Hif1α/Dhrs3a Pathway Participates in Lipid Droplet Accumulation via Retinol and Ppar-γ in Fish Hepatocytes

Excessive hepatic lipid accumulation is a common phenomenon in cultured fish; however, its underlying mechanisms are poorly understood. Lipid droplet (LD)-related proteins play vital roles in LD accumulation. Herein, using a zebrafish liver cell line (ZFL), we show that LD accumulation is accompanied by differential expression of seven LD-annotated genes, among which the expression of dehydrogenase/reductase (SDR family) member 3 a/b (dhrs3a/b) increased synchronously. RNAi-mediated knockdown of dhrs3a delayed LD accumulation and downregulated the mRNA expression of peroxisome proliferator-activated receptor gamma (pparg) in cells incubated with fatty acids. Notably, Dhrs3 catalyzed retinene to retinol, the content of which increased in LD-enriched cells. The addition of exogenous retinyl acetate maintained LD accumulation only in cells incubated in a lipid-rich medium. Correspondingly, exogenous retinyl acetate significantly increased pparg mRNA expression levels and altered the lipidome of the cells by increasing the phosphatidylcholine and triacylglycerol contents and decreasing the cardiolipin, phosphatidylinositol, and phosphatidylserine contents. Administration of LW6, an hypoxia-inducible factor 1α (HIF1α) inhibitor, reduced the size and number of LDs in ZFL cells and attenuated hif1αa, hif1αb, dhrs3a, and pparg mRNA expression levels. We propose that the Hif-1α/Dhrs3a pathway participates in LD accumulation in hepatocytes, which induces retinol formation and the Ppar-γ pathway.


Introduction
Currently, excessive accumulation of liver fat in cultured fish is a common phenomenon in the fish industry, representing a main risk factor for poor health status, high mortality, and decreased quality of fish fillets. This prevalent issue has become a bottleneck in the healthy development of the industry. In fish, fatty liver is mainly caused by the accumulation of lipid droplets (LDs) in hepatocytes. Thus, regulating the size and quantity of LDs is a key factor in controlling fatty liver disease. LDs are composed of a core enriched in triglycerides (TGs) and sterol esters (SEs), which are surrounded by a phospholipid monolayer due to the hydrophobicity of neutral lipids [1]. Many proteins are localized in the phospholipid membranes and are believed to function in the storage, transport, and metabolism of lipids, in signaling, and in providing a specialized microenvironment for metabolism [2]. LDs are highly dynamic organelles that are prone to changes depending

LD Accumulation Is Accompanied by Increased Dhrs3 Expression
To obtain the transcript profiles of ZFL cells during LD accumulation, we collected cells incubated in normal medium (NM) and high-fat medium (HFM), representing two types of fat-accumulating cells ( Figure 1A,B), and subjected them to RNA sequencing. In total, 320,061,096 raw reads were generated (Table S1). Among them, 314,128,116 high-quality clean reads (47.12 G) were obtained after filtering low-quality reads from the raw data. At least 90% of the reads matched the reference genome (Supplementary Table S1). A total of 936 and 850 differentially expressed genes (DEGs) were upregulated and downregulated, respectively, in the cells exposed to HFM compared to those exposed to NM ( Figure 1C; Table S2).
To verify the accuracy of the transcriptome data, the mRNA expression of dhrs3a and dhrs3b was determined using quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR). The mRNA expression of dhrs3a was significantly increased after the cells were incubated in the lipid medium for 1.5 h, with the expression peaking at 6 h and decreasing after 12 h ( Figure 1E). Notably, the mRNA expression of dhrs3b was only significantly increased at 12 h, and no expression was noted before 6 h ( Figure 1E). To clearly visualize the expression pattern of Dhrs3, we performed immunofluorescence staining of the protein. Cells in HFM showed an increase in Dhrs3 puncta. Dhrs3 was not observed on the enveloped LDs, but some presented punctate in the cytoplasm ( Figure 1F).

