Abstract

Background

   COVID-19 is a highly infectious respiratory disease that can manifest
   in various clinical presentations. Although many studies have reported
   the lipidomic signature of COVID-19, the molecular changes in
   asymptomatic severe acute respiratory syndrome coronavirus 2
   (SARS-CoV-2)-infected individuals remain elusive.

Methods

   This study combined a comprehensive lipidomic analysis of 220 plasma
   samples from 166 subjects: 62 healthy controls, 16 asymptomatic
   infections, and 88 COVID-19 patients. We quantified 732 lipids
   separately in this cohort. We performed a difference analysis,
   validated with machine learning models, and also performed GO and KEGG
   pathway enrichment analysis using differential lipids from different
   control groups.

Results

   We found 175 differentially expressed lipids associated with SASR-CoV-2
   infection, disease severity, and viral persistence in patients with
   COVID-19. PC (O-20:1/20:1), PC (O-20:1/20:0), and PC (O-18:0/18:1)
   better distinguished asymptomatic infected individuals from normal
   individuals. Furthermore, some patients tested positive for SARS-CoV-2
   nucleic acid by RT-PCR but did not become negative for a longer period
   of time (≥60 days, designated here as long-term nucleic acid test
   positive, LTNP), whereas other patients became negative for viral
   nucleic acid in a shorter period of time (≤45 days, designated as
   short-term nucleic acid test positive, STNP). We have found that TG
   (14:1/14:1/18:2) and FFA (4:0) were differentially expressed in LTNP
   and STNP.

Conclusion

   In summary, the integration of lipid information can help us discover
   novel biomarkers to identify asymptomatic individuals and further
   deepen our understanding of the molecular pathogenesis of COVID-19.

   Keywords: SARS-CoV-2, COVID-19, asymptomatic infection, lipids,
   long-term nucleic acid test positive

Introduction

   At the end of 2019, an acute respiratory disease caused by SARS-CoV-2
   continued to spread rapidly all over the world and attracted extensive
   attention ([41]1). Coronavirus disease 2019 (COVID-19) is a highly
   contagious disease that targets the respiratory tract. Asymptomatic
   infections (AS) are those who do not have any symptoms but carry
   SARS-CoV-2 ([42]2). The number of asymptomatic subjects has raised
   concerns worldwide because they are difficult to identify. In addition,
   some of the patients tested positive for the nucleic acid of SARS-CoV-2
   by RT-PCR but did not become negative for a longer period of time (≥60
   days, herein designated as long-term nucleic acid test positive, LTNP)
   whereas others tested negative for the viral nucleic acid in a shorter
   period of time (≤45 days, designated as short-term nucleic acid test
   positive, STNP). Asymptomatic and LTNP COVID-19 patients both threaten
   global public health. However, little is known about the detailed
   mechanisms responsible for the LTNP and STNP as well as AS and health
   controls.

   At present, research on virus–host interaction generally focuses on
   proteomic changes. However, metabolites, as the final product of
   cellular processes, provide different insights into the mechanism of
   viral infection ([43]3).. Previous studies have shown that the
   pathogenesis of viral infection is closely related to the lipid
   metabolism of infected cells. 25-Hydroxycholesterol (25-HC) inhibits
   viral invasion by inducing inflammatory and immune responses ([44]4).
   It has been found that the replication of SARS-CoV and SARS-CoV-2 is
   inhibited by 25-HC ([45]5). Omega-3 PUFA-derived lipid mediator
   protectins have been shown to inhibit influenza virus replication by
   blocking viral mRNA output ([46]3).

   Lipids play an important role in the life cycle of viruses. Their
   involvement in virus infection includes fusion of virus membranes and
   host cells, virus replication, endocytosis, and exocytosis ([47]6).
   SARS-CoV-2 is an enveloped virus surrounded by a lipid bilayer. Viruses
   manipulate host cells by targeting lipid synthesis and signal
   transduction pathways. They modify the host cells, enveloping them to
   produce lipids ([48]7). Therefore, understanding the alterations in
   host cell lipids can be beneficial in identifying potential biomarkers
   that can distinguish between healthy individuals, those with
   asymptomatic infections, and individuals with different types of
   symptoms caused by SARS-CoV-2. This knowledge can further aid in the
   development of targeted treatments.

   In this study, we aimed to investigate the potential pathogenesis of
   COVID-19 by analyzing the host response to SARS-CoV-2 infection in
   humans. We conducted lipidomics analysis on plasma samples obtained
   from a total of 220 individuals, including 104 COVID-19 patients and 62
   healthy controls. We identified 175 differentially expressed lipids
   that were associated with the SASR-CoV-2 infections, disease severity,
   and viral persistence in the COVID-19 patients. For example, we found
   that 81 lipids showed significant differences between AS and healthy
   controls, and 15 lipids could distinguish between LTNP and STNP.
   Moreover, we performed machine learning to further identify and verify
   the potential biomarkers for different disease severity. These striking
   results provided the potential biomarkers to learn about the mechanism
   of COVID-19, particularly in relation to AS, LTNP, and STNP.
   Furthermore, these findings have the potential to identify diagnostic
   and treatment targets for SARS-CoV-2 infection.

Methods

Ethics statement

   This study was approved by the Ethics Committee of the First Affiliated
   Hospital of Xi’an Jiaotong University (XJTU1AF2020LSK-015) and the
   Renmin Hospital of Wuhan University (WDRY2020-K130). All participants
   enrolled in this study provided written informed consent by themselves
   or their surrogates. The high throughput sequencing of plasma samples
   was performed on existing samples collected during standard diagnostic
   tests, posing no extra burden to patients.

