Human Cytomegalovirus pUL37x1 Is Important for Remodeling of Host Lipid Metabolism

Metabolome remodeling following HCMV infection.

Our work builds on the well-defined HCMV remodeling of the metabolome of primary human fibroblast cells under fully confluent, serum-free conditions (12–15, 21–23, 26, 27, 36). These conditions allow for the examination of metabolism while limiting potential lipid or metabolite contamination from bovine serum. Under confluent, serum-free conditions, subUL37x1 expresses pUL36 and pUL38 at WT-like levels while ablating pUL37x1 expression (Fig. 1A), similar to conditions with serum (9). The cell rounding and swelling associated with WT infection, e.g., cytopathic effect (CPE), were lessened in subUL37x1-infected cells at 1 to 2 days postinfection (dpi) (Fig. 1B). However, many subUL37x1-infected cells had WT-like CPE at later stages of infection (Fig. 1B, arrows). At 3 to 4 dpi, a larger amount of cellular debris was observed in cells infected with subUL37x1 than in WT-infected cells, consistent with a rise in cell death (Fig. 1B). Furthermore, we observed a 10-fold loss in virus yield with UL37x1 deletion (Fig. 1C). The loss of pUL37x1 also resulted in the release of fewer infectious virions per total released viral particles, i.e., a higher “particle-to-infectious unit” ratio (Fig. 1D). Overall, these observations are similar to those reported in previous studies of UL37x-1-null mutant viruses (9, 39, 40).

Next, we tested if pUL37x1 contributes to metabolic changes associated with HCMV infection. We examined the relative concentrations of intracellular metabolites in infected and uninfected cells using liquid chromatography–high-resolution tandem mass spectrometry (LC-MS/MS). Starting at 1 dpi, WT AD169 virus altered the intracellular concentrations of most metabolites measured (Fig. 2A). We observed an increase in glycolytic metabolites, amino acids, and TCA cycle intermediates in cells infected with WT virus (Fig. 2A), as previously described (14, 17, 21, 22, 25, 26). Infection with subUL37x1 mutant virus similarly altered metabolite levels starting at 1 dpi (Fig. 2A). The metabolic profiles of WT- and subUL37x1-infected cells were strongly correlated during mid- and late stages of replication (Fig. 2B). Importantly, glycolytic and TCA metabolites, e.g., hexose-phosphate, pyruvate, malate, α-ketoglutarate, and succinate, were increased in subUL37x1-infected cells compared to uninfected cells (Fig. 2A). However, these metabolites were slightly reduced compared to the levels in WT-infected cells at 2 to 3 dpi, suggesting that pUL37x1 may be important for achieving the overall high metabolite concentrations associated with infection but not required for HCMV to alter central carbon metabolism.

pUL37x1 is important for HCMV-induced FA elongation.

In addition to our metabolomic analysis, we quantitatively measured the effect of pUL37x1 on the cellular concentrations of fatty acids. HCMV infection enhances FA synthesis and elongation (12–14). FA synthase (FAS) forms long-chain FAs of up to 16 carbons in length (C₁₆) by connecting carbons, two at a time. The product of FAS is the 16-carbon saturated FA palmitate (C_16:0 [the number following the colon represents the number of double bonds in the FA]). Palmitate can be processed further to make longer chains and/or desaturated to introduce double bonds among the carbons. In humans, longer FAs are made by one or more of the seven FA elongases (ELOVL1 to -7). Similar to FAS, ELOVLs also add one 2-carbon unit per reaction cycle. We have previously shown that HCMV infection significantly increases the concentrations of saturated VLCFAs, specifically those with 26 or more carbons (≥C₂₆) (12, 13).

To determine if pUL37x1 is important for HCMV-induced FA elongation, we measured FAs extracted from lipids of HCMV-infected or uninfected cells. In cells infected with WT virus, saturated VLCFAs were increased by as much as 9-fold compared to the levels in uninfected cells, confirming our previous observations (Fig. 3) (12, 13). While the levels of some FAs were slightly higher in cells infected with subUL37x1 virus than in uninfected cells, the concentrations of C_26:0 and C_28:0 VLCFAs were significantly reduced in subUL37x1-infected cells relative to WT-infected cells at 2 and 3 dpi (Fig. 3).