Dhrs3a Knockdown Delays LD Accumulation in Cells in a Lipid-Rich Medium
To explore the role of Dhrs3a in LD accumulation in ZFL cells, we suppressed the expression of dhrs3a in cells using RNAi technology and incubated these cells with lipids. The mRNA expression of dhrs3a decreased by 52.70% after treatment with siRNA against dhrs3a ( Figure 2A). After incubation with the lipid-rich medium for 6 h, LD accumulation decreased in the Si-dhrs3a-treated cells compared to that in Si-N.C.-treated cells. However, this phenomenon was not observed in the cells incubated in the non-lipid medium ( Figure 2B,C). At the transcript level, incubation with the lipids significantly increased the mRNA expression of peroxisome proliferator-activated receptor gamma (pparg), sterol regulatory element-binding protein-1c (srebp-1c), and fatty acid synthase (fasn). However, treatment with Si-dhrs3a decreased the mRNA expression of pparg, srebp-1c, fasn, and fatty acid transport protein (fatp); among them, pparg was significantly upregulated ( Figure 2D). Notably, the reduction in LDs was almost completely reversed after incubation for 24 h ( Figure 2E,F), indicating that Dhrs3a delays LD formation in response to exogenous lipids at an early stage of development.

Exogenous Retinyl Acetate Maintains LD Accumulation
The ectopic expression of Dhrs3 in cells has been reported to increase retinyl ester concentrations [9], suggesting that Dhrs3 acts as a retinal reductase to generate storage forms of retinol ( Figure 3A). As expected, the retinol concentration in cells in HFM was significantly higher than that in cells in NM ( Figure 3B). To explore the role of retinol in the regulation of LD accumulation, the cells were treated with or without retinyl acetate. Under NM conditions, no obvious change in LD accumulation was observed in the retinyl acetatetreated cells ( Figure 3C,D). However, in the HFM, retinyl acetate-treated cells showed higher LD accumulation ( Figure 3C,D). TG content was detected only in cells treated for 24 and 48 h. Notably, as time increased from 6 to 48 h, the cellular TG content decreased in the control group but did not change in the retinyl acetate group ( Figure 3D), indicating that retinyl acetate maintained LD accumulation. At the molecular level, retinyl increased the expression of lipid synthesis-related genes, including pparg, fatty acid-binding protein 1 (fabp1), and fasn, especially in the HFM group ( Figure 3E).

Exogenous Retinyl Acetate Alters the Lipidome of Zebrafish
To explore the changes in lipid composition, we analyzed the lipidomes of cells treated with or without retinyl acetate in an HFM environment ( Figure 4). In the positive ion mode, 246 phosphatidylcholines (PCs), 83 phosphatidylethanolamines (PEs), 92 triacylglycerols (TAGs), and 35 sphingomyelins (SMs) were identified ( Figure 4A). Principal component analysis (PCA) revealed clear differences in the lipid compounds between the retinyl acetate and HFM groups ( Figure 4B), and a total of 84 upregulated and 58 downregulated lipid compounds were identified in the retinyl acetate-treated cells compared with those in the HFM group ( Figure 4C). The upregulated lipid compounds included 29 PCs, 25 TAGs, and 9 PEs, whereas the downregulated lipid compounds included 12 cardiolipins (CLs), 9 PCs, and 7 phosphatidylserines (PSs) ( Figure 4D). No CLs or PSs were upregulated in the retinyl acetate group compared to the HFM group. In the negative ion mode, 75 PCs, 74 PEs, 34 ceramides (Cers), and 33 phosphatidylglycerols (PGs) were identified in the cells (Supplementary Figure S2A). PCA clearly separated the different treatments (Supplementary Figure S2B). A total of 28 lipid compounds were upregulated and 37 were downregulated in the retinyl acetate group compared to the HFM group (Supplementary Figure S2C). Specifically, the upregulated lipid compounds included nine PCs, eight PGs, and seven PEs, whereas the downregulated lipid compounds included 12 CLs, 8 phosphatidylinositols (PIs), and 6 PS (Supplementary Figure S2D). . pparg, peroxisome proliferatoractivated receptor γ; srebp-1c, sterol regulatory element-binding protein-1c; fasn, fatty acid synthase; fatp, fatty acid transport protein. Statistical significance is denoted with asterisks as follows: * p < 0.05; ** p < 0.01; *** p < 0.001.

