Regulatory roles of external cholesterol in human airway epithelial mitochondrial function through STARD3 signalling

2.1 Efflux of cholesterol into lung epithelial cells enhances cytokines production

Plasma levels of cholesterol in patients with severe COPD were significantly higher than those in the heathy (Figure 1A, Table S1). To determine the mainly changed lipid metabolism pathways, we screened mRNA levels of the lipid metabolism-associated gene panel with 384 genes (details are listed in Table S2) in HBEs treated with 6% CS extract (CSE) or vehicle and found 67 differentially expressed genes (DEGs) (Figure S1A,B). Of those DEGs with statistical significance (Figure S1C), steroid hormone biosynthesis and cholesterol-related pathways were obviously changed (Figure 1B,C). To furthermore evaluate the effect of cholesterol overload in blood on the severity of lung injury and inflammation, mice were fed with high cholesterol diet (HCD) and developed more severe lung inflammation in CS/LPS-treated mouse models (Figure 1D) than those fed with normal diet (ND). CS/LPS-treated mice with HCD had more severe lung tissue destruction, inflammatory cell infiltration and elevated total cells in bronchoalveolar lavage fluid (BALF), rather than ND-treated mice (Figure 1E–G). The cholesterol level in BALF was higher in mice with HCD while decreased after CS/LPS treatment (Figure 1H).

Airway epithelial cells act as the primary receptor in exposure to external pathogens and stimuli and as the secondary initiator to induce local and systemic inflammation by releasing inflammatory mediators. The concentration of CSE at 5%–20% could significantly inhibit the proliferation of HBEs, and the concentration >20% showed significant cytotoxicity, which was validated and shown in Figure S2. We stimulated HBEs with CSE at different concentrations and found that intracellular levels of cholesterol elevated (Figure 1I,J). CSE and cholesterol combination (CSE/cholesterol) induced more severe inflammation in HBEs than CSE alone (Figure 1K). Two key regulators in cholesterol biosynthesis, 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) and sterol-regulatory element binding protein 2 (SREBP2) were found significantly down-regulated (Figure 1L). CSE induced higher mRNA level of interleukin (IL)-6 and IL-8 in HBEs, which was parallel with published ELISA data as inflammatory characteristics of CSE-stimulated HBEs²⁰ and partially inhibited by administration of atorvastatin in a dose-dependent pattern (Figure 1M). To evaluate roles of the liver X receptors (LXRs), which induce cholesterol efflux, LXR agonists T0901317 and GW3965 were applied. GW3965 significantly inhibited CSE-induced up-expression of IL-8 (Figure 1N), while T0901317 had more inhibitory effects on CSE-induced up-expression of IL-6 (Figure 1O).

2.2 Intracellular cholesterol overload changes mitochondrial transcriptome and oxidative phosphorylation

The exogenous cholesterol dynamically increased the accumulation of intracellular cholesterol in HBEs in a dose-dependent pattern (Figure 2A,B). mRNA expression of cholesterol metabolism genes was down-regulated 12 and 24 h after adding of exogenous cholesterol, including SREBP2, HMGCR and low-density lipoprotein receptor (LDLR, Figure 2C). Of transcriptomic profiles, 650 genes were up-regulated, and 1787 genes were down-regulated in HBEs with CSE/cholesterol, as compared to those without cholesterol (p < .05, |log2Fold Change|>0, Figure 2D). Oxidative phosphorylation (OXPHOS) and metabolic pathways ranked top 10 (Figure 2E). About 4292 DEGs significantly differed between cells with vehicle and cholesterol, 2437 between cells with CSE and CSE/cholesterol and 7214 between cells with cholesterol and CSE/cholesterol, respectively (Figure S3A,B). Figure S3C presented significant differences of mRNA expression between CSE-stimulated HBEs with vehicle or cholesterol. Of those DEGs, 242 mitochondria-associated genes had high confidence of mitochondrial localization in CSE-stimulated HBEs, of which top 50 mitochondria-associated transcriptional changes were closely related to protein-protein interactions (Figure S3D,E). Those mitochondria-associated genes were filtered by MitoCarta3.0 database²¹ and are listed in Table S3. Genes related to mitochondrial inner membrane and respiratory chain significantly changed in cellular component, genes related to mitochondrial translational process and electron transport were in biological process, and genes related to NADH activity in molecular function (Figure 2F). Of those biological processes, the OXPHOS ranked first, and mitochondria-associated DEGs were also enriched to fatty acid metabolism, citrate cycle and other metabolic pathways (Figure 2G).

