and CEBPb by transient knockdown and principal component analysis.

a) Schematic of the microarray pipeline for combinatorial investigation of the phenotype of HSCs as a function of 3ECMs, 2 stiffness, two transient gene knockdowns and a drug treatment. b) Box Plot of quantification of cellular eccentricity of HSCs on C4TC, treated with DMSO as function for the gene knockdown on each stiffness. c) Representative Image for the cellular morphology quantifications for NC and HSD KD HSCS on 1kPa, C4TC, DMSO. Scale Bar 100 microns. d) Box Plot of quantification of the nuclear area of HSCs on C4TC, treated with DMSO as function for the gene knockdown on each stiffness. e) Representative images of DAPI stain for the nuclear area quantifications for NC and HSD KD HSCS on 1kPa, C4TC, DMSO. Scale Bar 100 microns. f) Correlation plot of measured epigenetic and phenotypic readouts to PC1 and PC2 respectively. c) Principal component plane for 8 readouts as a function of the specific microenvironment (ECM, stiffness, gene knockdown and drug treatment). Every data point on the plane represents a specific microenvironmental condition. Boxplots: ‘*’: p-value <0.05; ‘****’: p-value <0.0001, calculated using Wilcox test in R, n>= 3 biological replicates (independent experiments) and n>= 10 technical replicates (individual islands).

To measure the fibrogenic phenotype of the HSCs, we performed cell microarray-based single cell quantitative immunostaining for the expression of Lox (lysyl oxidase) and intracellular Collagen 1. In addition, we determined the combined effect of ECM composition/stiffness and knockdown of CEBPb or HSD11b1 on both phenotypic and epigenetic markers. PCA analysis was performed to evaluate the comprehensive behavior of HSCs as a function of their microenvironment. The quality of representation of each measured readout on the principal component dimension is shown in Figure 6f. Lox expression, Collagen 1 expression, Nuclear Area and Cellular Area were the specific cellular measurements that correlated the best with PCA Dimension 1 (PC1). The expression of H3K9me and H3K9Ac were the measurements that correlated both with PCA Dimension 1 (PC1) and PCA Dimension 2 (PC2). Nuclear and Cellular Eccentricity correlated well with PC2. The position of each specific microenvironmental variable (stiffness, ECM, drug treatment and gene knockdown) is displayed on the principal component plane in Fig. 6g). The drug treatment DZNeP (purple color) corresponds to negative value on PC1 and PC2, amongst the lowest compared to every other condition, signifying reduction in multiple activation markers of HSCs. Furthermore, for the gene knockdown comparisons, the negative control (NC) exhibited the highest value for both PC1 and PC2, with subsequent reduction in PC1 for CEBPbKO and additional reduction in PC2 for HSDKO. For the underlying ECM conditions, C4TC has the lowest PC1 and PC2 value, signifying the least value for various readouts. It was also noted that stiffness conditions, 1kPa an 25kPa, were the farthest from each other on the PC plane, signifying the great influence on all the 8 phenotype and epigenetic readouts. Overall, HSDKO and DZNep treatment led to a decrease in various activation markers in HSCs, dependent on the underlying ECM and stiffness of the cells.