Case definition and study cohort

   The definition and classification of all COVID-19 patients in this
   study follow the Guidelines of the World Health Organization and the
   “Guidelines on the Diagnosis and Treatment of the Novel Coronavirus
   Infected Pneumonia” developed by the National Health Commission of
   People’s Republic of China ([49]8–[50]10). This study cohort included
   220 plasma samples derived from 166 individuals, consisting of healthy
   controls (HC, n=62), asymptomatic infections (AS, n=16), and
   symptomatic patients (SM, n=88). Symptomatic patients consisted of
   moderate diseases (MD, n=42) and severe diseases (SD, n=46). Then
   according to the time of a positive nucleic acid test, the individuals
   in the SM group were divided into 17 long-term nucleic acid test
   positive (LTNP, ≥60 days) and 34 short-term nucleic acid test positive
   (STNP, ≤45 days) individuals. In this study, based on the clinical
   observation that most of the COVID-19 patients hospitalized in the
   Renmin Hospital in Wuhan tested negative for the nucleic acid test
   within 45 days, we therefore defined the STNP as ≤45 days whereas the
   LTNP was ≥60 days. The demographic features, clinical laboratory
   testing results and other relevant information are provided in
   [51]Supplementary Table 1 .

Blood sample collection and plasma preparation

   The peripheral blood was collected into the standard EDTA-K2 Vacuette
   Blood Collection Tubes (Jiangsu Yuli Medical Equipment Co., Ltd, China;
   Cat.Y09012282) and stored at room temperature or 4°C until processed,
   and the sample storage time did not exceed 12 hours. The plasma was
   prepared after centrifugation of the whole blood sample at 2500 rpm for
   20 minutes and stored in the -80°C freezer until used for the studies.
   All the experimental procedures were completed inside a biosafety level
   2 (BSL-2) laboratory at the Department of Clinical Diagnostic
   Laboratories, Renmin Hospital of Wuhan University. A total of 220
   plasma samples were collected from 62 HC and 104 COVID-19 at different
   time-points from some of these individuals.

   For lipid compounds, samples were thawed on ice, whirled for around
   10s, and then centrifuged for 3000 rpm at 4°C for 5 min. We took 50 μL
   of one sample and homogenized it with 1 mL mixture (include methanol,
   MTBE and internal standard mixture). We whirled the mixture for 2 min
   and then added 500μl of water and whirled the mixture for 1 min and
   centrifuged it for 12,000 rpm at 4°C for 10 min. We extracted 500 μL of
   supernatant and concentrated it. We dissolved powder with 100 μL mobile
   phase B and then stored it at -80°C. Finally, we moved the dissolving
   solution into the sample bottle for LC-MS/MS analysis.

HPLC conditions

   For lipid compounds, the sample extracts were analyzed using an
   LC-ESI-MS/MS system (UPLC, Shim-pack UFLC SHIMADZU CBM 30A system,
   [52]https://www.shimadzu.com/; MS, QTRAP^® System,
   [53]https://sciex.com/). The analytical conditions were as follows
   UPLC: column, Waters ACQeITY UPLC HSS T3 C18 (1.8 µm, 2.1 mm*100 mm);
   column temperature, 40°C; flow rate, 0.4 mL/min; injection volume, 5μL;
   solvent system, water (0.04% acetic acid): acetonitrile (0.04% acetic
   acid); gradient program, 95:5 V/V at 0 min, 5:95 V/V at 11.0 min, 5:95
   V/V at 12.0 min, 95:5 V/V at 12.1 min, 95:5 V/V at 14.0 min. The
   effluent was alternatively connected to an ESI-triple quadrupole-linear
   ion trap (QTRAP)-MS.

ESI-Q TRAP-MS/MS

   LIT and triple quadrupole (QQQ) scans were acquired on a triple
   quadrupole-linear ion trap mass spectrometer (QTRAP), QTRAP^® LC-MS/MS
   System, equipped with an ESI Turbo Ion-Spray interface, operating in
   positive and negative ion mode, and controlled by Analyst 1.6.3
   software (Sciex). For lipid compounds, the following applied: source
   temperature 550 °C; ion spray voltage (IS) 5500 V (positive), -4500 V
   (negative); ion source gas 1 (GS1), gas 2 (GS2), curtain gas (CUR) were
   set at 55, 60, and 25 psi, respectively. The collision gas (CAD) was
   high. Instrument tuning and mass calibration were performed with 10 and
   100 μmol/L polypropylene glycol solutions in QQQ and LIT modes,
   respectively. DP and CE for individual MRM transitions was done with
   further DP and CE optimization. A specific set of MRM transitions were
   monitored for each period according to the metabolites eluted within
   this period in both hydrophilic and hydrophobic compounds.

