Gene network activity in cultivated primary hepatocytes is highly similar to diseased mammalian liver tissue
- 1.9k Downloads
It is well known that isolation and cultivation of primary hepatocytes cause major gene expression alterations. In the present genome-wide, time-resolved study of cultivated human and mouse hepatocytes, we made the observation that expression changes in culture strongly resemble alterations in liver diseases. Hepatocytes of both species were cultivated in collagen sandwich and in monolayer conditions. Genome-wide data were also obtained from human NAFLD, cirrhosis, HCC and hepatitis B virus-infected tissue as well as mouse livers after partial hepatectomy, CCl4 intoxication, obesity, HCC and LPS. A strong similarity between cultivation and disease-induced expression alterations was observed. For example, expression changes in hepatocytes induced by 1-day cultivation and 1-day CCl4 exposure in vivo correlated with R = 0.615 (p < 0.001). Interspecies comparison identified predominantly similar responses in human and mouse hepatocytes but also a set of genes that responded differently. Unsupervised clustering of altered genes identified three main clusters: (1) downregulated genes corresponding to mature liver functions, (2) upregulation of an inflammation/RNA processing cluster and (3) upregulated migration/cell cycle-associated genes. Gene regulatory network analysis highlights overrepresented and deregulated HNF4 and CAR (Cluster 1), Krüppel-like factors MafF and ELK1 (Cluster 2) as well as ETF (Cluster 3) among the interspecies conserved key regulators of expression changes. Interventions ameliorating but not abrogating cultivation-induced responses include removal of non-parenchymal cells, generation of the hepatocytes’ own matrix in spheroids, supplementation with bile salts and siRNA-mediated suppression of key transcription factors. In conclusion, this study shows that gene regulatory network alterations of cultivated hepatocytes resemble those of inflammatory liver diseases and should therefore be considered and exploited as disease models.
KeywordsGene arrays Bioinformatics Inflammation Metabolism Differentiation
Freshly isolated hepatocytes
Monolayer confluent culture
Monolayer subconfluent culture
Epithelial to mesenchymal transition
Hepatocyte in vitro systems represent a well-accepted tool in many fields of basic and applied research such as pharmacology and toxicology, tissue engineering and clinical hepatocyte transplantation (Godoy et al. 2013). However, despite of their widespread use, research with primary hepatocytes remains challenging (Godoy et al. 2013). Isolating hepatocytes from their physiological environment in the liver causes alterations in cell physiology and major gene expression alterations (Godoy et al. 2013; Zellmer et al. 2010). However, it has never been studied whether these changes only represent in vitro artifacts or whether they resemble disease-relevant processes. Such a situation has since long been acknowledged for liver fibrosis, where cultivated stellate cells undergo similar activation mechanisms as in the fibrotic liver (De Minicis et al. 2007). To address this question, we compared time-resolved, genome-wide data of cultivated hepatocytes and liver disease models. We report that alterations in cultivated human and mouse hepatocytes resemble those in inflammatory liver diseases, and that similar transcriptional networks and transcription factors are responsible for the identified changes. Since we performed the study under identical conditions for mouse and human hepatocytes, a systematic interspecies comparison was possible, identifying similar features such as HFN4-driven downregulation of metabolic functions, but also major interspecies differences such as Klf6-driven inflammatory processes. The resulting transcriptomic network directory offers a blueprint for interventions for improving the in vitro systems but also for interfering with disease-relevant processes.
Materials and methods
Hepatocyte isolation and cultivation
Primary mouse hepatocytes were isolated from male C57BL6/N mice (8–12 weeks old) by the two-step collagenase perfusion method (Godoy et al. 2013). Primary human hepatocytes were obtained under informed consent from patients undergoing surgical liver resection by a two-step collagenase I perfusion (Godoy et al. 2013). The cells were plated on six-well dishes, either onto dried collagen I (monolayer) or between two layers of soft-gel collagen (sandwich) as described in (Godoy et al. 2013). Details for the protocols can be found in the supplemental section.