Specifically, these studies demonstrated a combinatorial influence of ECM microenvironment, CEBPb and HSD11b1 expression, and treatment with DZNEP on Lox expression. In particular, a heatmap of Lox intensity as function of the combinatorial microenvironment is shown in Fig. 7a. Lox expression decreased significantly with CEBPb knockdown in C4TC, 1kPa, DMSO conditions, however with the addition of DZNep, the decrease was significant for HSD11b1 knockdown as well (Fig 7 a,b). For cells on C1FN and 25kPa, the gene knockdown conditions did not affect Lox expression, however Lox expression was significantly decreased for both the CEBPb KD and HSD KD with DZNep treatment (Fig. 7 a,d). A heatmap representation of intracellular Collagen 1 expression as a function of the combinatorial microenvironment is shown in Figure 7h. In particular, Collagen 1 expression was significantly reduced in CEBP KD and HSD KD HSCs with and without DZNep in HSCs on C4TC and 1kPa (Fig. 7f,g). However, with HSCs on 25kPa and C1FN a decrease in Collagen 1 expression was only observed with DZNep treatment and HSD11b1/CEBPb gene knockdowns (Fig. 7i,j). It is also important to note that for both Lox and Collagen 1, the 1 kPa condition exhibited higher expression compared to the 25kPa stiffness for every specific microenvironment combination. Further, we also analyzed cellular shape which was primarily dependent on the specific gene knockdown. With the HSD KD HSCs a significant reduction in cellular eccentricity (1 = elliptical, 0 = spherical) was observed on both stiffness, however significantly more reduced on the 1kPa compared to 25kPa (Supplemental Fig. 8). Overall, both the gene knockdowns, HSD11b1 and CEBPb, resulted in decrease of the fibrogeneic phenotype of the HSCs in a microenvironment specific manner.

Figure 7: Quantification of Lox and Collagen 1 expression as a function of HSD11b1 and CEBPb knockdown.

a) Heatmap of the Lox fluorescence intensity quantified in HSCs on the combinatorial microarrays b,d) Box Plot of quantification of Lox fluorescent intensity of HSCs on 1kPa,C4TC and 25kPa, C1FN respectively c) Representative Image for the Lox immunostain for NC and CEBPb KD HSCs on 1kPa, C4TC, DMSO. Scale Bar 100 microns e) Representative Image for the Lox immunostain for NC and HSD KD HSCs on 25kPa, C1FN, DZnep. Scale Bar 100 microns h) Heatmap of the Collagen 1 fluorescence intensity quantified in HSCs on the combinatorial microarrays g,j) Box Plot of quantification of Collagen 1 fluorescent intensity of HSCs on 1kPa,C4TC and 25kPa, C1FN respectively f) Representative Image for the Collagen 1 immunostain for NC and CEBPb KD HSCs on 1kPa, C4TC, DMSO. Scale Bar 100 microns i) Representative Image for the Collagen 1 immunostain for NC and HSD KD HSCs on 25kPa, C1FN, DZnep. Scale Bar 100 microns

CEBPb and HSD11B1 knockdown resulted in decreased H3K9Me/Ac under specific microenvironmental conditions

The CEBPb and HSD11B1 gene knockdowns were also evaluated for various epigenetic changes as a function of the combinatorial microenvironment. The heatmap with quantified intensity of H3K9Me as a function of the gene knockdowns and the combinatorial microenvironment is shown in Figure 8a. A significant decrease in H3K9Me presence was observed with both the CEBPb KD and HSD KD cells on 1kPa and C4TC (Fig. 8b,c). For cells on 25kPa and FNLN, a significant decrease in H3K9Me was observed only for HSD KD cells (Fig. 8d,e). Further, the decrease was enhanced by DZnep treatment for cells on 1kPa and 25kPa. The heatmap with quantified intensity of H3K9Ac as a function of the gene knockdowns and the combinatorial microenvironment is shown in Fig 8h. A significant decrease in H3K9Ac presence was observed with only for HSD KD cells on 1kPa, C4TC and 25kPa, FNLN without any drug treatment. However, for both ECM-stiffness combination, a significant decrease in H3K9Ac was also observed for the CEBPb KD cells with DZNep and further decrease for the HSD KD cells compared to control (Fig. 8f,g,i,j). Lastly, nuclear area was significantly affected by the gene knockdowns. With the HSD KD HSCs, a significant reduction in nuclear area was observed on both stiffness, however significantly more reduced on the 1kPa compared to 25kPa (Supplemental Fig. 8).

