Matrices with various stiffness were prepared as previously described [5] [13]. The following experimental procedures were performed in a sterile environment. Briefly, 40% (w/v) acrylamide (Sangon Biotech, China) and 2% (w/v) bis-acrylamide (Sangon Biotech, China) were mixed in different ratios, and crosslinked with tetramethylethylenediamine (TEMED, Klamar, China) and 10% ammonium persulfate solution (Table 1). Before polymerization, 100 μL mixture was added onto a glass plate pre-treated with gel slick (Lonza, Switzerland), and a circular silanizedcoverslip with a diameter of 20 mm was put onto the gel solution. After 20 min of polymerization, the coverslip with the polymerized matrix was pried up from the glass plate, and the matrix on the coverslip was rinsed with 2 mL 50 mM HEPES (Thermo Fisher, USA) for 3 times. Then the matrix was treated with Sulfo-SANPAH (Thermo Fisher, USA) and coated with 1 mg/mL Collagen I (Corning, USA). Finally, the matrix was sterilized with 70% (v/v) ethanol solution for 15 min, and washed with Dulbecco’s Phosphate-Buffered Saline (DPBS). The prepared matrices were stored in a 4˚C refrigerator for up to about 2 weeks and UV sterilized for 30 min before cell culture.

Human umbilical vein endothelial cells (HUVECs) were purchased from ScienCell Research Laboratories (ScienCell, USA) and cultured to passage 6 for experiments. HUVECs were cultured with EC medium supplemented with 5% fetal bovine serum, 1% EC growth factor and 1% penicillin/streptomycin (Sciencell, USA). In all experiments, HUVECs were cultured on matrices with different stiffness for more than 24 hours.

Total RNA was extracted using the TRIzol plus RNA Purification kit (Thermo Fisher Scientific, USA), and reverse transcribed using the FastQuant RT kit (with gDNase) (TIANGEN, China). The cDNA samples were analyzed by RT-qPCR (Thermo Fisher Scientific, USA) using SuperRealPreMix Plus (SYBR Green) (TIANGEN, China) and the specific primer pairs (Table 2). GAPDH was used as a normalization control.

Total RNA for transcriptome sequencing was extracted using the TRIzol reagent according to the manufacturer’s protocol. RNA purity and quantification were

evaluated using the NanoDrop 2000 spectrophotometer (Thermo Scientific, USA). RNA integrity was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Then the libraries were constructed using TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions. The transcriptome sequencing and analysis were conducted by OE Biotech Co. Ltd. (Shanghai, China).

The libraries were sequenced on an Illumina HiSeq X Ten platform and 150 bp paired-end reads were generated. Raw data (raw reads) were processed using Trimmomatic [19]. The reads containing ploy-N and the low-quality reads were removed. The clean reads were mapped to the human genome (GRCh38) using HISAT2 [20]. FPKM [21] of each gene was calculated using Cufflinks [22], and the read counts of each gene were obtained by HTSeq-count [23]. Differential expression analysis was performed using the DESeq (2012) R package [24]. P value < 0.05 and fold change > 2 or fold change < 0.5 was set as the threshold for significantly differential expression. Hierarchical cluster analysis of differentially expressed genes (DEGs) was performed to demonstrate the expression pattern of genes in different groups and samples. The identified DEGs were used subsequently for Gene Ontology (GO, http://geneontology.org/) and Kyoto Encyclopedia for Genes and Genomes (KEGG, http://www.genome.jp/kegg/) pathway analysis, which performed respectively using R based on the hypergeometric distribution.

In order to explore the interactions between DEGs, a protein-protein interaction (PPI) network was built using the online analysis database STRING (http://string-db.org/). Cytoscape (http://www.cytoscape.org/) software was used to visualize the network.

To explore the role of the TFs in mechanotransduction, we investigated the distribution of TF families. The relationship between TFs and their target genes was studied based on the hTFtarget database (http://bioinfo.life.hust.edu.cn/hTFtarget#!/tf).

Graph Pad Prism 6.0 software was used for statistical analyses and graphing. For all bar graphs, data were expressed as means ± SD. Differential analyses of mRNA expression between different matrix stiffness groups were carried out using Student’s t-test. Statistically significance was indicated by p < 0.05.

Samples were categorized by matrix stiffness into soft matrix group (1 kPa) and stiff matrix group (40 kPa), and were sequenced on an Illumina HiSeq X Ten platform. After removing the adapter sequences and discarding low-quality reads, the amount of data of each sample was distributed in the range of 6.22 - 7.09 G, the Q30 base was distributed in the range of 93.27% - 94.02%, and the average GC content was 50.57%. By aligning reads to the reference genome, the mapping rates of all samples were greater than 95%. In summary, the data acquired in this study were eligible to be used for later analysis (Table 3).

