Tgfb3 and Mmp13 regulated the initiation of liver fibrosis progression as dynamic network biomarkers

Abstract Liver fibrogenesis is a complex scar‐forming process in the liver. We suggested that the liver first responded to chronic injuries with gradual changes, then reached the critical state and ultimately resulted in cirrhosis rapidly. This study aimed to identify the tipping point and key molecules driving liver fibrosis progression. Mice model of liver fibrosis was induced by thioacetamide (TAA), and liver tissues were collected at different time‐points post‐TAA administration. By dynamic network biomarker (DNB) analysis on the time series of liver transcriptomes, the week 9 post‐TAA treatment (pathologically relevant to bridging fibrosis) was identified as the tipping point just before the significant fibrosis transition, with 153 DNB genes as key driving factors. The DNB genes were functionally enriched in fibrosis‐associated pathways, in particular, in the top‐ranked DNB genes, Tgfb3 negatively regulated Mmp13 in the interaction path and they formed a bistable switching system from a dynamical perspective. In the in vitro study, Tgfb3 promoted fibrogenic genes and down‐regulate Mmp13 gene transcription in an immortalized mouse HSC line JS1 and a human HSC line LX‐2. The presence of a tipping point during liver fibrogenesis driven by DNB genes marks not only the initiation of significant fibrogenesis but also the repression of the scar resolution.


| INTRODUC TI ON
Liver fibrosis is a common scar-forming process in the liver in response to various chronic insults. It is not an independent disease but a conceptually important stage of liver cirrhosis, during which the fibrotic scar is generally considered more soluble, and normal liver function and structure are likely preserved after effective treatment. The fibrogenic process is often triggered and accompanied by cell necrosis and inflammatory infiltration, which subsequently activate inflammatory and fibrogenic signalling cascades. 1,2 With injury and fibrosis progression, the normal anatomical lobules of the liver are gradually replaced by architecturally abnormal nodules separated by fibrous tissue, leading to cirrhosis and eventually functionally decompensated cirrhosis with fatal complications. 3 This occurs when the treatment is limited and overall prognosis is poor. Hepatic fibrosis itself and cirrhosis in compensated phase not only are lack of accurate non-invasive serum biomarkers but also have no distinct symptoms. The Metavir staging systems have been used for histological assessment of chronic liver diseases on the base of liver biopsy. Significant fibrosis refers to a METAVIR score of F2 or greater, whereas METAVIR F4 denotes cirrhosis. This is likely to have important implications especially in the early phases of the fibrogenic process when the fibrosis regression is higher because of the absence of significant changes in the tissue architecture. 4  Dynamic analysis based on high-throughput data provides new concepts and methods for identifying early warning signals of complex diseases, including liver fibrosis/cirrhosis and revealing basic mechanisms of disease initiation and progression at the network level. A novel model-free method based on nonlinear dynamic theory, termed dynamic network biomarker (DNB), has been developed to characterize the tipping point just before the critical transition during the progression of complex diseases. 6,7 Based on DNB theory, for many chronic diseases, there existed a sudden shift during the process of gradual health deterioration that results in a drastic transition to a disease state. In other words, the progression of a disease is divided into three phases, that is normal, pre-disease (tipping point or critical state) and disease states. [8][9][10][11][12][13][14] Unlike the normal and disease states, in which the complex systems are usually robust to perturbations in such stable states, a pre-disease state has low resilience and robustness. After crossing pre-disease state, it becomes difficult to return to the normal state because of state stability. Thus, to prevent disease deterioration, identifying such pre-disease state or tipping point is of great importance on both preventive and predictive medicine. This can be further exploited to detect the driving molecules or DNB genes during disease progression. DNB genes are a group of molecules with strong collective fluctuations, appearing only at the 'tipping point' of a homeostatic system, thereby considering as the key factors to drive the critical transition. In particular, the standard deviations (SD) and Pearson's correlation coefficients (PCCs) of DNB genes are both drastically increased at the tipping point just before the transition during the disease progression. A quantitative index, termed composite index (CI), for the DNB genes can be derived for quantifying the tipping point as well as its driving molecules. 9,10 The DNB method has been successfully applied to real-world biological data to identify the early warning signals of the sudden deterioration of several complex diseases such as cancer and metabolic disease. [15][16][17] Progression of liver fibrosis is generally a complex process, triggered by many related genes. Dysfunction of these genes is involved in a wide range of biological mechanisms, which can be used for characterizing the tipping point of critical transition from liver damage to liver fibrotic stage. This study aims to discover the tipping point with its key genes driving liver fibrosis progression by DNB analysis on liver transcriptomes. In addition, we further analyse the switching mechanism of the liver fibrosis by constructing its bistable model of key genes during the dynamic process.
Clearly, the results will help not only to reveal the molecular mechanism of liver fibrosis/cirrhosis but also to identify new potential therapeutic targets and molecular markers for non-invasive diagnosis of liver fibrosis.