Exogenous Retinyl Acetate Maintains LD Accumulation
The ectopic expression of Dhrs3 in cells has been reported to increase retinyl ester concentrations [9], suggesting that Dhrs3 acts as a retinal reductase to generate storage forms of retinol ( Figure 3A). As expected, the retinol concentration in cells in HFM was significantly higher than that in cells in NM ( Figure 3B). To explore the role of retinol in the regulation of LD accumulation, the cells were treated with or without retinyl acetate. Under NM conditions, no obvious change in LD accumulation was observed in the retinyl acetate-treated cells ( Figure 3C,D). However, in the HFM, retinyl acetate-treated cells showed higher LD accumulation ( Figure 3C,D). TG content was detected only in cells treated for 24 and 48 h. Notably, as time increased from 6 to 48 h, the cellular TG content

Hif1α Regulates dhrs3a and LD Accumulation
To explore the regulation of dhrs3a in zebrafish, we obtained the 2000 bp promoter sequence of dhrs3a from NCBI and predicted its transcription factor using hTFtarget. Hif1α showed the highest prediction score among all transcriptional factors, followed by P53, Ppar-γ, CCAAT enhancer binding protein α (Cebpα), and Srebp-1 ( Figure 5A). We next investigated whether Hif1α plays a role in controlling dhrs3a and LD accumulation.

Hif1α Regulates dhrs3a and LD Accumulation
To explore the regulation of dhrs3a in zebrafish, we obtained the 2000 bp promoter sequence of dhrs3a from NCBI and predicted its transcription factor using hTFtarget. Hif1α showed the highest prediction score among all transcriptional factors, followed by P53, Ppar-γ, CCAAT enhancer binding protein α (Cebpα), and Srebp-1 ( Figure 5A). We next investigated whether Hif1α plays a role in controlling dhrs3a and LD accumulation. The cells were incubated with an HIF1α inhibitor, LW6, and incubated in a fatty acid-containing medium. As expected, treatment with LW6 reduced LD accumulation in response to the high fatty acid content, especially by reducing the size of LDs ( Figure 5B,C). Moreover, incubation with the fatty acid-rich medium significantly increased the mRNA expression of hif1a, hif1ab, dhrs3a, and pparg, whereas the expression of these genes was suppressed by LW6 ( Figure 5D). Overall, these results suggest that dhrs3a is regulated by Hif1α. The cells were incubated with an HIF1α inhibitor, LW6, and incubated in a fatty acidcontaining medium. As expected, treatment with LW6 reduced LD accumulation in response to the high fatty acid content, especially by reducing the size of LDs ( Figure  5B,C). Moreover, incubation with the fatty acid-rich medium significantly increased the mRNA expression of hif1a, hif1ab, dhrs3a, and pparg, whereas the expression of these genes was suppressed by LW6 ( Figure 5D). Overall, these results suggest that dhrs3a is regulated by Hif1α.

Discussion
Research on the regulation of fat storage, including regulation from the perspectives of nutrition, genetic breeding, and cell culture, has become a popular topic. However, research on LD as an organelle has just begun, and the current research direction is focused on lipid catabolism [16][17][18] and adipocyte differentiation [19,20], whereas scant attention has been paid to the generation of LDs. In this study, we first showed that LD accumulation in ZFL cells is accompanied by an increase in LD protein expression levels, HIF1α inhibitor (LW6) and then transferred to HFM. Cells were stained with BODIPY (LDs, green) or DAPI (nuclei, blue). The LD size in cells was quantitated by ImageJ software (n = 20). Quantitative reverse transcription polymerase chain reaction analysis of the relative mRNA expression of lipid metabolism-related genes (n = 3). Statistical significance between HFM-treated cells and NM-treated cells was denoted with asterisks as follows: * p < 0.05; ** p < 0.01; *** p < 0.001; Statistical significance between HFM + LW6-treated cells and HFM-treated cells was denoted with pounds as follows: # p < 0.05; ## p < 0.01; ### p < 0.001.

