Co-regulation of hepatic steatosis by ferritinophagy and unsaturated fatty acid supply.

Both iron overload and iron deficiency have been reported in obesity and metabolic syndromes. Due to the presence of multiple intracellular iron pools and the dynamic nature of iron mobilization and use, the actual status and contribution of free and metabolically active iron toward metabolic syndrome remain to be established. The discovery of nuclear receptor coactivator 4 (NCOA4) as a ferritinophagy receptor provides an opening to address the connection between iron and metabolic diseases. This study aims to specifically dissect the role of hepatic ferritinophagy in lipid metabolism and hepatic steatosis. We conducted a series of Ncoa4 gain- and loss-of-function experiments to examine how ferritinophagy affects lipid metabolism through phenotypic and lipidomic analyses both in vitro and in vivo. We show that ferritinophagy is required to release iron from ferritin cages for biological use, and is induced by lipid loading in vitro and during the development of obesity in vivo. Ncoa4 knockdown impairs mitochondrial morphology and reduces palmitate-induced lipid droplet formation in cultured cells and the development of hepatic steatosis in obese mice models. Importantly, the effect of Ncoa4 deficiency on mitochondrial morphology and lipid accumulation is specifically linked to lipidomic reductions in unsaturated fatty acid content in triglycerides and cardiolipins, and an external supply of unsaturated fatty acids reverses these phenotypes.
Conclusion: This study shows that ferritinophagy-derived iron supports fatty acid desaturation and the synthesis of unsaturated fatty acid-rich lipids to reduce lipotoxicity. However, the continuous activation of ferritinophagy contributes to the development of hepatic steatosis and liver damage in obesity.

INTRODUCTION

Iron plays an essential role in many cellular processes as a prosthetic factor, including oxygen transport, mitochondrial respiration, fatty acid desaturation, and DNA biosynthesis.[ , ] However, the oxidative potential of free iron not bound to proteins is also highly detrimental through Fenton reaction and the formation of reactive oxygen species.[ ] Therefore, iron homeostasis is carefully managed in the body.[ , ]

Perturbation in iron homeostasis is often observed in patients with obesity, insulin resistance, or nonalcoholic fatty liver disease (NAFLD), collectively referred to as the dysmetabolic iron overload syndrome (DIOS).[ , ] Oxidative stress is one of the primary mechanisms linking iron overload with NAFLD.[ ] Not only does oxidative stress increase steatosis by causing apolipoprotein B100 degradation and impairing very‐low density lipoprotein secretion,[ , ] but it also can aggravate steatosis by activating inflammatory pathways and cell death.[ , , ]. Iron overloading may also directly activate AKT signaling, up‐regulate fatty acid and cholesterol synthesis, cause endoplasmic reticulum stress, and activate hepatic macrophage and stellate cells. As a result, iron overload may accentuate liver damage and promote the development of hepatic steatosis toward nonalcoholic steatohepatitis (NASH).[ , , , ]

However, iron overload in obese patients and patients with NAFLD is modest and often falls within the high end of the normal range.[ , ] Therefore, it is unclear whether this modest iron overload may have a direct, causal role in aggravating hepatic steatosis in vivo. Moreover, the assessment of iron status in vivo varies depending on the kind of marker used. For example, transferrin saturation decreases as body mass index increases, indicative of a state of iron deficiency,[ , ] and it may be caused by the increased blood volume in obese/overweight individuals.[ , ] In contrast, the serum hepcidin concentration, an indicator of systemic iron overload, is elevated in the obese populations.[ , ] Discrepancies among iron‐reported gene levels suggest the involvement of additional factors (e.g., inflammation) in regulating their expression.[ , , ] Thus, the actual status of free and metabolically available iron in obese and NAFLD animal models and patients warrants further study.[ ]

Most of the intracellular iron pool is stored in ferritin cages and is released in the lysosome following intracellular iron starvation, to support the synthesis of heme and iron–sulfur clusters.[ , , ] Nuclear receptor coactivator 4 (NCOA4) is the cargo receptor mediating autophagic turnover of ferritin (i.e., ferritinophagy).[ , ] Cell and animal models of NCOA4 deficiency confirm iron mobilization through ferritinophagy being essential for mitochondrial function, erythropoiesis, and even cell death.[ , , , ] Moreover, NCOA4‐driven ferritinophagy responds to intracellular iron availability to promote iron homeostasis; specifically, the ubiquitin E3 ligase HERC2 (HECT and RLD domain containing E3 ubiquitin protein ligase 2) binds to NCOA4 in an iron‐dependent manner to drive its proteasomal degradation and fine‐tune iron mobilization.[ ]

Here we report the identification of a ferritinophagy–fatty acid desaturation axis that governs iron and lipid homeostasis.

EXPERIMENTAL PROCEDURES

Animal studies

All animal protocols were approved by the Institutional Animal Care and Use Committee of Tsinghua University (Beijing, China). All mice were maintained on a 12‐h/12‐h light/dark cycle in a pathogen‐free barrier facility at Tsinghua Laboratory Animal Research Center at controlled temperature (23 ± 2°C) with free access to water and food. Male mice aged 8–12 weeks (except the high‐fat diet–fed mice) were used in respective experiments.

The Ncoa4 mouse model was generated by Cyagen with LoxP sites flanking the exon 2–6 of the mouse Ncoa4 gene. Hepatocyte‐specific Ncoa4 knockout (KO) mice were generated by crossing Ncoa4 mice with Albumin‐Cre transgenic mice (B6.Cg‐Tg[Alb‐Cre]21Mgn/J; The Jackson Laboratory). For the high‐fat diet (HFD)–induced fatty liver mouse model, 4‐week‐old wild‐type (WT) C57BL/6J mice were purchased from the Vital River Laboratory Animal Technology Co. and placed on HFD (D12492; 60 kcal% fat; Research Diets) for 16 weeks. Male leptin‐deficient (ob/ob) mice were purchased from Hua Fukang Biotechnology Co. All the virus administrations in vivo were carried out by tail‐vein injection. Ncoa4 knockdown in mice was generated by adenovirus‐driven expression of targeting short hairpin RNA. HFD and ob/ob mice with Ncoa4 knockdown were sacrificed after 6 h of food withdrawal (9:00 a.m.–3:00 p.m.) 10 days following virus administration. Liver‐specific green fluorescent protein (GFP) and NCOA4‐FLAG overexpression in WT mice were achieved by an intravenous injection of adeno‐associated virus (AAV 2/8; 1 × 10 11 genome copies per mouse).