| Animal models of hepatic fibrosis and preparation of time-series liver sample
The study protocol was approved by the committee on the use of  The assessment of liver fibrosis stages as judged by histological descriptions of METAVIR staging system were marked below the figures. Liver specimens of mice with TAA treatment for 9 weeks shown bridging fibrosis and were relevant to METAVIR stage 3. Magnification was 10× sequence reads, we performed sequence alignment using hierarchical indexing for spliced alignment of transcripts (HISAT) for mapping RNAseq reads to the reference (https://www.ncbi.nlm.nih.gov/assem bly/ GCF_00000 1635.26). 18 The raw expressed read counts of each sample at each time-point were then extracted by HTseq count program, which is a Python library to facilitate rapid development of scripts for processing and analysing high-throughput sequencing data. 19 Differentially expressed genes (DEGs) were calculated by DEseq2 for all adjacent time-points of mice (TAA-treated mice and normal control mice) and TAA-treated mice versus normal control mice at the same time-point, with a threshold of fold change > 2 or < 0.5, and false discovery rate (FDR) < 0.05. 20

| Dynamic network biomarker (DNB) analysis
To detect the tipping point and its leading genes in the TAAinduced animal model during liver fibrogenesis, we used DNB analysis methodology, 10,11 which can be implemented in timeseries high-throughput omics data such as RNA sequencing. where PCC i is the average PCC (absolute value) of the dominant group, PCC o is the average PCC (absolute value) between molecules in the dominant group and outside the group, and SD i is the average standard deviation of the dominant group. According to the DNB theory, the CI is expected to increase sharply once the biological system approaches a critical period or tipping point. Thus, it can serve as an effective early warning signal to identify the pre-disease state as well as the DNB genes.

| Functional analysis
Functional enrichment analysis aims to identify classes of molecules (genes or proteins) that are over-represented in a set of predefined molecules and to predict its association with disease phenotypes. In this study, we performed this method to uncover potential fibrosisassociated biological functions affected by the identified molecules by mapping the molecules into known molecule sets with functional annotation. Two databases, KEGG and Ingenuity Pathway Analysis (IPA, a database that contains millions of documented and published molecular interactions; http://www.ingen uity.com) were used for canonical pathway detection. 21 Enrichment of specific molecules in each biological process or pathway was estimated. Significantly enriched functions were chosen if the corresponding p-value was below a threshold (0.05). Significant DNBs in DEGs were ranked based on their enrichment in IPA and KEGG pathways associated with liver fibrogenesis.

| Histological staining and fibrotic staging
Liver histology and fibrosis stages were observed by haematoxylin and eosin (H&E), Masson's trichrome and reticulin staining. Formalinfixed liver tissues were embedded in paraffin. Morphological examination was performed with H&E staining for histological assessment of the progression of liver fibrosis. Masson's trichrome and reticulin staining was used to demonstrate fibrotic changes and collagen deposition.

| Cell Culture
An immortalized wild-type mouse HSC line JS1 and a human HSC line LX-2 were used to test the profibrogenic function of Tgfb3. 22,23 The cells were maintained in Dulbecco's Modified Eagle's Medium

| Real-time quantitative polymerase chain reaction (RT-qPCR)
Total RNA was isolated from liver tissues or cells using TRIzol rea-

| Construction of positive feedback loop between Tgbf3 and Mmp13 on collagen accumulation
Tgfb3 and Mmp13 have been found to inhibit each other based on the observed data as well as the available interaction information (Supplemental Figure S1). We next constructed the following dimensionless model for the network of Tgfb3 and Mmp13 similar to the toggle model: 24,25 Here, x and y represent the mean expression level of Tgfb3 and Mmp13, respectively. 1 and 2 are the effective rate of synthesis for Tgfb3 and Mmp13, while 1 and 2 are the cooperativity factors.
Obviously, it is a positive feedback loop as a result of two consecutive negative interactions. Phase diagram was built by solving the above equations, and clearly bistability occurred ( Figure 6A2 and 6A3). The parameters α and β were determined by appropriate values based on the ranges of available data and simulation.

