Hepatic kinome atlas: An in‐depth identification of kinase pathways in liver fibrosis of humans and rodents

Human subjects

All studies using human samples were approved by the Human Subjects Committee of the University of Kansas Medical Center.

Animal experimentation

All animal procedures were approved by the Cleveland Clinic Institutional Animal Care and Use Committee (IACUC) for the CCl₄ administration to mice and mouse feeding with choline‐deficient L‐amino acid–defined diet. The mouse fed with high fat supplemented with fructose diet were approved by the IACUC committee at the University of Kentucky.

Statistical and correlation analyses

The kinome array and upstream kinase identification, interaction network analysis, pathway and annotation enrichment analysis, and correlation analysis are described further in the Supporting Methods. The Student’s t tests were performed on quantitative real‐time PCR data to detect statistical differences between control and liver fibrosis groups, with p values adjusted for multiple testing corrections within each human and mouse using the Benjamain‐Hochberg method.^{[ 15 ]} Adjusted p alpha (false discovery rate) ≤ 0.05 was considered statistically significant.

All of the procedures, methods, and analysis are described in more detail in the Supporting Methods.

Interrogation of kinase pathways in humans and mice with fibrosis

To identify kinases that might drive fibrotic pathways among humans and rodents, we quantified the activity of 196 tyrosine (protein tyrosine kinase [PTK]) peptide substrates and 144 serine/threonine kinase (STK) peptide substrates in liver samples from humans with fibrosis/cirrhosis or healthy control and livers from rodents with fibrosis or healthy control. To visually represent the kinase activity demonstrated in the phosphorylation level of these 340 kinase peptide substrates that were changed with fibrosis relative to their healthy control, and an upstream analysis performed with the data deconvoluted into a kinase tree for humans (Figure 1A) and mice (Figure 1B). The higher mean final kinase scores determine a more substantial role in the observed phosphorylation changes between fibrosis and normal control samples. Mean final scores for all identified PTK and STK kinases were mapped according to paralogy. The kinase activity results were charted as bubble plots with node size based on Z scores in red (highest changed Z, mean node = 4), pink (mean node = 2), or gray (mean node = 1) circles to visually indicate potential clusters of kinase families that are changed with fibrosis in humans or rodents. The kinase tree data show significant clusters for the tyrosine kinase (TK) and the serine/threonine AGC kinase families. There was also some clustering in the CAMK and CMGC kinase families but to a lesser extent. The Tyrosine Kinase‐Like, STE, Cell Kinase 1, and atypical kinase families were primarily unchanged, but some specific targets in these families were changed. Next, we systematically analyze each kinase family and subfamilies to determine which pathways change most in humans and rodents with fibrosis.

Concordance between human and mouse data

Shapiro‐Wilk normality tests and quantile‐quantile plots indicate that the kinase activity data sets are not normally distributed (human kinase family data: W = 0.84641, p value = 8.825 × 10⁻⁶, rodent kinase family data: W = 0.82763, p value = 2.822 × 10⁻⁶, human individual kinase data: W = 0.82247, p value = 1.585 × 10⁻¹², mouse kinase family data: W = 0.79478, p value = 1.446 × 10⁻¹³). Figure 2A,B shows a sample–sample correlation heatmap of log‐transformed peptide phosphorylation data from both human and mouse analytical replicates. This correlation matrix analysis shows human and mouse fibrosis groups clustering closer together (0.89 median Information‐Content‐Informed Kendall‐tau Correlation [ICI‐kt]) than either of the two humans (0.84 median ICI‐Kt control vs. fibrosis) or mouse groups (0.88 median ICI‐Kt correlation control vs. fibrosis), demonstrating the high similarity between human and mouse liver elevated fibrosis kinase activity. Furthermore, all sample–sample correlations have a p value < 2.2 × 10⁻¹⁶ (i.e., below the machine precision limit).

