Interference with the HNF4-dependent gene regulatory network diminishes endoplasmic reticulum stress in hepatocytes

Abstract

Background:

In all eukaryotic cell types, the unfolded protein response (UPR) upregulates factors that promote protein folding and misfolded protein clearance to help alleviate endoplasmic reticulum (ER) stress. Yet, ER stress in the liver is uniquely accompanied by the suppression of metabolic genes, the coordination and purpose of which are largely unknown.

Methods:

Here, we combined in silico machine learning, in vivo liver-specific deletion of the master regulator of hepatocyte differentiation HNF4α, and in vitro manipulation of hepatocyte differentiation state to determine how the UPR regulates hepatocyte identity and toward what end.

Results:

Machine learning identified a cluster of correlated genes that were profoundly suppressed by persistent ER stress in the liver. These genes, which encode diverse functions including metabolism, coagulation, drug detoxification, and bile synthesis, are likely targets of the master regulator of hepatocyte differentiation HNF4α. The response of these genes to ER stress was phenocopied by liver-specific deletion of HNF4α. Strikingly, while deletion of HNF4α exacerbated liver injury in response to an ER stress challenge, it also diminished UPR activation and partially preserved ER ultrastructure, suggesting attenuated ER stress. Conversely, pharmacological maintenance of hepatocyte identity in vitro enhanced sensitivity to stress.

Conclusions:

Together, our findings suggest that the UPR regulates hepatocyte identity through HNF4α to protect ER homeostasis even at the expense of liver function.

INTRODUCTION

Liver diseases kill ~2 million people worldwide every year, the majority of these from the sequelae of fatty liver disease.¹ It is abundantly clear that endoplasmic reticulum (ER) stress is associated with fatty liver disease in humans and mouse models thereof, and contributes to disease pathogenesis.^2,3 The pathways by which ER stress affects hepatic function and exacerbates liver injury are still being characterized.

ER stress is caused by disruption to homeostasis in the organelle, most commonly in the form of impaired ER protein folding. Activation of the unfolded protein response (UPR) occurs in response to and is diagnostic of ER stress. UPR signaling is attenuated when ER homeostasis improves, while it is largely perpetuated during severe ER stresses that cannot be overcome.⁴ The essential function of the UPR is to restore ER homeostasis. The response accomplishes this directive by several mechanisms, the most widely conserved of which is the transcriptional upregulation of genes encoding ER chaperones and other factors that facilitate ER protein folding and trafficking.⁵ UPR signaling emanates from three ER-resident stress sensors—PERK (protein kinase R-like ER kinase), IRE1 (inositol-requiring enzyme 1), and ATF6 (activating transcription factor 6), each of which stimulates the production of 1 or more transcription factors that activate gene expression. The upregulation of ER chaperones occurs in every vertebrate organism and cell type in which the response has been examined. Yet, it is likely that additional pathways for protecting the ER emanate from UPR activation. Importantly, because the UPR is activated by ER perturbation and deactivated by the restoration of ER homeostasis, the response is likely “blind” to cellular conditions outside the organelle unless they impact ER function.

Despite the fact that transcriptional suppression is also observed by ER stress in every cell type, very little is known about the mechanisms by which this suppression is achieved and the purposes for which it is directed. We and others have observed that ER stress in the liver, commonly induced in vivo by the pharmacological agent tunicamycin (TM), leads to an extensive suppression of genes encoding metabolic regulators and is accompanied by fat accumulation, or steatosis, in the liver.^6–9 This regulation of metabolic genes by the UPR suggests that there is an adaptive benefit to the ER in doing so. This regulation being liver-specific suggests that gene suppression is mediated by the UPR-regulated transcription factors interacting and interfering with the liver-specific transcriptional machinery. Recently, it has been posited that the suppression of metabolic genes is part of a larger coordinated program by which the UPR suppresses the gene regulatory network (GRN) that controls hepatocyte identity.¹⁰

Our previous work has suggested that many genes suppressed by ER stress in the liver are putative targets of the transcription factor HNF4α,¹¹ which is essential for hepatocyte differentiation.^12,13 It regulates numerous essential hepatocyte functions, including metabolism, lipoprotein production, coagulation, and others.¹⁴ Loss of HNF4α causes hepatocytes to dedifferentiate and proliferate,^15,16 prevents the liver from functionally regenerating,¹⁷ and sensitizes the liver to diet-induced liver injury.¹⁸ Indeed, progressive loss of HNF4α function is a critical pathogenic component of a majority of liver diseases.¹⁹

Building upon our previous work linking UPR activation to the HNF4α-dependent GRN,¹¹ our goal here was to examine the relationship between these 2 pathways and determine whether such a relationship conferred any functional benefit toward ER homeostasis. Here, we describe complementary bioinformatic, in vivo, and in vitro approaches that together demonstrate that ER stress impairs the HNF4α-dependent GRN, and impairment of HNF4α in turn reduces the sensitivity of hepatocytes to ER stress.