2.3 Cholesterol enhances mitochondrial damage in exposure to CS

To investigate roles of cholesterol in mitochondrial function, we measured the mitochondrial mass and membrane potential (MMP) in HBEs and found that CSE elevated the mitochondrial mass (Figure 3A) and decreased the MMP (Figure 3B,C), as compared with cells with vehicle. Addition of cholesterol increased levels of mitochondrial mass. That increase did not culminate in compensatory improvement of MMP. We evaluated the oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) and found that basal respiration, proton leak, ATP production and spare respiratory capacity significantly decreased in HBEs with CSE, which were exacerbated by cholesterol (Figure 3D,E), while the ECARs were not changed (Figure S4). CSE led to swollen and fragmented mitochondria evidenced by ultrastructural alterations under electron microscope (Figure 3F,G, Figure S5A–C), and the ratio of length/width decreased in HBEs with CSE. Cholesterol exacerbated mitochondrial vacuolization and destroyed mitochondrial cristae and membrane structure (Figure 3F,G).

We analyzed mRNA expression of mitochondrial fusion genes including mitofusin (MFN) 1, MFN2, optic atrophy 1 (OPA1) and mitochondrial fission genes including dynamin-related protein 1 (DRP1), mitochondrial fission factor (MFF) and mitochondrial fission protein 1 (FIS1) by cell RNA sequencing (RNA-seq) and verified by RT-qPCR. Data from HBE RNA-seq demonstrated that CSE induced low-expression of MFN2 mRNA and high-expression of MFF and FIS1, which were altered by adding cholesterol (Figure S5D). In validation, levels of MFN2, OPA1 and MFF mRNA expression were higher in HBEs with CSE than those with vehicle, while cholesterol alone reduced the mRNA expression of OPA1 and DRP1, as compared with vehicle (Figure 3H). Of mitochondrial dynamics-associated proteins, MFN2 levels in cholesterol-treated cells were significantly lower than other groups, DRP1 in CSE or cholesterol-treated cells lower than those with vehicle or the combination, and FIS1 in cells with CSE, cholesterol, or the combination lower than those with vehicle, respectively (Figure 3I,J). Of mitochondrial fission/fusion proteins in the inner and outer membrane of mitochondria (DRP1, MFN2 and OPA1), protein levels of MFN2 in lung tissues harvested from animals were down-regulated by HCD and up-regulated by CS/LPS exposure (Figure 3K,L), which were consistent with results from the in vitro studies.

2.4 STARD3 regulates epithelial inflammation and mitochondrial function

To explore effects of intracellular cholesterol accumulation on mitochondrial function and inflammation, we measured transcriptomic profiles of HBEs 3 and 24 h after CSE stimulation. Our RNA-seq data described that STARD3 was the highest expressed gene among members of STARD family related to cholesterol homeostasis in HBEs (Figure 4A). mRNA levels of STARD3 or STARD4 were significantly lower or higher 3 or 24 h after CSE exposure, respectively (Figure 4A). mRNA expression of STARD5 and STARD6 decreased by time after CSE stimulation, especially at 24 h. Those biological functions of STARD members varied in response to alterations of cholesterol homeostasis in CSE-induced pathophysiology. We further evaluated transcriptional and translational levels of STARD3 in cells and animals and found that CSE up-regulated expression of STARD3 mRNA (Figure 4B) and protein (Figure 4C,D), while levels of STARD3 in CSE/cholesterol were significantly lower than those in CSE alone. Protein expression of STARD3 reduced in lung tissues of mice with HCD, as compared with those with ND or with CS/LPS (Figure 4E,F).