Next, to visually compare and quantitate the level of fibrosis in the human and mouse livers, we did sirius red and trichrome blue staining of the human cirrhotic and healthy livers and the murine livers treated with chronic CCl₄ or vehicle control. The results show that the healthy control human and vehicle control mouse livers had no fibrosis present and that cirrhotic human and chronic CCl₄‐treated mice had severe fibrosis to the level of a fibrosis score of 4.0 and 3.0, respectively (Figure 2C,D). Also, in comparison, both the human and murine livers with fibrosis had significantly higher TGF‐β mRNA expression, which is an established marker of fibrosis.^{[ 16 ]} These data indicate a closely related data concordance between human and murine fibrosis that has visual similarities among these models.

Therefore, we performed separate pathway and annotation enrichment analyses using the human and mouse differential kinase activity, respectively (Table S7). The ICI‐Kt correlation between human and mouse enrichment‐adjusted p values for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, REACTOME pathways, and Gene Ontology Biological Process annotations are 0.73 (p = 3.9 × 10⁻²⁴), 0.62 (p = 7.7 × 10⁻²⁶), and 0.78 (p = 4.3 × 10⁻¹⁶⁵), respectively (see Supporting Information for correlation plots in Figure 1 and full pathway enrichment tables in Table S6). This clearly shows that fibrotic versus control differential kinase activity in humans and mice is very similar at the pathway and biological process level. Moreover, the three most enriched KEGG pathways are the phosphatidylinositol‐3‐kinase (PI3K)–Akt signaling pathway (human‐adjusted p value = 5.9 × 10⁻¹⁷), Ras signaling pathway (human‐adjusted p value = 2.5 × 10⁻¹⁴), and EGF receptor tyrosine kinase inhibitor resistance pathway (human‐adjusted p value = 2.3 × 10⁻¹⁴). All three of these pathways have known associations with liver fibrosis.^{[ 17} , ^{18 ]}

Tyrosine kinase families changed with fibrosis

To determine TK families that might change with fibrosis, we compiled data from substrate phosphorylation events from the PamGene analysis into heatmaps (Figure 3A,B). We used bioinformatic analyses (described in Upstream Kinase Identification) from these data to identify upstream kinases responsible for the observed differential phosphorylation patterns of peptide substrates (Table S3). An aggregation of the data according to the TK family waterfall plot showed that DDR1 and INSR were the most differential active kinases (Figure 3C,D). We next compared the observed number of differentially phosphorylated peptide substrates targeted by each kinase or kinase family against 2000 random sampling iterations, selecting the same number of peptide substrates from the pool of 196 (PTK) or 144 (STK) (depending on whether TK or STK are being measured) peptide substrates found on the arrays. The number of experimentally determined differentially phosphorylated peptide substrates targeted by a given kinase or kinase family in relation to the number of randomly selected peptide substrates targeted by that same kinase or kinase family in each of the thousands of random sampling events indicates the significance of the kinase or kinase family of interest to the differential phosphorylation patterns observed between fibrosis and normal samples. These data are represented here as annotated histograms, herein referred to as peacock plots. The peacock plots for the DDR and INSR kinase families indicate significantly differential kinase activity in humans (Figure 3E) and rodents (Figure 3F). To determine whether DDR1 and INSR were changed at the gene level, we measured mRNA by real‐time PCR and compared this with the peacock plot of the substrate analysis. The mRNA expression was significantly increased for DDR1 (p = 0.0025) and INSR (p = 0.0392) in humans with cirrhosis/fibrosis (Figure 3E), but no statistically significant difference was detected in the mouse model of fibrosis (Figure 3F). The peacock plots indicate differential signaling of DDR and INSR kinase family activity on their substrates, which is represented by the number of kinase family peptide targets (x‐axis; hits) experimentally identified as differentially phosphorylated in fibrosis (red line) or identified during thousands of random sampling events (gray bars). The number of instances in which a given number of kinases or the family of peptide targets were selected from 2000 random sampling iterations is represented by the height of each gray bar (y‐axis; count).