Statistical analysis and machine learning

   The mass spectrum data were processed in Analyst 1.6.3 software. The
   characteristic ions of each substance were screened by QQQ. The
   MultiQuant software was used to open the mass spectrum file of the
   samples, and the chromatographic peaks detected by each metabolite in
   different samples were integrated and corrected according to the
   retention time and peak type information of metabolites. The
   statistical analysis was performed using the first blood sample
   collected from the participants. Orthogonal Partial Least
   Squares-Discriminant Analysis (OPLS-DA) was used to evaluate the
   statistical significance in different groups. Metabolites were selected
   using the following criteria: a variable importance in projection (VIP)
   value ≥ 1 and Fold change (FC)≥1.5 or ≤0.67. The normalized
   quantitation of metabolites was used as modeling data for every two
   compared groups. Then, 75% of the modeling data was selected as the
   training cohort, and the rest was used in the testing cohort. From the
   training set, we selected important lipids via machine learning using
   the xgboost method with the R package “xgboost”(version 1.4.1.1)
   ([54]11). For analysis of the five classes, 28 subclasses, carbon chain
   length, and degree of saturation of lipids in five compared groups, we
   used independent t-tests when the data were normally distributed and
   homoscedasticity; otherwise, the Mann-Whitney was used. The statistical
   analyses were performed in SPSS version 18.0 software. The data
   statistics used the mean values of raw intensities of lipid molecules
   with five classes, 28 subclasses, and different lengths of carbon chain
   (unsaturation) in different classes.

Results

Lipid profiling of COVID-19 plasma

   In this study, lipidomics analysis was conducted on 220 plasma samples
   obtained from a total of 166 individuals. The sample distribution
   included 62 healthy controls, 16 asymptomatic infections, 42
   individuals with moderate diseases, and 46 individuals with severe
   diseases. Additionally, there were repeated samples from 5 healthy
   controls, 12 individuals with moderate diseases, and 23 individuals
   with severe diseases, which were collected twice or more. The detailed
   descriptions of 166 individuals are shown in [55]Table S1 . We used
   OPLS-DA to analyze the 220 plasma samples. In total, we identified five
   major lipid classes [fatty acyls (FA), glycerophospholipids (GP),
   glycerolipids (GL), sphingolipids (SP), and sterol lipids (ST)]
   containing 28 lipid subclasses [carbocyclic fatty acids (CAR), free
   fatty acid (FFA), diglyceride (DG); monoglyceride (MG),
   lyso-phosphatidic acid (LPA), lyso-phosphatidylcholine (LPC),
   triglyceride (TG), ether-linked lyso-phosphatidyl-cholines (LPC-O),
   lyso-phosphatidylethanolamine (LPE), lyso-phosphatidylglycerol (LPG),
   lyso-phosphatidylinositol (LPI), lyso-phosphatidylserine (LPS),
   phosphatidic acid (PA), phosphatidylcholine (PC), cholesterylesters
   (CE), ether-linked phosphatidyl-cholines (PC-O),
   phosphatidylethanolamine (PE), phosphatidylglycerol (PG),
   phosphatidylinositol (PI), phosphatidylserine (PS), ceramides
   (Cer/Cert/Cerm), ceramides phosphate (CerP), and sphingomyelin],
   totaling 732 lipid molecules; detailed information of lipids is listed
   in [56]Table S2 . We analyzed differentially expressed lipids (DELs) in
   five compared groups, including COV vs HC, AS vs HC, SM vs AS, SD vs
   MD, and LTNP vs STNP ([57] Table S3 ).

Lipids associated with SARS-COV-2 infection

   The 50 lipids were at significantly different levels in COVID-19
   patients compared to healthy controls (VIP ≥ 1 and FC ≥ 1.5 or FC ≤
   0.67) ([58] Figures 1C, D ). Further, the DELs were used to perform the
   KEGG pathway analysis. The result suggested that the KEGG pathways of
   DELs were mainly involved in glycerophospholipid metabolism (hsa00564)
   and glycerolipid metabolism (hsa00561) ([59] Supplementary Figure 5A ).

Figure 1.

   [60]Figure 1
   [61]Open in a new tab

   Lipids associated with SARS-COV-2 infections. (A) The raw intensity of
   28 subclasses of lipids in COV vs HC. (B) The raw intensity of
   different degrees of unsaturation for FA, GL, GP, SP, and ST classes in
   COV vs HC. (C) The OPLS-DA scores plot of COV vs HC. (D) the volcano
   plot of COV vs HC. Red dots represent the upregulated lipids (FC≥1.5,
   VIP≥1); blue dots represent the downregulated lipids (FC ≤ 0.67,
   VIP≥1); gray dots represent lipids without significant changes
   (0.67<FC<1.5, VIP<1). (E) The important lipids prioritized by xgboost
   analysis. (F) The heatmap of 50 differentially expressed lipids in COV
   vs HC. (G) Receiver operating characteristic (ROC) and performance of
   the xgboost model in the training set. (H) the raw intensity of
   3-Hydroxy-decenoyl-carnitine and Lauroyl-carnitine.

   Moreover, compared with the healthy controls, 9 of 28 lipids subclasses
   showed a statistical difference in COVID-19 patients, including LPS,
   LPI, LPC-O, PA, PS, PC-O, CAR, Cert, and MG ([62] Figure 1A ). In
   addition, GL with unsaturation greater than 5, GP with unsaturation of
   6, and ST with unsaturation of 3, 5, and 6 showed high average levels
   in COV ([63] Figure 1B ). GP with unsaturation of 0, 2, and 4, as well
   as ST with unsaturation of 2, showed low average levels in COV. Our
   statistical results indicate that short-chain FA and GP, FA (C=14), GL
   (C=32), GP (C=34), and ST (C=18) show low average levels in COV. FA
   (C=12) and long-chain ST show high average levels in COV ([64]
   Supplementary Figure 3A ).

