Up-regulated SPP1 increases the risk from IPF to lung cancer via activating the pro-tumor macrophages

Abstract

The incidence of lung cancer (LC) in Idiopathic Pulmonary Fibrosis (IPF) patients is more than twice that in non-IPF. This study aims to investigate IPF-to-LC pathogenesis and to develop a predictor for detecting IPF predisposing patients to LC. We conducted unsupervised clustering to detect high-risk subtypes from IPF to LC. Subsequently, we performed single-cell RNA-seq analysis to characterize high-risk IPF by examining the immune microenvironment. We identified 42 common immune function-related pathogenic genes between IPF and LC. We developed an LC risk classifier for IPF patients, comprising five genes: SPP1, MMP9, MMP12, FABP4, and IL1B. The five-gene classifier can successfully distinguish the high-risk population from IPF patients. High-risk IPF patients exhibited an immunosuppressive microenvironment with higher oncogene expression than low-risk patients. Single-cell analysis revealed that SPP1⁺ macrophages at the terminal of macrophages' developmental trajectory may promote the progression from IPF to LC. The strong crosstalk between SPP1⁺ macrophages and inflammation-related cancer-associated fibroblasts promoted the tumorigenic process in IPF. In vitro, assays showed that co-culturing macrophages overexpressing SPP1 with MRC-5 cells induced the transition of fibroblasts into cancer-associated fibroblasts. SPP1 produced by macrophages promoted epithelial-mesenchymal transition in alveolar epithelial cells via stimulating the upregulation of N-cadherin and Vimentin in MLE-12 cells. This study provided a novel method to identify the LC risk population from IPF, revealing the cellular interactions involved in the transition from IPF to LC. Our findings highlighted SPP1 as a critical driver in IPF progression, offering a potential target for therapy in fibrosis.

Graphical Abstract

1. Background

Idiopathic pulmonary fibrosis (IPF), a chronic and progressive fibrotic interstitial lung disease of unknown etiology [1], is characterized by excessive extracellular matrix (ECM) protein deposition, the presence of fibroblastic foci, and the proximity of fibrotic areas to normal lung parenchyma [2]. Globally, the incidence of IPF ranges from approximately 0.33 to 4.51 cases per 10,000 individuals [3]. Patients with IPF show higher prevalence rates (4.8–48%) of lung cancer (LC) than patients without IPF (2.0–6.4%) [4]. As reported by Eisuke Kato et al., the 5-year all-cause mortality rate after LC diagnosis in IPF reaches 92.9% [5]. IPF patients with higher LC risk have limited treatment outcomes and poor prognosis [6]. Current studies lack exploration and interpretation focusing on the mechanisms of progression from IPF to LC. Deciphering the cancer risk subtype in IPF is of great significance [7].

Dysregulated innate and adaptive immune responses are important hallmarks of chronic lung conditions, including IPF and LC [8], [9]. An immunosuppressive tumor microenvironment is characteristically associated with the development of interstitial pneumonia associated with lung cancer [10]. Geffen et al. summarized the current knowledge on the potential role of regulatory immune cells [11]. The emerging evidence confirmed the essential role of immune cells in the fibrosis of IPF and anti-fibrotic therapy [12]. Macrophages and myeloid-derived suppressor cells may activate lung fibroblasts through TGF-β signaling [11]. Polarization of macrophages affects the pathogenesis of IPF and mediates inflammasome activation and macrophage responses [13]. M2 macrophages have been proven to promote the progression of pulmonary fibrosis [14]. Macrophages were associated with the progression of pulmonary fibrosis, with SPP1/MERTK-expressing macrophages displaying increased proliferative capacity in IPF [15]. Currently, there are diverse opinions among researchers regarding the relationship between SPP1 and M1/M2 macrophages. A previous study revealed that M1 macrophage polarization could induce SPP1 secretion [16]. The expression of SPP1 is associated with M1 and M2 macrophage phenotypes [17]. Notably, distinct from the traditional M1 and M2 phenotypes, macrophage polarity defined by SPP1 had an apparent correlation with prognosis [18]. We believe that examining differences from a diversity perspective can help us better understand the roles of macrophage status and immune cells in the progression and potential risk of IPF.

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful approach for the characterization of immune cells and revealing functions of diverse cell types [19]. Previously, Adams et al. confirmed the roles “profibrotic” of IPF-specific macrophage populations by scRNA-seq [20]. SPP1 expression is increased in the IPF-enriched macrophage [21]. Liu et al. demonstrated the high phenotypic heterogeneity of endothelial cells in bleomycin-induced lung fibrosis using scRNA-seq data sets [22]. It is proposed that the dysregulated epithelial cell interactions with immune cells contribute to pathologic IPF and trigger fibroblast and myofibroblast activation via signaling pathways [2]. The comprehensive analysis reveals the complexity and diversity of cell-cell communications in the process of IPF.

IPF and LC share similar risk factors, pathogenic factors, epigenetic and genetic alterations [23]. The mechanisms of aberrant intercellular interactions in the progression from IPF to LC urgently need to be elucidated. Here, we provided an integrated survey of transcriptomics and epigenomics. We applied an unsupervised clustering approach to distinguish the LC high-risk population from IPF patients. Furthermore, we performed scRNA-seq analyses on IPF cells to explore the impact of cell-cell communication on LC risk. Together, our work constructed an LC risk population prediction model for IPF and dissected the intercellular crosstalk between distinct immune cells and cancer-associated fibroblasts (CAFs), identifying potential targets for IPF patients with LC risk.

2. Results

2.1. Identification of common pathogenic genes between IPF and LC

This work hypothesized that genes dysregulated in both IPF and LC at transcriptional and epigenomic levels could be crucial modulators during the transition from IPF to LC. We profiled transcriptome differences between IPF or LC and normal samples (p < 0.05, Student’s t-test, Fig. 1A, Supplementary Table E1). 2123 common differentially expressed genes (DEGs) were identified between IPF and normal lung samples in the two cohorts (GSE32537 and GSE53845). 97.12% of DEGs were consistently differentially expressed in the two cohorts. 6268 consistent DEGs were identified between LC and normal lung samples in the three cohorts (GSE31210, GSE30219, and GSE19188). 97.88% of DEGs were consistently expressed in the three cohorts. Among IPF and LC, we obtained 923 common pathogenic DEGs. Similarly, we profiled epigenomic differences between IPF or LC and normal samples (p < 0.05, Student’s t-test, Fig. 1A, Supplementary Table E1). In GSE29895, 3664 differentially methylated genes (DMGs) were identified between IPF and normal lung samples. 14,524 DMGs were identified between LC and normal lung samples in the The Cancer Genome Atlas (TCGA) three lung adenocarcinoma (LUAD) cohort. Among IPF and LC, we obtained 2130 common pathogenic DMGs.