To validate the level of phosphorylation of DDR1 (pDDR1) and INSR (pINSR) and to determine whether they are enhanced in a specific liver‐cell type, we performed immunostaining using antibodies that are phosphorylation‐specific and for α‐smooth muscle actin (αSMA), a marker indicating hepatic stellate cells as previously identified.^{[ 19 ]} The pDDR1 and pINSR had low immunostaining in normal healthy human livers and vehicle‐treated mice livers, but the staining was much more pronounced in the human cirrhotic livers and mice with chronic CCl₄ treatment (Figure 3E,F). The pDDR1 and pINSR immunostaining were colocalized with the αSMA in human cirrhotic livers and the chronic CCl₄ treatment mice livers, indicating that this might be an effect that is specific to HSCs.

These data are presented above mRNA measures, demonstrating differential gene‐expression levels between human fibrosis samples and normal human samples, or fibrosis rodent samples and normal rodent samples. The TK families that were partly altered, as indicated by light gray circles, were fyn related Src family tyrosine kinase (FRK), ABL, and Platelet‐derived growth factor receptor. We therefore measured the mRNA levels of these genes to determine whether they were also changed. The PDGFRA, PDGFRB, and ABL1 were all higher in humans with fibrosis at a statistically significant level (adjusted p value < 0.018), but not FRK (adjusted p value = 0.124) or any of these genes measured in the mouse model of fibrosis (Figure 3G,H).

To further interrogate the TK group, we analyzed individual kinases consisting of the TK tree branches that were identified as significantly changed among kinases in humans and rodents (Figure 4A,B). As indicated by the upstream analysis shown by the waterfall plots in Figure 4C,D, considerable changes were observed for the Src family kinases (LCK, FYN, SRC, YES1, BLK, LYN, and HCK), Abl family (ABL1), Syk family (SYK), and Met family (RON/macrophage stimulating 1 receptor). To delve further into this evaluation, we performed random sampling analyses and quantification of gene‐expression changes for the kinases conserved among the two species. The peacock plots above each mRNA show the observed number of differentially phosphorylated peptide substrates (indicated by the red line) compared with thousands of iterations of random peptide selection (Figure 4E,F). The data for humans with fibrosis indicate that all TK assignments are not the result of chance, as indicated by the red line to the right, except one kinase, RON. These patterns were mostly conserved between humans and rodents. The human gene expression data for these targets showed altered Src family kinases (FYN, SRC, BLK, and HCK), ABL1, and SYK, which were all higher compared with human fibrosis. The analysis in mice showed that these kinases were also hyperactive, but no significant changes in gene expression were observed (Figure 4F).

Assimilation of STK families that are altered with fibrosis

The STK signaling events typically occur after a TK signaling event has been sent or by intracellular mechanisms that activate signaling mechanisms. We interrogated the STKs that might be hyperactive or hypo‐active in fibrosis. We organized the results of the substrate phosphorylation events in heatmaps to show how STK signaling events differ among humans (Figure 5A) and mice (Figure 5B) with fibrosis. Next, we compared the most differentially altered kinases in a waterfall plot based on the Z score for STKs that changed with fibrosis (Figure 5C,D). The highest families of STK events changed in our analysis were DMPK, PKA, and PKG. The peacock plots for DMPK and protein kinase cyclic adenosine monophosphate (cAMP)–activated catalytic subunit alpha (PRKACA/PKAα) in humans showed that their kinase activity was higher in humans and mice (Figure 5E,F). The mRNA expression levels of DMPK but not PRKACA were higher in humans, with DMPK expression being statistically significant. Mice with fibrosis had no statistically significant difference between these two genes. Next, we performed immunostaining using antibodies for DMPK (no phospho antibody available), phosphorylation of PKA (pPKA), and αSMA to validate levels and localization of liver cell type. The DMPK showed strong expression in the normal healthy human livers and vehicle‐treated mice livers (Figure 5E,F). The DMPK immunostaining was higher in the cirrhotic human livers but not the CCl₄‐treated mice livers. The DMPK was localized outside of the αSMA, indicating that it is likely expressed in hepatocytes. The pPKA had low immunostaining in normal healthy human livers and vehicle‐treated mice livers, similar to the pINSR and pDDR1. The staining was higher in the human cirrhotic livers and mice with chronic CCl₄ treatment (Figure 5E,F). The pPKA immunostaining was colocalized with the αSMA in human cirrhotic livers and the chronic CCl₄‐treated mice livers, indicating that this might also be an effect specific to HSCs.