Genome-wide analyses and bioinformatics
Affymetrix gene array analysis was performed as previously described (Godoy et al. 2009), using the Affymetrix GeneChip® Mouse Genome A430 2.0 (Santa Clara, CA, USA) (details in suppl. Section). Affymetrix gene expression data were preprocessed using ‘affyPLM’ packages of the Bioconductor software as previously described (Godoy et al. 2015). Data obtained from fresh hepatocytes were used as reference. A false positive rate of a = 0.05 with FDR correction and a fold change greater 2 were taken as the level of significance. Two samples, one M S day 1 and one S day 5, were identified as outliers by principal component and Pearson’s correlation analyses and were not included in the subsequent analyses (Suppl. Fig. 1; Suppl. Table 1). A list with all differentially regulated genes (DEG) in hepatocytes and liver disease models can be found in the supplemental section. Processing and visualization (principal component analysis) of data were performed using MATLAB tools (The MathWorks Inc., Natick, MA, USA). Clusters of correlated genes based on similar time-dependent fold change were generated by fuzzy c-means. A list with genes belonging to each fuzzy cluster is provided in the supplemental section. Gene set enrichment analysis (GSEA) was performed using the manually curated Gene Ontology of the BIOBASE Knowledge Library (BKL) of the ExPlain™ Web service (BIOBASE GmbH, Wolfenbüttel, Germany). Overrepresented transcription factors binding sites were identified using the algorithm PRIMA (PRomoter Integration in Microarray Analysis) of the Expander software 6.1 (EXPression ANalyzer and DisplayER) as previously described (Godoy et al. 2015). Metagenes were generated by calculating the mean expression value for each biological sample in any model system for each set of genes that belong to the respective metagene (Schmidt et al. 2008). Interspecies (mouse/human) gene expression comparison was performed by using orthologous genes (Yue et al. 2014). Correlation analyses (Spearman’s rank correlation and odds ratio) were performed using Statistics Toolbox of MATLAB. A classification probability and a metric of the gene regulatory networks (GRN) status, related to 16 specific mouse and human tissues and cells, were calculated using the CellNet platform (Cahan et al. 2014), using the locally available R version of the software CellNet http://pcahan1.github.io/cellnetr/.
Additional methods and gene expression profiles from mouse and human cell lines and disease models
Additional methods such in vivo models of liver disease, flow cytometry, Percoll-based purification, immunofluorescence, siRNA-mediated knockdown of Klf6 and MafF and LUMINEX assays, together with a complete list of chemicals, reagents, antibodies and tables containing gene expression data and bioinformatics, are provided in the supplemental section. Additional gene expression data (based on Affymetrix gene arrays) including mouse HCC (Dapito et al. 2012), mouse cell lines AML12 (Ventura-Holman et al. 2008) and iHep (stem cell-derived hepatocytes) (Morris et al. 2014), human cell lines HepG2 (Rodrigues et al. 2016) and HLC (stem cell-derived hepatocytes) (Godoy et al. 2015), human hepatitis B-infected liver (Farci et al. 2010), human non-alcoholic fatty liver (NAFLD) (Moylan et al. 2014), human cirrhosis and hepatocellular carcinoma (HCC) (Yildiz et al. 2013) were obtained from Gene Expression Omnibus http://www.ncbi.nlm.nih.gov/geo/ or ArrayExpress https://www.ebi.ac.uk/arrayexpress/ and provided in the supplemental section.
Stereotypic gene expression responses to different types of stress
Similarity of global gene expression changes induced during hepatocyte cultivation and various liver diseases
Identification of transcriptional control mechanisms in gene networks activated in cultivated hepatocytes
The same approach was used for human hepatocytes leading to four cluster groups (Fig. 5b). Cluster 1 represents loss of mature liver functions (i.e., metabolism) and therefore corresponds to the respective mouse cluster. Similarly, as for mouse the sandwich culture ameliorates the decrease in genes in Cluster 1A (Suppl. Table 12). The strongest induced genes in mouse Cluster 2 represent canonical inflammation genes (e.g., lipocalin 2, Fig. 5a). In the human Cluster 2 these genes were only slightly or not induced (Suppl. Table 12). The substantially milder induction of inflammation genes in human hepatocytes represents a major interspecies difference. Moreover, RNA processing and ribosomal genes are overrepresented in human Cluster 2 (Fig. 5b; Suppl. Table 13). These motifs are also found in the corresponding mouse Cluster 2 (Suppl. Table 11). Cluster 3 contains overrepresented proliferation and migration GO groups, similarly as the corresponding mouse cluster (Fig. 5b). Cluster 4 comprises genes involved in cholesterol metabolism which were upregulated in culture (Suppl. Table 13). The induction of Cluster 4 genes is human specific, since the corresponding mouse genes remain unaltered or are even downregulated. Key transcription factors overrepresented in Clusters 1, 2 and 3 are HNF4, ELK1 and E2F, respectively (Fig. 5b; Suppl. Table 13). Comparison of transcriptional control factors in human and mouse hepatocytes identified similarities (i.e., HNF4 in Cluster 1, ELK1 in Cluster 2 and ETF in Cluster 3) but also major differences, in particular Klf6 and Cebpd which are upregulated in mouse but not in human hepatocytes (Fig. 5a, b).