   We built an xgboost machine learning model based on intensity of DELs
   from 62 HC and 104 COVID-19 patients, leading to prioritization of 11
   important lipids ([65] Figure 1E ). This model reached an area under
   the curve (AUC) of 0.9989 in the training set, and only 4 of a total of
   124 samples (accuracy=0.9677) were classified incorrectly ([66]
   Figure 1G ). In the test set, 37 of 42 were correctly classified
   (accuracy=0.8810), and the AUC of this model can reach 0.9435 ([67]
   Supplementary Figure 4 ).

Lipids associated with asymptomatic infections of SARS-CoV-2

   For AS vs HC, we selected a total of 81 DELs (VIP ≥ 1 and FC ≥ 1.5 or
   FC ≤ 0.67) ([68] Figures 2B, C ). The KEGG pathways analysis using DELs
   suggested that pathways were mainly involved in glycerophospholipid
   metabolism (hsa00564) and glycerolipid metabolism (hsa00561) ([69]
   Supplementary Figure 5B ).

Figure 2.

   [70]Figure 2
   [71]Open in a new tab

   Lipids associated with asymptomatic infection of SARS-COV-2. (A) The
   raw intensity of 28 subclasses of lipids in AS vs HC. (B) The OPLS-DA
   scores plot of AS vs HC. (C) the volcano plot of AS vs HC. Red dots
   represent the upregulated lipids (FC≥1.5, VIP≥1); blue dots represent
   the downregulated lipids (FC ≤ 0.67, VIP≥1); gray dots represent lipids
   without significant changes (0.67<FC<1.5, VIP<1). (D) The important
   lipids prioritized by xgboost analysis. (E) The heatmap of 81
   differentially expressed lipids in AS vs HC. (F) Receiver operating
   characteristic (ROC) and performance of the xgboost model in the
   training set. (G) the raw intensity of PC(O-20:1/20:1), PC(O-20:1/20:0)
   and PC(O-18:0/18:1).

   Notably, in individuals with AS, there were significant decreases
   observed in various subclasses of lipids. Specifically, the SP
   subclasses, including Cer, Cert, CerP, Cer-Hex, and sphingomyelin; the
   ST subclasses, including CE; GP subclasses, including PS, PA, PI, PE,
   PG, PC-O, LPI, LPA, LPS, LPG, and LPC-O; the GL subclasses, including
   DG and MG; and the FA subclass, including CAR, showed marked decreases
   when compared to HC ([72] Figure 2A ). Compared to HC, the average
   levels of almost all GP with different unsaturation degrees in AS are
   significantly downregulated. At the same time, the average levels of ST
   and SP with unsaturation degrees equal to 1 and 2 are lower in AS ([73]
   Supplementary Figure 2A ). In terms of chain length, the average levels
   of medium-chain and long-chain GP, medium-chain GL, short-chain ST, ST
   (C=18), and FA (C=14) are significantly lower in the AS group ([74]
   Supplementary Figure 3B ). As for SP, the average levels of
   short-chain, medium-chain, and long-chain are all low in AS.

   We built an xgboost machine learning model based on intensity of DELs
   from 62 HC and 16 asymptomatic infections, leading to prioritization of
   eight important lipids ([75] Figure 2D ). This model reached an AUC of
   1 in the training set, and two of the training set samples
   (accuracy=0.9655) were classified incorrectly ([76] Figure 2F ). In the
   test set, 18 of 20 were correctly classified (accuracy=0.9000) ([77]
   Supplementary Figure 4 ) and the AUC is 0.9297.

Lipids associated with symptomatic infection of SARS-CoV-2

   The lipids were divided into five classes, four (GL, GP, SP, and ST) of
   which show statistical significance in SM and AS ([78] Supplementary
   Figure 1A ). Then, the 28 subclasses of five lipids classes were
   analyzed to explore the lipid subclasses differences and consisted of
   the results of five classes, CE, and all subclasses of SP (Cer,
   Cer-Hex, Cerm, CerP, Cert, and sphingomyelin), and GL (DG, TG, and MG)
   exhibiting differences in SM and AS ([79] Figure 3A ). Besides, 8 of 15
   SP subclasses showed a significant difference in SM and AS, including
   LPA, LPC-O, LPS, PC, PC-O, PG, PI, and PS. We also found that lipids
   with different carbon chain lengths and degrees of unsaturation show
   significant changes in SM and AS. Compared to AS, there are statistical
   differences in the unsaturation of GP, GL, SP, and ST in both SM and AS
   ([80] Supplementary Figure 2B ). The average levels of short-, medium-,
   and long-chain GL, GP, SP, and ST are significantly upregulated in SM
   ([81] Supplementary Figure 3C ).

Figure 3.

   [82]Figure 3
   [83]Open in a new tab

   Lipids associated with asymptomatic and symptomatic COVID-19. (A) The
   raw intensity of 28 subclasses of lipids in SM vs AS; (B) The OPLS-DA
   scores plot of SM vs AS. (C) the volcano plot of SM vs AS. Red dots
   represent the upregulated lipids (FC≥1.5, VIP≥1); blue dots represent
   the downregulated lipids (FC ≤ 0.67, VIP≥1); gray dots represent lipids
   without significant changes (0.67<FC<1.5, VIP<1). (D) The important
   lipids prioritized by xgboost analysis. (E) The heatmap of 65
   differentially expressed lipids in SM vs AS. (F) Receiver operating
   characteristic (ROC) and performance of the xgboost model in the
   training set. (G) the raw intensity of PE(16:0/22:4) and 9,10-DiHOME.

