Identification of secretory protein related Biomarkers for Primary Biliary Cholangitis based on Machine Learning and experimental validation

doi:10.21203/rs.3.rs-4907809/v1

Download PDF

Article

Identification of secretory protein related Biomarkers for Primary Biliary Cholangitis based on Machine Learning and experimental validation

https://doi.org/10.21203/rs.3.rs-4907809/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Primary biliary cholangitis (PBC) is challenging to diagnose and treat due to its insidious onset. This study aimed to identify effective diagnostic biomarkers for PBC by focusing on secreted proteins through bioinformatics approaches. Two PBC-related bulk datasets, GSE119600 and GSE61260, were retrieved from the GEO database for analysis and validation, respectively. Gene sets related to secreted proteins were sourced from the THPA database. The analysis of GSE119600 included differential expression analysis, WGCNA, immune infiltration analysis, and enrichment analyses. By intersecting differentially expressed genes (DEGs), WGCNA hub module genes, and genes related to secreted proteins, 18 candidate genes were identified. Machine learning techniques—LASSO, random forest, GMM, and SVM-RFE—narrowed these to four hub genes: CSF1R, PLCH2, SLC38A1, and CST7. The diagnostic performance of these genes was assessed using LDA, QDA, Bayesian test, and Nomogram methods, with internal and external validation AUC values of 0.867 and 0.722, respectively. Experimental validation in PBC model mice confirmed that the expression of these genes was significantly altered. These findings suggest that CSF1R, PLCH2, SLC38A1, and CST7 could serve as novel diagnostic biomarkers for early PBC detection and provide insights into its underlying mechanisms.

biomarkers

bioinformatics

machine learning

secretory proteins

Primary biliary cholangitis

Primary biliary cholangitis (PBC) is a chronic autoimmune liver disease with a complex pathogenesis. PBC is characterized by chronic inflammation of the small bile ducts, leading to bile duct injury and fibrosis, which may eventually progress to cirrhosis and liver failure [1–2]. In recent years, with a deeper understanding of PBC and the continuous progress and wide application of related antibody detection techniques, the global annual incidence of PBC has shown a yearly increase. According to current estimates, the number of new cases of PBC reaches at least 100,000 per year [3]. The exact etiology of PBC has not been fully elucidated, but studies have shown that both genetic and environmental factors play a role in the onset and progression of the disease [4]. The onset of PBC tends to be insidious, with patients often presenting no obvious symptoms in the initial stages. This makes early detection difficult, and many patients have already developed liver fibrosis or even cirrhosis by the time the diagnosis is confirmed [5]. Advanced PBC can lead to severe liver failure or cirrhosis, with liver transplantation being the only therapeutic option for end-stage patients [6]. Therefore, improving the diagnostic efficiency of PBC and the recognition of early symptoms is important for the survival and treatment of its patients.

Exploring the association of PBC with secreted proteins provides new insights into understanding the disease's molecular mechanisms. Secreted proteins such as bile acid transporters, anion exchangers, SIgA, antimicrobial peptides, and mucus play crucial roles in maintaining bile acid homeostasis, intestinal barrier integrity, and host-microbiome balance. Dysregulation of these secreted proteins promotes cholestasis, bile duct injury, intestinal flora dysbiosis, and metabolic disorders, leading to the pathogenesis of PBC [7–9].

Although existing bioinformatics studies have identified several diagnostic genes associated with PBC [10–11], they have not yet explored the relationship between these genes and secretory proteins, nor the mechanisms of action involved in the disease process. In this study, based on previous work, we further analyzed the PBC diagnostic genes associated with secretory proteins using machine learning techniques, optimized the diagnostic model, and improved the accuracy and efficiency of prediction. The study flow chart is shown in Fig. 1.

2.1Data sources

Two PBC-related Bulk datasets, GSE119600 and GSE61260, were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) using the search terms "primary biliary cholangitis" or "primary biliary cirrhosis". Secreted protein-related gene sets were obtained from the THPA database (https://www.proteinatlas.org/).

2.2 Acquisition of differentially expressed genes (DEGs)

The PBC-control dataset screened from GSE119600 was subjected to analysis of variance using the limma package in R (Ver. 4.3.3), with settings of p ≤ 0.05 and log2 ( fold change ) ≥ 0.25, to obtain the differential genes and visualize the results.

2.3 Differential gene function enrichment analysis

In order to explore the biological functions and specific mechanisms of PBC-related disease-causing genes, the screened PBC differential genes were subjected to GO function enrichment analysis, KEGG function enrichment analysis, and DO disease enrichment analysis by setting p ≤ 0.05 through the org.Hs.eg.db package, respectively, and their bar charts and bubble diagrams were plotted.

