HNF4α isoforms regulate the circadian balance between carbohydrate and lipid metabolism in the liver

Hepatocyte Nuclear Factor 4α (HNF4α), a master regulator of hepatocyte differentiation, is regulated by two promoters (P1 and P2) which drive the expression of different isoforms. P1-HNF4α is the major isoform in the adult liver while P2-HNF4α is thought to be expressed only in fetal liver and liver cancer. Here, we show that P2-HNF4α is indeed expressed in the normal adult liver at Zeitgeber time (ZT)9 and ZT21. Using exon swap mice that express only P2-HNF4α we show that this isoform orchestrates a distinct transcriptome and metabolome via unique chromatin and protein-protein interactions, including with different clock proteins at different times of the day leading to subtle differences in circadian gene regulation. Furthermore, deletion of the Clock gene alters the circadian oscillation of P2- (but not P1-)HNF4α RNA, revealing a complex feedback loop between the HNF4α isoforms and the hepatic clock. Finally, we demonstrate that while P1-HNF4α drives gluconeogenesis, P2-HNF4α drives ketogenesis and is required for elevated levels of ketone bodies in female mice. Taken together, we propose that the highly conserved two-promoter structure of the Hnf4a gene is an evolutionarily conserved mechanism to maintain the balance between gluconeogenesis and ketogenesis in the liver in a circadian fashion.


Introduction
Roughly 30% of human genes contain alternative promoters and yet the functional significance of the majority of those promoters, and the transcripts they generate, is woefully understudied.One such gene is the nuclear receptor (NR) Hepatocyte Nuclear Factor 4 alpha (HNF4a), a liver-enriched transcription factor (TF) best known as a master regulator of liver-specific gene expression and mutated in Maturity Onset Diabetes of the Young 1 (MODY1) (1,2).In mice, HNF4a is essential for fetal liver function (3) and liver knockout (KO) mice die within six weeks of birth with a fatty liver (4).
The human HNF4A and mouse Hnf4a genes are highly conserved and regulated by proximal P1 and distal P2 promoters.P1 drives the expression of transcripts containing exon 1A while P2 transcripts contain exon 1D, resulting in a loss of the N-terminal activation function 1 (AF-1).In the adult liver P1 is presumed to be the only active promoter, while during fetal liver development both P1 and P2 are active (5,6).The first P2-HNF4a transcript cloned, HNF4a7, was from the embryonal carcinoma cell line F9 (7), suggesting that it might play a role in cancer as well as fetal development.P1-HNF4a is downregulated in liver cancer and acts as a tumor suppressor (8)(9)(10)(11)(12), while overexpression of P2-HNF4a is linked to poor prognosis in hepatocellular carcinoma (HCC) (13).
The circadian clock regulates all aspects of physiology, including lipid metabolism.The liver is a major driver of the peripheral clock which is entrained by feeding and plays a role in fatty liver disease as well as hormonal homeostasis (14)(15)(16)(17)(18)(19).While P2-HNF4a is not typically found in normal adult liver, it is expressed under certain stress conditions that involve metabolic adaptation by cells, such as cancer and high fat diet (HFD) feeding (12,20).Furthermore, both cancer and HFD-induced obesity are known to be exacerbated by disruption of circadian rhythms (21,22).Since HNF4a is a known driver of liver metabolism and has been shown to play a role in hepatic circadian rhythms (2), we hypothesized that P2-HNF4a may play a unique role in liver metabolism and may intersect with the circadian clock.
To address the physiological role of P2-HNF4a, we employed exon swap mice (a7HMZ), which substitute exon 1A with exon 1D in the P1 promoter and demonstrate a role for the AF-1 domain in vivo (23).We compared the a7HMZ adult mice (express only P2-HNF4a) to wildtype (WT) mice (express P1-HNF4a) using RNAseq, ChIP-seq, rapid immunoprecipitation mass spectrometry of endogenous proteins (RIME), protein binding microarrays (PBMs) and metabolomics.An orchestrated, altered hepatic transcriptome in P2-HNF4a livers reveals large, significant differences in several cytochrome P450 transcripts and small but significant differences in key clock regulators, as well as expression of female-specific genes in the male livers.The distinct P2-HNF4a transcriptome appears to be due to altered protein-protein interactions, as well as altered chromatin binding but not differences in innate DNA binding specificity.Expression of P2-HNF4a is observed at Zeitgeber time (ZT) 9 and ZT21 in WT adult livers, and is upregulated in Clock-deficient mice.The P2-HNF4a hepatic metabolome is enriched in lipids and ketone bodies while mice expressing only P1-HNF4a exhibit enhanced gluconeogenesis and lack the elevated levels of ketone bodies normally found in females (24).Given that the P1/P2 promoter structure of the Hnf4a gene and many of its target genes are conserved across more than 90 million years (2,25), our results suggest that expression of P2-HNF4a in the liver is an evolutionarily conserved mechanism to balance carbohydrate and fatty acid metabolism during the circadian cycle.
The a7HMZ versus WT DEGs were compared with adult versus fetal (E14.5)liver DEGs (Figure 1G).More than two thirds of the genes upregulated in WT livers were also upregulated in the adult liver, while ~10% were enriched in fetal livers (562 versus 87, respectively).In contrast, a7HMZ-upregulated genes were more evenly split between adult and fetal liver (Figure 1G).Interestingly, alpha-fetoprotein (Afp) and other fetal liver genes were expressed at a lower level in a7HMZ (Supplementary Figure 1B), suggesting that the a7HMZ "program" is not simply a fetal one.Furthermore, while a7HMZ mice have a significantly (p < 0.01) higher liver-to-body weight ratio than WT or a1HMZ (exon 1A swapped for exon 1D in the P2 promoter) at postnatal day 14, the reverse (a1HMZ > a7HMZ) is observed at postnatal day 21 (Supplementary Figure 1C).Finally, proliferation genes Mki67 and Pcna were not upregulated in a7HMZ adult livers as one might anticipate for a predominantly fetal transcription factor (TF) (Supplementary Figure 1E).
To determine whether a7HMZ livers exhibit a cancer profile, the a7HMZ versus WT DEGs were plotted against DEGs of normal C57BL/6 livers versus murine hepatoma cell line Hepa1-6 (Figure 1H).As anticipated, genes upregulated in the WT liver were preferentially expressed at higher levels in normal liver compared to liver cancer (565 versus 69, respectively).In contrast, genes more highly expressed in a7HMZ livers were not enriched in liver cancer (Figure 1H and Supplementary Figure 1D).Taken together, these results indicate that P2-HNF4a drives a specific program of gene expression in the adult liver distinct from that of P1-HNF4a that is neither completely fetal-nor cancer-like, suggesting an alternative role for P2-HNF4a.