   Through OPLS-DA (VIP ≥1) and fold change (FC ≥ 1.5 or FC ≤ 0.67), we
   discovered a total of 65 differentially expressed lipids in SM and AS
   ([84] Figures 3B, C , [85]E ; [86]Table S3 ). To research the function
   of lipids, six significant KEGG pathways were found, including
   glycerophospholipid metabolism (hsa00564) and
   glycosylphosphatidylinositol (GPI)-anchor biosynthesis (hsa00563) ([87]
   Supplementary Figure 5 ; [88]Table S4 ). In addition, an xgboost
   machine learning model was built based on the 65 DELs derived from the
   88 symptomatic infections and asymptomatic infections, leading to
   prioritization of seven lipids ([89] Figure 3D ). This model reached an
   AUC of 0.9874 and an accuracy of 0.8267 in the training set ([90]
   Figure 3F ). We then tested the model on the test cohort, and the AUC
   value was 0.9545 ([91] Supplementary Figure 4 ).

Lipids associated with disease severity of COVID-19

   Significant differences were observed in three out of five lipid
   classes between SD and MD groups. These were GP (P=0.032), SP
   (P=0.014), and ST (P=0.020) ([92] Supplementary Figure 1A ).
   Furthermore, among the 28 lipid subclasses, 12 showed statistically
   significant differences between the SD and MD groups. These subclasses
   were MG, LPC, LPC-O, LPG, LPI, LPS, PA, PC-O, PE-P, CerP,
   sphingomyelin, and CE. ([93] Figure 4A ). The statistical results of
   unsaturation indicate that most FA and ST have low unsaturation levels
   (unsaturation ≤ 3). Additionally, FAs with unsaturation levels of 1 and
   2 have higher average levels in the SD group, while monounsaturated and
   polyunsaturated GPs and SPs, as well as STs with unsaturation levels of
   0 or 6, have higher average levels in the MD group ([94] Figure 4B
   ).Furthermore, our statistical analysis shows that compared to the SD
   group, the MD group has higher average levels of long-chain and
   short-chain GP and SP, SP (C=32), and ST (C=18) ([95] Supplementary
   Figure 3D ). On the other hand, the SD group has a higher average level
   of long-chain FA ([96] Supplementary Figure 3D ).

Figure 4.

   [97]Figure 4
   [98]Open in a new tab

   Lipids associated with the disease severity of COVID-19. (A) The raw
   intensity of 28 subclasses of lipids in SD vs MD; (B) The raw intensity
   of different degrees of unsaturation for FA, GL, GP, SP and ST classes
   in SD vs MD; (C) the OPLS-DA scores plot of SD vs MD. (D) the volcano
   plot of SD vs MD. Red dots represent the upregulated lipids (FC≥1.5,
   VIP≥1); blue dots represent the downregulated lipids (FC ≤ 0.67,
   VIP≥1); gray dots represent lipids without significant changes
   (0.67<FC<1.5, VIP<1). (E) The important lipids prioritized by xgboost
   analysis. (F) The heatmap of 21 differentially expressed lipids in SD
   vs MD. (G) Receiver operating characteristic (ROC) and performance of
   the xgboost model in the training set. (H) the raw intensity of 6
   keto-PGF1α and LPC(12:0/0:0).

   For the detailed compounds, we found 21 DELs in SD and MD via OPLS-DA
   (VIP ≥1) and FC (FC ≥ 1.5 or FC ≤ 0.67) ([99] Figures 4C, D ;
   [100]Table S3 ). Furthermore, these DELs were mainly enriched in
   glycerolipid metabolism (hsa00561) and glycerophospholipid metabolism
   (hsa00564) ([101] Supplementary Figure 5 ; [102]Table S4 ). Based on
   the significant lipids, the xgboost model was established using 17
   lipids with high importance scores ([103] Figure 4E ). This model
   reached an AUC of 0.9962, and the accuracy of the training set samples
   was 0.9076 ([104] Figure 4G ). We tested the model on the test cohort,
   and the AUC value was 0.8939 ([105] Supplementary Figure 4 ).

Lipids associated with the viral persistence of SARS-CoV-2 infection

   For LTNP vs STNP, we identified four classes (FA, GL, GP and SP) and
   six subclasses (Eicosanoid, FFA, PE, Cer, DG, and TG) totaling 15
   differentially expressed lipids, including 12 upregulated and 3
   downregulated lipids (VIP ≥ 1 and FC ≥ 1.5 or FC ≤ 0.67) ([106]
   Figure 5B ). In addition, we found that in the STNP group, there were
   higher average levels of ST with a carbon chain length of 20 as well as
   ST with an unsaturation value of 4 ([107] Supplementary Figure 3D ).
   Furthermore, KEGG results showed that these DELs were enriched in the
   glycerolipid metabolism (hsa00561) and glycerophospholipid metabolism
   pathways (hsa00564) ([108] Supplementary Figure 5 ; [109]Table S4 ).

Figure 5.