Fig. 1.

Pathogenic genes in IPF and LC. (A) DEGs and DMGs identified from different data sets (IPF vs. normal, LC vs. normal). “H-Expressed” represents highly expressed genes. “L-Expressed” represents lowly expressed genes. “HyperMethylated” represents hypermethylated genes. “HypoMethylated” represents hypomethylated genes. Statistical significance was tested by the Wilcoxon rank-sum test, FDR < 0.05. (B) 42 consistent pathogenic genes between IPF and LC. (C) Functional analysis of consistent pathogenic genes. Hypergeometric test, p-value < 0.05. Y-axis: name of the signaling pathway or function; X-axis: percentage of the number of genes assigned to a term among the total number of genes; Bubble size: number of genes assigned to a pathway or function; The color scale corresponds to -log10 (p-value). (D) A gene co-expression PPI network connected 16 low-expressed IPF and LC pathogenic genes and other genes. Pearson correlation test, p-value < 0.05, |r|> 0.3.

Compared with expression and methylation levels, 42 genes showed “high expression and hypo-methylation” (n = 26) or “low expression and hyper-methylation” (n = 16) in both IPF and LC samples (Fig. 1B). Function enrichment analysis revealed that 16 down-expressed genes participated in multiple immune-related biological processes, such as “positive regulation of leukocyte mediated immunity”, “regulation of leukocyte mediated immunity”, “positive regulation of immune effector process” and “positive regulation of mast cell activation involved in immune response” (p < 0.05, Hypergeometric test, Fig. 1C).

A co-expression sub-network consisting of 73 interactions between 16 low-expressed IPF and LC pathogenic genes and other genes was screened from the protein-protein interactions (PPI) network (p < 0.05, |r|>0.3, Pearson correlation test, Fig. 1D). Genes in the PPI sub-network were involved in the negative regulation of the apoptotic process, the cancer pathway, PD-L1 expression, and the PD-1 checkpoint pathway. These results indicated that immunity is involved in the transition process of IPF to LC.

2.2. Identification and characterization of the LC risk subtype from IPF

To dissect the LC risk subtype of IPF, we constructed a classifier based on 1811 immune genes from the ImmPort database. First, 43 immune-related genes with coefficient of variation (CV) values greater than 0.1 were screened in the IPF cohort GSE32537. Second, genes correlated with the expression of more than 50% of LC-related genes were defined as cancer risk features of IPF (Fig. 2A). Eventually, we screened seven gene features, including SPP1, MMP9, MMP12, FABP4, CXCL10, CCL20, and IL1B. Furthermore, unsupervised consensus clustering was performed on IPF samples using the seven-gene features, and two IPF subtypes were identified (Cluster 1 and Cluster 2) (Fig. 2B, Supplementary Fig. E1A-C). SPP1, MMP9, and MMP12 were significantly overexpressed in LC samples from cohorts GSE31210, GSE30219, and GSE19188, while IL1B and FABP4 were significantly underexpressed in LC samples (p < 0.05, Student’s t-test, Fig. 2C). Cluster 2 showed highly expressed SPP1, MMP9, and MMP12 genes, and therefore was inferred to be an LC high-risk subtype. Otherwise, cluster 1 with lowly expressed SPP1, MMP9, and MMP12 genes, was inferred to be a low-risk subtype (p < 0.05, Student’s t-test, Fig. 2C). As a result, the five genes, SPP1, MMP9, MMP12, IL1B, and FABP4 contributed to an IPF classifier. Gini indexes of five genes in the risk classifier ranked high among all genes (Supplementary Fig. E1D). This result supported the credibility and importance of feature genes in segregating high-risk and low-risk IPF patients. Afterward, we applied the classifier to three independent datasets (GSE53845, GSE33566, and GSE70866) and identified LC high-risk IPF samples in each cohort (Supplementary Fig. E2A-C).

Fig. 2.

Construction of a five-gene classifier for LC risk in IPF. (A) Conceptual illustration of the risk features of LC for IPF using transcriptomics from IPF and LC cohorts. CV, coefficient of variation; SD, standard deviation. (B) Hierarchical clustering heatmaps of seven classifier genes using k-means clustering. K = 2. (C) Determination of an LC high-risk IPF subtype according to the expression of classifier genes (IPF markers and LC markers) in IPF and LC samples. p-value < 0.05. (D, E) DEG functional enrichment analysis in IPF and LC high-risk samples. The top 10 terms in the biological process category from GO (D) and the top 10 terms in the KEGG (E). Y-axis: name of the signaling pathway or function; X-axis: percentage of the number of genes assigned to a term among the total number of genes; Bubble size: number of genes assigned to a pathway or function; The color scale corresponds to -log10 (p-value) in the Wilcoxon rank-sum test. (F) Differences in the immune microenvironment across various IPF subtypes (LC low-risk IPF vs. LC high-risk IPF). p-value < 0.05 in the Wilcoxon rank-sum test. (G) Analysis of differential expression in oncogenes and TSGs between LC high-risk IPF and IPF. TSG, tumor suppressor genes. p-value < 0.05, ** p-value < 0.01, *** p-value < 0.001 in the Wilcoxon rank-sum test.

We performed differential expression analysis between LC high-risk and low-risk and then identified 1927 differentially expressed genes. Gene Ontology database (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis showed these genes were enriched in various immune processes including T cell activation, positive regulation of cytokine production and PD-L1 expression, and PD-1 checkpoint pathway in cancer (p < 0.05, Hypergeometric test, Fig. 2D, E). CIBERSORT and EPIC algorithms estimated a higher presence of naïve B cells, macrophages M1, resting NK cells, CD4⁺ memory activated T cells, and CD8⁺ T cells in LC low-risk IPF samples than in high-risk IPF. Conversely, these algorithms indicated a reduced presence of memory B cells, macrophages M2, activated mast cells, neutrophils, and CAFs in LC low-risk IPF samples than in high-risk IPF (p < 0.05, Wilcoxon rank-sum test, Fig. 2F). High-risk IPF patients exhibited an immunosuppressive microenvironment. Moreover, LC high-risk IPF samples that were divided from three independent IPF cohorts (GSE53845, GSE33566, and GSE70866) exhibited consistent activated immune microenvironment characteristics (Supplementary Fig. E2D-F). Previous studies have demonstrated that reduced lung function is a risk factor for lung cancer [24]. Receiver operating characteristic (ROC) curve analysis showed that the five-gene risk classifier had moderate diagnostic ability for lung function-related index in patients with IPF (Area Under the Curve, AUC = 0.833, forced vital capacity in GSE33566, Supplementary Fig. E2G; AUC = 0.682, gender age physiology in GSE70866, Supplementary Fig. E2H). These results confirmed the efficacy of the five-gene classifier in predicting lung cancer risk or lung function.