FA elongation can be monitored by measuring the incorporation of ¹³C from labeled glucose into FAs using LC-MS, i.e., nontandem MS1-only analysis (12, 13, 42). Following a 1-h infection period, we fed cells medium that contained ¹³C-labeled glucose at positions 1 and 6 ([1,6-¹³C]glucose). Glycolytic breakdown of this labeled form of glucose will result in pyruvate containing one ¹³C atom (i.e., 1-labeled pyruvate) (Fig. 4A). Pyruvate can be further metabolized, generating 1-labeled citrate in the mitochondria. Citrate can be exported into the cytoplasm, where it can be converted to 1-labeled acetyl-CoA and unlabeled oxaloacetate. Alternatively, pyruvate may break down into acetate that can be used to generate 1-labeled acetyl-CoA (28, 43). The conversion of acetyl-CoA to malonyl-CoA by acetyl-CoA carboxylase 1 (ACC-1) is the rate-limiting step of FA synthesis and elongation. In our labeled cells, this will result in one ¹³C atom per 2-carbon unit used for FA synthesis and elongation.

Using LC-MS to measure the overall percentage of FA tails that have at least one ¹³C atom, we observed that uninfected cells have minimal labeling of FAs after correcting for the natural isotopic abundance of ¹³C. In WT-infected cells, FAs were labeled, and the greatest labeling occurred in longer FAs (e.g., 57% ± 2% of C_26:0 was labeled) (Fig. 4B). Relative to uninfected cells, the labeling in subUL37x1-infected cells was enhanced (e.g., 26% ± 6% of C_26:0 was labeled in subUL37x1-infected cells, compared to only 2% ± 3% in uninfected cells) (Fig. 4B). Moreover, the observed labeling in subUL37x1-infected cells was decreased relative to that in WT-infected cells (26% ± 6% compared to 57% ± 2% for C_26:0; P < 0.01).

In addition to a reduction in the percentage of labeling, the FAs isolated from cells infected with the subUL37x1 mutant virus had fewer ¹³C atoms than those from WT-infected cells (Fig. 4D). In WT-infected cells, we were able to observe up to 4 to 5 labeled atoms in C_26:0 (i.e., the incorporation of 8 to 10 new carbons). This observation was similar to that for cells fed uniformly labeled [¹³C]glucose, where two ¹³C atoms are added per 2-carbon unit (12, 13). However, in subUL37x1-infected cells, only 1 to 2 labeled atoms were observed in C_26:0, indicating that only one to two 2-carbon units were added (Fig. 4D).

Our glucose labeling strategy measured the carbon contributions of both citrate and acetate that is derived from glucose to FA elongation. Since HCMV infection increases the flow of carbons from acetate into lipid synthesis (28), we measured the contribution of acetate to HCMV-induced FA elongation using ¹³C-labeled acetate. In this case, both atoms of the 2-carbon unit were labeled. Again, cells were fed labeled acetate following a 1-h infection period. Similar to the labeled-glucose tracer studies, only minimal labeling of FAs from acetate was observed in uninfected cells (Fig. 4C). Infection increased acetate-derived atoms incorporated into VLCFAs (Fig. 4C). In WT-infected cells, 30% ± 1% of C_26:0 was labeled from [¹³C]acetate. Additionally, up to four labeled atoms from two 2-carbon units were incorporated into C_26:0 in WT-infected cells, a larger amount than in uninfected cells (Fig. 4E). Infection with the subUL37x1 mutant virus enhanced FA elongation from acetate compared to that in uninfected cells (Fig. 4C). However, the percent acetate labeling of C_26:0 was reduced in subUL37x1-infected cells relative to WT-infected cells (20% ± 6% versus 30% ± 1%; P < 0.05) (Fig. 4C). Additionally, fewer ¹³C atoms from acetate were incorporated into C_26:0 in subUL37x1-infected cells than in WT-infected cells (Fig. 4E). In summary, pUL37x1 is important for the high levels of FA elongation that are observed in HCMV-infected cells.

pUL37x1 expression enhances HCMV-induced ELOVL7 protein levels.