Primary hepatocytes isolation and culture

Primary hepatocytes were isolated from 8–10‐week‐old male mice. Briefly, mice were anesthetized with 250 mg/kg avertin, and the liver was perfused through the portal vein with 50 ml of prewarm Hank's balanced salt solution (HBSS) buffer (37°C) containing 0.25 mm ethylene glycol tetraacetic acid at a speed of 11 ml/min. The liver was then digested with 5–10 mg Collagenase IV (C5138; Sigma) prepared in an HBSS buffer supplemented with 5 mm CaCl2. The liver was then excised, minced, and filtered through a 100‐μm steel mesh. After washing twice with Medium 199 (MACGENE), primary hepatocytes were seeded into 0.1% gelatin‐coated and air‐dried 12‐well plates at a density of 2 × 105 cells/well in BFM199 medium (0.2% bovine serum albumin [BSA], 2% fetal bovine serum [FBS], and 1% penicillin/streptomycin) for 4 h at 37°C with 5% CO2. The culture was then washed and replaced with fresh Medium 199 supplemented with 1 mg/ml Primocin (ant‐pm‐2; InvivoGen) and then cultured for designated analyses.

Cell lines

HepG2 and HEK293FT cells were obtained from ATCC. HEK293A was purchased from Thermo Fisher Scientific. Cells were cultured in Dulbecco's modified Eagle's medium containing 10% FBS at 37°C in a 5% CO2 humidified atmosphere.

Statistical analysis

Data are presented as mean ± SEM. Statistical significance was evaluated by unpaired Student's t test or one‐way analysis of variance followed by Bonferroni or Tukey post hoc test. A value of p < 0.05 was regarded as statistically significant.

Additional methods

Details of additional assays are described in the Supporting Information.

RESULTS

Hepatic ferritinophagy ensures physiological iron sufficiency

Apart from being the primary site for storage of excess body iron in ferritin cages, the liver has a high metabolic requirement for iron as a prosthetic factor.[ ] The identification of NCOA4 as the receptor for lysosomal ferritin degradation[ ] made it possible to examine how iron mobilization rather than deposit in ferritin cages may regulate metabolism and affect the pathogenesis of the metabolic syndrome.

We first created a hepatocyte‐specific Ncoa4 deletion model (Ncoa4 ) to assess the role of ferritinophagy on local and systemic iron homeostasis (Figure 1A). Hepatocyte Ncoa4 deletion resulted in near‐complete ablation of Ncoa4 transcript and protein expression in the liver, but not in other tissues (Figure 1A and Figure S1A). In the absence of NCOA4‐mediated lysosomal degradation, ferritin accumulated in Ncoa4 liver tissues (Figure 1A). Hepatic non‐heme iron is primarily stored in ferritin cages, and it increased in the KO liver as the result of a blunted ferritinophagy (Figure 1B). Perl's Prussian blue staining also confirmed higher iron retention in Ncoa4 liver (Figure S1B). Therefore, ferritinophagy is actively engaged in the turnover of ferritin and ferritin‐stored irons under normal physiological conditions.

FIGURE 1

Hepatic ferritinophagy ensures physiological iron sufficiency. (A) Schematic illustration of the generation of hepatocyte‐specific Ncoa4 knockout (HepKO) mice, real time‐polymerase chain reaction (PCR) measurements of hepatic nuclear receptor coactivator 4 (Ncoa4) transcript levels, and immunoblot measurements of hepatic NCOA4 and ferritin heavy chain 1 (FTH1) protein levels in 10‐week‐old Ncoa4 and Ncoa4‐HepKO mice fed a normal chow diet (n = 6–7 mice/group). (B,C) Measurement of liver non‐heme iron levels (B), and real time PCR measurement of hepatic Hepcidin (Hamp1) transcript levels (C) in 10‐week‐old Ncoa4 and HepKO mice fed a normal chow diet (n = 6–7 mice/group). (D) Measurements of serum total iron‐binding capacity (TIBC) and serum iron (SI), and the calculated levels of unsaturated iron‐binding capacity (UIBC, TIBC‐SI) and transferrin saturation (Tf.Sat, SI/TIBC) in 10‐week‐old Ncoa4 and HepKO mice fed a normal chow diet (n = 7 mice/group). (E) Volcano diagram display of liver differential expression genes (DEGs) between Ncoa4 HepKO and Ncoa4 mice (n = 4 mice/group). (F) Gene‐set enrichment analysis (GSEA) enrichment score curve of genes involved in heme and ferrous iron binding in Ncoa4 HepKO mice. Benjamini and Hochberg (BH)–corrected p‐values are labeled on the bottom (n = 4 mice/group). (G–L) Schematic illustration of liver‐specific knockdown of Ncoa4 in C57BL/6J wild‐type (WT) mice, measurements of hepatic Ncoa4 transcript levels, and hepatic NCOA4 and FTH1 protein levels (G); liver non‐heme iron levels (H); hepatic Hamp1 transcript levels (I); serum TIBC, UIBC, SI, and Tf.sat (J); volcano diagram display of liver DEGs (K); and GSEA enrichment analysis of heme and ferrous iron binding (L) in shLacZ and shNcoa4 mice (n = 4–8 mice/group). (M–P) Schematic illustration of liver‐specific overexpression of Ncoa4 in C57BL/6J WT mice, measurements of hepatic Ncoa4 transcript levels, and hepatic NCOA4 and FTH1 protein levels (M); liver non‐heme iron levels (N); hepatic Hamp1 transcript levels (O); serum TIBC, UIBC, SI, and Tf.sat (P) in WT mice expressing adeno‐associated virus 8 (AAV8)–green fluorescent protein (GFP) or Ncoa4‐Flag (n = 6 mice/group). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Student's t test (see also Figure S1). mRNA, messenger RNA.

The lack of ferritinophagy‐derived iron was sensed by the hepatic iron homeostasis system. Hepcidin (Hamp1) is a hepatic hormone that inhibits the export of iron from the duodenum and spleen into the circulatory system under hepatic iron overload conditions. Its expression was down‐regulated in the Ncoa4‐deleted liver (Figure 1C). The liver also secretes transferrin into the circulation to transport iron from the periphery back to the liver. Circulating transferrin level measured as plasma iron‐binding capacity increased in the Ncoa4 mice (Figure 1D).