To investigate the roles of cancer-related genes in LC high-risk IPF samples, we further compared the expression patterns of oncogenes and tumor suppressor genes (TSG) in IPF subtypes. As expected, the expression of oncogenes (CARD11, CCR7, CD79B, et al.) in LC high-risk IPF was higher than in low-risk IPF (p < 0.05, Wilcoxon rank-sum test, Fig. 2G). On the contrary, tumor suppressive genes APC, CEBPA, and DNM2 were underexpressed in LC high-risk IPF than in low-risk IPF (p < 0.05, Wilcoxon rank-sum test, Fig. 2G). These observations suggested that activation of oncogenes and repression of TSGs promote the development of IPF to LC.

2.3. Identification of distinct cellular clusters for macrophages by scRNA-seq

To explore the cellular composition of LC high-risk IPF and elucidate the roles of LC risk genes SPP1, MMP9, and MMP12 in the immune microenvironment, we performed scRNA-seq analysis on 147,169 single cells from 32 IPF tissue samples. We identified 28 cell subsets in the GSE136831 (Fig. 3A, Supplementary Fig. E3A). The expression of genes SPP1, MMP9, MMP12, FABP4, and IL1B in the IPF classifier was primarily found in macrophages, including alveolar macrophages (Fig. 3B-F). Therefore, we conjectured those genes could modulate the function of specific types of macrophages to affect the progression from IPF to LC.

Fig. 3.

Identification of distinct macrophage clusters based on a five-gene classifier in IPF. (A) UMAP of 147,169 cells from the GSE136831 IPF cohort, colored by annotated cell type. (B-F) UMAP plots of three LC-markers (SPP1, MMP9, and MMP12) and two IPF marker genes (FABP4, IL1B) expressed in all cells. (G) UMAP plots of macrophages allow the visualization of three clusters (macrophages 1–3). Colors show the different clusters defined by the graph-based clustering method. (H, I) UMAP plots of two representative genes, SPP1 and FABP4 in each macrophage cluster. (J) UMAP plots are colored by reference to macrophage subtype annotations, using a color scheme consistent with that in (G). (K) UMAP with predicted cell fates colored by pseudotime inferred. Identified initial state nodes are outlined in yellow, and terminal state nodes are outlined in blue. (L, M) Top 15 enriched functions for top 20 DEGs in terminal (SPP1⁺ macrophage) and initial (FABP4⁺ macrophage) states. Wilcoxon test, FDR < 0.05. Y-axis: signaling pathway or function; X-axis: percentage of the number of genes assigned to a term among the total number of genes; Bubble size: number of genes assigned to a pathway or function; The color scale corresponds to -log10 (p-value).

According to the distribution of five-gene expression, we manually classified macrophages into three subgroups (macrophages 1–3, Fig. 3G, Supplementary Fig. E3B). SPP1 was enriched in macrophage 1 (SPP1⁺ macrophage, Fig. 3H), while FABP4 was enriched in macrophage 2 (FABP4⁺ macrophage, Fig. 3I). IL1B, MMP9, and MMP12 showed low expression in three macrophage subtypes (Supplementary Fig. E3C-E). To uncover the progression and regulatory coordination in macrophage development, we performed single-cell pseudotime trajectory analysis (Fig. 3J, K). This trajectory revealed nascent fractions of FABP4⁺ macrophages in progression and a transition from FABP4⁺ macrophages to low SPP1⁺ macrophages. In addition, we determined specific genes for each macrophage subset. The top 20 differentially expressed genes in SPP1⁺ macrophages were enriched in various metabolic processes, including the pyruvate metabolic process and regulation of the glycolytic process (Fig. 3L). Notably, SPP1 and MERTK genes were involved in the “ATP generation from ADP”. MERTK and FAM20C genes were involved in the “purine nucleoside diphosphate metabolic process”. In contrast, the top 20 differentially expressed genes in FABP4⁺ macrophages were related to the immunoglobulin-mediated immune response and other immune-related signaling pathways (Fig. 3M). There was no correlation between the expression of M1 macrophage markers (CXCL10, CD80, and CD86) and SPP1⁺ macrophages or FABP4⁺ macrophages (Supplementary Fig. E3F-H). Similarly, there was no correlation between M2 macrophage markers (CCL18 and CD163) and SPP1⁺ macrophages or FABP4⁺ macrophages (Supplementary Fig. E3I, J). This indicated the macrophage classification in this study is independent of the well-known M1/M2 macrophage types. Above results suggested that SPP1⁺ macrophage could contribute to pro-tumor effects, while FABP4⁺ macrophage exhibits pro-immune/anti-tumor effects.

2.4. Cell communication between inflammation-related CAF and macrophages

Crosstalks between CAFs and immune cells could be of value in exploring the progression and cancer risk of IPF [25]. We annotated 4036 CAFs based on the expression of markers COL1A2 and PDGFRB in the IPF cohort GSE136831 (Fig. 4A, B, Supplementary Fig. E4A). CellChat was used to infer inter-cellular communication by ligand-receptor analysis. The result indicated that interactions were enriched between CAFs, aberrant basaloids, and macrophages (macrophages 1–3) (Fig. 4C). Given the heterogeneity among CAFs, we next annotated CAFs into two clusters: inflammation-related CAFs (iCAFs, SLPI⁺C3⁺IGFBP6⁺) and matrix-related CAFs (mCAFs, POSTN⁺COL6A3⁺FN1⁺) (Fig. 4D, E). SPP1 showed low expression in all CAFs (Supplementary Fig. E4B). Genes specifically expressed in iCAFs were involved in ECM, thereby facilitating cell-cell communication (p < 0.05, Hypergeometric test, Fig. 4F). The mCAFs specific markers were mainly enriched in neutrophil activation and neutrophil-mediated immunity (p < 0.05, Hypergeometric test, Fig. 4G).