During infection, ELOVL7 produces saturated VLCFAs (13). HCMV infection increases ELOVL7 mRNA and protein expression (13). We examined if pUL37x1 impacts HCMV-induced ELOVL7 expression. In human foreskin fibroblast (HFF) cells, infection with WT virus enhanced ELOVL7 expression by 2 dpi (Fig. 5A), similar to our observations in MRC-5 primary fibroblasts (13). ELOVL7 protein levels were higher in subUL37x1-infected cells than in uninfected cells (Fig. 5A and B). However, the levels of ELOVL7 protein were reduced by approximately 40% in subUL37x1-infected cells relative to WT-infected cells (Fig. 5B and C).

We further investigated if pUL37x1 may affect other proteins important for FA elongation. HCMV infection increases the protein expression of ACC-1 (15), the enzyme producing the malonyl-CoA substrate for the FA elongation reaction (Fig. 6A). We found that ACC-1 levels were higher in subUL37x1-infected cells than in uninfected cells (Fig. 6B). Since FA labeling from [¹³C]glucose was reduced in subUL37x1-infected cells relative to WT-infected cells, we examined ATP citrate lyase (ACLY). ACLY converts citrate to acetyl-CoA in the cytosol. It was previously demonstrated that ACLY protein levels are slightly higher in infected cells than in uninfected cells at late time points (32). We found that ACLY levels were similar in WT- and subUL37x1-infected cells at 2 to 4 dpi (Fig. 6B). ACSS2 synthesizes acetyl-CoA from acetate. At 3 dpi, the cells infected with subUL37x1 virus had ACSS2 protein levels similar to those in WT-infected cells (Fig. 6C and D).

HCMV-induced metabolic remodeling requires AMPK activity, and infection increases AMPK protein levels (25, 36). Cells infected with subUL37x1 had levels of the catalytic subunit of AMPK, AMPKα, that were similar to the levels in WT-infected cells (Fig. 6B). AMPK reduces ACC-1 activity through phosphorylation. Cells infected with either WT virus or subUL37x1 virus have a larger amount of phosphorylated ACC-1 than uninfected cells (Fig. 6B), suggesting that pUL37x1 effects on FA elongation are mediated through an AMPK-independent mechanism.

PERK is important for HCMV-induced lipid metabolism (33). PERK protein levels are increased by HCMV infection (33, 44–46). We tested if pUL37x1 affects HCMV-induced PERK protein levels. At 3 dpi, the levels of PERK were increased by WT infection (Fig. 7A). However, the levels of PERK were 2-fold lower in subUL37x1-infected cells than in WT-infected cells (Fig. 7A and B).

HCMV infection remodels the host lipidome.

One function of FAs is to act as hydrophobic tails in lipids. Given our observations in Fig. 3 and 4, we hypothesized that pUL37x1 may help support the synthesis of lipids with VLCFA tails following HCMV infection. To test this hypothesis, we developed LC-MS/MS lipidomic methods to examine phospholipids, including those that contain ≥C₂₆ VLCFAs. We found that infection with WT virus markedly increased phospholipid concentrations, especially for those with VLCFA tails (Fig. 8A, left). Of the various classes of phospholipids examined, phosphatidylcholines (PCs) were among the lipid species most altered by infection (Fig. 8A and B). The concentrations of several PC species containing tails with a total of 44 to 48 carbons were significantly increased following HCMV infection (Fig. 8B). Of the top six PC species whose concentrations were increased by HCMV infection, their levels were higher in infected cells than in uninfected cells by 19- to 60-fold for PC(44:1) and by as high as 270- to 1,400-fold for PC(46:1). Since phospholipids contain two fatty acid tails, we examined the tail compositions of these six PC species. All contained a ≥C₂₆ tail (Fig. 8D, left). We found evidence of five different PC(44:1) species, with tail combinations of C_14:0 plus C_30:1, C_16:0 plus C_28:1, C_16:1 plus C_28:0, C_18:0 plus C_26:1, and C_18:1 plus C_26:0. PC(46:2) also contains five lipid species with tails ranging from C₁₆ to C₃₀ in length (Fig. 8D). For both PC(44:1) and PC(46:2), C₁₆-C₁₈ FAs were the most common tails observed in combination with ≥C₂₆ VLCFA tails. In this study, C₃₀ FAs were the longest tails observed.