Figure 8: Quantification of H3K9Me and H3K9Ac as a function of HSD11b1 and CEBPb knockdown.

a) Heatmap of the H3K9Me fluorescence intensity quantified in HSCs on the combinatorial microarrays b,) Box Plot of quantification of Lox fluorescent intensity of HSCs on 1kPa, C4TC and 25kPa, FNLN respectively c) Representative images for the H3K9Me immunostain for NC and CEBPb KD HSCs on 1kPa, C4TC, DZnep. Scale Bar 100 microns e) Representative Image for the H3K9Me immunostain for NC and HSD KD HSCs on 25kPa, FNLN, DZnep. Scale Bar 100 microns h) Heatmap of the H3K9Ac fluorescence intensity quantified in HSCs on the combinatorial microarrays g,j) Box Plot of quantification of H3K9Ac fluorescent intensity of HSCs on 1kPa,C4TC and 25kPa, FNLN respectively f) Representative images for the H3K9Ac immunostain for NC and CEBPb KD HSCs on 1kPa, C4TC, DZnep. Scale Bar 100 microns i) Representative images for the H3K9Ac immunostain for NC and HSD KD HSCs on 25kPa, FNLN, DZnep. Scale Bar 100 microns

Discussion

Epigenetic control, such as DNA methylation of HSCs during various stages of NAFLD and subsequent liver fibrosis has been widely explored [19]. In this work, we aimed to delineate the effects of extracellular matrix and substrate stiffness in the microenvironmental control of epigenetics and fibrogenicity of pre-activated HSCs. Based on the higher fibrogenic phenotype of pre-activated HSC when transitioned to soft compared to stiffer substrates in our previous investigation [16], we were motivated to examine the epigenetics of HSCs activation as a potential mechanism. High methylation and acetylation of H3K4 and H3K9 was quantified of HSCs on 1kPa compared to 6kPa and 25 kPa. Specific ECM regulation for all the markers was also found. Based on these cellular epigenetic readouts, we chose 3 ECMs, Collagen 1, Fibronectin; Collagen 4, Tenascin C and Fibronectin, Laminin and 2 stiffness, 1kPa and 25kpa for subsequent sequencing studies. ATACseq revealed higher chromatin accessibility in HSCs on 1kPa compared to 25kPa. Additionally, multiple regions were found to be differentially accessible based on the underlying ECM and stiffness. Promoter regions of CXCL8 and HSD11b1 were among the highly more accessible regions in HSCs on 1kPa compared to 25kPa. The differential accessibility of multiple regions led to differential expression of genes in HSCs based on the underlying ECM and stiffness. GSEA analysis revealed higher enrichment of gene expression of HSCs on 1kPa to NAFLD and fibrosis related genes compared to 25kPa. Further, combined analysis of ATACseq and RNAseq helped validate motif analysis results. Ultimately, two candidate genes HSD11b1 and CEBPb were chosen for further investigation of the mechanism of higher fibrogenicity of HSCs on 1kPa. Transient knockdowns created using siRNA treatment demonstrated microenvironmental controlled reduction in fibrogenic markers in HSCs.

The high-throughput microarray technology was utilized throughout these studies in different platform configurations to understand the combinatorial effect of various microenvironmental factors. Initial studies utilized 8 ECM combinations and 3 stiffnesses with immunostaining and nuclear morphology quantifications. These cell microarray-based analyses provided insights into the complex epigenetic landscape of the pre-activated HSCs as a function of the microenvironment, especially the stiffness. The second array configuration involved optimizing microarray technology for sequencing analysis, where it facilitated a controlled microenvironment, and allowed for efficient and good quality sample collection. The sequencing experiments acted as a further window into specific microenvironmental conditions where higher chromatin accessibility and subsequent gene expression highly enriched to NAFLD and fibrosis related genes was observed in HSCs on 1kPa compared to 25kPa. We also found how specific ECM conditions modulated both accessibility and expression of multiple genes. Lastly, microenvironmental control of drug treatment and 2 gene knockdowns was studied simultaneously using the coverslip-based microarrays, demonstrating the endless and efficient uses of this system.

