Results

Blood Routine and Circulating Markers of Systemic Inflammation Indicated Dysregulated Immune Response in COVID-19
Demographic and laboratory findings of the 50 recruited COVID-19 patients (Tables 1 and S1) were in good agreement with published literature on clinical characteristics of COVID-19 in China (Guan et al., 2020). Reductions in lymphocyte count (LC) (p < 0.0001) (Table S1), particularly in the numbers of T lymphocytes and CD4+ T lymphocytes (Figure S1A), were associated with increasing disease severity. Our observations were aligned with previous findings on laboratory-confirmed COVID-19 cases in Wuhan for which particularly drastic reductions in CD4+ T cell counts indicative of dysregulated immune response were noted, especially in severe COVID-19 patients (Qin et al., 2020). Indices of systemic inflammation, including C-reactive protein (CRP) (p = 0.0003), interleukin-6 (IL-6) (p = 0.0958), erythrocyte sedimentation rate (ESR) (p = 0.0008), serum ferritin (SF) (p = 0.0001), and procalcitonin (p = 0.0171), exhibited progressive increases as disease severity increased (Figure S1B; Table S1). We utilized a combination of targeted lipidomics (Lu et al., 2019) and untargeted metabolomics optimized in-house for screening human plasma samples. Our untargeted metabolomics detected an initial pool of 1,552 metabolite peaks with coefficients of variations <20% across quality control samples after subtraction of background noise. After proceeding to structural confirmation based on tandem mass spectrometry (MS/MS) spectra, the consolidated plasma metabolome finally contained 1,002 metabolites (598 lipids and 404 polar metabolites) quantitated using 71 internal standards.
Table 1 Demographics and Baseline Characteristics of COVID-19 Patients
Total (n = 50) Mild (n = 18) Moderate (n = 19) Severe (n = 13) p Value
Characteristics
Age, years 43.0 (34.3–53.8) 32.0 (22.3–40.0) 45.0 (38.0–53.5) 50.0 (40.0–78.0) 0.001
Onset of symptom to hospital admission, days 5.0 (3.0–8.0) 4.0 (2.0–5.0) 6.0 (4.0–7.0) 8.0 (5.0–10.0) 0.023
Duration of hospitalization, days 19.0 (11.0–27.0) 8.5 (5.3–19.0) 19.0 (13.0–27.0) 27.0 (20.0–36.0) 0.02
Sex 0.525
 Men 30 (60%) 9 (50%) 12 (63%) 9 (69%)
 Women 20 (40%) 9 (50%) 7 (37%) 4 (31%)
Exposure to Wuhan 26 (52%) 13 (72%) 7 (37%) 6 (46%) 0.087
Any comorbidity 18 (36%) 3 (17%) 7 (37%) 8 (62%) 0.037
 Hypertension 8 (16%) 0 3 (16%) 5 (38%) 0.016
 Diabetes 5 (10%) 1 (6%) 2 (11%) 2 (15%) 0.664
 Malignancy 1 (2%) 0 0 1 (8%) 0.234
 HIV 1 (2%) 0 0 1 (8%) 0.234
 Chronic liver disease 2 (4%) 0 0 2 (15%) 0.052
Death 2 (4%) 0 0 2 (15%) 0.052
Signs and Symptoms
Fever 42 (84%) 13 (72%) 18 (95%) 11 (85%) 0.175
Highest temperature,°C 0.702
 <37.3 8 (16%) 5 (28%) 1 (5%) 2 (15%)
 37.3–38.0 12 (24%) 4 (22%) 5 (26%) 3 (23%)
 38.1–39.0 22 (44%) 6 (33%) 10 (53%) 6 (46%)
 >39.0 8 (16%) 3 (17%) 3 (16%) 2 (15%)
Cough 31 (62%) 8 (44%) 12 (63%) 11 (85%) 0.075
Expectoration 15 (30%) 3 (17%) 5 (26%) 7 (54%) 0.076
Rhinorrhoea 1 (2%) 1 (6%) 0 0 0.404
Myalgia or fatigue 23 (46%) 5 (28%) 7 (37%) 11 (85%) 0.004
Nausea and vomiting 0 0 0 0
Sore throat 7 (14%) 3 (17%) 2 (11%) 2 (15%) 0.853
Shortness of breath 6 (12%) 0 1 (5%) 5 (38%) 0.003
Chest pain 1 (2%) 0 0 1 (8%) 0.234
Diarrhea 5 (10%) 3 (17%) 2 (11%) 0 0.31
Data were median (interquartile range, IQR), or n (%). p values comparing mild, moderate, and severe were computed using the χ2 test (sex, exposure to Wuhan, comorbidities, death, fever, highest temperature, cough, expectoration, rhinorrhea, myalgia or fatigue, nausea and vomiting, sore throat, shortness of breath, chest pain, and diarrhea), one-way ANOVA (onset of symptom to hospital admission, duration of hospitalization, and BMI), or Kruskal-Wallis H test (age). COVID-19, coronavirus disease 2019. See also Figure S1.