METHODS

k-means clustering

Microarray data were compiled from separate published experiments in which matched wild-type or Atf6α−/− (GSE48932),⁶ Ire1α ^LKO (GSE27038),⁹ or Perk ^LKO (GSE29929)⁷ animals were challenged with 1 mg/kg TM or vehicle for 8 hours, or in which Atf6α−/− or wild-type animals were challenged with 1 mg/kg TM for 34 hours (GSE48935).^6,11 For each probe set, raw expression values were log₂ transformed and were normalized against the average expression level for vehicle-treated wild-type animals. The data were then subjected to k-means clustering using the SimpleKMeans algorithm built into the Waikato Environment for Knowledge Analysis (Weka) machine learning software workbench.⁴⁵ The distances between clusters and instances were computed using the Euclidean distance. The genes in each cluster were subjected to pathway analysis using DAVID⁴⁶ and mined for enriched transcription factor binding sites using oPOSSUM.⁴⁷

Animal experiments

All animal procedures were approved by the University of Iowa Institutional Animal Care and Use Committee (IACUC) in accordance with NIH guidelines for vertebrate animal usage. Hnf4α ^fl/fl mice²³ were injected with AAV8-TBG-iCre (Vector Biolabs, cat# VB1724) or AAV8-TBG-eGFP (Vector Biolabs, cat# VB1743) i.p. using either 2.5 × 10¹¹ GC/mouse or 5 × 10¹¹ GC/mouse. One week after virus injection, mice were fasted for 4 hours before i.p. injection with 0.5 mg/kg TM or vehicle. Liver tissues and blood were harvested either 14 or 48 hours after the challenge. Aspartate aminotransferase (Biovision K753) and alanine transaminase (Biovision K752) values were determined using the manufacturer’s protocol. Quantitative-RT-PCR (including piloting of primer sets to confirm efficiency and specificity), immunoblot, and Xbp1 splicing assays were described.⁴⁸ Primer sequences are listed in Supplemental Table S9, http://links.lww.com/HC9/A558, and antibody sources and specificity controls in Supplemental Table S10, http://links.lww.com/HC9/A559. For transmission electron microscopy, Image J software (v.1.8.0_172) was used to quantify the circularity of the organelle. A value close to 1 represents a circular organelle, and a value close to zero (0) is considered noncircular.

RNA-Seq analysis

RNA-seq was performed by the Genomics Division, Iowa Institute of Human Genetics (IIHG) Core, University of Iowa. Following isolation, RNA was further purified by using an RNeasy Mini Kit (Qiagen), quantified using a Qubit machine (ThermoFisher), and 2 μg per sample was used for sequencing on an Illumina HiSeq 4000/75PE Sequencer, generating 23 million reads per sample. Data are deposited in NCBI GEO, Accession GSE235849.

Primary hepatocyte culture and 5C treatment

Primary hepatocytes were isolated as described.⁴⁸ Null media was prepared by supplementing William E media with B27 (50X; Gibco,17504044), Pen/Strep (Gibco, 15140-122), Amphotericin B (Thermo, 15290-018), and Glutamax (Gibco, 35050-061). 5C media was prepared by supplementing null media with Forskolin (20 μM; Enzo, BML-CN100), SB431542 (10 μM; Tocris, 1614), IWP2 (0.5 μM; Tocris, 3533), DAPT (5 μM; Tocris, 2634), and LDN193189 (0.1 μM; Tocris, 6053). Null media received equal volumes of solvents. After isolation, cells were resuspended in null media and plated on collagen-coated plates (C3867, Sigma). Four hours after isolation, fresh null media was replaced. On the next day and every 24 hours thereafter, null or 5C media was added. After 5 days of culture, cells were treated with ER stressors—TM (5 μg/mL), thapsigargin (50 nM), or vehicle (DMSO) for 16 hours, along with fresh 5C or null media. For pulse labeling, null- or 5C-treated cells (after 5 d) were switched to methionine-free/cysteine-free DMEM supplemented with dialyzed serum and 200 μCi/mL of ³⁵S-methionine/cysteine (Perkin-Elmer) and labeled for 30 minutes, and chases included 5 mM nonradioactive methionine/cysteine. The binding of lysates to and elution from ConA-sepharose (Cytiva) was described.²⁹ For albumin immunoprecipitations, lysates and media were precleared with protein G-agarose (Pierce), then incubated with antialbumin antibody overnight, bound to protein G for 1 hour, washed 3 times in RIPA buffer, and then eluted in SDS-PAGE loading buffer.

Statistics

Prism software (v.9.) was used for the statistical analysis. Except where noted otherwise, 1-way ANOVA was used with Tukey post hoc correction for performing multiple comparisons. “n” numbers represent either individual mice (for in vivo experiments) or independently treated wells (for cell culture experiments). All experiments were independently replicated. All image analysis was performed blinded.

RESULTS

k-means clustering identifies a distinct group of genes involved in hepatocyte identify that are suppressed by persistent ER stress

We first sought to gain insight into the process of gene suppression by the UPR by testing whether suppressed genes were coordinately regulated. To do this, we took advantage of published microarray data sets examining the global hepatic mRNA response to an 8 or 34 hours 1 mg/kg TM challenge, in animals with a deletion of either ATF6α,^6,11 IRE1α,⁹ or PERK⁷ (Supplemental Table S1, http://links.lww.com/HC9/A546, http://links.lww.com/HC9/A547, http://links.lww.com/HC9/A548, http://links.lww.com/HC9/A549). The premise of this approach was that genes whose regulation depended on a common factor would show a common pattern of regulation across these 4 experiments.