2. 4. Weighted gene co-expression network analysis ( WGCNA ) and hub module gene identification

The relationship between gene expression in PBC patients and controls was analyzed using weighted co-expression networks. The PBC-control dataset screened by GSE119600 was used as the input data with absolute median deviation treatment, and the co-expression network was constructed using the WGCNA package in R. Firstly, sample tree clustering was performed on the input data, andthe threshold was set to 60 to exclude outliers above the threshold. Then, network topology analysis was utilized to construct a one-step network by setting R2 to 0.85 and calculating the soft-threshold capacity, and a module cluster graph was drawn. Finally, in combination with PBC phenotypes, the expression patterns of the module genes in individual samples were demonstrated by the module eigenvalues. The specific modules that are significantly associated with the content of PBC were determined using module eigenvector gene analysis. Genes in the corresponding modules were selected and intersected with PBC differential genes and secreted protein-related genes to obtain candidate genes for thediagnosis of PBC, and their Venn diagrams were drawn.

2.5. Machine Learning

A series of machine learning algorithms were used to further screen the hub genes for diagnosing PBC, construct its diagnostic model, and validate it. Using a stratified sampling strategy, 80% of the samples were randomly selected as a training group through the sample function in R to participate in the training of the machine learning model. The remaining 20% of the samples were retained as a validation group to assess and verify the diagnostic efficacy of the model and to detect whether the model overfitted the training data. Finally, the GSE61260 dataset was used as an external test group to further assess the model's performance and diagnostic power on independent datasets.

Genes were screened and PBC diagnostic models were constructed using LASSO, RandomForest, SVM-RFE, and GMM machine learning algorithms. Discriminant analysis and naiveBayes methods were then used to validate the diagnostic effectiveness of the models.

2.6. Nomogram

Nomogram models were constructed using the hub genes. To assess the predictive accuracy of the column-line graph model, calibration curves were used. In addition, the validity of the model in practical clinical applications was measured by decision curve analysis (DCA). ROC curves were plotted and the area under the curve(AUC) was calculated to test the diagnostic performance of the candidate genes and the column-line diagram model.

2.7. Immune infiltration analysis

The gene expression data of PBC patients in GSE119600 were extracted, and the cibersort algorithm of the IOBR package in R was used to evaluate the types and numbers of PBC immune cells. The ggplot2 package was then used to plot the proportion of immune cells in each sample. The epic and quantiseq algorithms were used to evaluate and screen out the hub immune cells of PBC.

The PBC samples were categorized into high expression and low expression groups according to whether the hub gene expression was higher than the median of all samples. Their immune cell heat maps were plotted using the ggboxplot package. The GSVA package was used to perform a single-sample gene set enrichment analysis (ssGSEA), comparing the infiltration of 28 immune cells in the PBC samples.

A group comparison plot was drawn to verify the expression of CSF1R between the control and PBC groups. The PBC-control gene expression matrix in the GSE119600 dataset was then extracted, its ROC curve was plotted, and GSEA gene set enrichment analysis was performed on the hub genes, with their enrichment shubs plotted. A scatterplot was created using the ggscatterstats function to display the correlation between immune cell ratios and the expression of hub genes.

2.8.Animal model validation

Twenty mice were randomly divided into two groups, with ten in the model group and ten in the control group. The PBC model was established using 2-OA-BSA combined with intraperitoneal injection of PolyI[12]. After modeling, we euthanize laboratory animals in strict accordance with ethical and humane guidelines. We selected 1% sodium pentobarbital for intraperitoneal injection, with the dosage calculated based on the animal's body weight to ensure a rapid and painless loss of consciousness. After anesthesia, euthanasia is performed by cardiectomy, ensuring the rapid cessation of heartbeat and respiration, ultimately leading to the irreversible loss of brain function. This method avoids unnecessary pain and distress. All personnel performing euthanasia are professionally trained and hold the requisite qualifications and experience. Additionally, the protocol was reviewed and approved by the Institutional Animal Ethics Committee to ensure its ethical and scientific rigor. Post-euthanasia, a thorough examination is conducted by a qualified professional to confirm irreversible death. Our research design rigorously adheres to the 3R principle, aiming to minimize animal use while maximizing experimental efficiency and quality to ensure both scientific integrity and humane practices. The livers from both groups were then extracted for HE staining to confirm the successful induction of the PBC model. Subsequently, the differential genes identified through bioinformatics analysis and machine learning were validated for actual expression using qRT-PCR and Western blot.

2.9.Ethical Declarations

All animal experiments were conducted in accordance with the relevant guidelines and regulations. The experimental protocols were approved by the Wuhan Hualianke Biotechnology Co., Ltd. and the Animal Ethics Committee, under the approval number HLK-20230518-002. All procedures performed on animals were in strict compliance with the "Guidelines for the Care and Use of Laboratory Animals" published by the National Institutes of Health (NIH).

Furthermore, the study was conducted in accordance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines to ensure the proper and humane treatment of the animals used in the study.

3. 1.PBC differential genes

A total of 827 down-regulated genes and 639 up-regulated genes were obtained and their volcano maps were plotted (Figure. 2A).All differentially expressed up-regulated genes and down-regulated genes were extracted from the gene expression matrix, which were categorized into the PBC group and the control group and plotted in their heat maps (Figure. 2B).