In order to define regulatory roles of STARD3 in inflammatory response, we knocked down STARD3 in HBEs with siRNA against STARD3 (STARD3^kd) or over-expressed STARD3 in HBEs transferred with STARD3 (STARD3^oe) and measured mRNA expression of pro-inflammatory cytokines. mRNA expression of IL-6, IL-8 and IL-1β (Figure 4G) significantly decreased in STARD3^kd, as compared with negative control cells (STARD3^NC) (p < .05 or less, respectively). Levels of IL-6, IL-8 and IL-1β mRNA expression in STARD3^kd with CSE/cholesterol were higher than those with vehicle or CSE alone. Levels of IL-6 mRNA were significantly higher in STARD3^oe, as compared with negative control cells (STARD3^Ve) (Figure 4H). CSE/cholesterol significantly increased IL-6 mRNA expression in STARD3^oe, as compared with STARD3^Ve or in STARD3^oe with vehicle or CSE, respectively (Figure 4H). Similar effects of CSE/cholesterol to HBEs were observed in BEAS-2B cells (Figure S5E,F). We also evaluated roles of STARD3 in apoptosis and found that CSE increased the apoptosis, while STARD3^kd had lower amount of apoptosis after CSE stimulation, as compared with STARD3^NC (Figure 4I,J).

2.5 STARD3 down-regulation plays preventive roles in mitochondrial cholesterol overload

CSE/cholesterol significantly reduced mitochondrial mass (Figure 5A) and MMP (Figure 5B,C) in HBEs, while levels of MMP in STARD3^kd were higher than those in STARD3^NC, which was observed in BEAS-2B cells (Figure S6A–C). We measured the mitochondrial texture scores and morphology index using confocal microscopic imaging and performed the high content screening. The results illustrated that the short and fragmented mitochondria were partly recovered into long and tubular mitochondria in STARD3^kd as compared with STARD3^NC after CSE/cholesterol (Figure 5D). Levels of Filipin III intensity in mitochondria decreased (Figure 5E) and mitochondrial morphology increased (Figure 5F) in STARD3^kd, respectively. The length/width ratio increased in STARD3^kd after CSE/cholesterol (Figure S6D,E). Differences of mitochondrial ultrastructure were noticed between STARD3^NC and STARD3^kd with vehicle (Figure S7A,B), CSE (Figure S7C and S7D) or cholesterol (Figure S7E,F). Levels of basal respiration, proton leak, spare respiratory capacity and ATP production in HBEs with CSE were significantly lower than those with vehicle and became even lower in those with CSE/cholesterol (Figure 5G,H). Exogenous cholesterol did not exacerbate the respiration dysfunction in STARD3^kd, while the levels of mitochondrial respiration in STARD3^oe were lower than those in STARD3^Ve (Figure S6F,G). The overexpression of STARD3 exacerbated the dysfunction of mitochondrial respiration induced by CSE. Lower levels of STARD3 alleviated the mitochondrial cholesterol overload and partially rescued MMP function, morphology and respiration.

We also evaluated the mitochondrial dynamics-associated markers and found that mRNA expression of MFN2, OPA1, DRP1, MFF and FIS1 in HBEs with CSE were higher than those with vehicle or with CSE/cholesterol (Figure 6A). The expression of mitochondrial dynamics-associated genes was lower in STARD3^kd, as compared with STARD3^NC. MFN2 expression was significantly lower in STARD3^oe with CSE or CSE/cholesterol, as compared with STARD3^Ve (Figure 6B). STARD3^oe had lower expressions of DRP1 gene (Figure 6B) and protein (Figure 6E,F) than those of STARD3^Ve after challenges with vehicle, CSE or CSE/cholesterol. Protein levels of FIS1 and OPA1 elevated in STARD3^kd with CSE, while MFF and DRP1 elevated in STARD3^kd with CSE/cholesterol, when compared between STARD3^NC (Figure 6C,D). Levels of FIS1, OPA1, MFN2, MFF and DRP1 were down-regulated in STARD3^oe with CSE or CSE/cholesterol, as compared with STARD3^Ve (Figure 6E,F).