When we examined phospholipids from subUL37x1-infected cells, we observed a significant decrease in lipid levels compared to those in WT-infected cells (Fig. 8A, right). The most notably decreased lipids were PC lipids whose levels were greatly elevated in WT-infected cells. The levels of all six of the PC species discussed above were significantly decreased in subUL37x1-infected cells relative to WT-infected cells (Fig. 8B). However, the levels of five of the six PC lipids highlighted in this study were elevated in subUL37x1-infected cells relative to uninfected cells (Fig. 8B). Infection with the mutant virus raised the concentration of these lipids by 4- to 80-fold (Fig. 8B). Only PC(48:5) was not statistically significantly increased following infection with subUL37x1 (Fig. 8B).

The relative levels of some lipids were higher in subUL37x1-infected cells than in WT-infected cells (Fig. 8A). The levels of some phosphatidylserine (PS) lipids were increased by as much as 4-fold in subUL37x1-infected cells relative to WT-infected cells (Fig. 8C). None of these PS lipids contained a tail longer than C₂₂ (Fig. 8D, right). Furthermore, none of these species were increased following infection with WT virus, and one, PS(36:2), was slightly reduced in WT-infected cells compared to uninfected cells (Fig. 8C).

Inhibition of cell death does not alter lipid changes.

Since pUL37x1 is important for limiting cell death, the loss of PC-VLCFAs in subUL37x1-infected cells could be due to an increase in cell death. We tested this possibility by suppressing the cell death that occurs following subUL37x1 infection by treating cells with Nα-tosyl-l-lysine chloromethyl ketone (TLCK). TLCK is a serine protease inhibitor that blocks apoptosis. TLCK treatment at concentrations of up to 100 μM has previously been demonstrated to block cell death associated with UL37x1-null mutant viruses without affecting virus replication (47). Treatment with 35 μM TLCK restored the survival of subUL37x1-infected cells to WT-like levels at 3 dpi (Fig. 9A). At 3 dpi, most lipids were unaltered in TLCK-treated cells (Fig. 9B). However, treatment reduced the levels of some PC lipids with 44 carbons or more in their tails in WT-infected cells (Fig. 9B, left). The levels of PCs in subUL37x1-infected cells were largely unaltered by TLCK treatment (Fig. 9B, right). Similar to WT-infected cells, TLCK treatment in subUL37x1-infected cells decreased the levels of some PCs containing ≥C₄₄ tails. The levels of most PS lipids were also unaltered by TLCK treatment (Fig. 9C). These observations demonstrate that preventing the cell death associated with the loss of UL37x1 fails to restore lipid levels.

pUL37x1 has a limited ability to induce lipid synthesis without infection.

We next investigated if pUL37x1 is sufficient to alter lipid levels. We stably expressed pUL37x1 in uninfected fibroblasts using a lentivirus system. In these cells, pUL37x1 is downstream of the EF1α promoter. As a control, we used the same lentivirus system to express green fluorescent protein (GFP) using the EF1α promoter. We independently generated two pUL37x1-expressing cell lines (Fig. 10A). In this case, the second clone, UL37x1-c.2, had a slightly higher level of pUL37x1 expression. First, we examined PERK and ELOVL7 protein levels in these cells. The levels of both PERK and ELOVL7 were unaltered by the expression of pUL37x1 (Fig. 10A). Lipidomic analyses of the pUL37x1-expressing cells revealed that the levels of some PC and PS lipids were statistically increased in UL37x1-c.2 cells relative to those in GFP-expressing cells (Fig. 10B to E). However, the concentrations of lipids in cells expressing a lower level of pUL37x1 were similar to those in the GFP control cells. These results suggest that high levels of pUL37x1 expression may be sufficient to induce some lipid synthesis.

Cells, viruses, and experimental setup.

Human foreskin fibroblast (HFF) cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) containing 10% fetal bovine serum (FBS), 10 mM HEPES, and penicillin-streptomycin. Prior to infection, cells were maintained at full confluence for 3 days in serum-containing growth medium. Cells were switched to serum-free medium (DMEM, HEPES, and penicillin-streptomycin) the day before infection. Cells were infected with either the wild-type (WT) HCMV AD169 strain or subUL37x1, a UL37x1-null virus (9). Unless noted otherwise, infections were performed at a multiplicity of infection (MOI) of 3 infectious units per cell. Both WT and mutant viruses were kindly provided by Thomas Shenk. Both WT and subUL37x1 viruses were grown in HFF cells. Mutant virus was not grown in cells expressing pUL37x1. All virus stocks were made by pelleting virions from the supernatant of infected cells through 20% sorbitol. Viruses were resuspended in serum-free DMEM and stored at −80°C. Infectious virus titers were measured as the 50% tissue culture infectious dose (TCID₅₀) by visualizing infectious centers. Infectious centers were further confirmed by immunofluorescence microscopy. In this case, cells on the TCID₅₀ plate were fixed using methanol, and plaques were identified using an anti-pUL123 (IE1) monoclonal antibody (clone 1B12), followed by anti-mouse Alexa 488 secondary antibody. This method has similarly been used to quantify a mutant HCMV that lacks the UL38 gene (13). The particle-to-IU ratio was determined using UL123-specific primers to measure viral DNA (62).