P2-HNF4a livers are less sensitive to the circadian clock
Since NRs are known to play an important role in regulating the circadian clock in the liver and HNF4a has been shown to play a role in hepatic circadian rhythms (27-30), RNA-seq of WT and a7HMZ livers was performed at three different time points (10:30, 13:30, 20:30, equivalent to ZT3.5, ZT6.5 and ZT13.5, respectively).While the expression of ~250 to 500 genes was significantly altered (padj <0.01, absolute log2FC ≥ 1) between any two of the time points in WT mice, less than half that number was altered in a7HMZ livers (Figure 2A, top), suggesting a reduced sensitivity of a7HMZ livers to the circadian clock.There were also more genes down-than upregulated in a7HMZ livers (Figure 2A, bottom), consistent with the loss of AF-1 function in vitro (7,31).A volcano plot of DEGs highlights differential gene expression between WT and a7HMZ HNF4a is one of the most highly expressed TFs in the liver; P2-HNF4a promotes "feminization" of the mouse liver.Frontiers in Endocrinology frontiersin.orglivers at 10:30 AM (ZT3.5),including several cytochrome P450 (Cyp) genes and the NR gene CAR (Nr1i3) (Figure 2B).
To examine the impact of P2-HNF4a on other NR genes, we compared the FPKM values of all NR genes across all three time points.HNF4a was the most highly expressed NR in both WT and a7HMZ; the next most abundant NR, Rxra, was expressed at roughly 25% the level of Hnf4a (Supplementary Figure 2A).While most NRs displayed similar circadian oscillations in WT and a7HMZ livers, there were some notable exceptions: CAR (Nr1i3) was significantly downregulated in a7HMZ at all three time points (Supplementary Figure 2A, arrow).Rev-Erbb (Nr1d2), RORg (Rorc) and PPARa (Ppara), all involved in the transcriptional feedback loop that drives circadian expression in the liver (28), exhibited significantly reduced expression in a7HMZ livers at one time point (Supplementary Figures 2B), again suggesting a decreased responsiveness to the clock.

HNF4a is one of the most highly expressed TFs in the liver
Consistent with the relative abundance of HNF4a protein in the adult liver (1,32), HNF4a had one of the highest transcript levels of any TF, higher even than subunits of RNA polymerase II (e.g., Polr2m, Polr2b) (Figure 2C).The other liver-enriched TFs (LETFs, Cebpa, Onecut2, Foxa1, Hnf1a, HNF1b) had transcript levels at least 10-fold lower than Hnf4a (Figure 2C, inset, arrows), consistent with HNF4a being a major regulator of liver-specific gene expression.Several TFs showed statistically significant differences between WT and a7HMZ (Figure 2C, asterisk), including those known to play a role in sexual dimorphic gene expression (Stat5a, Stat5b, Ahr, Nr0b2) (Figure 2D and Supplementary Figure 2B) (33,34).Interestingly, Esr1 (estrogen receptor alpha, ERa) expression was significantly upregulated in a7HMZ while Ar (androgen receptor, AR) was downregulated (Figure 2E), suggesting a potential "feminization" of the a7HMZ liver.

P2-HNF4a dysregulates the expression of genes involved in fatty acid, steroid and xenobiotic/drug metabolism
HNF4a is a known regulator of Phase I and Phase II enzymes involved in the detoxification of drugs and xenobiotics (35) and has been computationally linked to sexually dimorphic and circadian expression of those genes (36).Therefore, we examined the level of expression of all cytochrome P450 (Cyp) genes (Phase I) as well as glutathione S-transferases (Gst) and UDP glucuronosyltransferases (Ugt) (Phase II).While the diurnal pattern of expression was generally the same in WT and a7HMZ, the absolute level of expression was often altered (Figure 3A).For example, the expression of Cyp2c50 and Cyp2c54, which encode enzymes that metabolize linoleic acid, the endogenous HNF4a ligand (37), was much lower in a7HMZ livers (Figures 1E, 3A).Several Ugt genes were dysregulated and metabolomic analysis revealed a significant (padj <0.01) decrease in UDP glucuronic acid in a7HMZ livers (Figure 3A, bottom).Since glucose is needed to make UDP glucuronic acid, this decrease could be linked to carbohydrate metabolism.
The NR CAR (Nr1i3) is downregulated 10-to 18-fold in a7HMZ mice (Figure 3B), as reported previously (23), and could explain some of the changes in Cyp gene expression observed in a7HMZ livers (38).In contrast, the expression of PXR (Nr1i2), which is known to co-regulate many Phase I and II genes with CAR and to be upregulated by HNF4a in fetal liver (38, 39), was not altered (Supplementary Figure 2C), suggesting that the primary role of P2-HNF4a in the adult liver may not be to regulate xenobiotic metabolism.
In addition to Cyp2c50/54, transcript levels of other fatty acid metabolic enzymes were also decreased in a7HMZ livers.Cyp2b10 and Ephx2 (Figure 3C), which convert arachidonic acid to oxylipins via a two-step process (40), were significantly downregulated, as were all four DiHETrE products of arachidonic acid in the CYP2B10-EPHX2 pathway (Figure 3D), confirming a phenotypic effect on fatty acid metabolism.Changes in gene expression in the steroid metabolism pathway were also observed in a7HMZ livers with an increase in Cyp17a1 and a decrease in Srd5a1 and Hsd3b5 (Figure 3E).CYP17A1 plays a predominant role in steroid hormone biosynthesis, while steroid 5-alpha-reductase (Srd5a1) metabolizes the conversion of testosterone into the more potent dihydrotestosterone (DHT) and 3 beta-hydroxysteroid dehydrogenase type 5 (Hsd3b5) is typically lower in female livers (41).Tellingly, several of the most significantly increased transcripts in a7HMZ livers, including Cyp2b9, Cyp2b13 and Cyp2a4 (Figure 3F), are female-specific, have testosterone hydroxylase activity and are known to be regulated by HNF4a (42).Furthermore, Ephx2 expression and activity is downregulated by estrogen (43), which could explain the observed decrease in DiHETrEs in a7HMZ livers.All told, these results are consistent with the "feminization" of the a7HMZ livers suggested by the increase in ERa and decrease in AR expression (Figure 2E).