The coordinated suppression of hepcidin expression and increase in circulatory iron transport masked the physiological importance of hepatic ferritinophagy. Transcriptome analysis revealed limited changes in the expression of iron‐binding and heme‐binding genes (Figure 1E,F and Figure S1J). Moreover, the Ncoa4 mice showed no gross defect in overall birth rate, body weight, plasma glucose levels, lipid profiles, spleen iron content, serum ferritin levels, and blood count (Figure S1C–I).

We then opted for an acute Ncoa4 liver‐specific KD mouse model to study the functional role of hepatic ferritinophagy (Figure 1G,H and Figure S1K). In the shNcoa4 mice liver, hepcidin gene expression was not down‐regulated (Figure 1I), and serum iron‐binding capacity did not increase (Figure 1J). There was also no significant difference in serum ferritin levels in control versus shNcoa4 mice (Figure S1L). In the absence of a compensatory regulation in hepatic iron uptake, ferritinophagy deficiency caused a broad dysregulation of the transcriptome (Figure 1K). Gene ontology analysis showed that the most significant down‐regulated functions in the shNcoa4 liver included iron‐binding and heme‐binding, monooxygenase, and oxidoreductase activities (Figure 1L and Figure S1J), indicative of functional iron deficiency.

We further created a gain‐of‐function model through AAV‐mediated expression of NCOA4 (Ncoa4 ) in mice liver (Figure 1M). In contrast to Ncoa4 deficiency, NCOA4 overexpression reduced hepatic ferritin levels (Figure 1M), and hepatic iron content decreased by about one third (Figure 1N), although Perl's staining was too weak to show the decrease accordingly (Figure S1M). In response to increased iron release through ferritinophagy, hepcidin gene expression in the Ncoa4 liver trended toward an increase (Figure 1O), and plasma iron‐binding capacity was reduced significantly (Figure 1P). The increase in transferrin saturation (Tf.Sat) might be attributed to high hepatic ferritinophagy and iron efflux (Figure 1P). Again, no obvious differences were observed for serum ferritin levels (Figure S1N).

Obesity up‐regulates ferritinophagy to promote hepatic steatosis

After demonstrating the functional relevance of hepatic ferritinophagy in regulating iron homeostasis, we went on to ask how ferritinophagy and iron mobilization may be controlled and contribute to the development of obesity and hepatic steatosis.

Transcript measurements showed that the expression of Ncoa4 was induced in the liver tissues of both the HFD‐fed and genetically induced (ob/ob) obese mice (Figure 2A and Figure S2A). Ncoa4 up‐regulation was accompanied by increased NCOA4 protein levels and NCOA4– ferritin heavy chain 1 (FTH1) interaction (Figure 2A,B and Figure S2A,B). Notably, the amount of ferritin pulldown increased in obese samples despite lower levels of NCOA4 in the precipitate (Figure 2B and Figure S2B). Consequently, hepatic iron deposits decreased in the obese mice (Figure 2C and Figure S2C). Intriguingly, the increase in hepatic ferritinophagy in the obese liver did not cause the iron overload response observed in lean, Ncoa4 mice. Instead, Hamp1 expression decreased (Figure 2D and Figure S2D).

FIGURE 2

Ferritinophagy is induced in obese mice liver to promote triglyceride storage. (A–D) Real‐time PCR measurements of Ncoa4 transcript levels and immunoblot measurements of NCOA4 and FTH1 protein levels (A); co‐immunoprecipitation (co‐IP) measurements and quantification of NCOA4‐FTH1 interactions (B); measurements of liver non‐heme iron levels (C); and real‐time PCR measurements of Hamp1 transcript levels (D) in the liver lysates of high‐fat diet (HFD; 4 months)–induced obese mice (n = 6 mice/group). Each lane of immunoblots in (A) and (B) are the mixture of two or six samples, respectively. (E) Schematic illustration of experimental design for Ncoa4 knockdown in HFD mice. (F–H) Immunoblot analysis of NCOA4, FTH1, and de novo lipogenesis enzyme protein levels (F); quantitative PCR measurements of Ncoa4 transcript levels (G); and quantification of the non‐heme iron levels (H) in the liver tissues of shLacZ and shNcoa4 HFD mice (n = 7–8 mice/group). (I–M) Hematoxylin and eosin (H&E) staining (I), liver triglyceride (TG) content (J), liver weight–to–body weight ratio (K), serum alanine aminotransferase (ALT) levels (L), and blood glucose levels (M) of shLacZ and shNcoa4 HFD mice (n = 7–8 mice/group; scale bar, 100 μm). (N–Q) Mass spectrometry quantification of the relative abundance of total cardiolipin (CL), phosphatidylcholine (PC), and phosphatidylethanolamine (PE) levels (N); the fatty acid compositions of CL, PC, and PE (O); calculated indices of fatty acid saturation (C16:0/C18:1) for CL, PC, and PE (P); and the measured C16:0‐containing or C18:2‐only CLs (Q) in the liver tissues of Ncoa4‐deficient HFD mice (n = 5 mice/group). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Student's t test (see also Figure S2). MUFA, monounsaturated fatty acid; NCD, normal‐chow diet; n.s., not significant; PUFA, polyunsaturated fatty acid; SFA, saturated fatty acid.

We conducted Ncoa4 knockdown experiments to understand the potential role of hepatic ferritinophagy up‐regulation in obese mice (Figure 2E–G and Figure S2E–G). Adenovirus‐mediated suppression of Ncoa4 expression on transcript and protein levels resulted in hepatic ferritin and non‐heme iron accumulation in both the HFD and ob/ob mice models (Figure 2F,H and Figure S2F,H). However, despite previously reported association between hepatic iron overload and hepatic steatosis, hepatic lipogenic enzyme levels were reduced in the iron‐accumulated, obese mice liver (Figure 2F and Figure S2F). Histology analysis and triglyceride measurement showed that hepatic steatosis in the obese mice was improved (Figure 2I,J and Figure S2I,J), and the liver weight was reduced (Figure 2K and Figure S2K). Hyperglycemia and indicators of liver damage also improved (Figure 2L,M and Figure S2L,M).

