System analysis of the functional cross-talk between PPARalpha, LXR and FXR in the human HepaRG liver cells

Transcriptional regulations exert a critical control of metabolic homeostasis. In particular, the nuclear receptors (NRs) are involved in regulating numerous pathways of the intermediate metabolism. The purpose of the present study was to explore in liver cells the interconnectedness between three of them, LXR, FXR, and PPARα, all three known to act on lipid and glucose metabolism, and also on inflammation. The human cell line HepaRG was selected for its best proximity to human primary hepatocytes. Global gene expression of differentiated HepaRG cells was assessed after 4 hours and 24 hours of exposure to GW3965 (LXR agonist), GW7647 (PPARα agonist), and GW4064 and CDCA (FXR synthetic and natural agonist, respectively). Our work revealed that, contrary to our expectations, NR specificity is largely present at the level of target genes, with a smaller than expected overlap of the set of genes targeted by the different NRs. It also highlighted the much broader activity of the synthetic FXR ligand compared to CDCA. More importantly, our results revealed that activation of FXR has a pro-proliferative effect and decreases polyploidy of hepatocytes, while LXR inhibits the cell cycle progression, inducing hepatocyte differentiation and a higher polyploidism. Conclusion: these results

endogenous ligands that are metabolites belonging to the intermediary metabolisms, and actively contribute to the regulation of metabolic pathways. The discovery of each receptor initially emphasized the specificity of each receptor in a given metabolic pathway, for example the peroxisome proliferator-activated receptors (PPARα , PPARβ/δ , PPARγ ; NRC1, NRC2, and NRC3, respectively in the international nomenclature) in lipid metabolism, the farnesoid X receptors (FXR and FXRβ; NR1H4 and NR1H5P, respectively) in bile acid metabolism, and the liver X receptor (LXRα and LXRβ; NR1H3 and NR1H2, respectively) in cholesterol metabolism (1,2). However the classical linear view with each NR engaged in modulating one or a few pathways is challenged by the numerous and complex interconnections between the metabolisms of glucides, lipids and amino acids, as well as by the numerous roles of NRs outside of metabolism. This highlights the need to delineate the regulatory network underlying homeostasis through systemic approaches. The aim of this study was to explore the connections between the three NRs mentioned above. More specifically, PPARα is activated by unsaturated fatty acids and involved in many facets of both lipid and glucose metabolism. LXRα and LXRβ are activated by cholesterol derivatives, but are also strongly lipogenic. Finally, FXR is bound by bile acids and is considered as a critical regulator of cholesterol metabolism (3). Thus, they clearly affect overlapping pathways. To better explore the interconnections, one must first assess the activity of each receptor in a given common and reproducible cellular context. For that purpose, we used the HepaRG hepatocarcinoma cell line, introduced in 2002 by Gripon et al. (4). HepaRG cells have been described as the closest in vitro model to human liver metabolism (5) and as a strong candidate for bioartificial liver applications (6). Our work revealed that, in contrast to our expectations, NR specificity is largely present at the level of target genes, with a smaller than expected overlap of the set of genes targeted by the different NRs. The connection and coordination then mainly occur at the level of pathways. Importantly, our observations also highlight an important role of FXR and LXR in regulating cell growth and death of liver cells, with opposite effects on cell cycle progression and on hepatocyte polyploidization.

Defining and validating the experimental set-up
The aim of the experiment was to evaluate which hypothesis could best explain the crossregulatory action of the nuclear receptors on metabolic pathways. Do the receptors regulate common target genes, or do they regulate different sets of genes belonging to common pathways? Differentiated HepaRG cells were chosen since their global gene expression patterns are much closer to liver or primary human hepatocyte cultures than the widely used HepG2 cell line (5) and include the expression of PPARα, LXRα/β, and FXR at mRNA levels. This expression is consistent with that observed in human primary cells. To activate each receptor, we used synthetic ligands known to be the most specific: GW4064 for FXR, GW3965 for LXRs, and GW7647 for PPARα, hereafter called FXR-L, LXR-L, and PPARa-L. We also used the bile acid CDCA, which is a natural ligand of FXR (11,12). CDCA is also known to affect LXR (13), PPARα (14) and HNF4 (15) pathways, even though the mechanisms are not yet well understood. For each compound, we selected the lowest concentration that triggered a good response of canonical targets in a pilot experiment (data not shown). Microarrays were run first on differentiated HepaRG, prior any treatment. A second batch of microarrays was then run 4 hours after the beginning of the cell exposure to each treatment, in order to identify direct target genes, whereas the last batch of microarrays, run at 24h, aimed at identifying global crosstalk. These analyses confirmed that the NRs of interest were expressed in HepaRG cells. We then identified all genes differentially expressed, when compared to the group of control samples, in at least one of the treatment conditions at either of the two time points. We used the RNA datasets of differentially expressed genes as features to perform a hierarchical clustering of the five different models analyzed by Hart et al. (5). As shown in supplemental Figure 1, the profile of our set of DE genes was found to be similar in human liver and in primary hepatocytes, whereas HepG2 cells have a gene expression pattern far from all the other models analyzed. The expression profile of HepaRG cells (differentiated and undifferentiated) was distinct from that of liver cells, but much closer to it than that of HepG2 cells. Altogether, these initial observations confirmed that HepaRG cells are in this context more pertinent than other in vitro models widely used in literature, such as HepG2 cells (16,17).