To identify potentially coordinately regulated genes among these data sets, we took an unbiased heuristic machine learning approach known as k-means clustering.²⁰ This approach divides the full dataset into a prespecified arbitrary number of clusters, which minimizes the average distance within each cluster. We empirically found that 12 clusters provided enough flexibility to identify groups of genes with meaningfully distinct regulation patterns while mitigating the risk of overspecification. Among the 12 clusters (Supplemental Table S2, http://links.lww.com/HC9/A550), upregulated genes comprised 8 (Supplemental Figure S1, http://links.lww.com/HC9/A551) and downregulated genes comprised 4 (Figure 1A). These clusters—particularly the downregulated ones—were most notably distinguished by the magnitude of their regulation in wild-type animals (solid-colored bars) and their responsiveness to longer term ER stress in animals lacking ATF6α (black hashed bars). In particular, Cluster 7 stood out because the expression of those genes was profoundly suppressed (approximately 16-fold for the centroid) specifically by longer term ER stress in Atf6α−/− animals (Figure 1A). This finding is notable because it is the context in which the extensive suppression of metabolic genes in the liver was first observed.^6,8

FIGURE 1.

k-means clustering analysis of matched gene expression datasets from unfolded protein response mutants identifies a cluster of genes as likely to be HNF4α-dependent. (A) Data were assembled from published microarrays examining RNA expression in the livers of wild-type or Perk-, Ire1α-, or Atf6α-deleted animals following a challenge with 1 mg/kg tunicamycin. k-means clustering was used to divide expression data from each microarray into 12 separate clusters based on the behavior of each gene in each group of each microarray. The centroids for each of the 4 clusters that captured downregulated genes are shown, along with the number of genes falling into each cluster. For each microarray, the expression for all groups was normalized against expression in wild-type animals treated with vehicle. The data are shown in log₂-transformed format, so the average expression in wild-type animals is “0.” As an illustration, the genes of cluster 7 are suppressed, on average, ~16-fold by 34 hour tunicamycin treatment in Atf6α−/− livers. Error bars represent standard deviations from the centroid value for each animal. Upregulated cluster centroids are shown in Supplemental Figure S1, http://links.lww.com/HC9/A551. (B–E) The genes of clusters 7 and 8 were subjected to pathway analysis using DAVID (B, C) and searched for enriched transcription factor binding sites (D, E) using oPPOSUM. Statistical cutoffs were FDR P < 0.01 for DAVID, Z-score > 10, and Fisher score > 7 for oPPOSUM. Abbreviations: Atf6, activating transcription factor 6; FDR, false discovery rate; Ire1, inositol-requiring enzyme 1; Perk, protein kinase R-like ER kinase.

Cluster 7 was enriched for genes involved in a wide spectrum of biological processes that characterize functional hepatocytes, including nutrient metabolism, coagulation and complement cascades, drug metabolism, and bile production and secretion (Figure 1B). By contrast, only 2 pathways were identified for cluster 8, and none of statistical significance was identified for clusters 3 and 5 (Figure 1C and Supplemental Table S3, http://links.lww.com/HC9/A552). In addition, the promoter regions (within 5 kb of the transcriptional start site) of cluster 7 genes were enriched for binding sites of the transcription factors HNF1α, HNF4α, and HNF1β (Figure 1D), which together drive hepatocyte differentiation—with HNF4α being higher in the GRN than HNF1α/β.²¹ Binding sites for these 3 transcription factors were also enriched among cluster 8 genes, as were binding sites for other transcription factors implicated in hepatocyte differentiation and function (Figure 1E).²² In contrast, neither cluster 3 nor cluster 5 genes were enriched for binding sites of transcription factors with obvious connections to metabolism or hepatocyte identity (Supplemental Table S4, http://links.lww.com/HC9/A553). Taken together, these results suggest that the genes comprising cluster 7 in particular are likely direct or indirect HNF4α targets. This regulation is also mediated through PERK signaling since only the loss of PERK abrogated their suppression (Figure 1A).

The transcriptional response to ER stress partially phenocopies loss of HNF4α

Since the genes of most interest in the above approach were putative HNF4α targets, we hypothesized that ER stress interferes with HNF4α activity, in which case we would expect that such HNF4α-dependent genes would be suppressed both by ER stress and loss of HNF4α. To test this idea, we injected Hnf4α ^fl/fl mice^17,23 with AAV-TBG-GFP (w.t.) or AAV-TBG-CRE^24,25 (Hnf4α ^LKO ) to delete Hnf4α specifically in hepatocytes (Figure 2A), followed 1 week later by injection with TM and sacrifice 14 hours thereafter. Global RNA expression was then profiled by RNA-sequencing (RNA-seq) (Supplemental Table S5, http://links.lww.com/HC9/A554). Principal component analysis showed that both TM treatment and loss of HNF4α had substantial effects on global gene expression (Figure 2B). A heatmap of differentially regulated genes revealed several groups of interest (Figure 2C): some genes [group (a)] were suppressed by ER stress independent of the presence or absence of HNF4α. Conversely, other genes [group (c)] were suppressed by loss of HNF4α independent of ER stress. However, a greater number of genes that were suppressed by loss of HNF4α were also suppressed by ER stress [group (b)]. Interestingly, most of the genes upregulated by ER stress [group (d)] were upregulated to a lesser extent when HNF4α was lost.