Down‐regulation in lipogenesis likely reduced hepatic steatosis in the shNcoa4 obese mice liver. Inefficient fatty acid desaturation might also contribute to such a phenotype, because (1) triglyceride synthesis in mice usually requires two molecules of unsaturated fatty acid and one molecule of saturated fat, and (2) fatty acid desaturation by the stearate‐CoA desaturase requires iron as a cofactor.[ , ] We performed lipidomic analysis for the shNcoa4 obese mice liver to explore this hypothesis.

On the lipid class level, improvement of hepatic steatosis in the shNcoa4 obese mice liver was accompanied with reductions in phosphatidylcholine (PC) content and increases in phosphatidylethanolamine (PE) content (Figure 2N), consistent with our previous observation that obesity increases the PC/PE ratio.[ ] On the fatty acid level, its composition varied greatly across different lipid species (Figure 2O). For example, cardiolipin (CL) was consisted primarily of linoleic acid (C18:2), a polyunsaturated fatty acid. In contrast, over 40% of the fatty acids in phosphatidylcholine were saturated (Figure 2O).

We calculated the ratio of palmitate over oleate as an indicator of fatty acid saturation, and it increased across all three phospholipids (Figure 2P). The increase was most prominent in CL, possibly due to its low baseline fatty acid saturation (Figure 2O). Detailed analysis showed that all palmitate‐containing cardiolipin species increased by more than 2‐fold, whereas linoleate‐only species decreased significantly (Figure 2Q).

Fatty acid desaturation mechanistically connects ferritinophagy to triglyceride storage

Work by the Schaffer group established that palmitate is preferentially incorporated into membrane phospholipids, and desaturation of palmitate by the stearate‐CoA desaturase (SCD1) quantitatively promotes triglyceride synthesis.[ ] Therefore, we reasoned that the improvement in hepatic steatosis and increase in palmitate content in the shNcoa4 obese mice lipidome might be attributed to defect in fatty acid desaturation. We used palmitate‐induced and oleate‐induced triglyceride synthesis system to elucidate the potential connection between ferritinophagy and fatty acid desaturation in hepatocyte cultures.

We first examined the engagement of ferritinophagy in cell culture. Similar to our observations in mouse liver, silencing the expression of the human NCOA4 gene in HepG2 cells resulted in ferritin and iron accumulation in vitro (Figure 3A,B). Additionally, NCOA4‐deficient HepG2 cells presented a series of iron‐deficiency hallmarks that include the induction of transferrin receptor and the down‐regulation of ferroportin 1 (Figure 3C), up‐regulation of iron‐responsive element‐binding protein 2 and binding to iron‐responsive element (Figure 3D), a reduction in the iron‐sensitive E3‐ubiquitin ligase FBXL5 (F‐box and leucine rich repeat protein 5) protein levels (Figure 3D), and a decrease in aconitase activity (Figure 3E). The mitochondria also became fragmented and swollen (Figure 3F), another sign of iron deficiency in siNCOA4 cells.

FIGURE 3

Knockdown of NCOA4 in HepG2 hepatocytes results in iron starvation. (A) Immunoblot measurements of NCOA4 and FTH1 protein levels in HepG2 cells transfected with either scrambled small interfering RNA (siRNA) or NCOA4 siRNA for 48 h. (B,C) Quantification of total non‐heme iron level (B), and real time‐PCR measurements of NCOA4, transferrin receptor (TFRC), and ferroportin 1 (FPN1) transcript levels (C) in siNCOA4 (48 h) or control HepG2 cells. (D) Immunoblot measurements and quantification of iron‐responsive element‐binding protein (IRP)/iron‐responsive element (IRE) binding, IRPs, F‐box and leucine rich repeat protein 5 (FBXL5) protein levels in siNCOA4 (48 h) or control HepG2 cells. HepG2 cells treated with deferoxamine mesylate (DFO; 100 μm) or ferric ammonium citrate (FAC; 25 μg/ml) for 24 h were lysed as the positive control. (E) Measurements of cellular aconitase activity in siNCOA4 (48 h) or control HepG2 cells. (F) Confocal imaging of mitochondrial morphology in siNCOA4 (48 h) or control HepG2 cells with MitoTracker Red (200 nm) staining. More than 100 cells were quantified for tubular or fragmented mitochondria. Scale bar, 20 μm. Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001,****p < 0.0001; Student's t test. NT, no treatment.

After confirming the presence and participation of ferritinophagy in regulating iron homeostasis in vitro, we examined how NCOA4 deficiency might affect saturated fat (palmitate)–induced and unsaturated fat (oleate)–induced lipid droplet formation differently.

As shown in Figure 4A, culturing HepG2 cells with 0.5 mm of palmitate increased intracellular lipid droplet formation, and knockdown of NCOA4 reduced lipid droplet formation by almost half (Figure 4A). The reduction in lipid droplet formation was not caused by reduced palmitate uptake based on the transcript and lipidomic analyses. The expression of fatty acid transporter clusters of differentiation 36 (CD36) and the lipidomic level of palmitate both increased in siNCOA4 cells (Figure 4B and Figure S3A,B). Instead, palmitate desaturation was specifically impaired in siNCOA4 cells. As shown in Figure 4B and Figure S3B, the addition of palmitate (C16:0) to the medium caused a modest lipidomic increase in palmitate levels and a much larger increase in its desaturation product, palmitoleate, in the control cells (Figure 4B and Figure S3B). In contrast, the addition of palmitate to the siNCOA4 cells caused a drastic increase in palmitate and its elongation product stearate, but not their desaturated product, palmitoleate and oleate (Figure 4B and Figure S3B). Consequently, the level of fatty acid saturation calculated by the ratio of (C16:0 + C18:0)/(C16:1 + C18:1) increased across all lipid species (Figure S3C,D).