An Integrated Panel of Plasma Lipids and Polar Metabolites Effectively Distinguished COVID-19 Patients from Healthy Controls with AUC = 0.975
To derive a plasma metabolite panel for distinguishing between COVID-19 patients and healthy controls, significant variables (p < 0.05) after adjustment of age, sex, and BMI were included in a starting pool. In an iterative process, ten sets of variables were established from the starting pool (Figure 1 ), and one representative variable was finally selected from each set based on (1) smallest p value and (2) reported biological function from a PubMed search. A final panel of ten metabolites was generated, which separated healthy controls from COVID-19 patients with AUC = 0.975 in a logistic regression model with leave-one-out (LOO) cross validation. Among these metabolites, sphingosine-1-phosphate (S1P) was reduced (p < 0.001) in COVID-19, and its level was raised (p = 0.0065) at hospital discharge relative to admission in a small subset of patients followed longitudinally (Figure S2). S1P generation via sphingosine kinase-2 in monocyte-derived macrophages was recently shown to promote the resolution of inflammation by alveolar macrophages in acute lung injury (Joshi et al., 2020). Biliverdin, the oxidized form of bilirubin, is part of the redox cycle constituting the primary physiologic function of bilirubin as a cytoprotective antioxidant (Baranano et al., 2002). Increases in biliverdin (p = 0.0077) in COVID-19 probably indicated enhanced oxidative stress in disease state, and its level was reduced longitudinally at hospital discharge with marginal significance (p = 0.0558) (Figure S2). Plasma 5-hydroxy-tryptophan was elevated in COVID-19 (p = 0.0203), and its depletion via induction of the indolamine 2, 3-dioxygenase pathway in human alveolar carcinoma type II-like cells was previously reported to suppress the growth of parainfluenza virus type 3 (Rabbani and Barik, 2017). As for lipids, increases in lysophopholipids including lysophosphatidic acid (LPA) 18:1 (p = 0.0249) and lysophosphatidylcholine (LysoPC) 18:1 (p = 0.0350) were observed in COVID-19, while neutral lipids including medium-chain TAG 48:1(18:0) (p = 0.0190), long-chain TAG 60:3(18:1) (p = 0.0390), and DAG 34:1(16:1/18:0) (p = 0.0010) were generally reduced. On the other hand, sphingolipids such as SM d18:1/18:1 (p = 0.002) and GM3 d18:1/25:0 (p = 0.0036) were generally increased with disease.
Figure 1 Plasma Panel for Differentiating COVID-19 Patients from Heathy Controls
Overview of selection scheme for plasma metabolite panel to differentiate COVID-19 (n = 50) patients from healthy controls (n = 26).
(A) From a total of 1,002 variables measured (598 lipids and 404 polar metabolites), variables with p < 0.05 between healthy controls and COVID-19 patients after adjustment for age, sex, and BMI were sieved out to form a starting pool comprising 322 variables.