FIGURE 2.

Liver-specific deletion of HNF4α partially phenocopies transcriptional suppression elicited by ER stress. (A) Deletion of HNF4α in the liver. Hnf4α ^fl/fl mice were treated with AAV8-TBG-eGFP (“w.t.”) or AAV8-TBG-CRE (“Hnf4α ^LKO ”) and then, 1 week later, challenged with 0.5 mg/kg TM, or vehicle control, for 14 hours as indicated. (B) Principal component analysis from RNA-seq. One sample each from the w.t. NT and Hnf4α ^LKO TM groups was excluded due to failure to pass presequencing quality control. (C) Heatmap of differentially expressed genes in the liver. Groups of interest are noted (a–d) and described further in the text. (D) A Venn diagram shows a substantial overlap between genes downregulated by TM (red) or by loss of HNF4α (blue) and also shows that most of the genes downregulated by TM in wild-type animals are not further significantly downregulated when HNF4α is deleted (yellow). (E) Biological processes downregulated by ER stress or loss of Hnf4α ^LKO identified by GO analysis. Abbreviations: ER, endoplasmic reticulum; FDR, false discovery rate; GO, gene ontology; KO, knockout; NT, not treated; TM, tunicamycin.

At least 20% of the genes significantly suppressed by ER stress were also significantly suppressed by loss of HNF4α and vice versa (Figure 2D). This overlap being not merely coincidental is attested by the fact that almost no overlap was observed among the genes upregulated by the 2 manipulations (Figure 4A). Among these overlapping genes, the vast majority (144/161) were not further suppressed by ER stress when HNF4α was deleted (Figure 2D). Pathway analysis was used to identify the biological processes represented among genes downregulated by either ER stress or loss of HNF4α. This analysis revealed that all but one of the processes suppressed by ER stress were also suppressed by loss of HNF4α (Figure 2E). Consistent with its identification as a group of likely HNF4α targets, Cluster 7 among the 4 downregulated clusters showed the greatest dependence on HNF4α, with the highest percentage of genes confirmed as HNF4α-responsive (Figure 3A and Supplemental Figure S2, http://links.lww.com/HC9/A551). This finding reinforces the idea that Cluster 7 identifies HNF4α targets that are suppressed by ER stress.

FIGURE 4.

Deletion of HNF4α in the liver diminishes unfolded protein response signaling during endoplasmic reticulum stress. (A) Analysis of genes upregulated by TM from RNA-seq data, analyzed similarly to Figure 2D. (B) Gene set enrichment analysis for protein secretion (top) or unfolded protein response (bottom) genes in unchallenged (left) and TM-treated (right) Hnf4α ^LKO animals compared to TM-treated wild-type animals. The blue vertical lines depict individual gene hits, and the red line depicts a “zero cross.” The negative ES and NES indicate that the gene sets skew toward the bottom (downregulated) portion of the gene ranking list for Hnf4α ^LKO animals relative to wild type. (C) Fold-induction by TM in both genotypes of the 20 genes most highly upregulated in wild-type livers. Statistical comparison was by 1-way ANOVA followed by Benjamini-Hochberg adjustment for multiple comparisons. #adjusted p between 0.05 and 0.1. (D) Immunoblot showing effective deletion of HNF4α and inhibition by TM of glycosylation of the endoplasmic reticulum–resident glycoprotein TRAPα, in an experiment conducted similarly to that shown in Figure 2. (E) quantitative-PCR analysis of the indicated unfolded protein response target genes. Abbreviations: ES, enrichment scores; FDR, false discovery rate; KO, knockout; NES, normalized enrichment score; NT, not treated; TM, tunicamycin.

FIGURE 3.

Hnf4α-dependent genes are selectively targeted by ER stress. (A) Venn diagram illustrating that a substantial portion of cluster 7 genes that were confirmed to be downregulated by TM in wild-type animals in the RNA-sequencing experiment were also downregulated by loss of HNF4α. Of these, the large majority (33/41) were not further downregulated by TM in Hnf4α ^LKO livers. (B) Expression of representative cluster 7 genes quantified by quantitative-RT-PCR from the livers of animals treated with TM similar to those shown in Figure 2 (Figure 4 and subsequent). Here and elsewhere, statistical comparison was by ANOVA with Tukey post hoc comparison unless otherwise noted. *p < 0.05; **p < 0.01; ***p < 0.001;****p < 0.0001. (C) Expression of the indicated genes in individual samples from the microarray data sets shown in Figure 1. (D) Quantitative-RT-PCR-determined expression of the genes in (C) determined as in panel (B). Abbreviations: Atf6, activating transcription factor 6; ER, endoplasmic reticulum; Ire1, inositol-requiring enzyme 1; NT, not treated; Perk, protein kinase R-like ER kinase; TM, tunicamycin.