Interspecies comparison identifies predominantly similar responses in human and mice
Interventions improving hepatocyte differentiation
A more generic approach to establish a microenvironment that better reflects the in vivo situation is offered by spheroids where hepatocytes form their own matrix (Landry et al. 1985), in contrast to collagen sandwich cultures where hepatocytes are kept between two layers of soft gel made of collagen type I isolated from rat tail (Uygun et al. 2010). In both cultivation systems, sandwich and spheroids, hepatocytes established a polar phenotype as evidenced by DPPIV staining of bile canaliculi (Fig. 8c). Analysis of representative genes from Cluster 1 (mature liver functions) and Cluster 2 (inflammation/RNA processing) showed qualitatively similar alterations in both culture systems (Fig. 8d). However, the degree of downregulation of some metabolism cluster genes (e.g., Bsep) is slightly ameliorated in spheroids as compared to sandwiches, which may prioritize the latter system for hepatocyte physiology (Fig. 8d). Consistently, expression of HNF4α, whose binding site is overrepresented in the metabolism cluster genes, was better maintained in spheroids than in sandwich cultures (Suppl. Fig. 17). However Cebpδ, a transcription factor responsible for expression of inflammation genes, is higher in spheroids, illustrating that this culture type does not suppress the inflammation response (Suppl. Fig. 17). Comparison of collagen sandwich cultures to other types of commercially available matrixes, such as matrigel and laminin demonstrated similar alterations of metabolism and inflammation gene clusters (Fig. 8d; Suppl. Fig. 18), indicated that the systemic alterations in gene expression occur independent from the specific type of extracellular matrix.
One of the results of the fuzzy clustering analysis was the identification of downregulated mature liver functions (Fig. 5) which includes a cluster of genes involved in bile salt metabolism. Since routinely used hepatocyte culture media do not include bile salts (Godoy et al. 2013), but on the other hand bile salts have been reported to induce mature liver functions (Avior et al. 2015) (Sawitza et al. 2015), we tested their influence in the present experimental system. Of seven bile salts, six significantly suppressed the diagnostic inflammation cluster markers Lcn2 (Fig. 8e) and Saa3, while Mt2 remained unaltered (Suppl. Fig. 19). Similarly, significant increases in the diagnostic metabolism cluster markers Bsep (Fig. 8e), Mrp2 and Cyp7a1 (Suppl. Fig. 19) were obtained.
One of the candidates identified by GRN analysis that potentially contributes to orchestrating the transcriptional inflammation response is Klf6 (Fig. 5). Klf6 is a binding partner of the overrepresented TF SP1 (Fig. 5), and it was transcriptionally upregulated in cultivated hepatocytes and in mouse liver disease (Fig. 5; Suppl. Table 1–3, 6). The biostatistical results were confirmed by qRT-PCR (Suppl. Fig. 20), immunostaining showed nuclear translocation of Klf6 (Suppl. Fig. 20) and western blot demonstrating upregulation of Klf6 in cultivated hepatocytes and in vivo after CCl4 administration (Suppl. Fig. 20). However, siRNA-mediated knockdown of Klf6 in cultivated mouse hepatocytes led only to a minor increase in Bsep expression (Suppl. Fig. 20). A second knockdown approach was performed for MafF, another transcription factor involved in the inflammation response. A pool of 4 oligos against MafF reduced expression in cultivated mouse hepatocyte by approximately 90 % (Suppl. Fig. 20). This led to a statistically significant reduction in the marker genes of the inflammation cluster Lcn2, Mt2, and Saa3, and increased the expression of metabolism cluster genes Bsep, Mrp2 and Cyp7a1 (Suppl. Fig. 20). However, it should be considered that the size of this recue effect is small and by far does not restore the expression levels to those of freshly isolated primary hepatocytes.