Fig. 4.

Identification of inter-cellular crosstalk among three macrophage subtypes and two CAF subtypes in IPF. (A) UMAP of CAFs from IPF cohort GSE136831, colored by annotated cell type (iCAF and mCAF). (B) UMAP plots show RNA levels of CAF specific surface markers (COL1A2 and PDGFRB). CAF, (C) Inferred cell-cell communication between CAFs and other cell types through ligand-receptor analysis using CellChat. (D, E) UMAP plots of RNA levels of iCAF-specific surface markers (SLPI, C3, and IGFBP6) or mCAF-specific surface markers (POSTN, COL6A3, and FN1). (F, G) Top 5 enriched functions for the top 20 DEGs in iCAF and mCAF (Wilcoxon test, FDR < 0.05). (H) Cell-cell communication between CAFs and other cell types is inferred by ligand-receptor analysis, as estimated using CellPhoneDB. (I) Cell-cell communication between distinct CAF subtypes and other cell types is inferred by ligand-receptor analysis, as estimated using CellPhoneDB. (J) Dot plot of ligand-receptor interaction scores shows predicted interactions between CAF cells and macrophages. The dot size represents -log 10 (p-value), and the color represents the interaction score, where dark red colors indicate greater predicted interactions. Ligand-receptor pairs are listed along the left axis. Ligand-expressing cells are listed to the left of the bottom axis labels. Receptor-expressing cells are listed to the right of the bottom axis labels.

Furthermore, we analyzed the inter-cellular communication using the CellPhoneDB database. The results supported the existence of a strong reciprocal cross-talk between CAFs and macrophages, especially SPP1⁺ macrophage (macrophage 1) (Fig. 4H). Focusing on the two CAF subtypes, iCAFs showed stronger interactions with macrophages compared to mCAFs (Fig. 4I). The dot plot of ligand-receptor interaction scores indicated that the C3-C3AR1 pair was enriched between iCAF/mCAF and macrophages 1–3. The receptor C3AR1, primarily expressed in macrophages, is known to enhance pulmonary fibrosis. Furthermore, the interaction of SPP1-CD44 was enriched in SPP1⁺ macrophage and iCAF. In SPP1⁺ macrophages, SPP1 is expressed as a ligand, while the receptor gene CD44 is predominantly found in iCAFs (Fig. 4J). The FN1-aVb5 complex was found to be enriched between SPP1⁺ macrophage and iCAF, as well as FABP4⁺ macrophage and iCAF and FABP4⁺ macrophage and mCAF (Fig. 4J). In summary, the single-cell analysis revealed that SPP1 and FN1 may affect inter-cellular communication via ligand-receptor interaction in the IPF immune microenvironment.

2.5. Communication between SPP1⁺ macrophages and CAFs in lung cancer

To investigate the correlation between SPP1⁺ and pro-tumor macrophages in LC scRNA-seq data, we collected scRNA-seq data for 208,506 cells and 29,634 genes derived from human LC tissues in the GSE131907 cohort. All single cells in GSE131907 were annotated into 10 types of cells (Fig. 5A). The sample sources of cells are shown in Supplementary Fig. E5A. We found high expression of SPP1, FABP4, and IL1B in myeloid cells (Fig. 5B, C, Supplementary Fig. E5B). MMP9 and MMP12 exhibited low expression in all cells (Supplementary Fig. E5C, D). Compared with cells isolated from normal lung tissues, SPP1 exhibited higher expression and FABP4 exhibited lower expression in cells from tumor tissues (Fig. 5D, E). In a further study, we identified seven different myeloid cell clusters, including alveolar macrophages, pleural macrophages, and monocyte-derived macrophages (mo-Mac) (Fig. 5F). In diverse types of myeloid cells, SPP1 expression was predominantly observed in mo-Mac, while FABP4 was remarkably expressed in alveolar macrophages (Fig. 5G-J). Inter-cellular communication analysis predicted strong crosstalk among mo-Mac, lung cancer cells, and CAF (Fig. 5K). The interaction of iCAF and mCAF with cancer cells was stronger than those of vascular CAF (vCAF) with cancer cells (Fig. 5L).

Fig. 5.

Intercellular crosstalk between distinct macrophage subtypes and CAF subtypes in LC. (A) UMAP of 208,506 cells from the GSE131907 LC cohort, colored by annotated cell type. (B, C) UMAP plots demonstrate the expression of the LC marker SPP1 and the IPF marker FABP4 in all cells. (D) UMAP plots for all cells from the LC cohort, color-coded by pathology type. PE, pleural effusion; mBrain, metastatic brain; mLN, metastatic lymph node; nLN, normal lymph node; nLung, normal lung; tL/B, lymph node or brain tumor; tLung, lung tumor. (E) Violin plots showing expression of SPP1 and FABP4 in cells from normal lung and lung tumor tissues. (F) UMAP of myeloid cells from the LC cohort colored by the distinct annotated cell type. Mo-Mac represents a monocyte-derived macrophage. (G, H) UMAP plots of myeloid cells expressing the LC marker SPP1 and the IPF marker FABP4. (I, J) Violin plots of SPP1 and FABP4 expression in different macrophage subtypes. (K) Cell-cell communication between CAF and other cell types (distinct macrophages and cancer cells) was inferred by ligand-receptor analysis, as estimated using CellPhoneDB. (L) Communication between different CAF subtypes and other cell types.

Mo-Mac has been shown to potentially promote tumor development. The enrichment of mo-Mac defined a more aggressive phenotype [26]. This finding further supported the notion that SPP1 could increase the risk of tumorigenesis in IPF via activating a special group of macrophages.