Discussion
Research on the regulation of fat storage, including regulation from the perspectives of nutrition, genetic breeding, and cell culture, has become a popular topic. However, research on LD as an organelle has just begun, and the current research direction is focused on lipid catabolism [16][17][18] and adipocyte differentiation [19,20], whereas scant attention has been paid to the generation of LDs. In this study, we first showed that LD accumulation in ZFL cells is accompanied by an increase in LD protein expression levels, which plays a role in LD formation via the generation of retinol, activation of the Ppar-γ pathway, and alteration of the lipid composition. Notably, we also showed that Hif1α is one of the main transcription factors for dhrs3a, regulating LD formation in fish liver cells.
LD formation is a complex biological process involving different steps in which many proteins play a role [6]. In this study, we showed that only 6 h of incubation with fatty acids induced an apparent LD accumulation phenotype in cells. A total of 1786 DEGs were identified, including well-known LD biogenesis marker genes, such as dgat and perilipins (Table S2). Many of these genes are responsible for LD morphology. For example, the gene prostaglandin-endoperoxide synthase 2a, which promotes the formation of prostaglandin 2α (PGF2α), was upregulated (Table S2). Recently, we have shown that PGF2α participates in the degradation of LDs and mitochondrial development in ZFL cells, which suggests that not only were the proteins involved in LD formation induced but LD degradation was also activated [21]. In the present study, we identified seven LD proteins using GO analysis, some of which are important for modulating LD morphology. For example, LDAH is a newly identified LD protein that has recently been shown to promote LD fusion and enhance adipose triglyceride lipase (ATGL) degradation and TG accumulation [22]. LDs are surrounded by a monolayer of phospholipids, mainly PCs, which are located in the LDs, and LDAH is the key enzyme that synthesizes PCs [23]. Notably, RDH10 is a short-chain dehydrogenase essential for retinoic acid biosynthesis and has been reported to be localized to LD, allowing them to serve as sites of retinoid homeostasis [24]. Similarly, DHRS3 is involved in retinal reduction and plays an important role in retinoid metabolism [9,12]. These results suggest that retinoids are important molecules in the regulation of LD homeostasis in fish cells.
In the present study, we focused on the protein Dhrs3, which is encoded by two gene subtypes in zebrafish: dhrs3a and dhrs3b. Dhrs3b did not appear to be sensitive in response to fatty acids in our study; however, it is not known whether it is more sensitive to other exogenous factors. Dhrs3a showed an obvious increase in response to fatty acids, and its expression subsequently decreased, which suggests that dhrs3a may be involved in the early stages of LD biogenesis. Indeed, our study demonstrates that dhrs3a knockdown can partly abolish LD accumulation after short-term incubation with a fatty acid-rich medium, but not after long-term incubation. In HepG2 cells, DHRS3 is an ER protein enriched at the focal points of LD budding, where it localizes to the phospholipid monolayer of ER-derived lipid droplets [12]. In the present study, we found that although Dhrs3 was significantly expressed in LD-containing cells, not all of the protein signals were located around LDs. These results suggest that LDs form in response to fatty acids and that Dhrs3 is involved in this process.
It has been speculated that once retinal is reduced to retinol by DHRS3 in mammals, it can be esterified and stored with long-chain fatty acids [25]. However, this hypothesis has not yet been verified. In the present study, we showed that dhrs3a knockdown decreased the mRNA expression of pparg. Retinoid X receptor (RXR) and PPAR-γ form heterodimers that regulate the transcription of genes involved in insulin action, adipocyte differentiation, lipid metabolism, and inflammation [26]. RXR ligands include naturally occurring retinoic acid and synthetic retinoids [26]. Thus, breaking the RXR/PPAR-γ heterodimers may be one of the reasons why dhrs3a knockdown alters LD accumulation in cells by decreasing RXR ligands. In our study, Dhrs3 production and retinol levels were significantly increased in LD-accumulated cells. Moreover, the retinol derivative, retinyl acetate, maintained the accumulation of LDs and upregulated pparg, as well as its downstream genes, fabp, fatp, and fasn, suggesting that retinyl acetate assists in LD biogenesis in fish cells, during which the activation of the Rxr/Ppar-γ heterodimers may play a vital role.
The structure of LDs is similar in all eukaryotic cells; it consists of a hydrophobic core formed by TAGs and steryl esters, which is surrounded by a phospholipid monolayer [27]. Moreover, this phospholipid monolayer consists of over a hundred different phospholipid molecular species, of which PC is the most abundant and is crucially important for LD stability [28,29]. Thus, it is not surprising that retinyl acetate increased the TAG and PC content in zebrafish cells in the present study, as these cells had higher LD accumulation. Moreover, we observed remarkable changes in the contents of CL, PI, and PS, which were significantly decreased in the retinyl acetate-induced LD-deposited cells. CL is a unique phospholipid that is almost exclusively located in the inner mitochondrial membrane where it is synthesized [30]. Thus, retinyl acetate may disturb mitochondrial homeostasis and lipid oxidation in cells. Moreover, changes in PI and PS may contribute to LD growth. For example, in the absence of LDs, the ER phospholipid composition is altered and displays an increase in PI content, and cells lacking seipin and Pex30 display higher levels of PI [3], suggesting that a decrease in PI may accelerate LD formation in ZFL cells.
Studies in mammals have shown that, although dhrs3 is regulated by P53, P63, RXR, and PPAR-γ, other transcription factors may also regulate the gene [31]. As indicated in this study, dhrs3 is regulated by P53; however, the identification of this transcription factor is based on the fact that dhrs3 was among the DEGs screened using microarray technology after the deletion of p53, which does not indicate that P53 is the only effective regulatory factor for dhrs3 [12]. In our study, the prediction of transcription factors in the 2000 bp promoter region upstream of dhrs3a in zebrafish showed that the score of Hif1α was higher than that of P53, Ppar-γ, Cebpα, Srebp-1, and other traditional lipid-promoting transcription factors. We further demonstrated that inhibition of Hif1α successfully decreased the expression of dhrs3a and LD accumulation in ZFL cells. Although the regulation efficiency of these TFs in the transcription of dhrs3a is unknown, our study demonstrates the role of Hif1α in the regulation of this gene for the first time. It should be noted that the above data demonstrate a correlation between disturbance of dhrs3a expression and its production accompanied by altered pparg expression. However, the findings suggest that Ppar-γ may potentially act as a transcription factor of dhrs3a. It is unclear whether Ppar-γ regulates dhrs3a expression or if there is a reciprocal relationship between Ppar-γ and dhrs3a production. Although our study showed that inhibition of Hif1α decreased both dhrs3a and pparg expression, there is no evidence to suggest that Hif1α directly regulates Ppar-γ. Therefore, further research is needed to clarify the role of Ppar-γ in dhrs3a expression to better understand the mechanism by which Hif1α influences Dhrs3a. HIF1 is closely related to lipid metabolism and is involved in the promotion of lipid accumulation and transport, regulation of fatty acid metabolism, steroid metabolism, TG synthesis, phospholipid metabolism, and LD formation [15,32]. In current aquaculture practices, high densities and fluctuating environmental factors can result in hypoxic stress. Studies have shown increased lipid accumulation in the livers of hypoxic stress-induced fish, such as turbot and golden pompano [33,34]. Therefore, our study provides new insights into the mechanism underlying the pathogenesis of fatty liver in fish in response to hypoxia.
For the LD formation experiment, the cells were incubated either with NM or with high-fat medium (HFM) containing 300 µM bovine serum albumin (BSA)-coated oleic acid (Sigma-Aldrich) for 6 h. The BSA-coated oleic acid is obtained by adding the oleic acid dissolved in ethanol dropwise to the BSA solution to obtain a homogenous mixture. For the retinyl acetate treatment experiment, 10 µM retinyl acetate (MedChemExpress, Monmouth Junction, NJ, USA) was added to cells cultured in NM or HFM, followed by incubation for 6 h.