3. 2. Differential enrichment analysis

DEGs were enriched in BP for cellular respiration, aerobic respiration, and the electron transport chain; in CC for the inner mitochondrial membrane, mitochondrial protein complex, and ribosome; and in MF for ribosome structural composition, primary active transmembrane transporter activity, and electron transfer activity. In the KEGG analysis, DEGs were significantly enriched in multiple neurodegenerative diseases, including amyotrophic lateral sclerosis and Alzheimer's disease. In the DO disease analysis, DEGs were significantly enriched in primary immune diseases, bronchial diseases, and bacterial infectious diseases. The visualization results are shown in Fig. 3.

3. 3. Immune infiltration analysis

Gene expression was extracted from the GSE119600 dataset for PBC and controls. The CIBERSORT algorithm was applied to evaluate the immune cell types of PBC patients and draw a scale diagram (Fig. 4A). Then, immune infiltration analysis was performed between PBC and control groups using the CIBERSORT (Fig. 4B), EPIC (Fig. 4C), and quanTIseq (Fig. 4D) algorithms, respectively. In the CIBERSORT algorithm, memory B cells, CD8 + T cells, resting memory CD4 + T cells, and regulatory T cells were lowly expressed in PBC patients, while immature CD4 + T cells, M0-type macrophages, and dendritic cell activation were highly expressed. In the EPIC algorithm, CD8 + T cells were likewise lowly expressed in PBC patients. In the quanTIseq algorithm, M2-type macrophages and neutrophils were highly expressed in PBC patients, and monocytes were lowly expressed.

3. 4.WGCNA

The PBC patients in the GSE119600 dataset were evaluated for scale independence and average connectivity after the outlier removal process (Figure. 5A, B). The optimal power value of 10 was selected based on a correlation coefficient greater than 0.85 (Figure. 5C), and the topological overlap matrix (TOM) was constructed with a soft threshold power of 10.A total of 12 gene modules were identified (Figure. 5E), and the topological overlap of the gene network was shown by a heat map (Figure. 5D). The correlation value of the blue module with gene significance was 0.18 (p = 1.2e-15) (Figure. 5F). Therefore, further studies focused on the genes within the blue module, which contained a total of 1,949 genes.

3. 5.Machine learning

Candidate genes for diagnosing PBC were obtained by taking the intersection of DEGs, secreted protein-related genes, and WGCNA blue module genes and plotting a Venn diagram (Fig. 6A). A series of machine learning analyses were performed on the expression matrices of the candidate genes to filter out the hub genes.

First, using Random Forest, we identified 18 candidate genes (Fig. 6D) and visualized the prediction accuracy (Fig. 6B) and gene importance shubs (Fig. 6C) of the Random Forest model. Second, by SVM-RFE analysis, we constructed the model and achieved the highest accuracy of 0.13 (Fig. 6F) and the lowest error rate of 0.87 when the model included 18 hub genes (Fig. 6E). Then, LASSO regression analysis identified 8 hub genes at the lowest lambda value and showed the regression coefficients (Fig. 6G) and the results of regression cross-validation (Fig. 6H). Finally, by GMM analysis, we identified 14 hub genes at the highest AUC value of the ROC curve (0.96) and visualized the AUC value of the individual regression models (Fig. 6I). The best models for diagnosing PBC constructed by the above machine learning methods were validated with the validation group, respectively. The corresponding ROC curves were plotted and the AUC values were calculated (Figure. 6J, K, L). The AUC values for RF, SVM-RFE, and LASSO were 0.919, 0.839, and 0.861,respectively.

The hub genes screened by each of RF, SVM-RFE, LASSO, and GMM were intersected to obtain the hub genes for diagnosing PBC: CSF1R, PLCH2, SLC38A1, and CST7, and their Venn diagrams were plotted (Fig. 7A). The hub genes were subjected to Bayesian testing and modeling, and the diagnostic efficacy of the model was tested using validation and test groups. Their ROC curves were plotted and the AUC values were calculated (Figure. 7B, C), which were 0.867 and 0.722, respectively. The diagnostic error rates were calculated using LDA and QDA algorithms. The total error rate was 0.32 for both, and heat maps were plotted to visualize the results (Figure. 7D, E).

3. 6.Hub gene expression and ROC

The expression matrix of hub genes was extracted from the GSE11960 dataset, divided into PBC and control groups. CSF1R, PLCH2, and SLC38A1 were highly expressed in PBC patients, while CST7 was lowly expressed. Box plots were drawn to show the differences in hub gene expression (Figure. 7F). The diagnostic potency of the respective hub genes for PBC patients in the GSE11960 dataset was calculated, and their ROC curves were plotted with AUC values calculated (Fig. 7G), with CSF1R having the highest AUC at 0.736.

3.7.Nomogram

Nomogram diagnostic plots of hub genes were drawn (Fig. 7H), and ROC curves (Fig. 7I), correction curves (Fig. 7J), and DCA curves (Fig. 7K) of the Nomogram were plotted. The AUC was 0.778 and the corrected curve as well as the DCA curve showed a good diagnosis.