(B) A starting variable with the lowest p value was selected, and variables with significant correlations (p < 0.05) to the starting variable selected were removed from consideration. From remaining variables in the starting pool, the next starting variable with the second lowest p value was identified, and the process was repeated in an iterative fashion until all variables in the starting pool were exhausted. This process generated a list of ten variables.
(C) Variables with significant correlations (p < 0.05) to each of the selected variables were added together to form ten established sets. To select a representative variable from each established set, the variable with the smallest p value and with reported biological function from a Pubmed search was chosen. A final panel of ten plasma metabolites, including S1P d18:1, SM d18:1/18:1, TAG60:3(18:1), LPA 18:1, biliverdin, TAG 48:1(18:0), DAG34:1(16:1/18:0), GM3 d18:1/25:0, lysoPC18:1, and 5-hydroxy-L-tryptophan, was generated, which distinguished between healthy controls and COVID-19 patients with an area under the curve (AUC) = 0.975 in a logistic regression model with leave-one-out (LOO) cross-validation. Boxplots for the ten selected metabolites in the final panel were illustrated and p values were indicated on top of each boxplot. Levels of polar metabolites measured using untargeted metabolomics were presented as corrected intensities, and lipids quantitated using targeted lipidomics were presented in nanomoles of lipids per liter (nmol/L) plasma.
See also Figures S2–S4.

Plasma Metabolome Changes Indicated Perturbed Oxidative Pathways of Cellular Energy Production
As baseline characteristics such as age, sex, and BMI were known to significantly influence plasma lipidomes and metabolomes, we constructed logistic regression models with these covariates to search for significant metabolites distinctly associated with different stages of COVID-19 (Figure S3). Several acylcarnitines, such as palmitoylcarnitine, stearoylcarnitine, and oleoylcarnitine, were reduced in COVID-19 (Figure S3, β-oxidation pathway). Reduced circulating levels of acylcarnitines may indicate attenuated entry of fatty acyls into the mitochondria for β-oxidation. Metabolites constituting the tricarboxylic acid (TCA) cycle were generally reduced in COVID-19 (Figure S3, TCA pathway). The reductions in polar metabolites participating in oxidative pathways of energy production (β-oxidation and TCA cycle), particularly in severe patients, may indicate metabolic response to declining lung functions and limiting blood oxygen to lower reliance on oxygen for cellular energy production. Lactate dehydrogenase was found to increase (p = 0.028) with increasing disease severity (Table S1), but plasma lactate was not significantly altered in COVID-19 compared to controls. The overall reductions in these polar metabolites might also reflect a response to change in nutrition, especially in severe patients. Although only 2 out of 13 severe patients were on mechanical ventilation at the point of blood collection (Table S3), a loss in appetite denotes a common general symptom of COVID-19 (Lechien et al., 2020). Interestingly, itaconic acid, a macrophage-specific metabolite derived from cis-aconitate, was progressively reduced with COVID-19 severity (Figure S3, TCA). Itaconate levels were previously reported to positively correlate with the expression of immune-responsive gene 1 (IRG1) in both human and mouse immune cells (Michelucci et al., 2013). Progressive reductions in sulfated steroids were also observed with increasing disease severity (Figure S3, steroid pathway). Numerous amino acids, including tryptophan, valine, proline, citrulline, and isoleucine, were significantly reduced in mild and moderate patients (Figure S3, amino acids). Decreases in specific amino acids were also previously observed in EVD fatalities (Eisfeld et al., 2017).