NR cross-talks happen at the pathway level and are mostly related to metabolism
The analyses of the set of genes differentially expressed at the 4 hours time point identified 681 differentially expressed genes, comparing the control to each of the treatments. A good part of these genes are associated to the FXR-L treatment (~500 genes) and to a lesser extent to CDCA (~160 genes), while fewer are associated to PPARa-L (~90), and to LXR-L (<10) (Fig 1A, left part). This important variation in the number of genes modulated by each treatment is consistent with separate gene expression or ChIP-seq studies (18)(19)(20). These differentially expressed genes include the main known targets of the selected nuclear receptors. For example, FXR-L and CDCA modulate ABCB11 (BSEP), ABCB4 (MDR3), CYP7A1 and PPARGC1 expression. Some known FXR target genes are found modulated only by FXR-L, and not by CDCA, like NR0B2 (SHP), HNF4A and PPARA. Others were only regulated by CDCA like UGTB4, RORA, HMGCS1 and NR1D1/2. PPARa-L upregulates its well-established target genes FABP4, HMGCS2, ACOX1, CD36 and CPT1A. At this point, LXR-L has a very modest effect, modulating less than 10 genes, which include known LXR target genes SCD, SREBF1 (also called SREBP), ABCA1, FASN, and MYLIP/IDOL. We then identified 1611 differentially expressed genes at the 24h time-point ( Fig 1A, right part). This strong increase compared to the early time-point is consistent with the identification of indirect effects. For example, SHP (NR0B2), a well-described FXR target gene very significantly increased at 4 hours by FXR-L, mediates indirect regulations, such as the strong down-regulation of Cyp7A1 seen at 24h (21). Also, SREBP1c, which is induced by LXR at 4h, is a regulator of key genes in lipid metabolism (22). These genes are in turn induced at 24h, as for example SCD1, ACLS1 and HMGCR and HMGCS. Like at 4h, most genes were associated to FXR-L (~1100) and CDCA (~350) and fewer to PPARa-L (~140) and LXR-L (<50). Intriguingly, less than half of the genes identified at 4h were still differentially expressed at 24h (312 out of 681) ( Table 1), suggesting that most direct effects are only transient. Comparing the different treatments shows surprisingly few crosstalks at the level of transcriptional regulation: among the 1980 differentially expressed genes, 75% are specific to a single treatment. As expected, FXR-L and CDCA have the biggest overlap, sharing 63% of the common genes (16% of all the differentially expressed ones). Nevertheless some genes are common targets for different NRs. For example PDK4, HMGCS2, and ACOX1, involved in lipid metabolism, are all up-regulated by PPARa-L and down-regulated by FXR-L; ABCA1 and MYLIP, well described target genes of LXR (3,23) are also down-regulated by FXR. Some other genes involved in lipid and glucose metabolism like, for example, FASN, FABP4, LPL, INSIG1, SCD, are found up-regulated both by PPARa-L and LXR-L ( Fig. 1A and Table  1). To explore the possibility of crosstalk at the level of pathways, we looked for enriched pathways shared across time-points and treatments. We identified 153 enriched pathways in our dataset. Like for differentially expressed genes, we observed an increase in the number of enriched pathways between the early and late time-points. Most enriched pathways (54%) were identified at both time-points (compared to 16% of the differentially expressed genes). Even though they remain predominantly associated to the FXR-L (140) or CDCA (114) treatments, PPARa-L (62) and LXR-L (46) treatments are also well represented. Importantly, 80% (123/156) of the enriched pathways are shared between several treatments, compared to 25% (500/1980) at the gene level. While 40 of the 123 shared pathways are still specific to the related FXR-L and CDCA treatments, which both target the same NR, the remaining 83 pathways are associated with two or even three treatments targeting separate NRs and thus point to crosstalk at the pathway level. In other words, more than half of all enriched pathways (83/153) are associated with at least two NRs ( Figure 1B; Table 1). The most striking observation at this point is that few genes are co-regulated by the NRs studied herein. In contrast, the cross-talk appeared in the associated pathways. We therefore explored further the shared enriched pathways to discover new NR crosstalks.