2.6 MFN2 down-regulation causes fatty acid oxidation and mitochondrial dysfunction

MFN2 was upregulated in the lung tissues of CS/LPS-treated mouse model and CSE-stimulated HBEs, while downregulated by exogenous cholesterol supplement. We knocked down MFN2 (MFN2^kd) and found the fragmentated mitochondria (Figure 7A), decreased MMP (Figure 7B), and altered mitochondrial morphology (Figure 7C) in MFN2^kd. mRNA expression of IL-6 and IL-8 elevated in MFN2^kd with CSE (Figure 7D), indicating that the downregulation of MFN2 may increase epithelial capacity of inflammatory reactions. We measured the lipid droplets (LDs) and found that MFN2^kd had more LDs than MFN2^NC and CSE increased LDs contents in MFN2^NC (Figure 7E,F). The lipidomic measurements of MFN2^NC and MFN2^kd showed that downregulation of MFN2 elevated level of free fatty acid (FFA), cholesterol ester (CE) and triacylglycerol (TAG) (Figure 7G). The mitochondrial dysfunction leads to the disruption in fatty acid oxidation (FAO), and excessive FFA in cytoplasm turns into triacylglycerol storing in LDs. We applied etomoxir (ETO), an inhibitor of carnitine palmitoyl-transferase 1α, for inhibiting FAO, and L-carnitine (L-CAR), an endogenous molecule for promoting FAO and interrupting the formation of LDs through the transport of long chain fatty acyl-CoAs into the mitochondria. ETO and CSE increased LDs in MFN2^NC (Figure 7H,I). L-CAR reduced the production of pro-inflammatory cytokines in CSE-stimulated MFN2^NC (Figure 7J), while L-CAR aggravated the inflammatory response of MFN2^kd with CSE. Levels of mTOR phosphorylation in MFN2^kd were significantly higher than in MFN2^NC, which was prevented by Rapamycin (Rap) (Figure 8A,B, p < .01, respectively). mRNA levels of lipid synthesis-related genes, including SREBP2, HMGCR, FASN and LDL-R, were higher in CSE-stimulated MFN2^kd (Figure 8C). Rap partially prevented the increase of LDs in MFN2^kd (Figure 8D) and the production of IL-6 in MFN2^NC and MFN2^kd (Figure 7D).

4.1 Patient population

The present study was approved by the Ethical Evaluation Committee of Zhongshan Hospital (B2019-197(2)) and designed using a case–control approach. Informed consent for lipids analysis was given by patients. The study involved 22 patients with severe COPD, which was diagnosed as described previously.⁶⁰ Patients were recruited between October 2019 and March 2020 with the medical history including admittance to the hospital due to acute exacerbation. Nineteen healthy volunteers were enrolled at Zhongshan Hospital. Plasma was harvested from healthy volunteers and patients with severe COPD on day one of admission. Details of patient population are listed in Table S1.

4.2 Cell culture, agents, transfection and RNA interference

HBE cells were derived from normal bronchus in a 1-year-old male heart-lung transplant patient, immortalized cell lines, which were first extracted from bronchus.⁵⁹ HBE cells were cultured in RPMI 1640 (SH30809.01, Hyclone, Logan, UT, USA) with 10% fetal bovine serum (#10099141, Gibco, ThermoFisher, MA, USA). BEAS-2B cells, another normal epithelial cell line, were purchased from ATCC (CRL-9609) and cultured in DMEM (SH30243.01, Hyclone, Logan, UT, USA). T0901317 (HY-10626), GW3965 (HY-10627), rapamycin and atorvastatin (HY-B0589) were purchased from MedChemExpress, NJ, USA. Transfection of RNA interference was conducted by Lipofectamine 2000 (#11668019, ThermoFisher Scientific, MA, USA). The sequencing is listed as below:

siRNA-STARD3-sense: CCUGGUUCCUUGACUUCAATT;

siRNA-STARD3-nonsense: UUGAAGUCAAGGAACCAGGTT;

siRNA-MFN2-sense: CCAUGAGGCCUUUCUCCUUTT;

siRNA-MFN2-nonsense: AAGGAGAAAGGCCUCAUGGTT.