ECM composition, a major component of the HSCs microenvironment, is constantly remodeling and restructuring during chronic liver disease. The ECM imparts both mechanical and biochemical signals to the cells. Collagens are a major fibrillar protein in the ECM that represents approximately 30% of the total protein content in the human body. Excessive deposition of Collagen 1, with upto a 6-fold increase in Collagen 1 accumulation is associated with chronic liver diseases [48]. Additionally, there are various non-collagenous protein in the ECM such as fibronectins, laminins and tenascins that interact with HSCs via cell membrane receptors such as the integrins, influencing cellular behavior [48]. We have previously reported multi-spectrum effect of ECMs on HSC Collagen 1 expression and proliferation [16]. In this study, we further demonstrated the role of chromatin accessibility and epigenetic modifications in these ECM mediated fibrogenic changes in HSCs. The HSCs demonstrated high methylation and acetylation on both H3K9 and H3K4 positions, the ECM combination specifically influenced methylation at H3K9 position on the 1kPa substrates. Collagen 1, Fibronectin on 1kPa exhibited significantly higher quantification of H3K9Me compared to Collagen4, Tenascin C on 1kPa. Furthermore, ECM combinations on 6kPa and 25kPa significantly influenced histone modification quantification at multiple histone positions. This demonstrated the complexity of the microenvironment in influencing cellular behavior, where unique combinations of both the ECM and substrate stiffness resulted in differential histone modifications.

For both the sequencing modalities, stiffness regulated a high number of regions/genes. We also noted a comparatively smaller number of genes significantly regulated by the underlying ECM. However, there were some interesting trends observed. From the ATACseq analyses, Collagen 1, Fibronectin was the ECM combination that was found to be most divergent relative to others on 25kPa stiffness, whereas Fibronectin, Laminin was most divergent in the context of 1kPa substrate stiffness. RNAseq analyses revealed similar trends, in which stiffness led to more than 6000 differentially expressed genes, compared to the role of ECM composition in the differential expression of 650 genes. The combination of RNAseq and ATACseq modalities enabled the correlation of gene expression variations with respective changes in chromatin accessibility. Hence, regions that were found differentially accessible were first mapped to a gene based on gene proximity. A common gene subset was subsequently defined based on the cross-comparison between differential expression and accessibility with the defined ECM composition and stiffness conditions. This intersection was greater than 70% for the genes regulated by stiffness. However, for the ECM regulated genes, this intersection was less than 15% of the total genes regulated by the ECM, signifying the complexity of the effect of the ECM compositional interactions. It has been shown that different levels of gene accessibility could lead to different levels of expression [49], which could be a reason why this common intersection varied for different microenvironmental comparisons. The common intersection gene list facilitated additional assessment using a number of publicly available databases (> 50 databases using the EnrichR tool). In addition to chromatin accessibility assessment, we performed motif analysis on selected regions found in the ATACseq data analysis. This enabled extraction of additional information of transcription factors binding to these regions and affecting the gene expression of the nearby gene. The motif sequences found were aligned to known motifs database and the RNAseq data became useful in verifying the potential findings from the motif analysis.