FIGURE 4

NCOA4 deficiency impairs fatty acid desaturation and PA‐induced triglyceride accumulation. (A) BODIPY 493/503 staining and quantification of lipid droplets in control or siNCOA4 HepG2 cells. Cells were transfected with scrambled or NCOA4 siRNA for 48 h, followed by 0.5 mm PA culture for 12 h. More than 100 cells were analyzed for the mean intensity of BODIPY signaling. Scale bar, 20 μm. (B) Heatmap display of perturbations of the fatty acid compositions of main lipid species in HepG2 cells treated as in (A). (C) Mass spectrometry quantification of the fatty acid compositions of diacylglycerol (DG), triglyceride (TG), CL, and phosphatidylglycerol (PG) in control (siNC) HepG2 cells cultured in the absence or presence of 0.5 mm PA for 12 h. (D) Mass spectrometry quantification of the relative abundance of DG, TG, CL, and PG in control and siNCOA4 (48 h) HepG2 cells cultured in the absence or presence of 0.5 mm PA for 12 h. (E) BODIPY 493/503 staining and quantification of lipid droplets in control or siNCOA4 HepG2 cells. Cells were transfected with either scrambled or NCOA4 siRNA for 48 h, followed by palmitate (PA; 0.5 mm), oleate (OA; 0.5 mm), or PA/OA (0.5 mm/0.5 mm) challenge for 12 h. Scale bar, 20 μm. Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001; Student's t test (see also Figure S3). DAPI, 4′,6‐diamidino‐2‐phenylindole; PI, phosphatidylinositol; TG, triglyceride.

The increase in fatty acid saturation was highest among diacylglycerides and phosphatidylglycerols (Figure S3D), lipid species that contain the highest levels of oleic acid (C18:1) (Figure S3B). As the HepG2 lipidome contains a limited amount of polyunsaturated fatty acid (Figure 4C), the limitation in oleate production shall limit the conversion of diacylglycerides to triglycerides and phosphatidylglycerol to cardiolipin. Consistent with this hypothesis, diacylglycerides and phosphatidylglycerols accumulated in siNCOA4 cells (Figure 4D). At the same time, a significant decrease in cardiolipin content was detected (Figure 4D). However, the reduction in triglyceride content was not evident in mass spectrometry analysis (Figure 4D), possibly due to signal saturation.

Oleate culture experiments provided further support for such a hypothesis. As shown in Figure S3E, siNCOA4 did not reduce oleate‐induced lipid droplet formation in the HepG2 cells (Figure S3E). There was even a trend toward increased lipid droplet formation (Figure S3E), possibly attributable to CD36 up‐regulation. Additionally, co‐treatment of palmitate with oleate rescued the lipid droplet formation defect in the siNCOA4 cells (Figure 4E). Therefore, NCOA4 deficiency specifically reduces fatty acid desaturation and impairs saturated fatty acid–induced lipid droplet formation.

Saturated fatty acid stabilizes NCOA4 to activate ferritinophagy and reduce cell death

Triglyceride synthesis is a major protective mechanism in response to palmitate exposure.[ ] Therefore, we asked whether cells specifically invoke ferritinophagy to protect them from palmitate toxicity through enhanced fatty acid desaturation and triglyceride synthesis.

Consistent with this view, palmitate, but not oleate, treatment of HepG2 cells induced the accumulation of the NCOA4 protein in a dose‐dependent manner (Figure 5A). Importantly, NCOA4 protein accumulation in palmitate‐treated cells was accompanied by a down‐regulation of intracellular ferritin levels (Figure 5A), indicative of ferritinophagy activation. Furthermore, despite lower ferritin levels, considerably more ferritin was bound to NCOA4 in palmitate‐treated cells than in control (Figure 5B).

FIGURE 5

Palmitate activates ferritinophagy to reduce lipotoxicity. (A) Immunoblot measurements and quantification of NCOA4 and FTH protein levels in HepG2 cells treated with bovine serum albumin (BSA)–conjugated PA or OA for the indicated dose for 12 h. (B) Co‐IP measurements and quantification of NCOA4‐FTH1 interactions in HepG2 cells treated with PA for the indicated dose for 12 h. Control immunoglobulin G (IgG) immunoprecipitation (IP) lysate is the mixture of BSA and PA‐treated samples. (C,D) Immunoblot measurements and quantification of NCOA4 and FTH protein levels (C) and co‐IP measurements and quantification of NCOA4‐FTH1 interactions (D) in mouse primary hepatocytes treated with PA for the indicated dose for 12 h. (E) Confocal imaging of the subcellular localization of exogenously expressed FTH1‐mCherry and lysosomal‐associated membrane protein 1 (LAMP1)–enhanced GFP (eGFP) in primary hepatocytes. Primary hepatocytes were infected with adenovirus‐expressing mouse FTH1 and LAMP1 fusion proteins for 4 h, followed by treatment with 0.5 mm PA or BSA vehicle for 12 h. Scale bar, 5 μm. (F) Schematic illustration of experimental design for PA‐induced lipotoxicity test in HepG2 cells. (G) Measurement of cell death by propidium iodide (PI) staining. HepG2 cells were transfected with either scrambled siRNA or NCOA4 siRNA for 48 h, followed by 0.5 mm PA or BSA vehicle challenge for 12 h, and then cells were trypsinized, stained, and analyzed with flow cytometry. (H) Measurement of cell death by Hoechst 33342 staining. HepG2 cells were treated as in (G); cells deaths were shown by microscopy and results were quantified and normalized to siNC‐BSA treatment (control). *Denotes statistical significance in comparison with control. Scale bar, 50 μm. (I) Measurement of cell death by detecting the level of lactate dehydrogenase (LDH) released into the culture medium. HepG2 cells were treated as in (G), and culture medium was collected for LDH analysis. Results were normalized to siNC‐BSA treatment (control). *Denotes statistical significance in comparison with control. Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ### p < 0.001, #### p < 0.0001, Student's t test (see also Figure S4).

Palmitate induced NCOA4 accumulation and ferritin‐NCOA4 binding in primary hepatocytes to a similar degree to HepG2 cells (Figure 5C,D). However, no reduction of ferritin levels was observed in palmitate‐treated primary hepatocytes (Figure 5C,D). The lack of lysosomal ferritin degradation in isolated primary hepatocytes allowed us to monitor palmitate‐induced ferritinophagy by demonstrating ferritin recruitment to lysosomal compartments. To this end, we co‐expressed the lysosome marker lysosomal‐associated membrane protein 1 (LAMP1)–GFP with FTH1‐mCherry in primary hepatocytes and examined their localization under normal and palmitate treatment conditions. As shown in Figure 5E, FTH1‐mCherry fluorescence signals were normally dispersed intracellularly in the control cells, and it showed little overlap with the lysosome marker LAMP1‐GFP. However, following the treatment of palmitate, FTH1‐mCherry was co‐localized with LAMP1‐GFP in puncta (Figure 5E), supporting the induction of ferritinophagy.