Cell growth and death is overrepresented in FXR and LXR cross-talks
Based on the KEGG pathway database (www.genome.jp/kegg/pathway.html), we manually classified enriched pathways into 4 groups: "Metabolism", "Cell Growth and Death", "Immune System" and "Other". 'Transport and catabolism', 'Endocrine, Digestive, Excretory Systems' and 'Endocrine and Metabolic diseases' were merged into the 'Metabolism' group. 'Genetic Information Processing', 'Cellular Processes' (other than 'Peroxisome') and 'Cancers' form the 'Cellular Growth and Death' group. Finally, 'Immune and Infectious Diseases' correspond to our 'Immune System' group. Unclassified pathways that we could not associate to one of these 3 groups form the 'Other' group (the names of the pathways in each category are made available in the supplementary archive "data_and_code.zip"). As a set of core genes may drive different pathways belonging to the same group, we then counted the genes belonging to the leading edge of the GSEA curves and involved in several . CC-BY-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/514976 doi: bioRxiv preprint pathways within each group (illustrated in supplementary Figure S2). In the metabolism group, only 14 genes are involved in at least 10% of the pathways. This amount increases for the two other groups with 121 genes being involved in at least 10% of pathways of cell growth and death and 96 for the immune system; 37 of them are common to both groups. As most DE genes are specific to one pathway, we are confident about the relevance of the enriched pathways. This analysis thus confirmed that metabolic regulations, which are the main functions of the NRs, are represented by the largest group of enriched pathways with the highest specificity of involved genes. The 'Metabolism' group covers 48% of the enriched pathways, 'Cellular Growth and Death' cover 30%, and 'Immune System' 14%. Unsurprisingly, all pathways shared by the three NR and 65% of the ones shared by FXR and PPARα are involved in metabolism (Table 2). However, 72% of the pathways associated to both FXR-L and LXR-L are related to cell growth and death, while this group does not represent more that 30% in the other combinations of treatments. The role of FXR and LXR in liver is considered as mostly related to metabolism regulation with main roles in lipid, cholesterol and glucose metabolism, FXR being anti-lipogenic and LXR pro-lipogenic (1). However they also play an important role in liver regeneration (24)(25)(26)(27)(28) or cancer development (29)(30)(31).

FXR and LXR oppositely regulate the cell cycle progression
To understand the mechanisms involving FXR and LXR in the regulation of cell growth and death, we focused on the "Cell cycle" pathway, which is enriched in four conditions: at 4h by FXR-L and CDCA, and at 24h by FXR-L and LXR-L. We further looked at the genes that belong to the leading edges of the cell cycle pathway at 24h for FXR-L (45 genes) and LXR-L (41 genes). Intriguingly, FXR-L predominantly activates gene expression while LXR-L inhibits it. Twenty-three of the 26 genes found in the leading edges for both treatments are oppositely regulated by FXR-L and LXR-L (Fig. 2). Using the KEGG diagrams to illustrate the corresponding pathway and genes ( Fig. 3 and supplementary Fig. S3), the patterns of FXR-L at 4h show an up-regulation of genes involved in the G1/S transition like C-MYC, and CCND1. At 24h, the cell cycle key players are upregulated by FXR-L and downregulated by LXR-L. The list comprises CCNE, CDK2, and P18 as well as genes involved in DNA replication such as CDC6 and members of the minichromosome complex (MCM) (32,33). These observations suggest a pro-proliferative role for FXR, with an up-regulation of genes involved in the G1/S progression at 4h and a more global positive regulation at 24h. LXR have the opposite effect inhibiting genes involved in the cell cycle progression and DNA replication.