Transfection of STARD3 overexpression plasmid (Genechem, Shanghai, China) was conducted by Lipofectamine 3000 (L3000008, ThermoFisher Scientific, MA, USA). The knock down efficiency was verified after 48 h. Targeted-STARD3 shRNA lentiviral vector was constructed in Genechem, Shanghai, China, and transfected into HBE cells according to the manufacturer's protocol. Stable HBE cells with lower STARD3 expression were verified by RT-qPCR as well as Western blot. All in vitro experiments were performed and triplicated, and there were six per group each time.

4.3 CSE preparation

Briefly, one cigarette was combusted through a modified 50-ml syringe apparatus into 5-ml RPMI 1640. The quality control of 100% CSE was performed by measuring the optical density at 320λ wavelength and a value of 1.158.⁶¹

4.4 Cholesterol preparation

Cholesterol was purchased from Sigma-Aldrich, Louis, USA (#C3045) and was dissolved in 100% ethanol at 65℃. Then, the liquid was added to methyl-β-cyclodextrin (HY-101461, MedChemExpress, NJ, USA) at a concentration of 5 mM cholesterol to 42 mg/ml methyl-β-cyclodextrin.⁶²

4.5 Cholesterol staining and high-content screening

Cholesterol was stained by Filipin III in cells fixed with paraformaldehyde. For the observation on colocalization of cholesterol and mitochondria, cells were incubated with mitotracker deep red (25 nM) for 30 min before fixation. Fixed cells were incubated with 50 μg/ml filipin III (SAE0088, Sigma, MO, USA) at room temperature for 1 h. The emission/excitation of filipin III were 340–380/385-470 nm. Cells were screened with Perkin-Elmer high-content automatic imaging system (40× air or 63× water objective) in confocal mode, and the results were analyzed by Perkin-Elmer Harmony image analysis software.

4.6 Analysis of mitochondrial function

The mitochondrial function was evaluated, including mitochondrial mass, MMP, morphology, respiration and mitochondrial dynamics. For mitochondrial MMP analysis, live cells were stained with mitotracker green (C1048, Beyotime, Shanghai, China, 100 nM) and mitotracker red CMXRos (C1049, Beyotime, Shanghai, China, 25 nM) in 1640 Medium at 37℃ for 30 min and evaluated by flow cytometry or high-content screening. For the analysis of mitochondrial morphology, the texture of mitochondria was extracted using Perkin-Elmer Operetta high-content automatic imaging system. Texture parameters of the mitochondria—SER features (mathematical basis is Gaussian derivative filter) including eight patterns (Spot, Edge, Ridge, Saddle, Valley Hole, Bright, Dark) at 1px scale with Kernel normalization were determined to measure mitochondrial morphology.^{63, 64} The scores of SER Ridge were calculated to quantify the mitochondrial morphology. In transmission electron microscopy, cells were collected ang fixed by 2.5% glutaraldehyde in 0.1 M sodium cacodylate pH 7.4. ImageJ was used to measure the ratio of length/width of the mitochondria as recently reported.⁶⁵ For the analysis of mitochondrial respiration, cell mitochondrial stress test kits (103015-100, Agilent Technologies, CA, USA) were used. OCR was measured using a Seahorse XFe96 Extracellular Flux Analyzer (Agilent Technologies, CA, USA).⁶⁶ Briefly, HBEs were plated on 96-well plates at 1*10⁵ cells/well in RPMI 1640 including 10 mM glucose, 2 mM l-glutamine and 1 mM sodium pyruvate. ECAR and OCR were detected at basal level and after injections of 2 μM oligomycin, 1 μM FCCP and 0.5 μM rotenone/antimycin mix. Cell number was averaged per condition, and OCR was normalized to these values. Basal respiration, proton leak, spare respiratory capacity and ATP production were calculated. For the analysis of mitochondrial dynamics, RT-qPCR and Western blot were used to evaluate MFN2, OPA1 as fusion markers and DRP1, MFF and FIS1 as fission markers.