   [110]Figure 5
   [111]Open in a new tab

   Lipids associated with the duration of a positive nucleic acid test for
   SARS-CoV-2 infection. (A) The OPLS-DA scores plot of LTNP vs STNP. (B)
   the volcano plot of LTNP vs STNP. Red dots represent the upregulated
   lipids (FC≥1.5, VIP≥1); blue dots represent the downregulated lipids
   (FC ≤ 0.67, VIP≥1); gray dots represent lipids without significant
   changes (0.67<FC<1.5, VIP<1). (C) The important lipids prioritized by
   xgboost analysis. (D) The heatmap of 15 differentially expressed lipids
   in LTNP vs STNP. (E) Receiver operating characteristic (ROC) and
   performance of the xgboost model in the training set. (F) the raw
   intensity of TG (14:1/14:1/18:2) and FFA (4:0).

   We also supposed DELs could be used as predictor for LTNP patients from
   STNP patients. A total of 17 LTNP and 34 STNP patients were included in
   this analysis. We built an xgboost machine learning model based on 15
   DELs and select 10 Lipids with high importance scores ([112]
   Figures 5A, C, D ). This model reached an AUC of 0.9535, and the
   accuracy was 0.8421 in the training set ([113] Figure 5E ). We then
   tested the model on an independent cohort of five LTNP and nine STNP
   patients; the AUC value was 0.7667 ([114] Supplementary Figure 5 ).

Discussion

   In this study, we applied LC-ESI-MS/MS to analyze the lipid profiles of
   220 plasma samples. They were also divided into five control groups and
   found that lipidome is highly dynamic in different groups. In total,
   732 lipids were found in our analysis ([115] Table S2 ), and 175 DELs
   were detected in five compared groups ([116] Table S3 ). In addition,
   combined with machine learning, multiple biomarkers were identified to
   distinguish the different COVID-19 severity from healthy controls.

   Compared with HC, though the raw intensity of five lipid classes showed
   no obvious change, the subclasses represented marked transformation in
   COV. To further explore GP in COV, we found that GP subclasses like LPS
   and PS are both downregulated in COV ([117] Figure 1A ).

   It was reported that PS was a global immunosuppressive signal in
   efferocytosis, infectious disease, and cancer ([118]12). LPS could
   increase endogenous TNF-α expression, generating tolerogenic dendritic
   cells to exhaust effector CD4 T cells ([119]13, [120]14). The possible
   explanation of the relatively lower intensity level of PS in COV might
   reflect the immune response in SARS-COV-2 infections. We also explored
   the unsaturation degree and carbon-chain length influence of FA in COV;
   we found fewer short-chain FA (SCFAs) in COV ([121] Supplementary
   Figure 3A ). Dr. Takabayashi cultured nasal epithelial cells and
   exposed them to SCFA and saw that the SCFAs had suppressed ACE2
   expression ([122]15). As is known, RBD of SARS-COV-2 can bind to ACE2
   to enter the host cell. For COVID-19 patients, downregulated SCFAs
   might encourage binding between ACE2 and SARS-COV-2.

   Combined with the fold change and machine learning result,
   acylcarnitine like 3-Hydroxy-decenoyl-carnitine (FC=0.49, C=12) and
   Lauroyl-carnitine (FC=0.57, C=12) play an important role in
   distinguishing COV and HC ([123] Figure 1H ). We also found the
   acylcarnitine intensity is obviously different in COV compared to HC
   ([124] Figure 1F ). Acylcarnitines serve as carriers to transport
   activated long-chain fatty acids into mitochondria for β-oxidation
   ([125]16) as energy powerhouse for cell activities. For COVID-19
   patients, downregulated acylcarnitines may suggest that infected people
   try to prevent virus replication by reducing ATP production, which is
   consistent with our result that show there are no obvious changes in
   long-chain FA ([126] Figure 1B ).

   In AS vs HC, raw intensity of GP, SP, and ST presented lower levels in
   AS. It is also disclosed that 20 of total 28 detected lipid subclasses
   show marked changes in AS. Though there are no obvious clinical
   symptoms in AS, the lipid profile changes dramatically ([127] Figure
   S1A ). For example, SP metabolites have key roles in the regulation of
   both trafficking and functions of immune cells ([128]17), and decreased
   levels of SP and GP were observed in both non-severe and severe
   COVID-19 patients. In the DELs, the fold change of 2-hexenoyl-carnitine
   (FC=4.42, C=6) is the highest. Unexpectedly, unlike downregulated
   acylcarnitines in COV vs HC, 2-hexenoyl-carnitine raw intensity is
   upregulated in AS, which is also present in COV vs HC DELs. One
   possible explanation is that the short-chain acylcarnitines are
   metabolites of lipids, and amino acids and carbohydrates are released
   in plasma ([129]18).The features, such as PC(O-20:1/20:1),
   PC(O-20:1/20:0) and PC(O-18:0/18:1), belonging to PC-O are selected by
   the xgboost model and are used to separate AS from HC ([130] Figure 2G
   ). From previous analysis we found serum levels of the PC-O were
   reduced in AS. As we known, PC was an important structural lipid in all
   cellular membranes, and PC-O was formed from the hydrolytic action of
   enzymes phospholipase A2, which was found in reasonable abundance in
   epithelial cells ([131]19). Meanwhile, budding is an essential step in
   the life cycle of enveloped viruses, and enveloped viruses acquire
   their lipid membrane from the cells from which they bud ([132]20).
   Therefore, deficiency of such PC lipids in AS may be the outcome of
   virus budding.