2.6. Macrophages overexpressing SPP1 promote transition of fibroblasts into CAFs

Fibroblasts are the key effectors of pulmonary fibrosis. Among all stromal cells in the tumor microenvironment, CAFs are the most abundant and are closely related to tumor progression. The communication between macrophages and fibroblasts can be mediated by cytokines. To explore whether SPP1, a protein secreted by macrophages, could activate the transformation of resting fibroblasts into CAFs, we established a co-culture model of MRC-5 with THP-1 cells. As shown in Fig. 6A, we induced the differentiation of THP-1 monocytes into macrophages using phorbol-12-myristate-13-acetate (PMA) and added recombinant human SPP1 (rhSPP1). After 48 h, we replaced the cell medium with a serum-free medium to facilitate co-culture with MRC-5 cells. Through quantitative reverse transcription polymerase chain reaction (qRT-PCR), we found that macrophages with overexpression of SPP1 induced ECM deposition and increased the mRNA levels of FN1, FAP, and ACTA2 in MRC-5 cells (Fig. 6B-D). This suggests that SPP1⁺ macrophages promote the differentiation of normal fibroblasts into CAFs. Consistent with the above results, upon rhSPP1 treatment, the co-cultured MRC-5 cells showed a significant increase in the protein expression of Fn1, FAP, and α-SMA (Fig. 6E, Supplementary Fig. E6). α-SMA is a marker of activated fibroblasts and is commonly used as a primary evaluation index for the therapeutic effects targeting CAFs. Immunofluorescence results indicated that SPP1 secreted by macrophages promoted the differentiation of fibroblasts into a-SMA positive CAFs (Fig. 6F).

Fig. 6.

Macrophage-secreted SPP1 promoted the differentiation of fibroblasts into CAFs. (A) Diagram of co-culture of THP-1 and MRC-5 cells. PMA, phorbol-12-myristate-13-acetate. (B-D) Detection of the relative mRNA expression of CAF markers by qRT-PCR; n = 6. (E) Western blot was used to determine the effects of rhSPP1 on the protein levels of FN1, FAP, α-SMA, and SPP1; n = 4. The blots were cropped and full-length blots are presented in Supplementary Fig. E6. (F) Immunofluorescence experiments were performed to detect the fluorescence intensity of α-SMA in MRC-5 cells (scale bar = 20 µm; n = 5). Data are presented as mean ± SEM; * p-value < 0.05, ** p-value < 0.01.

2.7. Macrophages secreted SPP1 promote EMT progression of alveolar epithelial cells

Then, we investigated the regulatory effect of SPP1 secreted by macrophages on epithelial-mesenchymal transition (EMT) during IPF-LC. THP-1 cells were co-cultured with MLE-12 cells using a transwell chamber with 0.4 µm pore size. This setup physically isolated the cells but allowed the diffusion of factors secreted from THP-1 to reach alveolar epithelial type II (ATII) cells (Fig. 7A). We found that an increase in SPP1 production by macrophages promoted mRNA expression of N-cadherin and Vimentin in MLE-12 cells, while the expression of epithelial marker E-cadherin was downregulated (Fig. 7B-D). Western blot results showed that macrophages treated with SPP1 enhanced the protein expression of N-cadherin, Vimentin, and SPP1, but inhibited the protein expression of E-cadherin (Fig. 7E, Supplementary Fig. E7). Using immunofluorescence experiments, we found that after co-culture with macrophages overexpressing SPP1, the fluorescence intensity of E-cadherin of ATII cells was significantly reduced, while that of N-cadherin showed the opposite effect (Fig. 7F). These results suggested that macrophages secreted SPP1 promoted alveolar epithelial to mesenchymal transition, a biological process that may play an important role in IPF-LC.

Fig. 7.

Recombinant human SPP1 promoted the EMT process of alveolar epithelial cells. (A) Diagram of co-culture of THP-1 and MLE-12 cells. PMA, phorbol-12-myristate-13-acetate. (B-D) mRNA expression of EMT-related genes by qRT-PCR experiments; n = 6. (E) Protein levels of N-cadherin, E-cadherin, Vimentin, and SPP1 after rhSPP1 administration; n = 4. The blots were cropped and full-length blots are presented in Supplementary Fig. E7. (F) Analysis of E-cadherin and N-cadherin expression in alveolar epithelial cells by immunofluorescence assay (scale bar = 20 µm; n = 5). Data are presented as mean ± SEM; ** p-value < 0.01.

3. Discussion

In this study, we investigated the common pathogenic molecular mechanism in IPF and LC and found that immune-related functions play essential roles in the progression from IPF to LC. We constructed an immune-based classifier for IPF samples and identified an LC high-risk cluster within human IPF. The classifier was constructed using five genes, including LC risk genes SPP1, MMP9, and MMP12, as well as low-risk genes IL1B and FABP4. Multiple oncogenes exhibited significantly increased expression in high-risk IPF. On the contrary, the expression of multiple TSGs was decreased in high-risk IPF. In this work, we characterized the immunosuppressive microenvironment of high-risk IPF, indicating that IPF development is associated with alterations in the immune microenvironment in vivo.

Single-cell transcriptome analysis validated the presence of two macrophage types: pro-tumor macrophages (SPP1⁺ macrophages) and pro-immune/anti-tumor macrophages (FABP4⁺ macrophages). Pro-immune macrophages were inferred to be in the initial state of the cellular developmental trajectory, while pro-tumor macrophages were at the terminal state. SPP1 is highly expressed in cancer tissues. High infiltration levels of SPP1 are associated with worse prognoses in cancer [27]. SPP1⁺ tumor-associated macrophages were enriched in an immunosuppressed microenvironment, exerting pro-angiogenic and pro-tumor metastatic functions, including decreased phagocytosis and inflammation and increased angiogenesis [27]. SPP1⁺ macrophages have been shown to promote angiogenesis and mediate lung cancer immune evasion by upregulating PD-L1 in lung adenocarcinoma [28]. Besides, some studies have demonstrated that FABP4 possesses tumor-suppressor effects in lung cancer, and overexpression of FABP4 inhibits tumor growth by activating PPARγ [29].

SPP1 could promote cellular communication between pro-tumor macrophages and CAFs through SPP1-CD44 ligand-receptor interaction, especially in iCAFs. The macrophage-derived SPP1-CD44 axis has been reported to accelerate the progression of malignant cells and promote cancer stemness by structuring interplays between CAFs and stem cell populations [30]. Ligand-receptor networks between SPP1⁺ macrophages and CAFs located near the tumor boundary promote tumor immune barrier structure formation [31]. Previous research demonstrated the levels of SPP1 and FN1 were increased in fibrotic lungs [32]. Single-cell transcriptome analysis in LC further validated the pro-tumor role of SPP1 via activating the interaction between mo-macrophages and CAF. Notably, the silencing of SPP1 or FN1 in fibroblasts attenuated the anti-apoptosis activity of fibrotic lung-derived fibroblasts [32]. SPP1 and FN1 were further confirmed to be associated with tumor progression [33], yet it does not reveal the underlying mechanisms of SPP1 in the cellular microenvironment. Cell assays confirmed that macrophage-derived SPP1 promotes the transition of fibroblasts into CAFs by activating the expression of FAP, Fn1, and α-SMA and promotes the EMT process of alveolar epithelial cells in IPF. This study provided a comprehensive analysis of the effect of SPP1 as a cancer risk factor in IPF.