To validate and extend these findings, we next carried out an identical ER stress experiment, this time using both sexes of mice (only males were used for the RNA-seq experiment). To examine gene regulation in this cohort with more granularity, we identified a group of 22 individual genes among the overlapping pathways identified in Figure 2E, which were suppressed by both ER stress and loss of HNF4α (Supplemental Table S6, http://links.lww.com/HC9/A555). We chose 5 of these from Cluster 7 for further examination—Pparα, Fads2, Dhcr24, Elovl2, and Elovl5 (Supplemental Figure S3, http://links.lww.com/HC9/A551). As we expected, all 5 of these genes are direct or indirect HNF4α targets, and none was further significantly suppressed by ER stress when HNF4α was absent (Figure 3B).

As a companion approach, we selected genes previously shown to be targets of HNF4α and to mark differentiated hepatocytes, including coagulation factors (F7, F9), urea cycle enzymes (Arg1, Otc), lipoprotein components (ApoC3), xenobiotic metabolism (Cyp3a11), and albumin (Alb),^13,23,26 without foreknowledge of how these genes would behave in either the k-means analysis or the targeted quantitative-RT-PCR analysis. Of these 7 genes, 4 were found within cluster 7 (Figure 3C). From the RNA-seq analysis, each of these 4 was suppressed by both ER stress and by loss of HNF4α, and only 1—F9—was significantly further suppressed by ER stress in Hnf4 ^LKO animals, albeit to a very modest extent (Figure 3D). Among the genes not identified as being in Cluster 7, Cyp3a11 was not significantly suppressed by ER stress, and Arg1 was not significantly suppressed by loss of HNF4α, while Alb behaved similarly to F9. Together, the data in Figures 1–3 suggest that the UPR suppresses a portion of the HNF4α-dependent GRN that regulates hepatocyte identity.

Loss of HNF4α diminishes ER stress sensitivity

Even though the loss of HNF4α is deleterious to the liver, the observation that its activity is seemingly impaired by the UPR implies that there is a functional benefit to the hepatocyte ER from this impairment during ER stress. The first observation in support of this idea was the effect of HNF4α disruption on bona fide UPR targets—those genes that are robustly upregulated by ER stress in wild-type animals. Notably, the majority of these were no longer significantly upregulated by ER stress in Hnf4α ^LKO livers (Figure 4A). This connection was supported by Gene Set Enrichment Analysis. Even in the absence of ER stress, the protein secretion and UPR gene sets were significantly—albeit modestly—suppressed in untreated Hnf4α ^LKO livers compared to wild-type untreated livers (Figure 4B, left and Supplemental Table S7, http://links.lww.com/HC9/A556). Thus, there appears to be a connection, manifested at the level of gene expression, between loss of HNF4α and ER homeostasis even when no exogenous ER stress is present. More strikingly, the UPR gene set was even further suppressed in TM-treated Hnf4α ^LKO livers relative to wild-type TM-treated livers; in fact, it was the most highly suppressed gene set under these conditions (Figure 4B, bottom right and Supplemental Table S8, http://links.lww.com/HC9/A557). Illustrating this point more concretely, of the 20 genes most strongly upregulated by ER stress in wild-type livers, a nearly uniform diminishment of that upregulation was seen in Hnf4α ^LKO livers (Figure 4C). These findings were supported by our companion experiment in the second cohort of animals (Figure 4D) where there was a lower upregulation in Hnf4α ^LKO livers of UPR target genes, including Bip, Edem1, Derl3, Erp72, and others (Figure 4E and data not shown). This broad suppression of the upregulation of target genes supports the conclusion that global UPR signaling is diminished in the livers of animals lacking HNF4α.

Diminished UPR signaling could be accounted for either by improved ER homeostasis in Hnf4α ^LKO livers or by disabled UPR signaling irrespective of ER homeostasis. In fact, disabled UPR signaling would be likely to exacerbate ER disruption caused by a stressor due to the lack of an adequate protective response. To discriminate between these possibilities, we examined ER ultrastructure by transmission electron microscopy. Loss of HNF4α on its own had no discernible effect on ER ultrastructure. Conversely, TM caused dramatic disruption of lamellar ER morphology in both genotypes. In wild-type animals, the ER is almost completely vesiculated. However, this disruption was less severe in Hnf4α ^LKO livers, as the ER did not vesiculate as completely as in wild-type animals and retained some lamellar organization, particularly in areas adjacent to mitochondria (Figure 5A). To measure this change, ER circularity was quantified from the experiment shown in Figure 5A, with the measurer blinded to sample identity; this analysis confirmed a less extensive circularization—that is, vesiculation—of the ER in Hnf4α ^LKO livers (Figure 5B). Also noted was a slight but significant elongation of mitochondria that was present upon ER stress only in Hnf4α ^LKO livers (Figure 5C).

FIGURE 5.