Further interventions to ameliorate the massive expression alterations in cultivated hepatocytes can be undertaken at the level of the signaling network. LUMINEX analysis of phosphoproteins in freshly isolated mouse hepatocytes and after cultivation for up to 7 days showed a strong increase in p-ERK1/2, p-JNK, p-MEK1, p-p38, and p-p70S6 compared to healthy mouse liver tissue (Suppl. Fig. 21). Consistently, we observed genes of the MAPK pathway upregulated in all cultures, including Map3k5, Map3k6, Mapk3, VRas and Rras (Suppl. Table 10, Cluster 2B). Perfusion of mouse livers with EGTA and collagenase either alone or in combination shows that EGTA alone is sufficient to induce the identified signaling activity (Suppl. Fig. 21). It is known since long that isolated hepatocytes can be kept in cold storage for more than 24 h without compromising their plating efficacy (Hengstler et al. 2000). Cold storage for 24 h reduced the phosphorylation levels of all aforementioned signaling proteins; however, this did not blunt the expression responses of representative genes of Clusters 1–3 (Suppl. Fig. 22). Therefore, ‘cooling down’ the signaling activities of freshly isolated ‘burning’ hepatocytes before plating does not prevent the massive isolation and cultivation-induced expression changes shown in Fig. 1.
In a genome-wide, time-resolved profiling study initially designed to improve human and mouse hepatocyte in vitro systems we made the surprizing observation that expression changes induced by hepatocyte isolation and cultivation resemble expression alterations in liver diseases. We demonstrate by genome-wide analysis that gene cluster responses in vitro, i.e., upregulation of inflammation/RNA processing and migration/cell cycle genes, downregulation of mature liver functions genes, also occurred in human HBV-infected liver tissue, cirrhosis and HCC, as well as CCl4 intoxication, partial hepatectomy, LPS intoxication and HCC in mouse, although the intensity varied and was generally weaker in diseased liver than in vitro. Our observations are in agreement with previous reports that also identified components of the acute phase response among the top upregulated genes in cultivated hepatocytes (Boess et al. 2003). However, our study goes beyond those observations because we establish precise correlations to disease models in vivo, and unravel control mechanisms responsible for the altered gene expression in vitro, allowing the design of molecular interventions. Gene network analysis identified a low HNF4 signature as one of the major reasons for loss of mature liver functions. Gain of inflammation/RNA processing was mostly driven by Klf6 and interaction partners, Cebp and ATF stress factors. Central control factors for the migration/cell cycle cluster include E2F family members.
Possible reasons for gene expression alterations in cultured primary hepatocytes are contamination with non-parenchymal cells, unphysiological extracellular matrix (ECM) or lacking soluble environmental cues present in normal liver such as bile salts. In our study, we addressed all these aspects individually. First, a strong reduction of NPCs contamination by Percoll gradient resulted in partial suppression of genes from the inflammation/RNA processing cluster. Second, different types of ECM were tested. We have previously observed that recombinant extracellular matrix proteins (e.g., laminins) significantly improve hepatocyte differentiation of HLCs compared to cancer-derived matrix (e.g., matrigel) (Cameron et al. 2015). In primary hepatocytes, however, matrigel or laminin ECM did not provide significant improvements in expression of either ‘inflammation’ or ‘mature liver function’ genes compared to collagen I cultures. Interestingly in hepatocyte spheroids, where cultivated hepatocytes generate their own matrix, a significant increase in genes from the ‘mature liver functions’ cluster was seen. Nonetheless, this increase is small when comparing to expression levels of healthy liver. Addition of bile salts increased expression of mature liver function genes (e.g., Bsep) although the achieved levels were again far below those of healthy liver. Interestingly, all tested bile salts decreased inflammation cluster genes. Knockdown of Klf6 and MafF, central transcription factors of the inflammation/RNA processing cluster, led only to a small decrease in inflammation cluster genes, suggesting that manipulation of only single transcription factors is not sufficient to tip the balance from inflamed livers back to the normal situation. Taken together, this illustrates that several mechanisms contribute to the altered cell state; hence, none of the individual interventions can result in a full rescue. However, the results strongly suggest that Percoll gradient purification and bile acid supplementation should be done routinely. Also, further factors of influence such as LPS contamination of collagenase or the strong induction of several signaling pathways in response to the isolation stress (‘burning hepatocyte phenomenon’) were individually not sufficient to explain the altered state of cultivated hepatocytes. In conclusion, the transcriptional state of hepatocytes in culture represents a multifactorial phenomenon that closely resembles gene expression patterns and transcriptional regulatory networks in liver disease. For pharmacological and toxicological routine work, this means that we should be aware that the frequently applied cultivated hepatocyte systems represent an inflamed liver. Importantly, this represents a disease model that can be exploited to develop anti-inflammatory strategies for liver disease. It is of interest that cultivated stellate cells, even when isolated from healthy livers, have been widely used as a model of liver fibrosis (Friedman 2008). Similarly, cultivated primary hepatocytes represent an inflammatory model in vitro.