SPP1⁺/FABP4⁺ macrophages were defined based on the expression of LC risk genes in IPF rather than the polarization state of macrophages. Notably, the SPP1⁺/FABP4⁺ macrophages subtype system was confirmed to be independent of previous M1/M2 typing systems and able to specifically identify macrophages associated with cancer risk in IPF. Previous reports support the transcriptional heterogeneity in macrophages, which does not conform to the binary M1/M2 paradigm [34]. The lung cancer risk-related SPP1⁺/FABP4⁺ macrophage classification does not comport with the M1/M2 polarization model, which was consistent with the findings in the single-cell analysis [35], [36]. Our finding emphasized that functional macrophage subtypes could be differentiated by gene signatures based on a combination of cancer risk suppression, and activation of related immune genes in bulk cohorts. The macrophage subtypes defined in this study are more representative than the traditional M1/M2 classification system. We raised the possibility of a combination of a basis for preliminary immune-related characterization of differential transcriptional and epigenetic in IPF and LC. Resultant interactions between macrophages and iCAFs are associated with the final function in cancer risk and transcriptional cell states of each macrophage subtype in the immune microenvironment.

We recapitulated an immune-based cancer risk model in single-cell transcriptomes from IPF patients and provided a new perspective to identify cancer risk-related cell subtypes. Meanwhile, some limitations should be noted in this study. First, the present work identified LC-risk patients at the population level. An optimized and improved model for individual classification is necessary. Given the detection cost and operability, developing a five-gene classifier kit might be a better option than RNA-seq for future applications. Second, transcriptomic data sampled from IPF and matched LC states in patients will be critical for further supporting the conclusions of this research and for detecting the dynamic carcinogenesis process in IPF. Third, this work will be followed by more analyses and experiments to further explore the impact of intercellular transcriptional regulation on IPF cancer risk. Strikingly, SPP1 remains plausible as a promising treatment target in IPF despite the lack of evidence from further experiments. We will conduct SPP1 knockout or silencing experiments in cells and mice to confirm the effectiveness of SPP1 in reducing the risk of cancer in IPF.

4. Conclusions

In summary, this work shows the heterogeneity within IPF, with various molecular alterations affecting the role of tumor immunity in the tumor microenvironment. Remarkably, we found a group of LC high-risk subtypes from IPF by applying an immune-based classifier. The ECM-related gene SPP1 contributed to the high-risk subtype and promoted the progression of IPF to the LC high-risk subtype via activating the communication between iCAF and pro-tumor macrophages. In future research, we may further explore the transformation process of IPF to LC, which may inhibit the progression to LC.

5. Materials and methods

5.1. Data sources

The bulk/single-cell RNA sequencing (RNA-seq) profiles, DNA methylation profiles, and clinicopathological data for IPF or LC patients in the study were obtained from Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo) and the TCGA (http://cancergenome.nih.gov/). Two bulk IPF RNA-seq cohorts (GSE53845 and GSE32537) and LUAD RNA-seq cohorts (GSE31210, GSE30219, and GSE19188) were used for screening common pathogenic genes between IPF and LC [37], [38], [39], [40], [41]. The IPF DNA methylation data set (GSE29895) and the LC DNA methylation data set from the TCGA were used for identifying common pathogenic methylations [42], [43]. IPF RNA-seq cohort GSE32537 was used for training the LC risk features. Two independent data sets (GSE33566 and GSE70866) were used for validation [44], [45]. Besides, IPF scRNA-seq data set GSE136831 and LC scRNA-seq data set GSE131907 were used to decipher the mechanism of intercellular interactions [20], [46]. The detailed information for data sets is described in Table 1.

Table 1.

Data sets were used in this study.

Accession	Disease	Type	Disease	Normal	Genes
Used for screening pathogenic genes
GSE53845	IPF	Bulk RNA-seq	40	8	19,596
GSE32537	IPF	Bulk RNA-seq	119	50	8466
GSE31210	LUAD	Bulk RNA	226	20	21,655
GSE30219	LUAD	Bulk RNA	85	14	21,655
GSE19188	LUAD	Bulk RNA	45	65	20,486
GSE29895	IPF	DNA Methylation	12	10	12,259
TCGA-LUAD	LUAD	DNA Methylation	458	32	19,254
Validation
GSE33566	IPF	Bulk RNA-seq	93	30	12,254
GSE70866	IPF	Bulk RNA-seq	112	20	20,189
scRNA-seq analysis
GSE136831	IPF	scRNA-seq	32	28	45,947
GSE131907	LUAD	scRNA-seq	58	11	29,634

5.2. Identification of features for IPF with high-risk of LC

LC-related genes were used to build an LC risk model in IPF and predict an LC high-risk subtype.

• (i)
  We calculated the CV for each immune-related gene (Eq. 1) in the training set GSE32537.

  $$\mathrm{CV}\left(\mathrm{g}\right)=\mathrm{SD}\left(\mathrm{g}\right)/\mathrm{Mean}(\mathrm{g})$$ (1)

  Where CV(g) represents the coefficient of variation for the immune-related gene g. SD(g) represents the standard deviation of expression values for gene g in all IPF samples. The mean (g) represents the mean of expression values for gene g in all IPF samples.
• (ii)We assessed the correlation of expression levels of immune-related genes and LC-related genes in LC data set GSE31210, which was tested by the Pearson correlation test with p < 0.05.
• (iii)
  For an immune-related gene g, we calculated the ratio of LC-related genes correlated with gene g, as shown in Eq. 2.

  $$\mathrm{Ratio}(\mathrm{g})=\mathrm{n}/\mathrm{N}$$ (2)

  Where n is the count of the LC-related genes that are correlated with gene g (p < 0.05, |r| > 0.3, Pearson correlation test). N is the count of all LC-related genes (N = 236).
• (iv)Last, gene g with a CV larger than 0.1 and a Ratio larger than 50% was defined as a risk feature of LC from IPF.

Finally, we identified a set of features for IPF patients with high risk from LC.

5.3. Bulk RNA-seq data processing

Preprocessing categories for bulk RNA-seq cohorts: (i) The probe-level expression of bulk RNA-seq cohorts was annotated to gene-level. (ii) Entrez gene IDs were used to map genes using the matched platform information. (iii) If a gene was mapped to multiple probes, the expression level for the gene was generated by averaging. (iv) Probes that failed to map to any gene ID or mapped to more than one gene ID were removed.