Unlike obesity‐induced NCOA4 expression in vivo, the elevation of NCOA4 protein levels in the palmitate‐treated HepG2 cells in vitro was not caused by transcriptional regulation (Figure S4A). Both NCOA4 and FTH1 were constitutively degraded through the lysosome and the proteasome, as treatment of cells with the proteasomal inhibitor MG132, the lysosome inhibitor chloroquine, and the autophagy inhibitor bafilomycin A1 all increased their protein levels (Figure S4B). Importantly, because co‐treatment of HepG2 cells with MG132 and palmitate did not increase NCOA4 protein levels beyond palmitate treatment alone (Figure S4B), palmitate might stabilize NCOA4 protein through the proteasomal pathway. Proteasomal degradation of NCOA4 was mediated by the iron‐dependent interaction between NCOA4 and the E3 ubiquitin ligase HERC2.[ ] We measured intracellular labile iron pool by Calcein‐AM and found about a 60% reduction in palmitate‐treated HepG2 cells compared with control cells (Figure S4C), consistent with a previous study.[ ] Therefore, palmitate may activate ferritinophagy downstream of iron depletion.

We then applied both gain and loss‐of‐function approaches to finally examine the functional role of ferritinophagy on palmitate‐induced toxicity (Figure 5F–I and Figure S4D–F). Measurement of cell death by propidium staining showed that NCOA4 knockdown or overexpression did not affect cell viability under basal conditions (Figure 5G and Figure S4E). Palmitate treatment in the culture medium increased the percentage of apoptotic cells from about 4% to 24% (Figure 5G). NCOA4 overexpression reduced palmitate‐induced cell death to 17.1% (Figure S4E), whereas siNCOA4 more than doubled cell death to 63.4% (Figure 5G). Cell death measured by Hoechst 33342 staining and cytosolic lactate dehydrogenase release into the medium reached the same conclusion (Figure 5H,I and Figure S4F).

Fatty acid desaturation quantitively correlates with cell survival under palmitate treatment conditions.[ ] As an additional confirmation, we carried out SCD1 knockdown experiments and examined the effect of SCD1 deficiency on palmitate‐induced cell death in a comparable system (Figure S4G). As shown in Figure S4H,I, SCD1 deficiency led to a modest increase in cell death even under basal, BSA‐treated cells, and palmitate‐induced cell death reached about 60%, comparable to siNCOA4 cells (Figure S4H,I). Therefore, ferritinophagy and desaturase play equivalent roles in protecting cells from saturated fatty acid–induced cell death.

Ferritinophagy is not required for re‐esterification of fatty acids in vivo

Fasting up‐regulates hepcidin expression, an indicator of iron overloading.[ ] We found that Ncoa4 was also up‐regulated by fasting both on the transcript and protein levels (Figure S5A). Because the liver re‐esterifies adipose tissue lipolysis–derived fatty acids under fasting conditions, we predicted that unsaturated fatty acid should not be limiting, and NCOA4 deficiency should not impair fasting‐induced hepatic steatosis.

Indeed, hepatic steatosis was not ameliorated in the overnight‐fasted shNcoa4 lean mice compared with their shLacZ controls (Figure 6A–C and Figure S5B). Instead, more triglyceride accumulated in the shNcoa4 liver and hepatocytes than in control (Figure 6C and Figure S5B), providing an in vivo support to our hypothesis that ferritinophagy‐controlled triglyceride synthesis is dependent on fatty acid composition.

FIGURE 6

Ferritinophagy is not required for hepatic re‐esterification of fatty acids under fasting conditions. (A) Schematic illustration of experimental design. Ten‐week‐old C57BL/6J male mice were tail vein–injected with adenovirus‐expressing shLacZ or shNcoa4. Mice were sacrificed 8 days following adenovirus administration after overnight fasting (5:00 p.m.–9:00 a.m.). (B,C) H&E staining (B) and triglyceride measurements (C) of the liver tissues of shLacZ and shNcoa4 fasted mice (n = 5–6 mice/group). Scale bar, 50 μm. (D–F) Transmission electron microscopy (TEM) analysis of hepatic mitochondrial morphology (D) from mice expressing shLacZ or shNcoa4 after overnight fasting; over 100 mitochondria in each group were quantified for average mitochondrial length (E) and area (F). (G–J) Mass spectrometry quantification of the relative abundance of total CL, PC, and TG (G), the fatty acid compositions of CL, PC, and TG (H), calculated indices of fatty acid saturation (C16:0/C18:1) for CL, PC, and TG (I), and the measured C16:0‐containing or C18:2‐only CLs (J) in the liver tissues of Ncoa4‐deficient, fasted mice (n = 5 mice/group). (K) Real‐time PCR measurements of hepatic Cd36, fatty acid–binding proteins (Fabps), glycerol‐3‐phosphate acyltransferase (Gpat), and diacylglycerol O‐acyltransferase 1 (Dgat1) transcript levels from mice treated as in (A) (n = 4 mice/group). (L) Measurements of liver triglyceride levels from mice expressing shLacZ, shNcoa4, shCd36, or double shLacZ and shCd36. Mice were sacrificed 8 days following adenovirus administration after overnight fasting (5:00 p.m.–9:00 a.m.) as in (A) (n = 6–8 mice/group). (M) Real‐time PCR measurements of hepatic Ncoa4, Cd36, Fabps, and Gpat transcript levels from mice treated as in (L) (n = 4 mice/group). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Student's t test (see also Figure S5).

Despite increased triglyceride synthesis, shNcoa4 lean mice hepatocytes presented swollen mitochondria (Figure 6D–F) and down‐regulation in cardiolipin content (Figure 6G), similar to palmitate‐treated siNCOA4 HepG2 cells. The differential effect of NCOA4 deficiency on cardiolipin and triglyceride synthesis may be attributed to the fact that hepatic triglycerides were synthesized overnight from adipose triglyceride–derived fatty acids, while cardiolipin and other phospholipids were the product of the hepatic fatty acid pool over a long period. Fatty acid composition analyses provided support for this proposition.