Loss of HNF4α aggravates liver injury while protecting the ER. (A) Representative transmission electron microscopy images showing ER and mitochondrial morphology in w.t. and Hnf4α ^LKO liver treated with vehicle or TM as in Figure 4. For each group, a region of perinuclear ER (nucleus indicated by “N”) and a region of nonperinuclear ER are shown, each from a different animal. In the TM-treated samples, orange arrowheads highlight dysmorphic ER, while yellow arrowheads indicate ER that is dilated but retains lamellar structure. (B, C) ER (B) and mitochondrial circularity (C) were quantified from EM images with the quantifier blinded to image identity. The degree of circularity was averaged over multiple images from each sample, with those averages shown as points within the violin plots. Statistical comparison was carried out on a per-animal basis, not a per-image basis. (D) Ki67 immunostaining shows hepatocellular proliferation in Hnf4α ^LKO livers. (E) Profound steatosis and likely cholecystitis are observed specifically in Hnf4α ^LKO livers after a 48 hour TM challenge. (F) The lipid droplet marker protein ADRP is elevated by both TM (14 h) and HNF4α deletion, and further elevated by their combination. (G, H) Plasma AST (G) and ALT (H) in w.t. and Hnf4α ^LKO animals treated with vehicle or TM for 48 hours. Abbreviations: ALT, alanine transaminase; AST, aspartate aminotransferase; ER, endoplasmic reticulum; NS, non-statistically significant; NT, not treated; TM, tunicamycin.

Despite apparently protecting ER homeostasis, loss of HNF4α exacerbated signs of liver injury upon ER stress challenge. As expected,¹³ loss of HNF4α led to hepatocellular proliferation, seen by Ki67 staining (Figure 5D), which was independent of ER stress treatment (data not shown). In addition, while both ER stress and loss of HNF4α led to hepatic lipid accumulation, the combined effect of both manipulations was, after a longer (48 h) challenge, profound hepatic steatosis and an enlarged and discolored gall bladder that was much more prominent in Hnf4α ^LKO TM-challenged animals than wild type (Figure 5E). Additional evidence for hepatic steatosis was seen in the expression of the lipid droplet marker protein ADRP, which was highest in Hnf4α ^LKO TM-treated animals (Figure 5F). While the loss of HNF4α did not affect plasma alanine transaminase (Figure 5G), aspartate aminotransferase was elevated only in Hnf4α ^LKO animals treated with TM (Figure 5H). Together, the data in Figure 5 suggest that loss of HNF4α protects the ER organelle from a TM challenge at the expense of the organ.

Maintenance of hepatocyte specification sensitizes cells to ER stress in vitro

Primary hepatocytes begin dedifferentiating immediately upon isolation and culture, as seen by their change to a more fibroblastic morphology (Figure 6A) and diminishment of HNF4α expression (Figure 6B). This propensity proved to be extremely vexing to our attempts to recapitulate in vitro the effects of in vivo deletion of HNF4α. However, it was recently demonstrated that a cocktail of 5 compounds (5C media) containing agonists and antagonists of various signaling pathways important in the maintenance of hepatocyte identity could forestall the dedifferentiation of primary human hepatocytes in vitro.²⁷ This 5C cocktail was partially effective in mouse primary hepatocytes as well, preserving the more cobblestone-like morphology of hepatocytes and reversing the decline in HNF4α expression seen upon culture (Figure 6A, B). The 5C-containing media also enhanced the expression of most (but not all) of the hepatocyte identity genes examined in Figure 3 (Figure 6C)—albeit to levels in most cases still well below those seen in intact liver (data not shown).

FIGURE 6.

Prodifferentiation 5C media sensitizes primary hepatocytes to ER stress. (A) Brightfield images of hepatocytes cultured for 5 days in either null or 5C-containing media, with the null cells showing a more fibroblastic appearance and the 5C cells retaining a more cobblestone-like morphology. (B) Immunoblot showing reversal of HNF4α suppression in hepatocytes cultured in 5C media. (C) Quantitative-RT-PCR showing expression of a set of genes diagnostic of hepatocyte differentiation after 5 days of null versus 5C-containing media. (D, E) 5C-cultured cells (5 d) are more sensitive to TM, as shown by quantitative-PCR analysis of unfolded protein response target genes (D) and Xbp1 mRNA splicing (E). (F, G) Same as (D, E) except using TG as the endoplasmic reticulum stressor rather than TM. Abbreviations: 5C, 5 compounds; Chop, C/EBP homologous protein; NS, non-statistically significant; NT, not treated; TG, thapsigargin; TM, tunicamycin.

We reasoned that this method could be used as an independent test of whether the differentiation state of hepatocytes affected cells’ sensitivity to stress. Indeed, cells grown in 5C-containing media were more sensitive to TM, as seen in the upregulation of UPR target genes including Erp72, C/EBP homologous protein (Chop), Derl3, and Bip (Figure 6D) and splicing of Xbp1 mRNA (Figure 6E). Similar results were obtained with the independent ER stressor thapsigargin (Figure 6F, G), which suggests that the link between differentiation and ER stress sensitivity is not confined to TM treatment. This difference occurs even though the expression of a wide array of ER-resident proteins involved in protein biogenesis did not differ between the 2 types of cells (Figure 7A).

FIGURE 7.