   We discovered four classes of lipids with significant differences in SM
   and AS. Among them, the dysregulation of GP and SP could result in a
   lipotoxic insult relevant to the pathophysiology of common metabolic
   diseases ([133]21). Li et al. discovered that the patients with
   metabolic diseases were at greater risk of developing into severe
   infections, and these comorbidities could also have an effect on the
   prognosis of COVID-19 ([134]22). In addition, there existed an
   interaction between the platelet activating factor (PAF) and
   sphingolipid biosynthetic pathways, which derived from a CoA
   independent transacetylase, mediating the transfer of an acetyl group
   from PAF in the synthesis of N acetylsphingosine ([135]23–[136]25). The
   PAF is an ether-glycerosphospholipid that is important for immune cell
   activation and plays a critical role in chronic inflammation, virus
   infection, and immune activation ([137]26). PE (18:1/16:0, 16:0/22:4
   and 22:6/16:0) was one of the leading important lipids in SM and AS. PS
   showed statistical significance in the 28-subclass analysis. It was
   shown that PE synergized with PS to enhance PS receptor-mediated virus
   entry ([138]27). The previous study suggested that the liposomes
   containing both PS and PE showed greater efficiency on the inhibition
   of the cells infected with a virus, such as Zika virus and Ebola,
   compared with the liposomes mixture separately composed of PS and PE
   ([139]28). In addition, our study found that 9,10-dihydroxyoctadecenoic
   acid (9,10-DiHOME) decreased in symptomatic infections compared with
   asymptomatic infections ([140] Figure 3G ), and the importance of
   9,10-DiHOME was also verified in machine learning. Lu et al. found that
   9,10-DiHOME was altered in patients with HBV-related liver diseases and
   derived from the cytochrome P450 (CYP450) pathways ([141]29).

   Through analysis of the severe infections and moderate infections,
   6-keto prostaglandin F1alpha (6 keto-PGF1α) was found to have the
   highest fold change among the compounds and the higher importance
   scores in machine learning. The deficiency of prostaglandin receptor
   may exacerbate the weight loss and delay in the viral clearance in the
   rats infected with the respiratory syncytial virus ([142]30). Besides,
   the murine model infected by HSV discovered that 6-keto-PGF1α had a
   correlation with HSV-infection and cytotoxicity, and HSV-infection
   could suppress the release of 6-keto-PGF1α ([143]31).
   Lysophosphatidylcholie (LPC) (12:0/0:0) had the highest VIP value and
   smallest fold change in SD and MD. Lysophosphatidylcholine had the
   function of reversibly arresting pore expansion during syncytium
   formation mediated by diverse viral fusogens ([144]32). In addition,
   LPC and some saturated fatty acids could induce dendritic cell
   maturation, and the dendritic cells showed impaired functionality in
   COVID-19 ([145]33, [146]34). Therefore, combined with the decreased
   expression level of LPC in SD compared with MD, it was suggested that
   the function and maturation of dendritic cells was destroyed in severe
   infections.

   For LTNP vs STNP, we detected four classes including six subclasses
   with a total of 15 DELs. For example, GL subclasses TG upregulated in
   LTNP, whereas FA subclasses FFA downregulated. GL is a major form of
   energy storage (mainly in the form of TG) ([147]35). As a hydrophobic
   component of the cell membrane, it plays an important role in
   structural function and has a dynamic impact on the signal mediators in
   intracellular and distant tissues ([148]35, [149]36). TG is associated
   with cardiovascular disease, atherosclerotic cardiovascular disease,
   and ischemic heart disease ([150]37–[151]39). It has been reported that
   increased TG levels have also been observed in patients with
   inflammatory disease and HIV-infected patients ([152]40, [153]41).
   Epidemiological studies have shown that elevated TG levels are
   independently associated with increased cardiovascular risk. Therefore,
   high intensity of TG might reflect pre-existing cardiovascular disease
   in LTNP which lead to the course time of COVID-19 more longer ([154]42,
   [155]43). FA are carbon chains with methyl groups at one end and
   carboxyl groups at the other end, and they are also the first line of
   defense against pathogens. Fatty acids are essential for viral
   infection because they provide the basis for various membrane lipids
   during viral proliferation, which plays an important role in immune and
   inflammatory responses ([156]44, [157]45). In our data, FFA
   downregulated in LTNP ([158] Figure 5F ), and thus dysfunctional immune
   or inflammatory responses might be the cause of positive long-term
   nucleic acid tests. In conclusion, our results suggest that the
   prolonged nucleic acid conversion time may be due to the patient’s own
   underlying disease or immune dysregulation.

   Furthermore, this study has some limitations. Firstly, there is a lack
   of follow-up on AS. Secondly, the study lacks subsequent experiments on
   cytokines. We plan to further conduct basic research based on our
   significant findings. Lastly, there is a lack of longitudinal studies
   on infected individuals. We hope that future research can delve deeper
   into these areas.

   In conclusion, our study presents a systematic lipidomic investigation
   of serum samples from 220 plasma samples of 62 healthy controls and 104
   COVID-19 patients and found multiple lipids that were associated with
   the virus infection, disease severity, and the viral persistence of the
   COVID-19 patients. Our findings may help us understand the potential
   contributions of lipid metabolites to the pathogenesis of COVID-19. We
   have demonstrated the potential of identifying novel biomarkers for the
   diagnosis and treatment of SARS-CoV-2 infection based on analysis of
   five compared groups of lipid metabolites. In conclusion, our results
   may provide useful diagnostic and therapeutic clues for COVID-19.

Data availability statement

   The datasets presented in this study can be found in online
   repositories. The names of the repository/repositories and accession
   number(s) can be found below: MTBLS7937 (Metabolights).