5.4. scRNA-seq data processing

We applied quality measures to remove less informative cells and finally obtained two data sets. Scanpy python package (version 3.0.1) was used to analyze the data set of GSE136831 with 147,169 cells from 32 IPF patients and GSE131907 with 208,506 cells from 58 LUAD patients [47]. The analyses included quality control, count normalization, and correction of mitochondrial influence.

Preprocessing categories for GSE136831: (i) Cells that expressed fewer than 200 genes were removed. (ii) Genes detected in fewer than three cells were filtered. (iii) Cells that expressed more than 5000 genes or more than 30% mitochondrion-derived unique molecular identifier counts were considered low-quality cells and were removed. (iv) The count matrix of cells was normalized and log2 transformed for subsequent analysis. (v) 3679 genes with highly variable expression values were screened using parameters “min mean = 0.0125, max mean = 3, min disp = 0.5”. (vi) We scaled each gene to unit variance and clip values exceeding the standard deviation of ten. Finally, 143,381 cells were available for this study.

Preprocessing categories for GSE131907: (i) Genes that were expressed in less than 0.1% of cells were removed. (ii) Cells that expressed fewer than 200 genes were removed. (iii) Cells that expressed more than 20% mitochondrion-derived were considered low-quality cells and then removed. (iv) The count matrix of cells was normalized and log2 was transformed for subsequent analysis. (v) 2347 genes with highly variable expression values were screened using parameters “min mean = 0.0125, max mean = 3, min disp = 0.5”. Finally, 208,506 cells were available for this study.

5.5. Pathogenic genes shared by IPF and LC

For bulk RNA-seq cohorts, firstly, two comparisons were performed using Student’s t-test: (i) We identified DEGs between IPF and normal samples in GSE32537 and GSE53845, respectively. (ii) We identified DEGs between LC and normal samples in GSE31210, GSE30219, and GSE19188, respectively. p values less than 0.05 adjusted by the Benjamini-Hochberg procedure were considered significant. Then, DEGs derived from two data sets (i) were defined as consistent IPF pathogenic DEGs. Data sets derived from three data sets in (ii) were defined as consistent LC pathogenic DEGs. Last, we obtained a set of common pathogenic DEGs among IPF and LC.

For DNA methylation cohorts, we identified a set of DMGs in GSE29895 (IPF vs. normal) and a set of DMGs in TCGA LUAD data (LC vs. normal) by Student’s t-test with adjusted p < 0.05. DMGs derived from two gene sets were defined as common pathogenic DMGs among IPF and LC.

Common pathogenic genes were defined as genes that are dysregulated in both IPF and LC at the transcriptional and epigenomic levels. Namely, differentially up-regulated genes showed hypo-methylation and vice versa.

5.6. Network of protein-protein interactions

PPI data was obtained from the functional protein association networks database STRING (version 11.5, https://string-db.org/). We constructed a PPI network based on the common pathogenic genes between IPF and LC. Then, we filtered the PPI pairs by screening co-expression relationships with the Pearson correlation test (p < 0.05 and |r| > 0.3). Cytoscape software version 3.6.0 (https://cytoscape.org/) was used to visualize the interaction network.

5.7. Functional enrichment analysis

Functional analysis was performed to annotate the function of genes of interest based on the six top categories (09100–09160) from KEGG (http://www.genome.jp/kegg/) and three categories (biological process, molecular function, and cellular component) from the GO (http://www.geneontology.org/) databases. The hypergeometric distribution model was used to perform pathway enrichment. The p-value was adjusted by the Benjamini-Hochberg procedure. The pathways with a false discovery rate (FDR) of less than 0.05 were significant. In this study, we used the R package “clusterprofiler” to perform GO and KEGG functional annotation. The functions of markers that are expressed in specific cells were annotated using “gseapy” in the Python environment.

5.8. Immune-related genes and cancer-related genes

1811 immune-related genes were retrieved from the ImmPort database (https://immport.niaid.nih.gov) (Supplementary Table E2). 236 LC-related genes were obtained from Online Mendelian Inheritance in Man (OMIM, https://www.omim.org/, Supplementary Table E3). 288 oncogene and TSG were downloaded from COSMIC (https://cancer.sanger.ac.uk/cosmic).

5.9. Estimation of immune cell infiltration

CIBERSORT and EPIC were applied to assess the proportions of infiltrating immune cells in the immune microenvironment, including 25 immune cell types. ESTIMATE was applied to determine the immune score of each sample using R software.

5.10. Single-cell population identification and clustering

Principal component analysis was used for dimensionality reduction with default parameters. Unsupervised clustering of cells was performed using the Leiden algorithm [48]. For visualization, Uniform Manifold Approximation and Projection (UMAP) was applied. All single cells in GSE136831 were annotated by gene markers from the study by Taylor et al. [20]. Cells in GSE131907 were annotated by gene markers from the study by Kim et al. [46]. Macrophages were annotated as mo-Mac, alveolar macrophages, or pleural macrophages. Differentially expressed genes with adjusted p < 0.05 were identified as specific gene signatures by Student’s t-test between cell clusters (one cluster vs. all other clusters).

5.11. Single-cell trajectory analysis

We constructed differentiation trajectories for each cell lineage based on the connectivity between cells. Partition-based Graph Abstraction (PAGA) is a spatially based algorithm for extracting the “skeleton” of cell differentiation, which is used to display the differentiation trajectory of cells and assess the closeness of relationships between clusters. We applied the PAGA algorithm to measure and quantify the degree of connectivity between cell populations [49]. Unsupervised graph-based clustering was performed using the Leiden algorithm and visualized with the UMAP algorithm.

5.12. Cell-cell communication

CellPhoneDB was performed to infer signaling interactions between two cell populations by considering the minimum average expression of the members of ligand–receptor complexes [50]. Based on enriched intercellular ligand-receptor interactions, we determined the strength of cell-cell communication. CellPhoneDB v.2.0 was used to predict cell-cell communication in this study. A scRNA-seq cohort and cell annotation information are needed for CellPhoneDB. CellPhoneDB software is available on the Linux system. CellChat was used to identify intercellular signaling molecule interactions and visualize cell-cell communications atlases from scRNA-seq data [51]. CellChat was implemented through the R package.