P1-and P2-HNF4a isoforms have similar but non-identical DNA binding profiles both in vivo and in vitro
To determine whether the P2-HNF4a transcriptional program is due to alterations in chromatin binding, ChIP-seq analysis was performed at 10:30 AM (ZT3.5) using an antibody (a445) that recognizes both isoforms (Figure 1A).Consistent with the high level of expression of the Hnf4a gene, there was a large number of HNF4a binding events in both WT and a7HMZ livers (~40,000 peaks).While the vast majority of peaks were similar in the two sets of mice, ~1.4 to 2.6% of the peaks were enriched for a particular isoform (WT unique: 572 peaks; a7HMZ unique: 1067 peaks) (Figure 4A).Analysis of the feature distribution of the ChIP peaks shows that both WT-and a7HMZ-unique peaks were less frequently located in the promoter region (≤ 2kb from +1) than the common peaks and the a7HMZ-unique peaks were enriched in intronic regions (Figure 4B).
Motif mining showed that the most common motif in both the WT and a7HMZ unique peaks was an HNF4a motif (xxxxCAAAGTCCA).
To determine whether there might be additional TFs bound in those peaks, we analyzed the DNA sequence of the uniquely bound peaks with an HNF4a-trained support vector machine (SVM) algorithm and categorized the peaks into one of four categories (>2, 2 to 1.75, 1.75 to 1.5 and 1.5 to 1.25 SVM score) based on the single highest-scoring SVM motif within the peak.All but a few peaks fell into one of these categories suggesting that the isoform-specific peaks are likely due to direct binding to the DNA.Nonetheless, de novo motif calling with MEME-ChIP revealed different TF motifs in some of the isoformspecific peaks.CEBPA and FOX were the only motifs significantly enriched in WT-unique peaks, but several motifs were found in a7HMZ-unique peaks, including SOX, GATA5, SMAD2, ETS, CEBPB, FOX and PAR bZIP (Figure 4C).
To investigate the innate DNA binding specificity of the HNF4a isoforms, we designed PBMs with variations on HNF4a consensus motifs (a direct repeat with a spacing of 1, DR1, AGGTCAxAGGTCA, or DR2, AGGTCAxxAGGTCA), as well as genomic sequences mined from HNF4a ChIP-seq peaks (Figure 4D, top middle).In total, ~44,000 test sequences were spotted in quadruplicate on a glass slide and probed with human HNF4a2 or HNF4a8 ectopically expressed in COS-7 cells or with liver nuclear extracts (NEs) from WT and a7HMZ mice (HNF4a2/  5 for data plotted in these graphs. HNF4a1 has been found to interact with SP1 both on and off chromatin, an interaction that involves the N-terminal domain of HNF4a1 (44-46).

