Deconvolution analysis identified altered hepatic cell landscape in primary sclerosing cholangitis and primary biliary cholangitis

Abstract

Introduction: Primary sclerosing cholangitis (PSC) and primary biliary cholangitis (PBC) are characterized by ductular reaction, hepatic inflammation, and liver fibrosis. Hepatic cells are heterogeneous, and functional roles of different hepatic cell phenotypes are still not defined in the pathophysiology of cholangiopathies. Cell deconvolution analysis estimates cell fractions of different cell phenotypes in bulk transcriptome data, and CIBERSORTx is a powerful deconvolution method to estimate cell composition in microarray data. CIBERSORTx performs estimation based on the reference file, which is referred to as signature matrix, and allows users to create custom signature matrix to identify specific phenotypes. In the current study, we created two custom signature matrices using two single cell RNA sequencing data of hepatic cells and performed deconvolution for bulk microarray data of liver tissues including PSC and PBC patients.

Methods: Custom signature matrix files were created using single-cell RNA sequencing data downloaded from GSE185477 and GSE115469. Custom signature matrices were validated for their deconvolution performance using validation data sets. Cell composition of each hepatic cell phenotype in the liver, which was identified in custom signature matrices, was calculated by CIBERSORTx and bulk RNA sequencing data of GSE159676. Deconvolution results were validated by analyzing marker expression for the cell phenotype in GSE159676 data.

Results: CIBERSORTx and custom signature matrices showed comprehensive performance in estimation of population of various hepatic cell phenotypes. We identified increased population of large cholangiocytes in PSC and PBC livers, which is in agreement with previous studies referred to as ductular reaction, supporting the effectiveness and reliability of deconvolution analysis in this study. Interestingly, we identified decreased population of small cholangiocytes, periportal hepatocytes, and interzonal hepatocytes in PSC and PBC liver tissues compared to healthy livers.

Discussion: Although further studies are required to elucidate the roles of these hepatic cell phenotypes in cholestatic liver injury, our approach provides important implications that cell functions may differ depending on phenotypes, even in the same cell type during liver injury. Deconvolution analysis using CIBERSORTx could provide a novel approach for studies of specific hepatic cell phenotypes in liver diseases.

Keywords: CIBERSORTx; deconvolution analysis; hepatic cell landscape; primary biliary cholangitis; primary sclerosing cholangitis.

Figure 1

Outline of cellular deconvolution using CIBERSORTx in this study. Data of scRNA-seq (GSE185477 and GSE115469) were downloaded and used to create custom signature matrices (sc18sig and 11sig, respectively). Microarray data of human liver tissues (GSE159676) were downloaded and used as mixture, which is the target data for deconvolution. CIBERSORTx calculates cell composition in liver tissues of GSE159676 using sc18sig and 11sig.

Figure 2

Characteristics of created custom signature matrices. Two signature matrices were created in this study: one using scRNA-seq data of GSE185477 (sc18sig), and one based on data of GSE115469 (11sig). (A) Heatmaps of sc18sig and 11sig generated by CIBERSORTx in the process of custom signature matrix creation. (B) Validation of sc18sig and 11sig. Deconvolution was performed using signature matrices and validation mixture files. Estimated cell fraction was multiplied by 100 to obtain estimated cell population, and Pearson correlation analysis was performed with actual population in validation mixture data.

Figure 3

Cell composition for biliary phenotypes estimated by CIBERSORTx. Deconvolution was performed using CIBERSORTx with signature matrices, sc18sig and 11sig, and the mixture file of GSE159676. (A) Estimated fraction for the “Cholangiocytes” phenotype. (B) Microarray data of GSE159676 for genes associated with cholangiocytes. (C) Estimated fraction for “BCL2+ Cholangiocytes,” which are recognized as small cholangiocytes. (D) Expression for progenitor-associated genes in GSE159676. Mean ± SEM, ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 vs. Healthy. Sample numbers are 7 for Healthy, 12 for PSC, and 3 for PBC.

Figure 4

Cell composition for mesenchymal phenotypes estimated by CIBERSORTx. (A) Estimated fraction for myofibroblast-like HSCs (Mes2 and Mes4) using sc18sig and for “Hepatic_Stellate_Cells” using 11sig. (B) Estimated fraction for quiescent HSCs (Mes1) using sc18sig. (C) Expression for genes associated with quiescent and activated HSCs in GSE159676. Mean ± SEM, ^*p < 0.05, and ^****p < 0.0001 vs. Healthy (n = 7 for Healthy, n = 12 for PSC, n = 3 for PBC).

Figure 5

Increased fractions of specific immune cells estimated by CIBERSORTx. (A) Estimated composition for inflammatory macrophages using sc18sig and 11sig. (B) Expression levels of genes associated with inflammatory macrophages. (C) Estimated fractions for NK cells. (D) Marker expression associated with NK cells. (E) Estimated fractions for plasma cells. (F) Expression levels for plasma cell markers. Mean ± SEM, ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 vs. Healthy (n = 7 for Healthy, n = 12 for PSC, n = 3 for PBC).

Figure 6

Decreased composition of periportal and interzonal hepatocytes in PSC/PBC livers. (A) Estimated composition for periportal and interzonal hepatocytes using sc18sig. (B) Estimated composition for periportal/interzonal hepatocytes using 11sig. (C) Marker expression associated with periportal hepatocytes. (D) Expression levels of marker genes for interzonal hepatocytes. Mean ± SEM, ^*p < 0.05, ^**p < 0.01, ^***p < 0.001 vs. Healthy (n = 7 for Healthy, n = 12 for PSC, n = 3 for PBC).

Figure 7

Alteration of cell composition during cholestatic liver injury. Previous studies and estimation using CIBERSORTx indicate that some specific hepatic cell phenotypes are increased or decreased in PSC and/or PBC compared to healthy conditions. Increased large cholangiocytes lead to ductular reaction, and increased activated HSCs and inflammatory macrophages lead to hepatic fibrogenesis or inflammation, respectively.