Plasma Lipidome Distinctly Associated with COVID-19 Resembles Exosomal Membrane Lipid Compositions
Lipids with overall false discovery rate (FDR) <0.05 were shortlisted and grouped according to major lipid classes for visual clarity in a forest plot (Figure 2 ). We observed reductions in major classes of plasma glycerophospholipids including phosphatidic acids (PAs), phosphatidylinositols (PIs), and PCs, with accompanying increases in their corresponding lysophospholipids (i.e., LPAs, LPIs, and LPCs) that indicate enhanced phospholipase A2 activity in COVID-19 patients. Cytosolic phospholipase A2α (cPLA2α) activation was reported to trigger pulmonary inflammation following pathogen infection (Bhowmick et al., 2017). PUFA-PEs were the only diacyl forms of glycerophospholipids that increased in COVID-19. Changes in phospholipidome (i.e., reductions in PCs and PIs and increases in PEs) observed in COVID-19 corroborated a previous study on plasma lipid alterations in EVD fatalities compared to survivors (Kyle et al., 2019). Among glycerophospholipids, PCs constitute the major membrane components of circulating lipoproteins (Cole et al., 2012), and PC-transfer protein (PC-TP) promotes cellular lipid efflux in nascent high-density lipoprotein (HDL) formation mediated by apolipoprotein A1 (apo-A1) (Baez et al., 2002), the major protein component of HDLs. PI was found to exhibit selective enrichment in plasma HDL fractions and was not detected in plasma lipoprotein-free fractions (Dashti et al., 2011). Changes in phospholipidome therefore suggested reductions in circulating HDLs as COVID-19 progresses, which was in agreement with the observed reductions in apo-A1 (p = 0.0687) as disease severity increases (Figure S1C). As for neutral lipids, marked reductions in several DAGs, with concomitant increases in FFAs (e.g., FFA18:1 and FFA 18:2), were observed in mild and moderate COVID-19 (Figure 2), while TAGs were significantly reduced only in mild cases (Figure S4). Increases in C18-FFAs and diminished TAGs were in agreement with previously reported circulating lipid changes associated with ARDS (Bursten et al., 1996; Maile et al., 2018). In stark contrast to the glycerophospholipid and neutral lipid pathways, sphingolipid classes of SMs and GM3s, e.g., GM3 d18:1/16:0, GM3 d18:0/16:0, GM3 d18:1/25:0, and GM3 d18:0/25:0, displayed progressive increases with increasing severity (Figures 2 and S4), possibly reflecting the augmented secretion of these lipids into the circulation. Of particular interest, lipidomes of exosomal membranes were previously shown to be specifically enriched in SMs (Stoorvogel et al., 2002) with diminished DAGs that cumulatively give rise to enhanced membrane rigidity (Laulagnier et al., 2004). Exosomes from lymphocytes also exhibited cell-type-specific enrichments in GM3s (Wubbolts et al., 2003) and PSs (Brügger et al., 2006). Thus, a gross overview of plasma lipidomic signatures distinctly associated with COVID-19, taking into account baseline cofounders, revealed a close resemblance to that of exosomal lipid compositions (i.e., enriched in SMs and GM3s with reduced DAGs).
Figure 2 Plasma Lipids Associated with Severity of COVID-19
Logistic regression model with covariates BMI, age, and sex was built with each lipid to search for significant variables that could predict disease severity of subjects (i.e., healthy control, and mild, moderate, and severe COVID-19). Only lipids with false discovery rate (FDR) <0.05 were shortlisted and presented. Forest plots illustrate the magnitude of odds ratios with indicator of significance of the estimate in the model; ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05. For non-significant lipids, the estimates were plotted as zeros. Lipids were broadly classified according to major classes of neutral lipids, glycerophospholipids, and sphingolipids. See also Figure S4.