HNF4a isoforms have unique interactomes
To assess the contribution of differential chromatin binding to changes in gene expression, we cross-referenced the ChIP-seq and RNA-seq datasets at the 10:30 AM time point and found that ~22% of WT-specific (62 out of 294) and a7HMZ-specific genes (41 out of 181) have one or more unique ChIP peaks within 50 kb of the transcription start site (TSS, +1) (Figure 5A).WT-specific genes matching these criteria include Nr1i3, Cyp2c50, Cyp2c54, Rarres1, Fmn1, Cdhr5, and Camk1d, while a7HMZ-specific genes include Cyp2b9, Fgfr1, Wnk4, Cyp4a14, Ppl, Vnn1, Acot1, and Cyp17a1 (Supplementary Table 2).Many of the most dysregulated genes contained differentially bound peaks within ~5 kb of +1 -Nr1i3, Apoa4, Cyp2c50, Cyp2c54, Cyp2b9, Cyp4a14, Acot1, Cyp17a1,Ucp2, Cyp2d26 and Treh (Figures 5B, C and Supplementary Figure 4A, B).While the differential peaks were typically not the only nor the largest peak in the gene, they could reflect rapid cycling on and off the DNA with functional consequences.
Since the majority of dysregulated genes had no nearby HNF4a isoform-specific ChIP peak, we examined HNF4a protein-protein interactions in WT and a7HMZ livers by RIME at 10:30 (ZT3.5) and 20:30 (ZT13.5).Both time points yielded a considerable number of interacting proteins at least eight-fold above the background, including many proteins that bound a single isoform   5D, bottom, middle).Many NRs interacted with HNF4a in both WT and a7HMZ livers, including NR3C1(GR) and NR0B2 (SHP), both of which have been shown previously to functionally interact with HNF4a, further validating the RIME results (47,48).There were also isoform-specific interactions, mostly with WT at 10:30 (NR1H2, NR2C1, NR2C2, PPARA).Interestingly, xenobiotic receptor PXR (NR1I2) interacted with HNF4a but only in a7HMZ livers at 10:30 (Figure 5D, bottom left and right).While the expression of Nr1i2 was not changed in a7HMZ livers (Supplementary Figure 2C), an environmental estrogen that activates PXR has been shown to increase the expression of two female-specific Cyp genes (Cyp2b9 and Cyp2a4) in male mice: both are significantly upregulated in a7HMZ livers and bound by HNF4a (Figure 5C and Supplementary Table 2) (49).HNF4a in a7HMZ livers also uniquely interacted with ESRRG (ERRg, Nr3b3) but only at 20:30: ERRs play important roles in mitochondrial biogenesis and function, including fatty acid oxidation (50).Interestingly, ERR DNA binding motifs were found in a7HMZ ChIP peaks but not WT (Supplementary Figure 3B).
There were many other TFs and co-regulators that interacted with a single HNF4a isoform, often in a circadian fashion (Figure 5D, bottom), which could explain the observed differential gene expression between WT and a7HMZ.Several of these proteins were previously confirmed by more conventional means (31,(51)(52)(53).Finally, there were several signaling molecules that interacted uniquely with the isoforms and at distinct time pointsone or more of these could also contribute to isoform-specific gene expression, independent of ChIP peaks (Supplementary Figure 4D).

HNF4a isoforms differentially impact circadian gene expression
Interactions with circadian TFs suggest that HNF4a may play a role in the hepatic clock.Analysis of all DEGs between any two time points (10:30, 13:30 or 20:30; padj<0.01 and log2FC>2) for either WT or a7HMZ yielded 53 genes, including commonly known circadian genes (Cry1, Rorc, Dbp, Bhlhe41, Usp2, Per2, Per3, Arntl, and Nr1d1) as well as many metabolism-related genes (Fmo3, Lpl, Car3, Corin, Npas2, Hmgcs1, Mme, Slc45a3, Hsd3b4, Hsd3b5, Slc10a2) (Figure 6A top).While most of these circadian-regulated genes showed the same general profile in WT and a7HMZ livers, there were some differences in the magnitude of the circadian effect between the genotypes.For example, Fmo3, a drug metabolizing gene whose expression varies greatly between individuals, had a much higher expression in a7HMZ livers at 20:30.In contrast, Aqp8, a water channel protein important for mitochondrial respiratory function (54), had much lower levels of expression in a7HMZ at all time points (Figure 6A, arrows).
While the majority of the clock machinery maintained cyclic expression in both genotypes (Figure 6A bottom), there were significant differences in expression between WT and a7HMZ in core clock components Arntl, Clock, Cry1, Nfil3, Npas2 and Per3 at one or more time points (Figure 6B), as well as Rorc and Ppara (Supplementary Figure 2B).The fact that other core components of the clock machinery did not show differences between the two genotypes (e.g., Per1, Per2, Rora, Bhlhe40) (Supplementary Figure 5A) suggests that the effect of P2-HNF4a on the clock is a specific one.
A sample distance matrix further confirmed a subtle yet real effect of P2-HNF4a on the hepatic clock.While the WT replicates at a given time point are much more similar to each other than they are to other time points, a7HMZ replicates show strong self-identity only in the 13:30 samples (Figure 6C).This is despite the fact that a principal component analysis (PCA) showed a good separation and categorization of each sample group (Supplementary Figure 7A).

P2-HNF4a is expressed at discrete times in the normal adult liver
While expression of P2-HNF4a protein in the normal adult liver has not been previously reported, this could be due to the time of day that livers are typically harvested (before midday).Since the current results show links between P2-HNF4a in a7HMZ livers and the circadian clock, we harvested livers from WT mice at four time points (ZT3, ZT9, ZT15, ZT21) and looked for P2-HNF4a mRNA by RT-qPCR and protein by immunoblot (IB).The results show expression of P2-HNF4a at ZT9 (4 PM) and ZT21 (4 AM) and a further increase in Clock KO livers.In contrast, P1-HNF4a RNA and protein levels did not oscillate in either WT or CLOCK KO mice (Figure 6E, Supplementary Figures 5B, C).