| Animal model of liver fibrosis
Images

| Gene expression profiling
After multiple comparisons, 3602 differentially expressed genes

| DNB analysis identifies the tipping point just before liver fibrosis transition
Based on the DNB theories, 10,11,27 the progression of a specific disease was divided into three phases as follows: normal, pre-disease (tipping point or critical state) and disease states ( Figure 3A). To accurately determine the critical stage of fibrosis progression in TAA-induced animal

| Functional analysis revealed that fibrosisassociated pathways are influenced by DNBs
As week 9 in the experimental model of TAA-induced liver fibrosis was determined as the tipping point just before significant fibrosis transition, DEGs were calculated before and after week 9 using DEseq2. 729 differentially expressed genes were identified and 52 of them existed in DNBs ( Figure 4A, Supplemental Table S1). Functional analyses were then performed on DEGs and DNBs separately. Two

| Tgfb3 plays a key role in the progression of liver fibrosis based on DNB ranking
To uncover which of the 12 candidate DNBs was more important in fibrosis progression, we ranked the DNB genes based  Figure 5A. Tgfb3 ranked first, followed by up-regulated DNB. This gene encoded TGF-β3, a cytokine belongs to TGF-β superfamily that is involved in ECM formation and wound healing. 32

| Functional analysis of Tgfb3 on fibrogenic gene expression in HSCs
Mmp13 was also found to be a key DNB for the critical transition of fibrogenesis. This gene encoded collagenase 3 that is known to play a role in ECM protein degradation. 33,34 To further delineate the interaction within Tgfb3 and Mmp13, and type I collagen that is the main substrate of Mmp13, all fibrosis-related pathways from KEGG and IPA for Tgfb3 were collected ( Figure 5B) for analysing pathway-based interaction between them ( Figure 5C). By searching the whole interaction networks in IPA, all paths that started with Mmp13 and ended RT-qPCR analyses revealed that the Tgfb3 expression in the liver of TAA-treated mice increased after TAA administration and first peak at 9 weeks, whereas the increase of Tgfb1 expression was found to lag behind Tgfb3 and peaked at 17 weeks post-TAA administration ( Figure 6A1).
The functional influence of Tgfb3 on fibrogenic gene (eg Mmp13, collagen type 1, CTGF) expression was further investigated by an in vitro study with an immortalized mouse HSC line JS1. As shown in Figure 6B Figure S2).

| Bistable model of Tgfb3 and Mmp13 during fibrotic process
Based on the theoretical model for network, the phase map on the regulation between Tgfb3 and Mmp13 during fibrogenic process was generated, which is clearly a bistable system from a viewpoint of dynamics ( Figure 6A2, Supplemental Figure S3). The steady (stable) state 1 is the normal stage with relatively higher expression of Mmp13 and lower expression of Tgbf3 ( Figure 6A3), so that collagen can be removed properly. When environmental burden (ie TAA stimulation) exists, the production of Tgfb3 increases and collagen accumulates as a scar repairing response. As the injury further continues, the production of Tgfb3 would pass the tipping point (near unstable steady state) and eventually reach another stable state (steady state 2) ( Figure 6A3), where the expression and activity of Mmp13 inversely decrease and the fibrillar collagen would be unremovable.

| D ISCUSS I ON
In the present study, DNB analysis was applied to our time series of liver transcriptomes of mice with TAA-induced liver fibrosis to identify the critical state or tipping point during liver fibrogenesis process, which was found to be at week 9, pathologically relevant to METAVIR stage 3, that is bridging fibrosis, with a set of DNB genes that may exert critical function driven by significant activation of fibrogenic signalling pathways at a network level. This result was in line with a previous report that an increase in METAVIR stage is associated with a progressive increase in the fibrosis area and the increase of fibrous tissue accumulation is not linear. 35 Less reversibility was found in late stages of fibrosis as evidences obtained from experimental studies and clinically liver biopsied patients. 36,37 Significant increase and accumulation of fibrotic tissue happens in stage 3 during fibrogenesis. The treatment of high fibrosis staging requires 'dissolution' of more fibrous tissue; hence, it is more difficult. The results indicated that clinically reliable diagnosis before stage 3 and early intervention are important to achieve a complete fibrosis reversion.
In this study, 153 DNB genes were obtained, of which 52 were   50 and attributed a key role in the pathogenesis of glioblastoma. 47 Up-regulation of Tgfb3 expression was found in cirrhotic livers of primary biliary cirrhosis patients 51 and in CCl4-treated rat livers. 52 Higher Tgfb3 level in patients without non-alcoholic fatty liver diseases (NAFLD) was associated with higher chance of future development of NAFLD. 53 In Con A-induced autoimmune liver-damaged mice, abnormal HSC activation was accompanied by imbalance of Tgfb1 and Tgfb3 expressions in the early stage, which can line immune and hepatic fibrosis processes, leading to further switching and progressing of liver fibrosis. 54 In this study, XDB38040400). The authors also wish to thank professors Scott L.
Friedman of Mount Sinai hospital, New York, the USA, and Derek A.
Mann of Newcastle University, England, for their helpful discussion on the study.

CO N FLI C T S O F I NTE R E S T
The authors declare no conflict of interest.