The high-throughput combinatorial studies to study the transient knockdown of the two candidate phenotypic regulator genes, HSD11b1 and CEBPb, with and without the drug DZnep, a methyl transferase EZH2 inhibitor, revealed both stiffness and ECM dependent reduction in Lox expression, Collagen 1 expression, H3K9Me, H3K9Ac and changes in cellular and nuclear morphology. Within the stiff substrate conditions, a substantial reduction in all of the phenotypic and epigenetic markers that we tested was only observed with gene knockdown in combination with DZnep treatment. On soft substrates, the gene knockdowns exhibited a strong effect towards reducing the phenotypic and epigenetic markers, which was further enhanced by DZNeP treatment. In addition, ECM composition acted in a cooperative manner to modulate the HSC phenotypic responses to both gene knockdown and small molecule inhibitors. HSD11b1 encodes a dehydrogenase protein that plays an important role in cortisol metabolism in the liver [42]. Genetic variations in the gene HSD11b1 has been linked to NAFLD, visceral obesity [50] and targets for HSD11b1 have explored in clinical trials [43]. C/EBPb is a transcription factor that has a major role in the liver maintaining liver homeostasis and managing stress response [44, 51] and elevated Collagen 1 and IL-6 expression in HSCs has been linked to C/EBPb activity [52, 53]. Furthermore, it was previously found that the phosphorylation of C/EBPb and its subsequent translocation to the nucleus was critical for the progression of liver fibrosis in a mice model [54]. Most studies evaluating the role of HSD11b1 and C/EBPb in the liver have studied the whole liver collectively, and have not yet systematically tested the role of these genes, and potential gene-targeting therapies, in hepatocytes versus various non-parenchymal cells. Our study provides additional evidence and motivation for targeting both these proteins in hepatic stellate cells for therapeutic application, specifically pointing towards a microenvironment specific role, where the specific ECM composition and stiffness of the liver might influence the overall outcome. For future studies, the incorporation of unactivated (not previously cultured) HSCs or earlier passages of HSCs, as well as a broader range of stiffnesses could be employed to further refine the dynamics of HSC phenotypic alterations, including the contributions of genes such as HSD11b1, C/EBPb, and other candidates from our studies in this process.

There have been multiple studies of fibrosis in different tissues and a concept of persistence of activated fibroblasts once they are exposed to a high stiffness environment has begun to emerge [55–57]. Reversal of activation of valvular fibroblasts on soft hydrogels was found to be time-dependent on the prior culture of fibroblasts on stiff hydrogels [58] and this persistence was dependent on chromatin remodeling [56]. Dysregulated inflammation regulated by activated fibroblasts in response to constantly changing microenvironment has been attributed to chronic fibrosis in multiple contexts [59]. Our results support this paradigm of persistence of activated HSCs having a myofibroblast-like phenotype in the liver. Activation markers in HSCs cultured on tissue culture plastic were also compared to HSCs cultured on Fibronectin coated 2kPa and 25kPa substrates which demonstrated higher expression of various activation markers (Col1a1, TIMP1 and Acta1) on 2kPa substrates (Supplementary Figure 9). Collectively, we have observed enhanced ECM-dependent fibrotic response of pre-activated HSCs on soft substrates compared to stiff substrates, mediated by chromatin remodeling. Additionally, our results highlighted the multifaceted nature of the pre-activated HSCs and how simultaneous targeting of numerous targets could be more beneficial for treatment of fibrosis in NAFLD. Furthermore, the high-throughput microarray technology could be useful in personalized medicine applications, where the presence of a unique combination of ECMs and underlying levels of epigenetic markers in a patient’s biopsy sample could be a guiding factor for therapeutic selection.

Statement of Significance.

Hepatic stellate cells (HSCs) are one of the primary drivers of liver fibrosis in non-alcoholic fatty liver disease. Although HSC activation in liver disease is associated with changes in extracellular matrix (ECM) deposition and remodeling, it remains unclear how ECM regulates the phenotypic state transitions of HSCs. Using high-throughput cellular microarrays, coupled with genome-wide ATAC and RNA sequencing within engineered ECM microenvironments, we investigated the effect of ECM and substrate stiffness on chromatin accessibility and resulting gene expression in activated primary human HSCs. Overall, these findings were indicative of a microenvironmental adaptation response in HSCs, and the acquisition of a persistent activation state. Combined ATAC/RNA sequencing analyses enabled identification of candidate regulatory factors, including HSD11B1 and CEBPb. siRNA-mediated knockdown of HSD11b1 and CEBPb demonstrated microenvironmental controlled reduction in fibrogenic markers in HSCs.