Metabolomic profiling indicates a role for P2-HNF4a in ketogenesis
Since metabolism is tightly linked to the clock in the liver (55-57), we performed metabolomic analysis of primary metabolites and complex lipids on WT and a7HMZ livers at 10:30 AM (ZT3.5).Nearly one-quarter of the primary metabolites (100 out of 369 total) were significantly down-regulated (p <0.05) in a7HMZ livers (Figure 7A, up in WT).Metabolite Set Enrichment Analysis showed that the top four enriched categories in WT are involved in carbohydrate metabolism and protein biosynthesis (Supplementary Figure 6A).Glucose and pyruvate were both significantly down in a7HMZ livers (Figure 7B), as was PEPCK (Pck1), an important enzyme in gluconeogenesis (Supplementary Figure 6B).Genes in pathways downstream of pyruvate were also significantly decreased in a7HMZ, including lactate dehydrogenase (Ldha, Ldhd), pyruvate carboxylase (Pcx) and citrate synthase (Cs) in the Kreb's cycle (Supplementary Figure 6C), as was citric acid, a key intermediate in the cycle (Figure 7B).Kreb's intermediates oxalic and succinic acid were also reduced although they did not reach significance (Supplementary Figure 6D).In contrast, genes involved in the formation of ketone bodies were upregulated in a7HMZ (Hmgcs2, Hmgcl) (Supplementary Figure 6E), as was the ketone body b-hydroxybutyric acid (Figure 7B), as previously reported (23).Levels of hundreds of complex lipids were altered  , 7. See Supplementary Table 4 for complete metabolomics data.
(up or down) in a7HMZ livers, including a notable increase in total triglycerides, diacylglycerides and acylcarnitines in the a7HMZ liver and a decrease in phospholipid species (Figures 7A, C and Supplementary Figure 6F).Since ketone bodies are elevated upon fasting, we performed RNA-seq on livers from 12-hr fasted WT and a7HMZ mice.The transcriptomes for both WT and a7HMZ fasted livers were quite distinct from the fed time points as well as from each other (Supplementary Figures 7A, B): WT mice had more genes altered upon fasting (673 versus 531 in a7HMZ) as well as more WTspecific genes either up-or downregulated (Figure 7D).Liver-tobody weight ratios were significantly lower in a7HMZ versus WT fed mice; in contrast, in fasted livers the ratio was lower in WT (Figure 7E).Since WT mammals are known to store fat in their liver during periods of fasting and since fasted a7HMZ livers accumulate more fat than WT mice (Figure 7C) (23), these results suggest that P2-HNF4a might promote a "fasting-response" program, consistent with the expression of P2-HNF4a at ZT9, near the end of the daily fasting period.
However, when the mice were subjected to a prolonged fast, unexpectedly, 50% of the a7HMZ mice died after ~60 hrs; in contrast, a1HMZ and WT mice survived a full 72 hrs without food (Figure 7F).Mortality was not due to hypoglycemia as blood glucose levels did not drop below 65 mg/dL; in fact, they increased after 48 hrs of fasting at ZT11, especially in a1HMZ (Supplementary Figure 7C).In contrast, circulating ketone bodies were highly elevated in the a7HMZ mice that survived the 60-hr fast (4.25 mM) (Figure 7G), and suggested that the a7HMZ mice undergoing a prolonged fast might have died of ketoacidosis.
Since the a7HMZ transcriptome showed signs of "feminization" and since females tend to have higher levels of ketone bodies than males (24, 58), we examined whether the elevated levels of ketone bodies in WT females is due to the ability to express P2-HNF4a.As anticipated, in WT mice ketone bodies were higher near the end of the daily fast (ZT11, 7 PM) than at the end of the feeding period (ZT23, 7 AM), in both males and females (Figure 7G).In contrast, a7HMZ males had ketone bodies at ZT23 nearly as high as at ZT11.Importantly, a7HMZ and WT females had much higher levels of ketone bodies at ZT11 than their male counterparts, whereas the a1HMZ females had levels similar to a1HMZ males and much lower than either WT or a7HMZ females (Figures 7G).This suggests that P2-HNF4a is required for the elevated levels of ketone bodies in females.

Discussion
While many mammalian genes have multiple promoters that drive expression of proteins with alternative N-termini, the physiological relevance of those different isoforms is seldom known.Using exon-swap mice and omics approaches, we show for the first time that the alternative isoform of the Hnf4a gene (P2-HNF4a), previously thought to be expressed only in fetal liver and liver cancer, plays an important metabolic role in the adult liver and is implicated in both the circadian clock and sex-specific gene expression.

Both P1-and P2-HNF4a are required for metabolic homeostasis in males and females
The "P2-HNF4a program" is characterized by a decrease in carbohydrate metabolism and an increase in hepatic fat storage, as well as ketogenesis, which typically occur during periods of fasting (59).Altered expression of genes involved in fatty acid oxidation or oxidative phosphorylation in the mitochondria are consistent with a shift from carbohydrates to fatty acids as an energy source (e.g., Hmgcs2, Acot1, Ucp2, Figure 1F and Supplementary Table 2).In contrast, it appears that P1-HNF4a drives gluconeogenesis and is required to temper the P2-HNF4a response to avoid ketoacidosis under conditions of stress, such as fasting.Only WT mice, which express both HNF4a isoforms in the liver, achieve homeostatic balance between carbohydrate and lipid metabolism (Figure 7H).We propose that this balance is achieved on a daily basis by upregulating P2-HNF4a at the end of the fasting period (~ZT9) (Figure 7), resulting in the well characterized elevation of ketone bodies right before feeding (60).Intriguingly, P2-HNF4a is also required for the elevated levels of ketone bodies in female mice (Figure 7G), consistent with a "feminization" of the a7HMZ livers and previous results showing that KO of HNF4a in the adult liver leads to a loss of male-specific genes and an increase of femalespecific genes (61).Finally, P2-but not P1-HNF4a interacts with the NAD-dependent deacetylase SIRT1, which is activated upon fasting and is associated with fatty acid oxidation, ketogenesis and fatty liver (Figure 5D) (62).