3. 8. ssGSEA

Immunoinfiltration analysis using ssGSEA was performed for each hub gene, categorizing PBC patients into high and low expression groups based on whether their expression levels were above the median. Effector memory CD8 T cells, natural killer T cells, plasma-like dendritic cells, T follicular helper cells, and type 2 T helper cells were more predominant in the CSF1R high-expression group than in the low-expression group. In the CST7 high-expression group, activated dendritic cells were more predominant than in the low-expression group. Central memory CD4 T cells, immature dendritic cells, and type 2 T helper cells were more predominant in the CST7 low-expression group than in the high-expression group. Natural killer T cells and T follicular helper cells were more predominant in the PLCH2 high-expression group than in the low-expression group, and macrophages, activated dendritic cells, and neutrophils were more predominant in the PLCH2 low-expression group than in the high-expression group. Activated B cells, activated CD4 T cells, activated CD8 T cells, central memory CD4 T cells, effector memory CD8 T cells, young B cells, type 1 T helper cells, and type 17 T helper cells were more predominant in the SLC38A1 high-expression group than in the low-expression group. Eosinophils, mast cells, neutrophils, and myeloid-derived suppressor cells were more predominant in the SLC38A1 low-expression group than in the high-expression group. The results are visualized in the heat map shown in Fig. 8.

3. 9.CSF1R

The best-expressed CSF1Rs among the hub genes were subjected to GSEA analysis. PBC patients were divided into CSF1R high and low expression groups based on whether their CSF1R expression was higher than the median, and were subjected to GSEA enrichment analysis using hallmark gene sets. The absolute value of NES was taken as the top five positive values and the two negative values to plot the GSEA enrichment analysis shubs, as shown in Fig. 9. The CSF1R high-expression group was mainly enriched in allograft rejection, inflammatory response, interferon α response, interferon γ response, and unfolded protein response. The CSF1R low-expression group was mainly enriched in heme metabolism and oxidative phosphorylation. CSF1R gene expression was extracted, and the results of the CIBERSORT immune infiltration analysis of PBC patients were plotted as immune component correlation scatter plots, as shown in Fig. 10. Figure 10 showed that CSF1R gene expression was significantly correlated with B memory cells, CD4 + memory resting T cells, CD8 + T cells, and regulatory T cell components.

3. 10. Animal experiment results

HE staining results showed that in PBC mice, multiple inflammatory cell aggregates could be seen in the hepatic portal vein area. Granulomas were observed in the area of inflammatory cell aggregates, and a large number of lymphocytic infiltrations were present in the biliary epithelium. The results of Western blot and qRT-PCR showed that the mRNA and protein levels of PLCH2, CSF1R, p-CSF1R, and SLC38A1 were significantly increased compared to the Normal group, whereas those of CST7 were decreased, consistent with previous bioinformatics analysis results, As shown in Fig. 11.

In this study, a diagnostic model of PBC, as well as an in-depth study of its pathogenesis, was constructed through the integrated application of bioinformatics analysis and machine learning techniques. The diagnostic model was verified through experiments. CSF1R, SLC38A1, PLCH2, and CST7 together constructed a model for the diagnosis of PBC, which can be used for the early diagnosis of PBC and facilitate its clinical prophylactic treatment.

The results of the DO enrichment analysis suggest a potential link between DEGs in PBC and mitochondrial dysfunction, autoimmune diseases, inflammatory responses, and abnormalities in immune regulation. Mitochondrial dysfunction may be involved in the pathogenesis of PBC by exacerbating cellular damage and inflammatory responses through mechanisms such as oxidative stress and calcium ion homeostasis imbalance [13–14]. In addition, the emergence of asthma and bronchial diseases further emphasizes the role of chronic inflammation in the pathogenesis of PBC. Inflammatory responses are not only involved in the early course of PBC but may also influence the disease progression of PBC by promoting the development of hepatic fibrosis and cirrhosis. Inflammatory mediators, such as cytokines and chemokines, play a hub role in the pathogenesis of PBC and may be potential targets for future treatment of PBC [15–16].

The results of the GO enrichment analysis showed that PBC is closely related to mitochondrial function, especially cellular respiration, the electron transport chain, oxidative phosphorylation, and the assembly and function of the mitochondrial respiratory chain complex I. Impairment of the mitochondria's central role in energy metabolism may lead to decreased energy production, which in turn affects the normal function of liver cells [17–18].

The results of the KEGG enrichment analyses revealed potential links between PBC and immunomodulation, energy metabolism, inflammatory responses, and neurodegenerative processes. In particular, the enrichment of oxidative phosphorylation pathways emphasizes the potential role of mitochondrial dysfunction in energy metabolism in PBC, while pathways in diabetic cardiomyopathy and non-alcoholic fatty liver disease may be associated with the metabolic disturbances observed in PBC patients [19]. In addition, the pathway enrichment of reactive oxygen species in chemical carcinogenesis may indicate a role for oxidative stress in the pathogenesis of PBC [20–22]. Although the direct association of pathways in neurodegenerative diseases such as Huntington's disease, Parkinson's disease, and Alzheimer's disease with PBC is not clear, these pathways may suggest potential neurological implications of PBC [23–24].