Secretory pathway capacity in null-treated versus 5C-treated cells. (A) Immunoblots of β-actin (loading control), HNF4α, and the indicated endoplasmic reticulum–resident proteins in null-treated and 5C-treated cells. Each lane represents an independently cultured well. (B) Cells cultured for 5 days in null- or 5C media were pulse-labeled for 30 minutes with ³⁵S-methioine/cysteine followed by chase in nonradioactive complete media for the indicated times. Albumin was immunoprecipitated from lysates and media and visualized by autoradiography. (C) Cells were cultured and labeled as in (B), and lysates were collected immediately after the labeling period. Equal protein masses were adsorbed to ConA-sepharose beads and eluted with the competitor sugar α-d-methyl-mannopyranoside. Samples were visualized by coomassie stain for total high-mannose N-glycoproteins or autoradiography for specifically newly synthesized N-glycoproteins. (D) An experiment similar to (C) except that total protein lysates were visualized. Abbreviation: 5C, 5 compounds.

We speculated that null-treated cells might have a lower overall load of nascent proteins within the secretory pathway since many proteins, such as albumin, which characterize differentiated hepatocytes are secretory pathway clients. Indeed, a pulse-chase with ³⁵S-methionine/cysteine confirmed that the synthesis of albumin was greatly diminished in null-treated cells. Yet, this protein was secreted equally rapidly in both cell types—within an hour in both cases—showing that there is no gross impairment of secretory pathway traffic in 5C-treated cells (Figure 7B). To assess ER client protein synthesis more globally, we pulse-labeled cells with ³⁵S-methionine/cysteine and then bound the lysates to ConA-sepharose, which binds only high-mannose glycoproteins (ie, newly synthesized pre-Golgi or ER-resident secretory pathway clients; Supplemental Figure S4, http://links.lww.com/HC9/A551) and, thus, serves as a readout for ER protein synthesis specifically.^28,29 Contrary to our prediction, null-treated cells were characterized not by diminished synthesis of ER client proteins but rather by elevated synthesis (Figure 7C) to an extent commensurate with a global increase in protein synthesis (Figure 7D). These results suggest that dedifferentiation in vitro renders cells relatively resistant to ER stress through a mechanism other than bulk changes in the load of nascent client proteins on the organelle.

DISCUSSION

It has long been understood that the major function of the UPR is the restoration of ER homeostasis, which it brings about primarily through transcriptional regulation. Across eukaryotic organisms and cell types, the core of this response is the upregulation of ER chaperones and other factors that facilitate ER protein folding, trafficking, and degradation.^30,31 Yet, the transcriptionally suppressive facet of the UPR is much less understood, and much less conserved from one cell type to another. In the liver, this response includes the suppression not only of genes involved in metabolism, as we and others have shown, but also more broadly of genes that demarcate hepatocyte identity as shown more recently¹⁰ and here by us.

The principal goal of this study was to understand the nature and consequences of this suppressive program for the hepatocyte. Our major conclusions are that ER stress interferes with a substantial portion of the HNF4α-dependent GRN, and this interference likely confers to the cell a degree of resistance to ER stress even at the potential cost of the hepatocyte functions to which HNF4α activity is crucial. Although HNF4α has previously been linked to the ER stress response by us and others,^10,11 this work is the first direct demonstration that HNF4α affects ER stress sensitivity in vivo. That the interference of hepatocyte identity protects the ER was seen using 2 complementary models—1 based on in vivo deletion of HNF4 and 1 based on the pharmacological forestalling of dedifferentiation in vitro. In both cases, diminished ER stress signaling was seen when HNF4α was absent or attenuated.

Although the responsiveness of numerous HNF4α-dependent genes to ER stress suggests that there is a discrete pathway by which the UPR interferes with HNF4α activity, the nature of this pathway is not yet clear. The simplest possibility, suggested elsewhere,¹⁰ is if HNF4α is itself a target of UPR-directed transcriptional suppression. However, in both this paper (Figures 2A, 4D) and our previous work,¹¹ ER stress-induced suppression of HNF4α targets preceded any detectable changes in HNF4α protein expression. Therefore, it seems more likely that the UPR interferes with HNF4α activity, and eventual effects on HNF4α expression are a consequence of the autoregulatory feedback loop by which HNF4α regulates its own expression.³² The major remaining possibilities are that the UPR prevents HNF4α from binding to its target promoters/enhancers or that it impairs its activity when bound. The observation that most but not all of the genes suppressed by loss of HNF4α are also suppressed by ER stress [groups (b) vs. (c), Figure 2C] suggests that HNF4α retains its activity toward some of its target genes, and thus, that wholesale loss of its DNA binding capacity during ER stress is unlikely. For this reason, we favor, on principle, models whereby the UPR interferes with HNF4α activity even when it is capable of binding its target sites.