4.7 LDs staining

Cells were stained with BODIPY 493/503 (GC42959, GLPBIO, CA, USA, 10 μM) for 60 min to visualize LDs in fixed cells. HCS (63X water object, confocal mode) was used to scan the cells and calculate the mean fluorescence intensity.

4.8 GC-MS/MS metabolomics to evaluate cholesterol level

Plasma from healthy volunteers and COPD patients or BALF from mouse models were obtained. Regarding derivatization, methoxyamine with N-methyl-N(trimethylsilyl)trifluoroacetamide (MSTFA) was applied.⁶⁷ Briefly, 100 μl of fresh plasma sample was treated with 400 μl of cold methanol-water (4:1 v/v, 5ug/ml tridecanoic acid was included). The mixture was vortexed, shacked at 37°C, 1200 rpm for 30 min and then centrifuged. The supernatant was evaporated under a stream of nitrogen gas. Methoxyamination of samples was carried out with 50 μl MeOX (20 mg/ml) in pyridine. Each sample was sonicated for 1 min and shaked at 30°C, 1200 rpm for 90 min, trimethylsilylated in 40 μl MSTFA and shook at 37°C, 1200 rpm for 30 min. The supernatant was clarified by centrifugation at 12000 rpm for 5 min and transferred to a vial for GC-MS/MS analysis. The Agilent Fiehn GC/MS Metabolomics RTL Library (version 2013) was employed for metabolite identifications. Tridecanoic acid was added as an internal standard. The formula: standard sample peak area/standard sample concentration = sample peak area/sample concentration, was used to calculate the metabolites content in the sample.⁶⁸

4.9 Quantitative RT–PCR

TRIzol (#15596026, Invitrogen, CA, USA) was used to extract total RNA. Then, mRNA was reverse-transcribed to cDNA by PrimeScript RT Master Mix (RR036A, Takara, Japan). The real-time PCR was performed by TB Green Premix Ex Taq (RR420A, Takara, Japan) with primers listed in Table S4.

4.10 Western Blot

Ice-cold lysis buffer (with protease inhibitors) was used to lyse cells. After centrifugation, total proteins in the supernatants were quantified using BCA kit (P0010, Beyotime, Shanghai, China). Note that 30-μg protein was loaded per lane in 10% SDS-PAGE and electrophoresed on a PVDF membrane for detection with antibodies. Primary antibodies are as follows: β-tubulin (#2128, 1:1000, Cell Signalling Technology, MA, USA); β-actin (#4970, 1:1000, Cell Signalling Technology, MA, USA), p-mTOR (phosphor S2448, 1:1000, Cell Signalling Technology, MA, USA), MFN2 (ab124773, 1:1000, Abcam, Cambridge, UK), DRP1 (ab184247, 1:1000, Abcam, Cambridge, UK), OPA1 (ab157457, 1:1000, Abcam, Cambridge, UK), MFF (#17090-1-AP, 1: 5000, Proteintech, IL, USA) and FIS1 (#10956-1-AP, 1:1000, Proteintech, IL, USA). Secondary antibodies were HRP linked anti-mouse/rabbit IgG (1:2000, #7076/#7074, Cell Signalling Technology, MA, USA). The bound antibodies were visualized with ECL detection (#32106, ThermoFisher Scientific, MA, USA). The gray values were calculated by Image J (NIH, USA).