The results of immune infiltration analysis showed low expression of memory B cells, CD8 + T cells, resting memory CD4 + T cells, and regulatory T cells in patients with PBC, which may indicate suppression of immune response and imbalance of immune tolerance, potentially promoting the development of autoimmune response [25–26]. Meanwhile, the high expression of immature CD4 + T cells, M0-type macrophages, and dendritic cells, as well as an increase in M2-type macrophages and neutrophils, reflect the activation of inflammatory responses and the activated state of immune cells in PBC patients [27–28]. In addition, the low expression of monocytes may be associated with immunomodulation or immunosuppression in the disease [29].

The CSF1R gene encodes the macrophage colony-stimulating factor 1 receptor (CSF1R), which plays a crucial role in macrophage growth, differentiation, and survival [30]. CSF1R is a transmembrane receptor tyrosine kinase that regulates immune cell function and response by activating downstream signaling pathways, mainly through binding to its ligand CSF1 (M-CSF) or IL-34 [31]. In recent years, numerous studies have shown that CSF1R plays a hub role in a variety of immune-related diseases, especially in autoimmune and inflammatory diseases [32]. In PBC, CSF1R may have an important impact on the onset and progression of the disease by regulating the activities of macrophages and other immune cells [33]. A large number of infiltrating macrophages were found in the livers of patients with PBC [34]. Whether the activation and survival of these macrophages are dependent on the CSF1R signaling pathway remains to be investigated. Activation of the CSF1R signaling pathway not only promotes macrophage survival and proliferation but also enhances the pro-inflammatory function of these cells [35], thereby exacerbating liver inflammation and fibrosis [36].

The PLCH2 (Phospholipase C Eta 2) gene encodes a member of the phospholipase C family, a family of enzymes that play hub roles in intracellular signaling. PLCH2 generates two important second messenger molecules, inositol triphosphate (IP3) and diacylglycerol (DAG), by hydrolyzing phosphatidylinositol-4,5-bisphosphate (PIP2), and these messenger molecules regulate intracellular calcium ion release and protein kinase C (PKC) activation, respectively, thereby affecting a variety of cellular processes, including cell proliferation, differentiation, and metabolism [37]. Changes in intracellular calcium ion concentration are important regulators of a variety of immune cell functions, including the activation and differentiation of T and B cells [38]. The PLCH2-mediated hydrolysis process of PIP2, which generates IP3, induces the release of calcium ions from the endoplasmic reticulum, thereby affecting immune cell function [39]. In PBC, aberrant calcium signaling may lead to overactivation of immune cells and overproduction of inflammatory mediators, which in turn exacerbate liver injury and fibrosis [40]. PLCH2 affects the pathological process of PBC by regulating bile duct epithelial cell tight junctions and barrier properties, and may influence the function and barrier integrity of the bile duct epithelial cells by affecting the bile duct epithelial cell function and barrier integrity [41].

The SLC38A1 (Solute Carrier Family 38 Member 1) gene encodes a sodium-dependent neutral amino acid transporter protein also known as SNAT1. SLC38A1 is responsible for the transport of a wide range of neutral amino acids in cells, including glutamine, alanine, and leucine, which play important roles in protein synthesis, cell signaling, and energy metabolism [42]. SLC38A1 is expressed in many tissues, particularly in the brain and liver, and its function is critical for the regulation of cellular nutrition and metabolism [43]. By regulating the uptake of hub amino acids by immune cells, SLC38A1 may influence their metabolic state and function, which in turn may impact the immune response to PBC [44]. For example, glutamine is an important energy source and precursor for immune cells, and its inadequate supply may affect their activity and function [45]. SLC38A1, by regulating the uptake of neutral amino acids by hepatocytes, may play a hub role in maintaining the metabolic homeostasis and function of hepatocytes, thus affecting the progression of PBC [46].

The CST7 (Cystatin F, also known as Leukocystatin) gene encodes a cysteine protease inhibitor, and Cystatin F is a member of the Cystatin family of proteins involved in regulating various physiological processes through the inhibition of intra- and extracellular degradation of cysteine proteases (e.g., histones C and L) [47]. These processes include apoptosis, protein degradation, and immune regulation [47]. CST7 is mainly expressed in immune system cells, such as natural killer (NK) cells and macrophages, and plays an important role in regulating immune responses and inflammatory processes [48]. The low expression of CST7 may lead to increased cysteine protease activity in the immune system of PBC patients, triggering a series of pathological changes. Increased cysteine protease activity may result in hyperactivity of immune cells (e.g., NK cells and macrophages) and enhanced attacks on bile duct epithelial cells, exacerbating bile duct injury and destruction [49]. Cysteine proteases play an important role in the inflammatory response, and their increased activity may lead to excessive release of pro-inflammatory cytokines and mediators in the liver, further exacerbating the progression of inflammation and fibrosis [50].