Metabolic tracer studies were performed using [1,6-¹³C]glucose and fully labeled [1,2-¹³C]acetate sodium salt from Cambridge Isotopes (catalog numbers CLM-2717 and CLM-440, respectively). Following a 1-h infection period, cells were washed with warm phosphate-buffered saline (PBS) three times prior to the addition of serum-free DMEM containing either 4.5 g/liter glucose (25 mM) or 0.0035 g/liter acetate (60 μM). For 3-dpi samples, the labeling medium was replenished at 2 dpi.

UL37x1 expression in uninfected cells.

The UL37x1 gene from BAC-AD169 was cloned into a lentiviral pLVX-EF1α-puro vector between the EcoRI and BamHI sites (TaKaRa). Since UL37x1 contains a BamHI site, PCR was used to generate UL37x1 that was flanked by EcoRI and BglII sites. Digestion and ligation destroyed the BamHI site in the final plasmid. The pLVX-EF1α-UL37x1 plasmid was verified by sequencing following cloning. A pLVX-EF1α-puro plasmid expressing GFP was used as a control. Stable expression of pUL37x1 in HFF cells was done using lentiviral transduction as previously described (13). For lipidomic analyses, 8 × 10⁴ cells were seeded per well on a 6-well plate. Cells were grown for 48 h in DMEM containing 10% FBS, HEPES, and penicillin-streptomycin. The cells were washed with PBS prior to lipid extraction as described below.

TLCK treatment.

Nα-Tosyl-l-lysine chloromethyl ketone (TLCK) was purchased from Sigma-Aldrich. Following a 1-h infection or mock infection period, cells were treated with 35 μM TLCK or the dimethyl sulfoxide (DMSO) vehicle. At 2 dpi, the drug was refreshed. At 3 dpi, cell survival was determined using a lactate dehydrogenase (LDH) release cytotoxicity assay kit (Pierce, Thermo Scientific). Lipids were extracted at 3 dpi and analyzed as described below.

Protein analysis.

Proteins were examined by Western blotting of whole-cell lysates. SDS-PAGE was performed with Tris-glycine-SDS running buffer using Mini-Protean TGX anyKD or 4 to 20% gels (Bio-Rad). Proteins were transferred to an Odyssey nitrocellulose membrane (Li-Cor). Western blot analyses were performed using 5% bovine serum albumin (BSA) in Tris-buffered saline with 0.05% Tween 20 (TBS-T) for blocking and antibody incubation, except for analyses of anti-ELOVL7, anti-ATP citrate lyase, and anti-PERK blots, which were performed using 3% milk in TBS-T. The following antibodies were used: mouse monoclonal anti-pUL37x1 clone 4B6-B (9), mouse monoclonal anti-pUL36 clone 10C8 (63), mouse monoclonal anti-pUL38 clone 8D6 (64), rabbit polyclonal anti-ELOVL7 (catalog number SAB3500390; Sigma-Aldrich), rabbit monoclonal anti-AMPKα (catalog number 2603; Cell Signaling), rabbit polyclonal anti-ATP citrate lyase (catalog number PA5-29497; Thermo), rabbit polyclonal anti-ACC-1 (catalog number 3662; Cell Signaling), rabbit polyclonal phospho-ACC-1 Ser79 (catalog number 3661; Cell Signaling), rabbit monoclonal anti-PERK (catalog number 3192; Cell Signaling), rabbit monoclonal anti-ACCS2 (catalog number 3658; Cell Signaling), mouse monoclonal anti-actin (catalog number 929-42212; Li-Cor), and mouse monoclonal anti-α-tubulin (catalog number T6199; Sigma-Aldrich). Blots with mouse monoclonal anti-HCMV, anti-actin, and anti-tubulin antibodies were incubated for 1 h at room temperature. All others were incubated overnight at 4°C. Quantification of Western blots was performed using a Li-Cor Odyssey CLx imaging system.