As shown in Figure 6H, lean mice lipidome had more polyunsaturated fatty acids and less saturated fatty acids than their obese counterparts (Figure 6H). Knockdown of Ncoa4 selectively increased fatty acid saturation in cardiolipins and phosphatidylcholines (Figure 6I,J), although to a less degree than shNcoa4 obese mice liver and siNCOA4 cells, but not in triglycerides (Figure 6I). Instead, fatty acid saturation in triglycerides was lower in the shNcoa4 liver than in control (Figure 6I).

Mitochondrial dysfunction might increase hepatic steatosis observed in the Ncoa4‐deficient liver. Additionally, transcript analysis found that the fatty acid transporter CD36 was also up‐regulated in the shNcoa4 liver tissues as in the siNCOA4 HepG2 cells (Figure 6K and Figure S3A). Knockdown of CD36 rescued the hepatic steatosis phenotype observed in overnight‐fasted shNcoa4 mice liver (Figure 6L,M and Figure S5C). Thus, ferritinophagy deficiency has the potential to increase triglyceride accumulation through CD36 induction.

DISCUSSION

Together, the results presented in this study indicate that ferritinophagy is essential for mobilizing iron from ferritin stores to support fatty acid oxidative desaturation and the synthesis of triglycerides and cardiolipins to reduce lipotoxicity. Although beneficial in the short term, the chronic induction of ferritinophagy likely contributes to the development of hepatic steatosis and severe liver disease in the long term.

Although the enzymatic requirement for iron in the proteome is well‐established, our findings shed light into the seemingly non‐deterministic relationship between iron and metabolic syndrome.[ , ] First, ferritinophagy activity may distort the relationship between iron accumulation and iron availability. As shown in our Ncoa4 gain and loss‐of‐function studies, iron accumulation is inversely correlated with functional iron sufficiency (Figure 1B,C for Ncoa4 knockout; Figure 1H,L for Ncoa4 knockdown; Figure 1N,O for Ncoa4 overexpression). Although the status of ferritinophagy in dysmetabolic iron overload syndrome (DIOS) has not been established, it is possible that iron accumulation may not quantitatively indicate functional iron sufficiency and overload. Second, dietary unsaturated fatty acid content may modify the relationship between ferritinophagy and hepatic steatosis/obesity. When the dietary fatty acid content is low, the body will be more dependent on ferritinophagy to desaturate fatty acids for triglyceride synthesis. However, under circumstances of ample dietary unsaturated fatty acids, high ferritinophagy activity may improve mitochondrial function through enhanced cardiolipin synthesis to reduce obesity and hepatic steatosis. Therefore, monitoring ferritinophagy status and dietary/plasma fatty acid composition may be important to dissect the relationship between iron and DIOS. Although it is challenging to measure ferritinophagy in clinical samples, the circulating ferritin secreted from the lysosome[ ] may serve as a surrogate marker. There are early indications that hepatic NCOA4 expression correlates with serum ferritin levels: obese patients with diabetes have higher levels of serum ferritin levels than patients without diabetes,[ ] and they often present higher hepatic NCOA4 expression.[ ] Future research is needed to establish their quantitative correlation.

Initially, we were concerned about the possibility of Ncoa4 deficiency in exacerbating lipotoxicity in obese mice. Feedback inhibition of lipogenesis may play a role in ameliorating liver damage in the shNcoa4 obese mice. Alternatively, ferritinophagy is an essential component of ferroptosis and many diseases.[ , , , ] We may speculate that the manifestation of metabolic syndrome requires ferritinophagy‐derived iron to generate oxidative stress and inflammation. Along this line, transcriptome analysis from clinical samples does show that obese patients without diabetes have a lower hepatic expression of NCOA4 compared with lean individuals or obese patients with diabetes, and NCOA4 expression is positively associated with gluconeogenic and lipogenic gene expression in obese patients.[ ] Therefore, chemical inhibition of ferritinophagy may be a viable approach to tackle NAFLD and NASH. The lack of effect of hepatic Ncoa4 deficiency on erythropoiesis reported herein and elsewhere[ ] suggests a good safety profile in this approach. Nevertheless, dietary fatty acid composition, CD36 induction, and compensatory up‐regulation in iron absorption may all affect the long‐term utility of such a strategy.

FUNDING INFORMATION

Supported by the National Key Research and Development Program of China (2017YFA0504603 and 2016YFA0502002), National Natural Science Foundation of China (NSFC 31671229 and 81,471,072), the Tsinghua‐Peking Center for Life Sciences, and Guangzhou Laboratory to S.F.

CONFLICT OF INTEREST

Nothing to report.

Figure S1 Hepatic ferritinophagy ensures physiological iron sufficiency, related to Figure 1A. Immunoblot measurements and quantification of nuclear receptor coactivator 4 (NCOA4) and ferritin heavy chain 1 (FTH1) protein levels in spleen, kidney, pancreas, muscle, and visceral fat of 10‐week‐old fl/fl and HepKO mice (n = 6 mice/group; each lane is a mixture of two samples). (B) Dimethylaminoazobenzene (DAB)–enhanced Perl’s staining of liver sections of 10‐week‐old fl/fl and HepKO mice (n = 3 mice/group). (C) Calculated offspring birth rate of fl/fl ad HepKO mice when intercrossing Ncoa4 mice with Ncoa4 and Alb‐Cre mice. (D) Body weight of 10‐week‐old fl/fl and HepKO mice fed normal chow diet (n = 7 mice/group). (E) Measurements of random and overnight fasting blood glucose level of 12‐week‐old fl/fl and HepKO mice (n = 6–7 mice/group). (F) Measurement of plasma triglyceride levels in 10‐week‐old fl/fl and HepKO mice fed a normal chow diet (n = 7 mice/group). (G,H) Measurement of spleen non‐heme iron levels (G) and serum ferritin levels (H) in 10‐week‐old fl/fl and HepKO mice fed a normal chow diet (n = 7 mice/group). (I) Complete blood counts of 12‐week‐old fl/fl and HepKO mice (n = 7 for fl/fl and n = 6 for HepKO). (J) Gene ontology (GO: molecular function) analysis of liver differential expression genes (DEGs) from Ncoa4 HepKO versus fl/fl mice, and shNcoa4 versus shLacZ mice (n = 4 group/mice). (K,L) DAB‐enhanced Perl’s staining of liver sections (K; n = 3 mice/group) and measurement of serum ferritin levels (L; n = 8 mice/group) of shLacZ and shNcoa4 mice. (M,N) DAB‐enhanced Perl’s staining of liver sections (M; n = 3 mice/group) and measurement of serum ferritin levels (N; n = 6 mice/group) in WT mice expressing adeno‐associated virus 8 (AAV8)–green fluorescent protein (GFP) or Ncoa4‐Flag. Scale bar, 100 μm. Data are presented as mean ± SEM (*p < 0.05; Student’s t test). Abbreviations: HCT, hematocrit; HGB, hemoglobin; MCH, mean corpuscular hemoglobin; MCV, mean corpuscular volume; n.s., not significant; PLT, platelet; RBC, red blood cell; WBC, white blood cell

Click here for additional data file.