5.13. Cell culture and treatment

The human fetal lung fibroblast cell line MRC-5 and mouse alveolar epithelial cell line MLE-12 were obtained commercially from the Cell Bank of the Chinese Academy of Sciences. The cell lines were cultured in DMEM (Sigma, Germany), supplemented with 10% FBS (Biological Industries, Israel), and 1% Penicillin-Streptomycin-Amphotericin B (Solarbio, Beijing, China) under sterile condition, and cultured at 37 °C in a 5% CO₂ incubator.

To investigate the effect of cytokines secreted by macrophages on MRC-5 and MLE-12. Suspended THP-1 cells were cultured in the upper chamber of a transwell co-culture system and treated with 100 ng/ml PMA (Beyotime, Jiangsu, China) for 48 h. Subsequently, adherent THP-1 cells were exposed to rhSPP1 (R&D Systems, U.S.A) for an additional 48 h. In addition, MRC-5 cells or MLE-12 cells were seeded in the lower chamber, and adherent cells were co-cultured with fresh serum-free medium for subsequent experiments.

5.14. Quantitative real-time RT-PCR

Total RNA from cells was extracted using a TRIzol reagent. The concentration and purity of RNA were determined by NanoDrop 8000 (Thermo, U.S.A). Total RNA was reverse transcribed using the cDNA reverse transcription Kit (Mona, Wuhan, China) using random primers. mRNA was quantified by SYBR Green on a LightCycler® 96 system (Roche, Switzerland). Data were filtered through a 2^-ΔΔ Analysis of CT methods.

5.15. Western blot

Total cellular protein was extracted using RIPA lysis buffer and protease inhibitors (Beyotime, Shanghai, China). Proteins were separated on 10% SDS polyacrylamide gels and transferred to nitrocellulose membranes (Pall Life Sciences, Ann Arbor, MI, USA). Anti-Fn1 antibody (1:500, Proteintech, Wuhan, China), FAP antibody (1:500, Wanlei bio, Shenyang, China), a-SMA antibody (1:1000, Proteintech, Wuhan, China), E-cadherin antibody (1:500, Proteintech, Wuhan, China), N-cadherin antibody (1:500, Proteintech, Wuhan, China), Vimentin antibody (1:500, Proteintech, Wuhan, China), SPP1 antibody (1:500, Wanlei bio, Shenyang, China). β-actin (1:1000, Proteintech, Wuhan, China) was used as a loading control. Membranes were incubated with secondary antibodies (1:8000, Abcam, USA) for 1 h at room temperature. Finally, Odyssey (Odyssey CLX, biosciences, USA) detection protein quantification was performed.

5.16. Immunofluorescence staining

After MRC-5 cells or MLE-12 cells were treated with the co-culture system, the cells were fixed in 4% PFA at room temperature and then penetrated and blocked. Cells were incubated with anti-α-SMA (1:200, Abcam), anti-E-cadherin (1:200, Proteintech) and anti-N-cadherin (1:200, Proteintech) were incubated overnight at 4 °C. The following day, FITC conjugated Goat anti-mouse antibody (1:500, Alexa fluor 488, life technology) or Goat anti-rabbit antibody (1:500, Alexa fluor 594, life technology) or Goat anti-mouse antibody (1:500, Alexa fluor 594, life technology) was applied in dark conditions. Nuclei were counterstained with DAPI for 5 min. Then, immunofluorescence was observed and analyzed under a fluorescence microscope.

5.17. Statistical analysis

P values less than 0.05 were considered significant. Patients who did not have an “event” during follow-up were censored. A Student's t-test was used to detect differentially expressed genes between two groups with p < 0.05 adjusted by the Benjamini-Hochberg procedure. All statistical analyses were carried out using R software version 4.0.4 (http://www.r-project.org/) or Python version 3.6.1.2 (https://www.python.org/).

Ethics approval and consent to participate

Not applicable.

Funding

This study was supported by the National Natural Science Foundation of China (No. 32270710 and 32171127); the Scientific Fund Project of Heilongjiang Province (No. JQ2022H001 and YQ2021H005).

CRediT authorship contribution statement

Tingting Chen: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing – original draft, Visualization. Jiayu Guo: Visualization, Writing - original draft. Liqiang Ai: Methodology, Validation, Formal analysis, Investigation, Writing – original draft, Visualization. Yuquan Wang: Methodology, Investigation. Yan Wang: Writing – original draft, Visualization. Bo Chen: Formal analysis, Investigation, Writing – original draft, Visualization. Mingyue Liu: Validation, Formal analysis, Investigation, Writing – original draft, Visualization. Shuping Zhuang: Data curation, Visualization. Kaidong Liu: Data curation, Visualization. Zhangxiang Zhao: Data curation, Visualization, Writing – original draft. Haihai Liang: Supervision, Writing – original draft. Yunyan Gu: Supervision, Project administration, Writing – original draft.

Declaration of Competing Interest

The authors declare that they have no conflict of interest.

Biographies

Tingting Chen, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China; Department of Surgery, The Chinese University of Hong Kong, Shatin, Hong Kong SAR, China.

Jiayu Guo, Department of Pharmacology (State-Province Key Laboratories of Biomedicine-Pharmaceutics of China, Key Laboratory of Cardiovascular Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China.

Liqiang Ai, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China; Center for Medical Genetics, School of Life Sciences, Central South University, Changsha, China.

Yuquan Wang, Department of Pharmacology (State-Province Key Laboratories of Biomedicine-Pharmaceutics of China, Key Laboratory of Cardiovascular Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China.

Yan Wang, Department of Pharmacology (State-Province Key Laboratories of Biomedicine-Pharmaceutics of China, Key Laboratory of Cardiovascular Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China.

Bo Chen, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China.

Mingyue Liu, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China.

Shuping Zhuang, Department of Pharmacology (State-Province Key Laboratories of Biomedicine-Pharmaceutics of China, Key Laboratory of Cardiovascular Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China.

Kaidong Liu, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China.

Zhangxiang Zhao, The Sino-Russian Medical Research Center of Jinan University, The Institute of Chronic Disease of Jinan University, The First Affiliated Hospital of Jinan University, Guangzhou, China.

Haihai Liang, Department of Pharmacology (State-Province Key Laboratories of Biomedicine-Pharmaceutics of China, Key Laboratory of Cardiovascular Research, Ministry of Education), College of Pharmacy, Harbin Medical University, Harbin, China.

Yunyan Gu, Department of Systems Biology, College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, China.

Appendix A. Supplementary material

Supplementary material.

.

Supplementary material.

.

Supplementary material.

.