Metabolomics and metabolite analysis.

All metabolite levels were measured using liquid chromatography–high-resolution tandem mass spectrometry (LC-MS/MS). At various times after infection, metabolic reactions were rapidly quenched using cold 80% methanol, and extraction was performed as previously described (21, 27). Extracted water-soluble metabolites were dried under nitrogen gas and resuspended in Optima high-performance liquid chromatography (HPLC)-grade water (Fisher Chemical) for reverse-phase liquid chromatography or 50% methanol for hydrophilic interaction chromatography (HILIC). Mass spectrometric analysis was performed by using four Thermo Scientific mass spectrometers: Quantum Ultra triple-quad, Quantum Max triple-quad, Exactive orbitrap, and Q-Exactive Plus orbitrap. Metabolites were measured in both positive and negative modes using reverse-phase or HILIC separation (65, 66). Metabolites were identified using mass spectral and ultra-high-performance liquid chromatography (UHPLC) retention time data that were generated from external standards. Peaks were examined, and metabolites were identified using the Metabolomic Analysis and Visualization Engine (MAVEN) (67, 68). Each metabolite was measured using two technical replicates per experiment. Each metabolic experiment contained controls for determining ionization artifacts and sample degradation. All chemicals and solutions for metabolomics, fatty acid analysis, and lipidomics were of ultrapure HPLC or MS grade.

Fatty acid analysis.

Fatty acid (FA) analyses were performed as previously described (12, 13). Briefly, lipids were extracted from cells lysed using cold 50% methanol containing 0.05 M HCl. Cells were scraped and transferred to glass vials. Lipids were extracted in chloroform, and FA tails were chemically cleaved via saponification at 80°C for 1 h in basic methanol buffer. The FAs were isolated using hexane and dried under nitrogen gas. Following resuspension in a 1:1:0.3 solution of methanol-chloroform-water or a 1:1:1 solution of methanol-chloroform-isopropanol, FAs were analyzed by UHPLC and high-resolution mass spectrometry (LC-MS). FAs were examined using a Luna C₈ reversed-phase column (Phenomenex) and an Exactive or Q-Exactive Plus orbitrap mass spectrometer (13). In the latter case, FAs were separated using the LC buffer solutions described in “Lipidomics and lipid analysis” below. FA spectra were collected at a resolution of 140,000 (at m/z 200). For labeling experiments, the instrument was calibrated immediately before the start of the analysis. Experiments using metabolic tracers and FA labeling were additionally analyzed at a resolution of 280,000. The data were analyzed, and FA levels were quantified using MAVEN. For labeling experiments, the data were corrected for naturally occurring carbon 13 using MATLAB (13). Each sample was measured using two technical replicates per experiment.

Lipidomics and lipid analysis.