A question of high practical relevance is mouse-to-human extrapolation, since most therapeutic concepts are initially tested in rodents. However, the relevance of mouse inflammation data for humans has been challenged up to the extreme position that there are no relevant similarities at all (Leist and Hartung 2013). The present study using mouse and human hepatocytes under identical pro-inflammatory culture conditions offers a blueprint for rational interspecies extrapolation. Qualitatively similar is the inflammation/RNA processing response that in both species is driven by increased Krüppel-like factors, Sox4, Myc, Tead2 (ETF) and ELK1. It should be considered that despite of this overall correlation, important differences can be identified, such as the key inflammation-associated transcription factor Cebpd which is upregulated in mouse, but repressed in human primary hepatocyte cultures. A further common feature of both species is downregulation of mature liver function genes such as Pck1, Adh4 and G6pc that are more than tenfold downregulated in both species, with HNF4 as a dominant control factor. In conclusion, the knowledge of interspecies conserved versus distinct transcriptional networks allows predicting which features observed in mice can or cannot be extrapolated to humans.
The present study leads to the following practical recommendations: (1) Sandwich cultures or spheroids provide the most effective culture conditions to sustain expression of genes associated with mature liver functions. However, when an in vivo-like metabolism is critical, for instance in the study of metabolism-associated toxicity, compound exposures should be performed within the first 24 h of cultivation, due the time-dependent suppression of mature liver function genes. This is even more critical for mouse than human hepatocytes. (2) Collagen monolayers provide a system where proliferation-associated genes can be induced. This feature is also more substantial in mouse hepatocytes which can engage in cell proliferation. (3) When primary hepatocytes are cultivated over longer periods, they approach transcriptional states of hepatocyte cell lines and stem cell-derived hepatocyte-like cells. In conclusion, the present genome-wide study of primary human and mouse hepatocytes demonstrates that the broadly applied culture systems represent models of inflamed livers.
We are grateful for the excellent technical assistance of Katharina Rochlitz, Brigitte Begher-Tibbe and Georgia Günther.
This study was supported by the SEURAT-1 Projects NOTOX (EU-Project FP7-Health Grant Agreement No. 267038) and DETECTIVE (EU-Project FP7-Health Grant Agreement No. 266838), the BMBF (German Federal Ministry of Education and Research) Project Virtual Liver 0313854 to JGH and 0315753 to TSW.
- Boess F, Kamber M, Romer S et al (2003) Gene expression in two hepatic cell lines, cultured primary hepatocytes, and liver slices compared to the in vivo liver gene expression in rats: possible implications for toxicogenomics use of in vitro systems. Toxicol Sci 73(2):386–402CrossRefPubMedGoogle Scholar
- Botella LM, Sanz-Rodriguez F, Komi Y et al (2009) TGF-beta regulates the expression of transcription factor KLF6 and its splice variants and promotes co-operative transactivation of common target genes through a Smad3-Sp1-KLF6 interaction. Biochem J 419:485–495. doi: 10.1042/Bj20081434 CrossRefPubMedPubMedCentralGoogle Scholar
- Campos G, Schmidt-Heck W, Ghallab A et al (2014) The transcription factor CHOP, a central component of the transcriptional regulatory network induced upon CCl4 intoxication in mouse liver, is not a critical mediator of hepatotoxicity. Arch Toxicol 88(6):1267–1280. doi: 10.1007/s00204-014-1240-8 CrossRefPubMedGoogle Scholar
- Farci P, Diaz G, Chen Z et al (2010) B cell gene signature with massive intrahepatic production of antibodies to hepatitis B core antigen in hepatitis B virus-associated acute liver failure. Proc Natl Acad Sci USA 107(19):8766–8771. doi: 10.1073/pnas.1003854107 CrossRefPubMedPubMedCentralGoogle Scholar
- Godoy P, Hewitt NJ, Albrecht U et al (2013) Recent advances in 2D and 3D in vitro systems using primary hepatocytes, alternative hepatocyte sources and non-parenchymal liver cells and their use in investigating mechanisms of hepatotoxicity, cell signaling and ADME. Arch Toxicol 87(8):1315–1530CrossRefPubMedPubMedCentralGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.