To determine if these gene modulations reflect an actual regulation of the cell cycle, we analyzed the distribution of cell cycle phases by propidium iodide staining and flow cytometry. We treated HepaRG cells with FXR-L and LXR-L for 24, 48 or 72 hours. Knowing the pro-proliferative effect of serum in the culture media, the experiment was made both in the presence and in the absence of serum. Flow cytometry analyses revealed two well separated populations corresponding to diploid and tetraploid populations (2C and 4C DNA content, respectively), characteristic of mammalian hepatocytes (Fig. 4A) (34). In serum free conditions, FXR-L treatment triggered a decrease of the tetraploid cell population, visible at 24h and significant at 48h and 72h, with 25% and 36% less 4C cells respectively, while LXR-L treatment has no significant effect (Fig. 4D). In presence of serum, the 4C population also decreases at 72h with FXR-L (22%), whereas it significantly increases with LXR-L (17%). The mechanisms involved in the mammalian hepatocyte polyploidism are not yet well understood. However polyploidy is known to appear at the terminal differentiation during late . CC-BY-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/514976 doi: bioRxiv preprint foetal development, while proliferating cells are mainly diploid. Indeed, by sorting the cells by their cell cycle phase status, we observe in the diploid population an accumulation of cells at G1 after 48 hours treatment with LXR-L (suppl. Fig. S5). This accumulation is further increased at 72 hours, while S and G2 populations are decreasing (Fig. 4B). In opposite manner, FXR-L tends to decrease the G1 population and increase the S and G2 populations at 24h, even if these changes are statistically significant only at 72h in serum-free conditions ( Fig. 4B and suppl. Fig. S5). The tetraploid population shows the same pattern with a stronger effect of FXR-L, decreasing significantly the G1 population in both conditions (Fig. 4C). Hence, our results are consistent with a correlation between polyploidy and hepatocyte differentiation stage (35): FXR-L has a pro-proliferative effect and decreases polyploidy, while LXR-L inhibits the cell cycle progression, inducing hepatocyte differentiation and a higher polyploidism. Positionning the genes identified in the GSEA results with the BIOCARTA pathways show that FXR-L affects the G1/S and G2/M checkpoints, while LXR-L only affects the G1/S checkpoint, blocking the cells in the G1 phase (Fig. 4). To better assess the key players involved in these regulations, we evaluated the activity of some cell cycle regulators at the transcriptional and post-transcriptional levels (Fig. 5). The LXR-L treatment affects the G1/S checkpoint by inhibiting the expression of CDC25A and CCNE2, and upregulating the expression CDKN2B (p15). Quantitative PCR also shows a downregulation of E2F1 and E2F2 mRNA expression by qPCR (Fig. 5A). The treatment also has effects at the posttranscriptional level inhibiting the CCND3 (Cyclin D3) expression and decreasing RB phosphorylation (Fig. 5B&C). On the other hand, FXR-L also deeply affects the key players of the G1/S checkpoint, CCNE1 and SKP2, which are upregulated by FXR-L (Fig. 5A). This is consistent with the observed hyper-phosphorylation of RB (Fig. 5B&C) and should result in an activation of the cell cycle progression. Finally, FXR-L also affects the mRNA expression of the G2/M checkpoint regulators CDC25C or CCNA2 (Fig. 5A).

Discussion
Nuclear receptors are important transcriptional regulators of metabolism (2), acting on both specific and shared pathways. The numerous side effects associated to novel therapeutics for metabolic disorders based on NRs activation or inhibition highlight the need for understanding their regulatory crosstalks. Herein, we demonstrated that the cross-talk between the three selected receptors, FXR, LXR and PPARα is largely due to a distribution of their respective specific target genes in shared pathways, rather than to their effects on common genes. This study further allowed identifying an important role of FXR and LXR in controlling, in opposite manner, some key factors of cell cycle progression in liver cells and of hepatocyte polyploidization. A strength of this study is the use of HepaRG cells, which exhibit a gene expression profile close to the primary human cells. By using a simplified, well standardized, and human-related model, the results obtained are likely to be informative with respect to human responses to the ligands used herein. Another strength of this study was to obtain the profile of gene expression upon activation of FXR, LXR and PPARα, respectively, using the very same experimental conditions. This was the condition to be able to interpret the specificity of each receptor. Indeed, existing gene expression profiles for these receptors that have already been published in the GEO repository cannot be directly compared due to experimental setups and cell line (for example, HepG2 vs HepaRG, vs primary hepatocytes, see (5) and Suppl. Fig.  S1).