In the present study, we successfully validated the expression changes of hub genes in PBC using a mouse model to assess their value as potential diagnostic markers. Using Western blot and qRT-PCR techniques, we observed that the mRNA and protein levels of PLCH2, CSF1R, p-CSF1R, and SLC38A1 were significantly elevated in the mouse model of PBC, whereas the mRNA and protein levels of CST7 were decreased. The results of these two techniques were highly consistent, suggesting that these hub genes may be reliable markers for the disease.

In this study, CSF1R, PLCH2, SLC38A1, and CST7 secretory protein-related genes were screened as potential diagnostic markers for PBC using bioinformatics analysis and machine learning techniques. Abnormal expression of these genes may be closely related to the pathological process of PBC, revealing their hub roles in immune regulation, metabolic homeostasis, and inflammatory response. However, this study has some shortcomings, such as a relatively small sample size, insufficient validation experiments, and the need for further research to translate the bioinformatics predicted results into clinical applications. In the future, in-depth study of the functions and mechanisms of these genes is expected to develop precise therapeutic strategies for PBC, improving the prognosis and quality of life for patients.

Funding

This research was financially supported by National Natural Science Foundation of China (Project No. 82274367).

Availability of data and materials

Publicly available datasets were analyzed in this study. GSE119600 and GSE61260, were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/).Secreted protein-related gene sets were obtained from the THPA database (https://www.proteinatlas.org/).

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Author contributions

ZX and YC designed the research, analyzed data, LHZ,SG and JX performed statistical analysis. ZX and YL wrote the paper. YJ,LZ and LC revised the manuscript. All authors have read and approved the final manuscript. ZX,YC and YL share first authorship.