Multiple mechanisms are responsible for HNF4a isoform-specific gene regulation
Our results indicate that P2-HNF4a drives its unique transcriptional program via multiple mechanisms.Differential recruitment of co-regulators to target gene promoters (Figure 5D) could explain altered expression of genes such as Ces2e, which encodes carboxylesterase 2, an enzyme that hydrolyzes triacylglycerols (Figure 7I top).Both P1-and P2-HNF4a bind the Ces2e promoter in a similar fashion but Ces2e is expressed at much lower levels in a7HMZ livers compared to WT, which could explain the elevated levels of triglycerides in a7HMZ livers (Figures 1E, 7C, Supplementary 4C).A second potential mechanism is differential binding to regulatory regions.An enriched ChIP-seq peak in a7HMZ livers, for example, could explain the upregulation of a key enzyme in b-oxidation of fatty acids, Acot1 (Figures 7I middle and Supplementary Figure 4A).Similarly, a reduction in ChIP peaks could explain the decrease in expression of Apoa4 and Nr1i3 (CAR) in a7HMZ livers (Figures 5B, C).A third mechanism involves differential recruitment and/or interaction of TFs with a given HNF4a isoform (Figure 7I bottom).For example, PPARa is known to be a major player in ketogenesis, activating the expression of the mitochondrial enzyme HMGCS2 which catalyzes the first step in ketogenesis (59).P1-HNF4a has been shown to decrease Hmgcs2 expression by repressing PPARadependent activation (63).This repression could be facilitated by a unique protein-protein interaction between P1-HNF4a and PPARa (Figure 5D).In contrast, in a7HMZ livers Hmgcs2 expression is elevated and HNF4a ChIP-seq peaks in a7HMZ livers are similar to those in WT (Supplementary Figures 4C, 6E).Similarly, specific interactions between P2-HNF4a and TFs involved in sex-specific gene expression (e.g., NR1I2, SP1 family/ KLF) could contribute to increased expression of female-specific genes such as Cyp2b13 (Figures 3F, 5D, Supplementary Figure 4C) (49).Additional mechanisms driving the P2-HNF4a program include differential interaction with signaling molecules, altered expression of other TFs, such as those that play a role in sexspecific gene expression (Stat5b, Stat5a, estrogen and androgen receptors) (33,36), and elevated levels of ketone bodies which can impact histone deacetylase activity, as well as the circadian clock (64) (Figures 2, 7G, Supplementary Figure 4D).

Physiological and pathological triggers of the P2-HNF4a program
There are now three known physiological conditions in which P2-HNF4a is expressed in the liverfetal liver and ZT9 and ZT21 in adult liver (Figure 6).Increased expression of P2-HNF4a expression right before birth (E17.5),followed by a sharp decline after birth (5, 6), could explain why the a7HMZ transcriptome is not more similar to that of the E14.5 fetal liver: rather than promoting early liver development, the role of P2-HNF4a appears to be a metabolic one, perhaps preparing the fetus to survive the birthing process and immediate postnatal period by increasing fat in the liver.The subsequent decrease in P2-HNF4a expression after birth could be mediated by GR which is induced by stress hormones released during parturition (65): GR preferentially increases the expression of P1-HNF4a (7, 66) which would in turn repress the P2 promoter (6).
Factors responsible for increased expression of P2-HNF4a at ZT9 have not been identified, but its expression seems to be required for the increased expression of ketogenic genes and ketone bodies in response to the daily fast (Figures 7B, G, Supplementary Figure 6E).The role of P2-HNF4a at ZT21 is more difficult to explain as ketone bodies are low at that time (Figure 7G) (60).Total protein synthesis is increased at ~ZT22 (67), as well as both P2-and P1-HNF4a-specific targets (Supplementary Figure 7D), so expression of P2-(and P1-)HNF4a at ZT21 could be the result of a global effect on protein synthesis.
In addition to physiological triggers, there are now four pathological conditions in which P2-HNF4a is known to be elevated in the adult livercancer (11,12,68), high fat diet (12, 20), disrupted clock (Figures 6D, E) and alcoholic hepatitis (68).In terms of cancer, our results indicate that P2-HNF4a is not oncogenic per sethe P2-HNF4a transcriptome shows only a partial overlap with HCC, key proliferation markers (Ki67 and PCNA) are not upregulated in a7HMZ livers and there is no evidence of hepatomegaly (Figures 1H, 7E, Supplementary Figures 1D,E).Furthermore, no increase in spontaneous, macroscopic tumors has been observed in a7HMZ livers, even in older mice (unpublished observation).While HCC patients with increased P2-HNF4a have a poor prognosis (13), rather than acting as an oncogene per se, P2-HNF4a may be upregulated simply due to a decrease in the expression of the tumor suppressor P1-HNF4a (7, 66) and inadvertently promote liver cancer progression via metabolic effects.For example, acylcarnitines are elevated in a7HMZ livers (Figure 7C, Supplementary Figure 6F) and have been identified as potential diagnostic and prognostic biomarkers for HCC (69,70).Given the renewed interest in cancer metabolism, including in HCC, it will be of interest to determine exactly how a metabolism altered by unopposed P2-HNF4a might contribute to cancer progression and/or treatment (71).This is particularly true considering that elevated ketone bodies may trigger a protective mechanism against oxidative stress (72), which would suggest that elevated levels of P2-HNF4a in HCC may actually play a protective role.Finally, several matrix metalloproteinases (Mmp14, Mmp15, Mmp19), which are linked to poor prognosis of liver or colorectal cancer patients (73)(74)(75), are also upregulated by P2-HNF4a (Supplementary Table 1) and dysregulation of genes involved in drug metabolism could impact treatment of liver cancer (Figure 3).
The second condition that leads to expression of P2-HNF4a in the adult liverhigh fat diet (HFD)could be related to both cancer and the third condition, disrupted clock.We recently reported that P2-HNF4a expression is increased in the livers of mice fed a HFD and that the circadian regulator BMAL1 represses P2-HNF4a expression in HCC (12).Consistently, P2-but not P1-HNF4a interacts with BMAL1 (ARNTL) and CLOCK and the Clock KO increases P2-but not P1-HNF4a expression (Figures 6D,  7D, E).Dysregulation of the clock, such as during jet lag, could potentially contribute to liver cancer by upregulating P2-HNF4a (Figures 6D, E) (21).While HNF4a, including exon 1A and exon 1D, is highly conserved between mouse and human (96% identity on the protein level in BLAST) (2), given the genetic variation in regulatory regions between species, it will be of interest to determine whether P2-HNF4a expression impacts the expression of genes involved in the human hepatic metabolome (and hepatic circadian clock) in the same fashion as the mouse (76).
The fourth pathological condition where P2-HNF4a is expressed in the liverhuman alcoholic steatohepatitis (68)is consistent with increased fat in a7HMZ livers and an enrichment of genes associated with alcoholism in a7HMZ mice (Figures 1F, 7C).The TGFb pathway is implicated in P2-HNF4a expression under this scenario; SMAD binding motifs were found in a7HMZ ChIPseq peaks but not WT peaks (Supplementary Figure 3B).
In summary, the results presented here strongly suggest that the function of P2-HNF4a is to modulate the hepatic metabolic response in general, rather than to solely promote proliferation during fetal development and liver cancer.While the elevated levels of circulating ketone bodies in a7HMZ mice suggest that P2-HNF4a may be a player in the fasting response, other results suggest that the role of the alternative isoform of HNF4a may be more complex.For example, expression of CAR (Nr1i3) is significantly decreased in a7HMZ livers and yet CAR is known to be increased during fasting, due to the action of PPARa and PGC1a, along with HNF4a, on the Nr1i3 promoter (77).Additionally, while expression of PEPCK (Pck1), a major driver of gluconeogenesis, did not increase after a 12-hour fast in a7HMZ livers as it does in WT animals, others have shown that P2-HNF4a can activate the Pck1 promoter more effectively than P1-HNF4a, at least in the presence of PGC1a (20).Taken together, the findings presented here indicate that P2-HNF4a plays an important physiological role in the normal adult liver and intersects with the circadian clock in a complex fashion that merits further investigation.