LD Staining
For LD staining, ZFL cells were seeded in 6-well plates at a density of 1.2 × 10 6 cells per well and allowed to grow for 24 h, with three replicates per assay. LDs and nuclei were stained with BODIPY™ 493/503 (Invitrogen) and DAPI (Invitrogen), respectively. The methods used for fluorescence staining, image acquisition, and LD quantification were based on a previous study [21]. Briefly, the cells were fixed in 10% formalin, stained with BODIPY for 30 min and DAPI for 10 min, and imaged using an inverted fluorescence microscope with a ×40 objective (Nikon TS2, Tokyo, Japan). BODIPY puncta were quantified using ImageJ software (version 1.53q; National Institutes of Health, Bethesda, MD, USA).

TG and Retinol Content Assay
Cells were seeded in 6-well plates at a density of 1.2 × 10 6 cells/well. Three replicates were used for each treatment group. After treatment, the cells were collected by trypsinization. The TG content was assayed using an enzymatic kit (Applygen, Beijing, China). The retinol content was determined using an ELISA kit (CED051Ge; Cloud-Clone Co., Houston, TX, USA).

Immunofluorescence Staining
Immunofluorescence staining was performed as previously described [21]. Briefly, the cells were seeded in 6-well plates at a density of 1.2 × 10 6 cells/well and incubated for 24 h. After treatment, the cells were fixed with 4% paraformaldehyde and incubated with Triton X-100 (0.5%; Leagene, Beijing, China). Bovine serum albumin (5%) was then added for blocking. The cells were then incubated overnight with a primary antibody against DHRS3 (ab236603; Abcam, Cambridge, UK; diluted at a concentration of 1:1000 in fetal bovine serum) at 4 • C, and subsequently with a secondary antibody conjugated to Alexa Fluor 488 (Cell Signaling Technology, Danvers, MA, USA) for 2 h. The LDs and nuclei were stained, and the cells were photographed.