Levy C, Manns M, Hirschfield G. New Treatment Paradigms in Primary Biliary Cholangitis. Clin Gastroenterol Hepatol. 2023;21(8):2076-2087. doi:10.1016/j.cgh.2023.02.005
Sarcognato S, Sacchi D, Grillo F, et al. Autoimmune biliary diseases: primary biliary cholangitis and primary sclerosing cholangitis. Pathologica. 2021;113(3):170-184. doi:10.32074/1591-951X-245
Lleo A, Wang GQ, Gershwin ME, Hirschfield GM. Primary biliary cholangitis. Lancet. 2020;396(10266):1915-1926. doi:10.1016/S0140-6736(20)31607-X
Gulamhusein AF, Hirschfield GM. Primary biliary cholangitis: pathogenesis and therapeutic opportunities. Nat Rev Gastroenterol Hepatol. 2020;17(2):93-110. doi:10.1038/s41575-019-0226-7
Trivella J, John BV, Levy C. Primary biliary cholangitis: Epidemiology, prognosis, and treatment. Hepatol Commun. 2023;7(6):e0179. Published 2023 Jun 2. doi:10.1097/HC9.0000000000000179
Akamatsu N, Sugawara Y. Primary biliary cirrhosis and liver transplantation. Intractable Rare Dis Res. 2012;1(2):66-80. doi:10.5582/irdr.2012.v1.2.66
Prieto J, Banales JM, Medina JF. Primary biliary cholangitis: pathogenic mechanisms. Curr Opin Gastroenterol. 2021;37(2):91-98. doi:10.1097/MOG.0000000000000703
Li H, Guan Y, Han C, et al. The pathogenesis, models and therapeutic advances of primary biliary cholangitis. Biomed Pharmacother. 2021;140:111754. doi:10.1016/j.biopha.2021.111754
Zhao Y, Wei S, Chen L, Zhou X, Ma X. Primary biliary cholangitis: molecular pathogenesis perspectives and therapeutic potential of natural products. Front Immunol. 2023;14:1164202. Published 2023 Jun 30. doi:10.3389/fimmu.2023.1164202
Pham HN, Pham L, Sato K. Bioinformatic analysis identified novel candidate genes with the potentials for diagnostic blood testing of primary biliary cholangitis. PLoS One. 2023;18(10):e0292998. Published 2023 Oct 16. doi:10.1371/journal.pone.0292998
Wang H, Zhang J, Liu J, Jiang Y, Fu L, Peng S. Identification of AKR1B10 as a hub gene in primary biliary cholangitis by integrated bioinformatics analysis and experimental validation. Front Mol Biosci. 2023;10:1124956. Published 2023 Feb 9. doi:10.3389/fmolb.2023.1124956
Wakabayashi K, Lian ZX, Leung PS, et al. Loss of tolerance in C57BL/6 mice to the autoantigen E2 subunit of pyruvate dehydrogenase by a xenobiotic with ensuing biliary ductular disease. Hepatology. 2008;48(2):531-540. doi:10.1002/hep.22390
Prieto J, Banales JM, Medina JF. Primary biliary cholangitis: pathogenic mechanisms. Curr Opin Gastroenterol. 2021;37(2):91-98. doi:10.1097/MOG.0000000000000703
Kadaja L, Kisand KE, Peet N, et al. IgG from patients with liver diseases inhibit mitochondrial respiration in permeabilized oxidative muscle cells: impaired function of intracellular energetic units?. Mol Cell Biochem. 2004;256-257(1-2):291-303. doi:10.1023/b:mcbi.0000009876.23921.e6 2021;12:591236. Published 2021 Mar 25. doi:10.3389/fimmu.2021.591236
Yang Y, He X, Rojas M, Leung PSC, Gao L. Mechanism-based target therapy in primary biliary cholangitis: opportunities before liver cirrhosis?. Front Immunol. 2023;14:1184252. Published 2023 May 30. doi:10.3389/fimmu.2023.1184252
Medford A, Childs J, Little A, et al. Emerging Therapeutic Strategies in The Fight Against Primary Biliary Cholangitis. J Clin Transl Hepatol. 2023;11(4):949-957. doi:10.14218/JCTH.2022.00398
Hollingsworth KG, Newton JL, Taylor R, et al. Pilot study of peripheral muscle function in primary biliary cirrhosis: potential implications for fatigue pathogenesis. Clin Gastroenterol Hepatol. 2008;6(9):1041-1048. doi:10.1016/j.cgh.2008.04.013
Xu Q, Zhu W, Yin Y. Diagnostic value of anti-mitochondrial antibody in patients with primary biliary cholangitis: A systemic review and meta-analysis. Medicine (Baltimore). 2023;102(45):e36039. doi:10.1097/MD.0000000000036039
Hao J, Yang T, Zhou Y, et al. Serum Metabolomics Analysis Reveals a Distinct Metabolic Profile of Patients with Primary Biliary Cholangitis. Sci Rep. 2017;7(1):784. Published 2017 Apr 11. doi:10.1038/s41598-017-00944-9
Dubreuil MM, Morgens DW, Okumoto K, et al. Systematic Identification of Regulators of Oxidative Stress Reveals Non-canonical Roles for Peroxisomal Import and the Pentose Phosphate Pathway. Cell Rep. 2020;30(5):1417-1433.e7. doi:10.1016/j.celrep.2020.01.013
Rusz M, Del Favero G, El Abiead Y, et al. Morpho-metabotyping the oxidative stress response. Sci Rep. 2021;11(1):15471. Published 2021 Jul 29. doi:10.1038/s41598-021-94585-8
Afzal S, Abdul Manap AS, Attiq A, Albokhadaim I, Kandeel M, Alhojaily SM. From imbalance to impairment: the central role of reactive oxygen species in oxidative stress-induced disorders and therapeutic exploration. Front Pharmacol. 2023;14:1269581. Published 2023 Oct 18. doi:10.3389/fphar.2023.1269581
Mosher VAL, Swain MG, Pang JXQ, et al. Primary Biliary Cholangitis Alters Functional Connections of the Brain's Deep Gray Matter. Clin Transl Gastroenterol. 2017;8(7):e107. Published 2017 Jul 27. doi:10.1038/ctg.2017.34
Grover VP, Southern L, Dyson JK, et al. Early primary biliary cholangitis is characterised by brain abnormalities on cerebral magnetic resonance imaging. Aliment Pharmacol Ther. 2016;44(9):936-945. doi:10.1111/apt.13797
Shi TY, Zhang FC. Role of autoimmunity in primary biliary cirrhosis. World J Gastroenterol. 2012;18(48):7141-7148. doi:10.3748/wjg.v18.i48.7141