Principal component analysis (PCA) was performed to show the linear combination of the original high-dimensional data after dimensionality reduction [25]. In this study, PCA results showed that the principal component 1 (PC1) and principal component 2 (PC2) had 69.5% and 16.5% variations, respectively, which indicated that there were obvious differences between different matrix stiffnesses and that the same matrix stiffness had good repeatability. Based on this, the samples were distinguished into 2 groups of different dimensions (Figure 1(a)). The RNA-seq results showed that gene expression was significantly different between the two groups. Therefore, the heatmap of the gene expression profile based on different gene expression values between the two groups was acquired (Figure 1(b)). Totally, there were 1775 DEGs identified, among which 450 DEGs were up-regulated and 1325 DEGs were down-regulated in HUVECs cultured on soft matrix (1 kPa) compared to those cultured on stiff matrix (40 kPa) (Figure 1(c)).

Based on DEGs, GO analysis was conducted to evaluate the biological significance of matrix stiffness on ECs. In this study, we counted the number of differential genes in each GO term, and used the hypergeometric distribution algorithm to calculate their significance. We found that enriched GO terms were within three main categories: biological process, molecular function, and cellular component.

In the biological process category, most enriched genes were related to positive regulation of transcription by RNA polymerase, suggesting that changes in

matrix stiffness would exert extensive impact on many aspects of the ECs. In the molecular function category, most DEGs enriched among protein binding, ATP binding and metal ion binding. In the cellular component category, most DEGs were associated with nucleoplasm, cytosol, nucleus, centriole and membrane (Figure 2(a)).

Moreover, we also compared the gene distribution of GO level 2 containing 64

terms (Figure 2(b)). In each of the above category, the enriched terms were associated with cell-cell interaction and cell-ECM interaction, including biological adhesion and cell junction. In the molecular function category, binding showed distinct expression differences. These results implied that changes in matrix stiffness can affect cell junction, which in turn may disrupt vessel integrity and endothelial barrier function.

KEGG pathway enrichment analysis was performed to further explore the signaling pathways involved in mediating the effects of matrix stiffness on cells. In this study, we show the top 20 signaling pathways with the highest enrichment scores after screening (Figure 3(a)). And, the gene expression involved in each pathway is shown in Figure 3(b). Among these pathways, focal adhesion and regulation of actin cytoskeleton, which play an important role in cell-ECM interactions and cell-cell communications, may indicate the potential targets of mechanotransduction [26] [27]. In addition, DEGs were also enriched in intercellular junctions (including adherens junction, tight junction and gap junction) and endocytosis, which play an important role in endothelial cell barrier function [28].

TFs play an indispensable role in regulating gene expression, are ubiquitous in all organisms and differences in their expression may provide important clues for understanding mechanotransduction. Therefore, the distribution of differential TF families was counted followed by their target genes (Figure 4). By comparing the distribution of TF families (up- and down-regulated), we found clear differences between the two groups (Figure 4(a)). Among these TF families,

zf-C2H2, chromosome remodeling factor, bHLH and ARID, are involved in mechanotransduction [29] [30] [31]. In order to further study the regulation of TFs on gene expression, we extracted the differential target genes corresponding to differential TFs according to the relationship list between TFs and target genes, and screened out the top 25 differential TFs (Figure 4(b)). In addition, we also found that some transcription factors like BRD4, BRD2, ARID1A and ARID2, which have been shown to be involved in mechanotransduction, were associated with a large number of DEGs [30] [32] [33].

PPI analysis was performed based on the top 35 DEGs. In order to explore the interactions between DEG, we established a PPI network using an online tool STRING database (https://stringdb.org/), and then used Cytoscape (http://www.cy-toscape.org/) software to visualize the network (Figure 5). We

used node size to represent the fold of differential expression, color to distinguish up- or down-regulated genes, and line width to indicate interaction score (>0.7). In this network, some genes encoding ECM proteins, which play a considerable role in maintaining the integrity of ECM, such as fibronectin 1 (FN1), collagen α2(IV) chain (COL4A2), matrix metalloproteinase 2 (MMP2) and matrix metalloproteinase 14 (MMP14), have obvious expression differences and interactions. Moreover, cadherin 5 (CDH5), which plays a critical role in endothelial adherens junction assembly and maintenance, presents the same variation. In addition, ITGA5 and ITGB5, which encode integrin subunits that mediate cell-ECM interactions and function as mechanosensor, have the same results.

Based on the above bioinformatics analysis results, we found that the genes encoding some ECM proteins such as FN1 and COL4A, some genes encoding junction proteins such as CDH5 and F11R, the genes mediating cell-ECM communication such as ITGA5 had obvious differences in different matrix stiffness. ECM maintains the structural integrity of tissues and regulates cell and tissue functions. Dysregulation of the ECM composition and structure can lead to several human diseases [34]. The integrity of cell junctions is the basis for cells to function as a barrier. Therefore, six genes (including CDH5, F11R, FN1, COL4A2, MMP14 and ITGA5) were subject to PCR validation. As shown in Figure 6, the expression of these genes was reduced on soft matrix (1 kPa), which is consistent with the RNA-seq results. These results suggested that matrix stiffness may affect the endothelial cell barrier function as well as the structural integrity of ECM.