4.11 Animal models

Wild-type C57BL/6J mice (6–8 weeks old) were purchased from Shanghai Jiesijie Animal Experiment Co., Ltd. (Shanghai, China) and housed in a temperature- and humidity-controlled room in Zhongshan Hospital Animal Experiment Center. CSE/LPS-induced mouse models were induced as previously reported.^{69, 70} Briefly, mice were intraperitoneally injected with 200 μl 2.5% avertin (T48402, Sigma, MO, USA) for anesthesia and intratracheally instilled with 25 μg LPS (L2880, Sigma, USA) in 50 μL saline on days 0 and 15. On days 1∼14 and 16∼29, the mice were exposed to CS (1 h/time, 20 cigarettes/day) in a closed organic glass box (80 × 60 × 50 cm). Each cigarette contains 10-mg tar, 0.8-mg nicotine and 12-mg carbon monoxide. HCD was supplied by Research Diets (D12109, Research Diets, NJ, USA).⁷¹ Mice were sacrificed at day 30, and the blood, BALF and lung tissue were collected for further analysis. The mouse lung tissues were processed for protein and/or RNA measurements. For RNA, 50-mg tissues in liquid nitrogen were added 1-ml TRIzol and homogenized by tissue grinder, followed by RNA extraction using chloroform and isopropanol. For protein, 50-mg tissues in liquid nitrogen were added 300-μl ice-cold lysis buffer plus protease inhibitors and homogenized by tissue grinder, followed by the procedures mentioned in Western blot section.

4.12 HBE RNA sequencing

Cellular RNA was extracted by TRIZOL, and the integrity was assessed using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies, CA, USA). AMPure XP system (Beckman Coulter, Beverly, USA) was used for purification. Illumina Novaseq platform was used for sequencing. Data were aligned to the reference genome using Hisat2 v2.0.5. FeatureCounts v1.5.0-p3 was used to count the reads numbers mapped to each gene, and Fragments Per kilobase Million (FPKM) of each gene was calculated. DEGs were analyzed by DESeq2 R package (1.20.0). Gene ontology enrichment analysis and KEGG pathways were implemented by DAVID Bioinformatics Resources 6.8. 1. Differential mitochondria-associated genes from HBE RNA sequencing were filtered by Human MitoCarta 3.0 datasets.

4.13 mRNA expression of STARD family in RNAseq data

RNAseq library preparation and sequencing were processed following the instruction in the section ‘HBE bulk RNA sequencing’. RNAseq analysis was performed from three independent replicates for each condition or group. Then, we obtained the RNAseq BAM files in the RNAseq database. Gene expression analysis of STARD family was performed in six independent experiments of different conditions. For heatmap, RNAseq reads were normalized (RPKM) for all conditions, and variations were scored as a Log ratio between different conditions or groups. One-way ANOVA test was used to test the significance of differences between conditions by R language and Prism 8.0.

4.14 Analysis of intracellular lipidomics

Cultured cells were collected and normalized to 2*10^6 cells/sample, and centrifuged. Cell pellets were added with 0.9 ml double distilled water and disrupted by repeated freeze-threw cycles for three times. Lipid internal standard mixture was added. Lipid extraction was performed using a modified Bligh and Dyer procedure, as described preciously.⁷² The extracts were dissolved by 200 μl chloroform/methanol (1:2, v/v, including 10mM NH₄Ac) and stored at −20℃. We used normal phase liquid chromatography and coupled triple-quadrupole mass spectrometer (QTRAP 5500, AB SCIEX, USA) to detect the lipid by positive and negative electrospray ionization mode. Q-Trap in the multiple reaction monitoring mode operation was applied.²² Lipidomics data were analyzed by MultiQuant software (AB SCIEX, USA).

4.15 Statistical analysis

All quantitative data were calculated using GraphPad Prism Version 7.0 (Graph Pad Software, CA, USA) and presented as mean ± standard error of the mean. An unpaired two-tailed Student's t-test or ANOVA test was used to calculate the p values according to the different data types. p < .05 was considered significant.