. CC-BY-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/514976 doi: bioRxiv preprint An important observation in the present study is that the three receptors we tested regulate relatively few genes in common. There are several levels of promiscuousness between NRs: they share the same binding partner RXR, they have promiscuous binding sites, and they act on common metabolic pathways. However, our results emphasize their specificity in activating genes. This specificity does apply at the very early time point (4h), where we expect only direct target genes to be activated, but also extend to the late time point (24h), where new genes are activated, in part as a consequence of the first set of gene activation. Similar observations have been reported when analyzing the gene regulatory profile of PPARγ and LXR receptors in cancer cells. While they both exert an inhibitory action on cancer cell proliferation, via a common metabolic reprogramming, the genes they activated remain specific to each receptor (36). Another feature of interest is the overlap but also specificity of the action of CDCA and FXR-L. CDCA is considered to be a natural ligand for FXR (11). However, a large number of genes are activated with the synthetic ligand but not with the natural one. This may be explained by the so-called selective NR modulator effect (37), by which the conformational changes that occur in NRs upon ligand binding may provoke a different interface for co-factor recruitment, depending on the nature of the ligand. It may also be caused by a different affinity of the ligand, a different accessibility, and/or different metabolism of the ligand within the cell. These facts are particularly important when searching for appropriate ligands aimed at therapeutic usage, implying that a full exploration of the gene expression profile for each new molecule proposed is required. Unsurprisingly, the main pathways enriched by the three NR agonists are related to metabolism. More interestingly, 72% of the pathways shared by FXR and LXR are involved in regulations of cell growth and death. In addition, 56% of genes belonging to the leading edge of the cell cycle pathway (evaluated by GSEA) are common to both NRs but with an opposite modulation. The anti-proliferative activities of LXR has been observed in a number of cellular systems, including vascular smooth muscle cells, cancer cells, T lymphocytes, pancreatic islet beta cells and mouse liver (Reviewed in (38)). This brought up the claim that LXR might be a target for anti-tumoral treatments in different tissues, mainly documented in breast, prostate and intestinal cancer cells. LXR mediated cell-cycle inhibition has been mechanistically correlated with the expression of lipogenic and triglyceride accumulation (31), but could also be related to the inhibition of SKP2 transcription (33,39). In the present study, SKP2 is not affected by LXR-L, suggesting that LXR's anti-proliferative activity is due to alternative mechanisms, such as LXR-mediated repression of AP1 signaling (36,40). The role of FXR is more controversial, or complex. FXR was demonstrated to be involved in the bile-acid dependent progression of intestinal metaplasia (41,42), possibly through a SHPdependent increase of CDX2 gene expression (43). In contrast, the deletion or downregulation of FXR favors hepatocarcinogenesis (29,44,45), suggesting that FXR plays a protective role. Finally, FXR is also required for liver regeneration (25,28,46). The mechanism by which FXR can both favor liver regeneration and protect from hepatocarcinogenesis remains unsettled (44). The following three main findings of the present study are noteworthy in this context. First, FXR-L significantly upregulates IL6 and increases the expression of a number of pro-inflammatory cytokines, such as IL1b, CSF1, CSF2, and TGFb3. These signals are indeed important in triggering the early response for liver regeneration (44), even though these results are in contradiction with the reported antiinflammatory activities of FXR, via down-regulation of the NF-KB pathway (47), not observed in our dataset. Second, we showed the positive effect of FXR on cell cycle progression, increasing the activities of key regulators of the G1/S and G2/M checkpoints. Third, we show that FXR-L inhibits hepatocyte polyploidization. This effect could involve the insulin-signaling pathway, which is sitting at the cross-road of many metabolic pathways, and . CC-BY-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/514976 doi: bioRxiv preprint contributes to the formation of binucleated tetraploid liver cells through the PI3K/Akt pathway (48). Indeed, the insulin pathway is affected by all treatments in this study, but more particularly by FXR-L at 24h (p-value<0.001), with an inhibition of 17 key genes including INSR, IRS and PI3K (Suppl. Fig. S4). This suggests that FXR may prevent the polyploidization of the HepaRG cells via inhibition of the insulin INSR/PI3K pathway.. Interestingly, polyploidy was shown to trigger cell transformation and tumor formation in p53-null cells (49). Albeit it remains difficult to generalize this observation to other cellular context, the inhibition of polyploidization mediated by FXR-L may contribute to its protective role against carcinogenesis. Thus, besides illustrating the high levels of target gene specificity of each NR tested herein, our results revealed an important antagonistic effect of FXR and LXR on cell cycle progression in hepatocytes. Assessing these roles in different cells and tissues might be of importance when contemplating the therapeutic applications that these two receptors are conveying.      Table 2: Distribution of categories of enriched pathways depending on their association with treatments. Enriched KEGG pathways were grouped manually into 4 wide categories ("Metabolism", "Cell growth and death", "Immune system" and "Other"). Top four rows: Number and percentage of positively enriched KEGG pathways from all treatments combined, found either at a single time point (4h, 24h) or at both time points (4h and 24h) in each category; totals from both time points. Remaining table: Numbers and percentages of pathways found per ligand treatment (at any or both time points), with a breakdown of how many DE genes and enriched pathways are exclusive to one treatment and how many are shared between two or multiple treatments. Bottom section: Numbers and percentages of pathways found if the pathways lists from CDCA and FXR--L treatments are merged.
. CC-BY-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/514976 doi: bioRxiv preprint