Zhao Y, Wei S, Chen L, Zhou X, Ma X. Primary biliary cholangitis: molecular pathogenesis perspectives and therapeutic potential of natural products. Front Immunol. 2023;14:1164202. Published 2023 Jun 30. doi:10.3389/fimmu.2023.1164202
Wang C, Shi Y, Wang X, et al. Peroxisome Proliferator-Activated Receptors Regulate Hepatic Immunity and Assist in the Treatment of Primary Biliary Cholangitis. Front Immunol. 2022;13:940688. Published 2022 Jul 8. doi:10.3389/fimmu.2022.940688
Tsuneyama K, Baba H, Morimoto Y, Tsunematsu T, Ogawa H. Primary Biliary Cholangitis: Its Pathological Characteristics and Immunopathological Mechanisms. J Med Invest. 2017;64(1.2):7-13. doi:10.2152/jmi.64.7
Di Marco GS, Chasan AI, Boeckel GR, Beul K, Pavenstädt H, Roth J, Brand M. Monocytes as Targets for Immunomodulation by Regional Citrate Anticoagulation. International Journal of Molecular Sciences. 2024; 25(5):2900. https://doi.org/10.3390/ijms25052900
Stanley ER, Chitu V. CSF-1 receptor signaling in myeloid cells. Cold Spring Harb Perspect Biol. 2014;6(6):a021857. Published 2014 Jun 2. doi:10.1101/cshperspect.a021857
Hagan N, Kane JL, Grover D, et al. CSF1R signaling is a regulator of pathogenesis in progressive MS. Cell Death Dis. 2020;11(10):904. Published 2020 Oct 23. doi:10.1038/s41419-020-03084-7
Chen D, Xiong L, Zhang L, et al. CSF1R is a Prognostic Biomarker and Correlated with Immune Cell Infiltration in the Gastric Cancer Microenvironment. Pharmgenomics Pers Med. 2021;14:445-457. Published 2021 Apr 13. doi:10.2147/PGPM.S301303
Wen Y, Lambrecht J, Ju C, Tacke F. Hepatic macrophages in liver homeostasis and diseases-diversity, plasticity and therapeutic opportunities. Cell Mol Immunol. 2021;18(1):45-56. doi:10.1038/s41423-020-00558-8
Wen Y, Lambrecht J, Ju C, Tacke F. Hepatic macrophages in liver homeostasis and diseases-diversity, plasticity and therapeutic opportunities. Cell Mol Immunol. 2021;18(1):45-56. doi:10.1038/s41423-020-00558-8
Mun SH, Park PSU, Park-Min KH. The M-CSF receptor in osteoclasts and beyond. Exp Mol Med. 2020;52(8):1239-1254. doi:10.1038/s12276-020-0484-z
Zong C, Meng Y, Ye F, et al. AIF1 + CSF1R + MSCs, induced by TNF-α, act to generate an inflammatory microenvironment and promote hepatocarcinogenesis. Hepatology. 2023;78(2):434-451. doi:10.1002/hep.32738
Katan M, Cockcroft S. Phosphatidylinositol(4,5)bisphosphate: diverse functions at the plasma membrane. Essays Biochem. 2020;64(3):513-531. doi:10.1042/EBC20200041
Izquierdo JH, Bonilla-Abadía F, Cañas CA, Tobón GJ. Calcium, channels, intracellular signaling and autoimmunity. Reumatol Clin. 2014;10(1):43-47. doi:10.1016/j.reuma.2013.05.008
Vig M, Kinet JP. Calcium signaling in immune cells [published correction appears in Nat Immunol. 2009 Feb;10(2):223]. Nat Immunol. 2009;10(1):21-27. doi:10.1038/ni.f.220
Seegren PV, Harper LR, Downs TK, et al. Reduced mitochondrial calcium uptake in macrophages is a major driver of inflammaging. Nat Aging. 2023;3(7):796-812. doi:10.1038/s43587-023-00436-8
Rao RK, Samak G. Bile duct epithelial tight junctions and barrier function. Tissue Barriers. 2013;1(4):e25718. doi:10.4161/tisb.25718
Gauthier-Coles G, Vennitti J, Zhang Z, et al. Quantitative modelling of amino acid transport and homeostasis in mammalian cells. Nat Commun. 2021;12(1):5282. Published 2021 Sep 6. doi:10.1038/s41467-021-25563-x
Qureshi T, Sørensen C, Berghuis P, et al. The Glutamine Transporter Slc38a1 Regulates GABAergic Neurotransmission and Synaptic Plasticity. Cereb Cortex. 2019;29(12):5166-5179. doi:10.1093/cercor/bhz055
Sugiura A, Beier KL, Chi C, et al. Tissue-Specific Dependence of Th1 Cells on the Amino Acid Transporter SLC38A1 in Inflammation. Preprint. bioRxiv. 2023;2023.09.13.557496. Published 2023 Sep 13. doi:10.1101/2023.09.13.557496
Xiao Y, Hu B, Guo Y, et al. Targeting Glutamine Metabolism as an Attractive Therapeutic Strategy for Acute Myeloid Leukemia. Curr Treat Options Oncol. 2023;24(8):1021-1035. doi:10.1007/s11864-023-01104-0
Nwosu ZC, Megger DA, Hammad S, et al. Identification of the Consistently Altered Metabolic Targets in Human Hepatocellular Carcinoma. Cell Mol Gastroenterol Hepatol. 2017;4(2):303-323.e1. Published 2017 May 31. doi:10.1016/j.jcmgh.2017.05.004
Sawyer AJ, Garand M, Chaussabel D, Feng CG. Transcriptomic Profiling Identifies Neutrophil-Specific Upregulation of Cystatin F as a Marker of Acute Inflammation in Humans. Front Immunol. 2021;12:634119. Published 2021 Apr 1. doi:10.3389/fimmu.2021.634119
Magister S, Kos J. Cystatins in immune system. J Cancer. 2013;4(1):45-56. doi:10.7150/jca.5044
Chen J, Zhang S. The Role of Inflammation in Cholestatic Liver Injury. J Inflamm Res. 2023;16:4527-4540. Published 2023 Oct 13. doi:10.2147/JIR.S430730
Ruiz-Blázquez P, Pistorio V, Fernández-Fernández M, Moles A. The multifaceted role of cathepsins in liver disease. J Hepatol. 2021;75(5):1192-1202. doi:10.1016/j.jhep.2021.06.031

No competing interests reported.

Download PDF

Editor invited by journal
29 Aug, 2024
Submission checks completed at journal
28 Aug, 2024
First submitted to journal
13 Aug, 2024

You are reading this latest preprint version

Identification of secretory protein related Biomarkers for Primary Biliary Cholangitis based on Machine Learning and experimental validation

Status:

Version 1

Abstract

Figures

1 Introduction

2 Methods and materials