Transcriptome Analysis
Cells were seeded in 6-well plates at a density of 1.2 × 10 6 cells/well and treated with NM or HFM in three replicates. The cells were then collected using the RNA TRIzol reagent (Life Technologies Inc., Thermo Fisher Scientific) and sent to Novogene Biotechnology (Beijing, China) for transcriptome analysis. The methods for RNA extraction, RNA quality assessment, transcriptome library preparation, Illumina sequencing, transcriptome assembly, identification of DEGs, and annotation can be found in a previous report [35]. The DEGs were selected based on the following criteria: |log2 ratio| > 1 and padj < 0.05.

qRT-PCR
For qRT-PCR analysis, cells were seeded in 6-well plates at a density of 1.2 × 10 6 cells/well in three replicates. qRT-PCR was performed as described in a previous report [21]. Relative gene expression was calculated using the comparative CT method (2 −∆∆Ct ) [36,37]. We used β-actin as the reference gene. The sequences of the primers used for analysis are listed in Supplementary Table S3.

Lipidomic Analysis
Cells were inoculated in 25 cm 2 plastic bottles at a density of 1.0 × 10 6 cells/well in 5 replicates, this density can result in a range of cell density of 70-80%. After treatment, precooled phosphate-buffered saline was used to wash the cells, and a 60% aqueous methanol solution (chromatographic grade) was added to collect the cells. The cell suspension was placed in a glass centrifuge tube with a Teflon-lined cap, precooled methanol was added, and the mixture was vortexed. Next, precooled methyl tert-butyl ether was added and the mixture was incubated at room temperature (~25 • C) in a shaker for 1 h. Mass spectroscopy-grade water was added to the mixture, and organic phases were layered; the mixture was incubated at room temperature for 10 min, and centrifuged (1000× g, 10 min). The upper organic phase (MTBE) was collected, and its nitrogen concentration was determined using a nitrogen-blowing apparatus. Analysis of the redissolved isopropyl alcohol was performed using liquid chromatography with a tandem mass spectrometry (LC-MS/MS) system (Thermo Fisher Scientific). An equal amount of supernatant was mixed from each processed sample to prepare a quality control sample.
Raw data generated using ultra-high-performance LC-MS/MS were imported into Compound Discoverer 3.01 (CD3.1; Thermo Fisher Scientific). For accurate identification, peak alignment was performed for different samples according to a retention time deviation of 0.2 min and mass deviation of 5 ppm, and the peak area was quantified. The molecular formula was predicted based on molecular ion peaks and fragment ions and compared with the Lipidmaps, Lipidblast, and HMDB databases. The statistical software R (R version r-3.4.3), Python (Python 2.7.6), and CentOS (CentOS release 6.6) were used for the statistical analyses. Blank samples were used to remove background ions, quantitative results were normalized, and the lipid data were qualitatively and quantitatively analyzed.
PCA was performed using MetaX, a flexible and comprehensive metabolomic data processing software. The differentially regulated lipid compounds were set based on the following criteria: VIP > 1, p < 0.05, fold change |FC| ≥ 2, or ≤0.5. The metabolites of interest were screened using volcanic maps with Log2(FC) and −log10(p-value) of the metabolite content. The clustering heatmap was normalized by z-scores in the differential metabolite intensity region and drawn using the Pheatmap package in R.

Statistical Analysis
All data are expressed as mean ± SD (standard deviation). Differences between NM-and HFM-treated cells or control and retinyl acetate-treated cells were determined using an independent sample t-test. Two-way analysis of variance (ANOVA), followed by Bonferroni's post hoc test, was used to compare differences between experimental treatments. All analyses were performed using the PASW Statistics 18 software (SPSS, Chicago, IL, USA).

Conclusions
In the present study, we propose a new mechanism for lipid formation in fish cells in which Dhrs3a is the main LD protein that acts on LD growth at a relatively early stage. The metabolite, retinol, may contribute to this process by activating Ppar-γ and decreasing the lipid classes CL, PI, and PS. We also show that Hif1α is one of the main transcription factors regulating dhrs3a. Further studies should focus on the comparison of Hif1α and other TFs in the regulation of dhrs3a and on validating this LD formation pathway in vivo.