How might the UPR interfere with HNF4α in ways besides preventing its DNA binding? The fact that the suppression of genes of cluster 7 is—like most of the other stress-suppressed genes—PERK-dependent (Figure 1A) points to a role for either translational repression mediated by eIF2α phosphorylation or to 1 or more of the factors that are translationally upregulated by eIF2α phosphorylation such as ATF4, CHOP, or others^33,34 (or possibly to both mechanisms). As we have shown, the stress-dependent repression of the cluster 7 gene Ppara is lost when CHOP is deleted.³⁵ As an inhibitory C/EBP-family transcription factor, CHOP could act by dimerizing with another C/EBP-family member that would ordinarily costimulate gene expression in concert with HNF4α and squelching it.³⁶ An attractive candidate would be C/EBPα, which is known to coregulate many genes with HNF4α.³⁷ However, we found that liver-specific deletion of C/EBPα alone had only modest effects on the expression of cluster 7 genes (data not shown). An alternate possibility is that CHOP binds to C/EBP binding sites (as we have shown in Chikka et al³⁵) and exerts a dominant negative influence on the transcription of genes that are ordinarily coregulated by C/EBPα and HNF4α. We also note that ER stress suppresses genes that are not HNF4α targets [group (a), Figure 2C], showing, unsurprisingly, that there are additional stress-dependent mechanisms for repressing hepatic gene expression as suggested.¹⁰

Our findings show that loss of HNF4α is beneficial to ER functionality at least during a TM challenge. They suggest that, more broadly, partially unwiring the hepatocyte gene expression program might be exercised by the UPR in order to protect ER function during physiological challenges. It stands to reason that the ER in hepatocytes—with their need for abundant protein secretion—might be intrinsically more taxed than in cells in which the HNF4α-dependent program has been (temporarily) silenced. We cannot yet be certain whether the differential stress sensitivity seen in the in vivo deletion of HNF4α and in vitro manipulation of hepatocyte identity through 5C treatment reflect the effects of the same cellular processes or merely give concordant results coincidentally. However, we found that cells cultured in null media, though they were relatively resistant to ER stress (Figure 6), did not, in fact, have a reduced total load of nascent ER client proteins compared to 5C-treated cells (Figure 7). This finding suggests that some other mechanism must drive the increased stress sensitivity at least of 5C-treated cells.

A key question is how permanent the apparent interruption of hepatocyte identity is in hepatocytes exposed to chronic ER stress and whether that phenotype underlies at least a portion of the liver injury seen in conditions such as obesity, alcoholism, and exposure to toxins, which have been shown to have an ER stress component.² In this study, we have referred to the gene expression changes that accompany ER stress in the liver as a loss of hepatocyte identity rather than as dedifferentiation, simply because we found no evidence that ER-stressed hepatocytes achieve a truly mesenchymal fate nor that they transdifferentiate. The transience of a single TM challenge in a wild-type animal might be insufficient to truly effect a permanent cell fate change. However, it is worth noting that cluster 7 genes in particular—those that most strongly encode hepatocyte identity and are most significantly tied to HNF4α—are also the most profoundly suppressed at a longer time point when the adaptive capacity of the UPR is impaired by deletion of ATF6α (Figure 1A). These genes are not ATF6α targets per se, in which case their regulation by ER stress would be lost when ATF6α is deleted. Rather, their enhanced suppression in Atf6α−/− animals is most likely a consequence of the fact that, without full upregulation of ER chaperones and other protective factors, Atf6α−/− cells experience much more persistent ER stress in the face of a particular challenge.³⁸ This finding ties hepatocyte identity proportionally to the persistence of ER stress being experienced by the cell. It suggests that ER stresses that are truly chronic in nature might result in the permanent dedifferentiation of at least a subset of hepatocytes. The consequence of this dedifferentiation would be at best an impairment of normal liver function and at worst an increased propensity toward hepatocellular transformation. The association of dedifferentiation with exacerbated ER stress has also been observed in other cell types, most notably secretory cells including pancreatic beta cells^39–41 and thyrocytes,⁴² suggesting that it could be a conserved feature of UPR signaling in professional secretory cells.

It is widely accepted that ER stress contributes to liver injury through hepatocyte cell death, which causes inflammatory signaling that ultimately stimulates liver fibrosis and transformation.⁴³ Indeed, during persistent stress, cells will presumably be more likely to become overwhelmed by the stress burden and die. Acute exposure to particularly hepatotoxic substances such as acetaminophen might result in a substantial level of ER stress-dependent cell death.⁴⁴ However, for the most common forms of chronic liver injury including obesity and alcoholism, it is conceivable that the transcriptional and presumptively adaptive response to ER stress might contribute as much or more to liver dysfunction as cell death.

Supplementary Material

hc9-7-e0278-s001.xls ^{(8.3MB, xls)}

hc9-7-e0278-s002.xls ^{(8.3MB, xls)}

hc9-7-e0278-s003.xls ^{(9.7MB, xls)}

hc9-7-e0278-s004.xls ^{(9.7MB, xls)}

hc9-7-e0278-s005.xlsx ^{(68.6KB, xlsx)}

hc9-7-e0278-s006.pdf ^{(4.7MB, pdf)}

hc9-7-e0278-s007.xlsx ^{(20.1KB, xlsx)}

hc9-7-e0278-s008.xlsx ^{(19.8KB, xlsx)}

hc9-7-e0278-s009.xlsx ^{(4.2MB, xlsx)}

hc9-7-e0278-s010.xlsx ^{(9.4KB, xlsx)}

hc9-7-e0278-s011.xlsx ^{(10.8KB, xlsx)}

hc9-7-e0278-s012.xlsx ^{(9.9KB, xlsx)}

hc9-7-e0278-s013.docx ^{(16.3KB, docx)}

hc9-7-e0278-s014.docx ^{(15.5KB, docx)}