Figure S2 Ferritinophagy is induced in obese mice liver to promote triglyceride storage, related to Figure 2. (A–D) Real‐time polymerase chain reaction (PCR) measurements of Ncoa4 transcript levels and immunoblot measurements of NCOA4 and FTH1 protein levels (A); co‐immunoprecipitation (co‐IP) measurements, and quantification of NCOA4‐FTH1 interactions (B); measurements of liver non‐heme iron levels (C); and real‐time PCR measurements of hepcidin (Hamp1) transcript levels (D) in the liver lysates of genetically (ob/ob, 10 weeks)–induced obese mice (n = 5–6 mice/group). Each lane of immunoblots in (A) and (B) are the mixture of two or six samples, respectively. (E) Schematic illustration of experimental design for Ncoa4 knockdown in ob/ob mice. (F) Immunoblot analysis of NCOA4, FTH1, and de novo lipogenesis enzyme protein levels in the liver tissues of control and Ncoa4‐deficient ob/ob mice (n = 5 mice/group). (G–M) Real‐time PCR measurements of Ncoa4 transcript levels (G), quantification of the liver non‐heme iron levels (H), hematoxylin and eosin (H&E) staining (I), liver triglyceride (TG) content (J), liver weight–to–body weight ratio (K), serum alanine aminotransferase (ALT) levels (L), and blood glucose levels (M) of control and Ncoa4‐deficient ob/ob mice (n = 6 mice/group; scale bar, 100 μm). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, Student’s t test

Click here for additional data file.

Figure S3 NCOA4 deficiency impairs fatty acid desaturation, related to Figure 4. (A) Real‐time PCR measurements of clusters of differentiation 36 (CD36), stearoyl‐coenzyme A desaturase 1 (SCD1), diacylglycerol O‐acyltransferase 1 (DGAT1), and DGAT2 transcript levels in control and siNCOA4 (48 h) HepG2 cells cultured in the absence or presence of 0.5 mm palmitate (PA) for 12 h. (B) Mass spectrometry quantification of the fatty acid compositions of main lipid species in control and siNCOA4 (48 h) HepG2 cells cultured in the absence or presence of 0.5 mm PA for 12 h. (C,D) Calculated indices of fatty acid saturation ([C16:0 + C18:0]/[C16:1 + C18:1]) of main lipid species in HepG2 cells treated as previously for both bovine serum albumin (BSA; C) and PA (D) challenged conditions. (E) BODIPY 493/503 staining and quantification of lipid droplets in control and siNCOA4 (48 h) HepG2 cells cultured in the presence or absence of 0.5 mm oleate (OA) for 12 h (scale bar, 20 μm). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; Student’s t test

Click here for additional data file.

Figure S4 Palmitate activates ferritinophagy to reduce lipotoxicity, related to Figure 5. (A) Real‐time PCR measurements of NCOA4 transcript levels in HepG2 cells treated with 0.5 mm PA for 12 h relative to BSA vehicle (n = 3). (B) Immunoblot measurements and quantification of NCOA4 and FTH1 protein levels in HepG2 lysates. Cells were treated with 0.5 mm PA in the presence or absence of bafilomycin A1 (BafA1, 40 nm), CQ (10 μm), or MG132 (10 μm) for 12 h. (C) Measurements of the intracellular labile iron pool (LIP) in HepG2 cells. Cells were treated with 0.5 mm PA or BSA vehicle for 12 h before Calcein‐AM (125 nm) staining. (D–F) HepG2 cells were transfected with either vector control or NCOA4 OE plasmids for 48 h, followed by 0.5 mm PA challenge for 12 h and immunoblot analysis of NCOA4 and FTH1 protein level (D); cell death was detected by propidium iodide (PI) staining (E) and lactate dehydrogenase (LDH) release (F). (G) Real‐time PCR measurements of SCD1 transcript levels in control and siSCD1 (48 h) HepG2 cells. (H) Microscopy imaging and quantification of cell deaths in control and siSCD1 (48 h) HepG2 cells cultured in the presence or absence of 0.5 mm PA for 12 h. *Denotes statistical significance in reference to siNC‐BSA control (scale bar, 100 μm). (I) Measurement of LDH release in control and siSCD1 (48 h) HepG2 cells cultured in the presence or absence of 0.5 mm PA for 12 h. *Denotes statistical significance in reference to siNC‐BSA control. Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ### p < 0.001; Student’s t test

Click here for additional data file.

Figure S5 Ferritinophagy is not required for hepatic re‐esterification of fatty acids under fasting conditions, related to Figure 6. (A) Measurements of liver Ncoa4 transcript levels and immunoblot measurements of NCOA4 and FTH1 protein levels in the liver tissues of fed and overnight‐fasted (5:00 p.m.–9:00 a.m.) WT mice (n = 5–6 mice/group). Each lane of immunoblots is the mixture of two samples. (B) BODIPY staining and quantification of lipid droplets in primary hepatocytes from control and Ncoa4‐deficient mice after overnight fasting (scale bar, 50 μm). (C) H&E staining of mice liver for the indicated group mice after overnight fasting (scale bar, 50 μm). Data are presented as mean ± SEM, **p < 0.01, ***p < 0.001; Student’s t test

Click here for additional data file.

Table S1 Synthesized oligos for small interfering RNA or short hairpin RNA, and primers for real‐time PCR

Click here for additional data file.

Appendix S1: Supporting Information

Click here for additional data file.

Appendix S2: Supporting Information

Click here for additional data file.