Materials and methods
(See Supplemental Methods for additional methods and details).

Animal models
Young adult (16 to 20 weeks) male WT and a7HMZ mice in a mixed 129/Sv plus C57BL/6 background (23) were fed a standard lab chow (LabDiet, #5001:13.6%fat from pork lard; 28.9% protein; 57.5% carbohydrates) and used for RNA-seq, CHIP-seq, RIME analysis (all samples from the same set of mice), and oxylipin analysis.The a7HMZ male mice used for primary metabolite and complex lipid metabolomic analysis were backcrossed to C57BL/6N for 10+ generations and used with C57BL/6N WT controls (n=8, 35 weeks of age).a7HMZ and a1HMZ (backcrossed 10+ generations into C57BL/6N) were compared to scientific C57BL/6N (WT) controls for newborn liver analysis (mixed-sex) and glucose/ ketone body analysis (males and females; ~16 to 20 weeks of age).Clock-deficient (Clock KO) male mice were provided by Dr. David Weaver (78) and fed a standard rodent diet (PicoLab Rodent Diet #5053: 13.1% fat from soybean oil; 24.5% protein; 62.4% carbohydrates).All mice were fed ad libitum and kept in 12-hr light/dark conditions in a specific pathogen-free (SPF) facility, unless indicated otherwise, and euthanized by CO 2 asphyxiation followed by tissue harvest at the indicated time points.Care and treatment of the animals were in strict accordance with guidelines from the Institutional Animal Care and Use Committee at the University of California, Riverside (UCR), or the McGovern Medical School, UT Health.
Reads were aligned to the mouse reference genome (mm10) with TopHat v2.1.1 using default parameters except for allowing only 1 unique alignment for a given read.Raw read counts were calculated at the gene level for each sample using HTSeq v0.6.1.Library normalization was performed with EDASeq (80); withinlane normalization on GC content was performed with the LOESS method and between-lane normalization was performed with nonlinear full quantile method.Normalization factors from EDASeq were used for differential expression analysis with DESeq2.Normalized read counts, FPKM (fragments per kilobase per million), and rlog (regularized log transformation) results were generated for downstream analysis.

Chromatin immunoprecipitation sequencing (ChIP-seq) and SVM analysis
ChIP-seq of isolated liver cells from WT and a7HMZ males (n=3, aged 16-18 weeks) was performed as previously described (79) using 4.2 mg of affinity-purified anti-HNF4a (a445) (1) or rabbit IgG control (Santa Cruz, cat#sc-2027).Libraries were submitted for 50-bp single end sequencing by Illumina HiSEQ 2500 at the UCR IIGB Genomics Core.Reads were aligned to the mouse reference genome (mm10) with Bowtie2.Peaks were called with MACS2 for individual samples, as well as a pooled peak dataset using the SPMR (signal per million reads) parameter.Aligned reads and MACS2 peak-sets were analyzed with DiffBind (81) with DESeq2 and library size equal to total aligned reads to identify common and uniquely bound regions of the genome.Default parameters were used unless noted otherwise.ChIP-seq peaks were called with MACS2 and then filtered on -log10(p-value) ≥ 10, to approach six-fold enrichment above control.Differentially bound peaks were identified using DiffBind with MACS2 output.Curated peak lists were generated by filtering all results on peaks with "concentration" ≥ 5; defined by DiffBind as the "mean (log) reads across all samples" in contrast.The kernel-based SVM was trained as previously described using results from independent HNF4a PBM experiments (82).