Ethics statement

   The studies involving human participants were reviewed and approved by
   the First Affiliated Hospital of Xi’an Jiaotong University
   (XJTU1AF2020LSK-015) and the Renmin Hospital of Wuhan University
   (WDRY2020-K130). The patients/participants provided their written
   informed consent to participate in this study.

Author contributions

   Conceptualization: CZ and BS. Methodology and Investigation: XL, XQ,
   JY, XW, HQ, FH, HB and YL. Resources: CZ, BS, BHZ, and KH. Writing: XL
   and XQ. Review and editing: XL, XQ and BS. Supervision: BS and JY.
   Funding: BS. Project administration: BS. All authors contributed to the
   article and approved the submitted version.

Funding Statement

   This study is supported in part by the Department of Science and
   Technology of Shaanxi Province (Grant No. 2020ZDXM2-SF-02) (CZ and BS)
   and operational funds from The First Affiliated Hospital of Xi’an
   Jiaotong University (CZ and BS).

Abbreviations

   DELs, Differentially Expressed Lipids; KEGG, Kyoto Encyclopedia Of
   Genes And Genomes; FA, Fatty Acyls; GP, Glycerophospholipids; GL,
   Glycerolipids; SP, Sphingolipids; ST, Sterol Lipids; CAR, Carbocyclic
   Fatty Acids; FFA, Free Fatty Acid; DG, Diglyceride; TG, Triglyceride;
   MG, Monoglyceride; LPA, Lyso-Phosphatidic Acid; LPC,
   Lyso-Phosphatidylcholine; LPE, Lyso-Phosphatidylethanolamine; CE,
   Cholesterylesters. LPG, Lyso-Phosphatidylglycerol; LPI,
   Lyso-Phosphatidylinositol; LPS, Lyso-Phosphatidylserine; PA,
   Phosphatidic Acid; PC, Phosphatidylcholine; PE,
   Phosphatidylethanolamine; PG, Phosphatidylglycerol; PI,
   Phosphatidylinositol; PS, Phosphatidylserine; Cer/Cert/Cerm, Ceramides;
   CerP, Ceramides Phosphate; LPC-O, Ether-linked
   lyso-phosphatidyl-cholines; PC-O, Ether-linked phosphatidyl-cholines.

Conflict of interest

   The authors declare that the research was conducted in the absence of
   any commercial or financial relationships that could be construed as a
   potential conflict of interest.

Publisher’s note

   All claims expressed in this article are solely those of the authors
   and do not necessarily represent those of their affiliated
   organizations, or those of the publisher, the editors and the
   reviewers. Any product that may be evaluated in this article, or claim
   that may be made by its manufacturer, is not guaranteed or endorsed by
   the publisher.

Supplementary material

   The Supplementary Material for this article can be found online at:
   [159]https://www.frontiersin.org/articles/10.3389/fimmu.2023.1221493/fu
   ll#supplementary-material
   Supplementary Figure 1

   (A) The raw intensity of FA, GL, GP, SP and ST classes in 5 compared
   groups. (B) The raw intensity of 28 subclasses for lipids in LTNP vs
   STNP. (C) The heatmap of fold change for FA, GL, GP, SP and ST classes
   in 5 compared groups; (D) the rising sun map of 732 lipids in 28
   subclasses and 5 classes.
   [160]Click here for additional data file.^ (629.7KB, jpg)
   Supplementary Figure 2

   The raw intensity of different degrees of unsaturation for FA, GL, GP,
   SP and ST classes in 3 compared groups, including AS vs HC (A), SM vs
   AS (B) and LTNP vs STNP (C).
   [161]Click here for additional data file.^ (621.4KB, jpg)
   Supplementary Figure 3

   The raw intensity of different carbon chain lengths of FA, GL, GP, SP
   and ST classes in 5 compared groups. The lipids of each class are
   defined according to our data as follows, FA: short chains (C<12),
   medium chain (C=12 or C=14), long chain (C>14); GL: short chain (C<32),
   medium chain (C=32 or C=34), long chain (C>34); GP: short chain (C<32),
   medium chain (C=32 or C=34), long chain (C>34); GL: short chain (C<32),
   medium chain (C=32 or C=34), long chain (C>34); SP: short chain (C<32),
   medium chain (C=32 or C=34), long chain (C>34); ST: short chain (C<18),
   medium chain (C=18 or C=20), long chain (C>20).
   [162]Click here for additional data file.^ (537.3KB, jpg)
   Supplementary Figure 4

   Receiver operating characteristic (ROC) and performance of the xgboost
   model in the test set for 5 compared groups.
   [163]Click here for additional data file.^ (296.4KB, jpg)
   Supplementary Figure 5

   The KEGG analysis of the differentially expressed lipids in 5 compared
   groups.
   [164]Click here for additional data file.^ (282.5KB, jpg)
   Supplementary Table 1

   the detailed characteristic information of samples used in this study.
   [165]Click here for additional data file.^ (31.2KB, xlsx)
   Supplementary Table 2

   Summary of lipids in this study.
   [166]Click here for additional data file.^ (1.1MB, xlsx)
   Supplementary Table 3

   the differentially expressed lipids in 5 compared groups.
   [167]Click here for additional data file.^ (230.9KB, xlsx)
   Supplementary Table 4

   the KEGG results for differentially expressed lipids in 5 compared
   groups.
   [168]Click here for additional data file.^ (16.1KB, xlsx)

References