Similar to the FA analysis, cells were washed with PBS prior to analysis. For lipidomics, cells were lysed in cold 50% methanol, and lipids were extracted by using chloroform. The chloroform was removed by drying the samples under nitrogen gas. Lipids were resuspended in a 1:1:1 solution of methanol-chloroform-isopropanol. For each sample, 3 wells on a 6-well plate were used. Lipids were extracted in parallel from two wells and independently analyzed (e.g., technical duplicates). The cells in the third well were counted. Samples were normalized according to the total number of live cells at the time of lipid extraction. Each sample was resuspended in 125 μl of a 1:1:1 solution per 100,000 cells. In parallel with the biological samples, the extraction procedure was performed on wells that lacked cells. These “no-cell” controls were used to determine contaminants that were removed from the analysis. For initial experiments, 1 μl of Splash Lipidmix lipidomic mass spectrometry standards (Avanti Polar Lipids) was added to 50% methanol prior to extraction. These standards were used to establish our lipidomic methods to determine the extraction efficiency and proper data analyses/interpretations. For the final data included in this paper, the Lipidmix standards were not included in the samples to eliminate the possibility of contamination. After resuspension, the samples were immediately analyzed. During data collection, the samples were stored at 4°C in an autosampler. Lipids were separated using a Kinetex 2.6-μm C₁₈ column (catalog number 00F-4462-AN; Phenomenex) at 60°C on a Vanquish UHPLC system (Thermo Scientific). Two solvents were used for UHPLC: solvent A (40:60 water-methanol plus 10 mM ammonium formate and 0.1% formic acid) and solvent B (10:90 methanol-isopropanol plus 10 mM ammonium formate and 0.1% formic acid). UHPLC was performed at a 0.25-ml/min flow rate using the following gradient conditions: 75% solvent A–25% solvent B for 2 min, 35% solvent A–65% solvent B for 2 min at a curve value of 4, a hold at 35% solvent A–65% solvent B for 1 min, 0% solvent A–100% solvent B for 11 min at a curve value of 4, and a hold at 0% solvent A–100% solvent B for 4 min. The column was briefly washed and reequilibrated before the next run. In total, each run lasted 30 min. All lipids were examined using a Q-Exactive Plus mass spectrometer operating in a data-dependent full MS/dd-MS2 TopN mode. MS1 data were collected at a resolution of either 70,000 or 140,000 (at m/z 200) using an automatic gain control (AGC) target of 1e6 with transient times of 250 ms and 520 ms, respectively. Spectra were collected over a mass range of m/z 200 to 1,600. MS2 spectra were collected using a resolution of 35,000, an a AGC target of 1e5, and a 120-ms maximum injection time. Each sample was analyzed twice, using 10 μl of sample for negative mode and 8 μl of sample for positive mode. In negative mode, a normalized collision energy (NCE) value of 20 was used. In positive mode, the NCE value was increased to 30. All lipids were ionized using a heated electrospray ionization (HESI) source. The following HESI parameters were used in negative mode: sheath gas flow rate of 44, auxiliary gas flow rate of 14, sweep gas flow rate of 2, spray voltage of 3.3 kV, S-lens radio frequency (RF) of 76, auxiliary gas temperature of 220°C, and capillary temperature of 320°C. Similar settings were used in positive mode except that the spray voltage was increased to 3.5 kV and the S-lens RF level was decreased to 65. The instrument was calibrated weekly.

As with metabolites and FA, lipids were measured using two technical replicates per experiment. Lipidomic data were analyzed using MAVEN and Xcalibur (Thermo Scientific). In addition to the Splash Lipidmix, the following lipid standards were used for lipidomic method development and to establish LC retention times and MS2 fragmentation patterns for diacylglycerol (DG)(20:0/18:0), lysophosphatidic acid (lysoPA)(16:0), lysoPA(18:1), lysophosphatidylcholine (lysoPC)(16:0), lysoPC(18:0), lysoPC(22:0), lysophosphatidylethanolamine (lysoPE)(16:0), lysoPE(18:0), lysophosphatidylglycerol (lysoPG)(16:0), lysophosphatidylinositol (lysoPI)(16:0), lysophosphatidyserine (lysoPS)(16:0), lysoPS(18:1), monoacylglycerol (MG)(16:0), MG(18:1), phosphatidic acid (PA)(16:0/18:1), PA(18:0/20:4), PA(18:0/22:6), PC(16:0/18:1), PC(22:1/22:1), PC(24:0/24:0), phosphatidylethanolamine (PE)(16:0/18:1), PE(18:0/18:0), PE(18:0/20:4), phosphatidylglycerol (PG)(16:0/18:1), PG(18:0/20:4), PG(18:0/22:6), phosphatidylinositol (PI)(16:0/18:1), PI(18:0/20:4), PS(16:0/18:1), PS(18:0/20:4), triacylglycerol (TG)(16:0/18:1/16:0), TG(18:0/18:0/18:1), TG(18:1/16:0/18:1), and lipid reference material in human serum from the NIST.

Lipids were initially identified according to their MS1 mass using a 5-ppm cutoff and LC retention times. Lipid identification was confirmed using the following MS2 data: PC fragment of 184.074 (positive mode), PE fragment of 196.038 or the tail plus 196.038 (negative mode) and loss of 141.019 (positive mode), PG fragment of 152.996 plus the identification of the FA tails (negative mode), PI fragment of 241.012 (negative), and PS neutral loss of 87.032 (negative). PC lipids were further confirmed in negative mode using PC lipid ions containing a formic adduct. Both PC and PS tails were identified in negative mode.

Data availability.

The MS data files were deposited in the MetaboLights database (https://www.ebi.ac.uk/metabolights) under accession number MTBLS1073.