Protein binding microarrays (PBM)
Protein binding microarrays (PBMs) were carried out as previously described (82).Nuclear extracts (NE) were prepared from COS-7 cells transiently transfected via CaPO 4 with HNF4a expression vectors for human HNF4a2 (NM_00457) and HNF4a8 (NM_175914) essentially as previously described (83).Liver NE from WT and a7HMZ mice were prepared as previously described (37).A custom-designed array was ordered from Agilent (SurePrint G3 Custom GE 4x180k), which contained oligonucleotides ~60 nucleotides (nt) in length comprised of: sequences within 100 bp of the center of HNF4a ChIP-seq peaks from human colon cancer cells (proliferative Caco-2) (84) were taken in 30nt windows moving 5 nt at each step (~25,000 sequences); 17,250 permutations of canonical HNF4a DR1 motifs (5'-AGGTCAAA GGTCA -3'); 500 permutations of DR2 motifs with variable spacer (5'-AGGTCNNNNGGTCA -3'); ~900 random control 13-mer DNA sequences and ~170 positive controls.A total of ~44,000 test sequences were spotted in quadruplicate on the slide as single-stranded DNA for a total of ~176,000 spots of DNA.The DNA was made double-stranded and transiently transfected Cos-7 cells expressing human HNF4a2 or human HNF4a8 or liver NEs from adult male mice fed a standard rodent diet were applied.HNF4a binding was imaged with 2-µm resolution using Agilent G2565CA Microarray Scanner at the UCLA DNA Microarray Core.Extraction and normalization of the data were as described previously (82) using gProcessedSignal from Agilent software with background correction.PWMs were generated using seqLogo.See Supplementary Table 5 for PBM results from the Caco-2 ChIPseq peaks (~25,000 unique sequences) presented in Figure 5C and the PBM Project at Synapse.org for the entire dataset.

Rapid immunoprecipitation and mass spectrometry of endogenous proteins (RIME)
RIME was performed as previously described (85) with slight modifications.Livers from the same mice used for the RNA-seq and CHIP-seq -WT and a7HMZ males n=3, 16-18 weeks of age sacrificed at 10:30 (ZT 3.5) or 20:30 (ZT13.5)werecrosslinked and IP'd with the P1/P2 antibody.Multidimensional protein identification technology (MudPIT) analysis was performed by the UCR IIGB Proteomics Core.Raw MS1 and MS2 spectra were processed with Proteome Discoverer 2.1 (Thermo Scientific) and submitted to Mascot search engine to match against NCBI nonredundant mouse protein database.Only proteins with 1% FDR cut-off (q ≤ 0.01) were considered for subsequent analysis.Area under the curve, as reported by Proteome Discoverer, was averaged together for WT and a7HMZ samples (n=3) at each time point.IgG samples (n=3) from both WT and a7HMZ were averaged together to create a background sample.Areas were converted to log2 scale and the fold-change above IgG background was calculated for the WT and a7HMZ samples.Proteins with less than 8-fold change above background were omitted.Similarly, a 8-fold difference between WT and a7HMZ samples was used to identify unique protein interactions.

Primary metabolite, complex lipids and oxylipin analysis
All metabolomic analysis was performed at the West Coast Metabolomics Center at the University of California Davis as described previously (86) using liver tissue rinsed in cold PBS, snap frozen and stored in liquid nitrogen.Data (pmol/gm tissue or peak height) are presented as mean +/-standard error of mean (SEM).Student's T-test was used to determine statistical significance (p < 0.05) using GraphPad Prism v6.

Quantification and statistical analysis
Differential gene expression (DEG) was measured using raw read counts with DESeq2: statistical significance was defined as adjusted p-value (padj) ≤ 0.01, unless otherwise noted.Legends denote thresholds using log2 fold change (log2FC) cutoffs.R library "gage" was utilized to identify differentially enriched KEGG pathways in Figure 1.Heatmaps were generated with pheatmap package in R; data were row-normalized before plotting, except for NR heatmap in Supplementary Figure 2. Transcription Factor (TF) rankings for Cleveland plots were ordered at the 13:30 (peak HNF4a expression) then manually curated with the aid of PANTHER (Mi et al., 2017).Venn diagrams were generated by the VennDiagram package in R. Unique and common RIME results were submitted to DAVID for ontology analysis.Statistical significance for primary metabolite and complex lipid data defined as p ≤ 0.05 by Mann-Whitney U-test or Benjamini-Hochberg padj <0.05, as indicated.All barplots represent mean ± SEM; significant differences are noted between genotypes at a given time point, unless indicated otherwise.For FPKM plots, padj values are from DESeq2; in other plots, p-values are from two-way Student's T-test or One/Twoway ANOVA, as indicated.Student's Ttest was used while comparing two groups/conditions.One-way ANOVA was used for analyzing data for more than two groups that were compared for only one factor and two-way ANOVA was used for analyzing data from more than two groups that were compared for two factors.Posthoc analysis was applied for the ANOVAs to account for multiple comparisons.Standard statistical tests/programs were used for analyzing the metabolomics and transcriptomics data.External expression datasets and analysis are described in Supplemental Methods.
The raw metabolomics data (primary metabolites and complex lipids) have been deposited in Metabolomics Workbench (www.metabolomicsworkbench.org)under Project #PR000461.
FIGURE 5HNF4a isoforms have unique protein-protein interactions.(A) Number of genes with one or more WT-or a7HMZ-unique ChIP-peaks within a 50kb of +1 of differentially expressed genes in WT and a7HMZ livers (padj ≤ 0.01).(B, C) UCSC Genome Browser view of dysregulated genes with a unique ChIP-signal and RNA-seq from 10:30 AM.Axes for WT and a7HMZ signals are set to the same scale but may differ between genes.(D) Top, number of proteins bound to HNF4a in WT vs. a7HMZ livers at 10:30 (ZT3.5 left), at 20:30 (ZT13.5 right) and in both WT and a7HMZ at 10:30 vs. 20:30 (middle).Bottom, select proteins involved in transcription regulation bound only in WT, a7HMZ or both genotypes.See Supplementary Figure4Dfor interactions with signaling proteins and Supplementary Table3for all interacting proteins.

Table 3
for all interacting proteins.(Figure5D,top).There was considerable overlap between the common groups for the two time points (96 proteins bound HNF4a in both WT and a7HMZ livers at both time points), underscoring the robustness of the method.There were also many proteins that bound both isoforms but only at a single time point(10:30: 141; 20:30: 401).