The STK events, unlike the PTK, also had some kinases that had negative Z scores. The kinase families with the most negative Z scores (DYRK, PDK1, ERK [mitogen‐activated protein kinase 8 (MAPK8)], JNK [MAPK1]) are represented as expression levels (Figure 5G,H), and peacock plot analysis is shown for each in Figure S2. The results in humans show that gene expression for DYRK1A, PDK1, and MAPK1 was significantly higher. The gene expression for these four kinases was not significantly different for mice with fibrosis. However, there was a trend to be lower. Further analysis of the subfamily STKs revealed that some pathways were conserved between humans and rodents with fibrosis (Figure 6A,B). Comparing the most changes in the STK category among humans and mice with fibrosis, several kinases in the PK family were significantly enhanced (Figure 6C,D). The AGC group contains PKA, PKG, and PKC STKs, which is one of the highest bubble plot clusters in the paralog phylogenetic tree in Figure 1. The top‐ranked STK kinases include protein kinase C alpha (PRKCA/PKCα), protein kinase C beta (PRKCB/PKCβ), protein kinase C delta (PRKCD/PKCδ), protein kinase C theta (PRKCQ/PKCθ), protein kinase B beta PKBβ/AKT2, protein kinase cGMP‐dependent 1 (PRKG1/PKG1), protein kinase cGMP‐dependent 2 (PRKG2/PKG2), protein kinase X‐linked (PRKX), Pim1 and Pim2, which were the most differentially active STKs between humans and rodents (Figure 6E,F).

Comparison of animal models of fatty liver with emerging fibrosis

To validate our findings from the livers of humans with severe fibrosis from cirrhosis and the CCl₄‐induced animal model of fibrosis, we included two additional mouse models that have MAFLD with emerging fibrosis. As we show in Figure 2, we performed sirius red staining in mice fed a choline‐deficient (CD) diet compared with chow control or a high‐fat plus fructose (HFD+F) diet compared with chow control. As expected, the results show that both of the chow control groups had no fibrosis present (Figure 7A,B). The CD diet had a fibrosis score of 2.0 and a statistically significant induction of Tgfb1 (p = 0.0392) (Figure 7A). The HFD+F diet had a lower fibrosis score of 1.0 and a near (but not significant) induction of Tgfb1 (p = 0.0503) (Figure 7B). These two models are indicative of hepatic steatosis with developing fibrosis, with the CD diet having more developed fibrotic events compared with the HFD+F diet model.

We next wanted to determine whether similar signaling events to the humans with cirrhosis and the CCl₄‐induced fibrosis mouse model were also occurring in these two MAFLD mouse models. The kinome analysis for the TK substrate phosphorylation levels was compiled into heatmaps for the CD and HFD+F diets and their chow controls. The upstream kinase results were aggregated to form the TK family waterfall plots as described previously. We found that INSR was the most differential active kinase in the CD diet compared with chow and that DDR1 was the third highest (Figure 7C). However, the HFD+F diet compared with the chow‐fed group showed that INSR and DDR1 were lower in the livers (Figure 7D). The peacock plots of DDR1 and INSR showed that they had hyperactivity in the CD diet, but both had lower activity in the HFD+F compared with chow diets. The mRNA of Ddr1 was significantly reduced in the CD diet compared with chow‐fed, and Insr had no significant difference. For the HFD+F diet compared with chow‐fed, neither Ddr1 nor Insr mRNA levels were changed. The STK peptide substrate phosphorylation levels were quantitated for both models and prepared in heatmaps and waterfall plots as described previously. Similar to the humans with cirrhosis and CCl₄‐induced fibrosis models, the CD diet compared with chow control had higher DMPK and PKA kinase activity and no significant differences in Dmpk or Prkaca mRNA levels (Figure 7E). Interestingly, the HFD+F diet compared with chow control had enhanced DMPK and PKA kinase activity (Figure 7E). The Dmpk mRNA expression was significantly (p = 0.0033) raised in the HFD+F group, and no difference in Prkaca mRNA levels.