Association of Plasma Lipids with Pathologically Relevant Clinical Indices
We then evaluated if plasma lipids altered in COVID-19 were significantly correlated with relevant clinical indices. Spearman correlations were performed and only correlations with p < 0.05 were indicated as colored circles on the correlation plots (Figure 3 ). We observed that PCs, particularly PUFA-PCs and PCps, displayed significant negative correlations with clinical indices of systemic inflammation (IL-6, CRP, procalcitonin [PCT], ESR, and SF) (Figure 3A; Table S1). This suggested that reductions in plasma PUFA-PCs and PCps were associated with aggravated systemic inflammation. Corroborating these observations, it was shown in a small cohort of cystic fibrosis patients that serum PUFA-PCs were positive indicators of lung function (measured in terms of predicted forced expiratory volume in 1 s) and negative indicators of systemic inflammation (Grothe et al., 2015). In contrast to PCs, only PEps, but not PUFA-PEs, were significantly and negatively associated with clinical indicators of systemic inflammation (Figure 3B). PCps were also specifically and positively associated with apo-A1 (Figure 3A). Of outstanding interest, we observed that plasma GM3s represented the only pathologically altered lipid class that was strongly and negatively correlated with T cell count and CD4+ T cell count (Figure 3C), which progressively decreased as disease severity increased (Figure S1A). The negative correlations suggest that increases in plasma GM3s were associated with reductions in circulating CD4+ T cell counts in COVID-19. Reductions in circulating CD4+ T cells constitute an important feature of dysregulated immune response reported in COVID-19 (Qin et al., 2020).
Figure 3 Correlation of Plasma Lipids with Clinical Indices
Correlation plots illustrate spearman correlations between clinical indices with phosphatidylcholines (PC) (A), phosphatidylethanolamines (PE) (B), and multivesicular body-related lipids including bis(monoacylglyero)phosphate (BMPs), monosialodihexosyl gangliosides (GM3), and sphingomyelins (SMs) (C). Only correlations with p < 0.05 were indicated with colored circles. Negative correlations were shown in red and positive correlations were shown in blue, with sizes of circles representing the magnitude of the correlations. LeuC, leukocyte count; NC, neutrophil count; LC, lymphocyte count; PC_clinic, platelet count; Hb, hemoglobin; aPPT, activated partial thromboplastin time; PT, prothrombin time; D.dimer, D-dimer; ALB, albumin; ALAT, alanine aminotransferase; AST, aspartate aminotransferase; TBIL, total bilirubin; Serum Cr, serum creatinine; LDH, lactate dehydrogenase; IL.6, interleukin-6; CRP, C-reactive protein; PCT, procalcitonin; ESR, erythrocyte sedimentation rate; SF, serum ferritin; LA, lactic acid; TCellC, T cell count; CD4.TCellC, CD4+ T cell count; VLDL-Cho, very low-density lipoprotein cholesterol; HDLCho, high-density lipoprotein cholesterol; total cho, total cholesterol; TG, triglycerides; Fe, iron; ApoA1, apolipoprotein A1; ApoB, apolipoprotein B.

Multiscale Embedded Correlation Networks to Uncover Pathologically Relevant Lipid Modules
Co-regulated genes often display similar patterns of gene expression, which translates to strong correlations between their gene expression levels (Williams, 2015). Under the same analogy, strong correlations between lipid levels can imply that these lipids lie along a common metabolic pathway and are co-regulated, and changing correlation patterns between lipid-pairs in disease compared to healthy states can potentially indicate pathologically relevant metabolic dysregulation. We had previously shown in a cohort of antecedent diabetes that such a systems approach to interrogate lipidomics data based on differential correlations can sieve out pathway aberrations even before actual changes in metabolite levels set in (Lu et al., 2019). Thus, in order to decipher lipid pathway dysregulation at early stages of infection, we then looked for pathologically relevant lipid modules in mild COVID-19 relative to healthy controls using MEGENA R to construct networks from differentially correlated lipid pairs calculated via the R package DGCA. Only differential correlations with empirical p < 0.05 were displayed (Figure 4 ). Four notable modules in the global network were circled and enlarged for emphasized discussion.
Figure 4 Differential Correlation Analyses of Plasma Lipids in Mild COVID-19 Relative to Healthy Controls
Multiscale embedded correlation network analysis illustrates the differential correlation of lipids in mild COVID-19 relative to healthy controls to reveal changes in lipid metabolic pathways upon early stage of viral infection. Only lipid pairs with significant differential correlations (empirical p < 0.05) were included. Sign/sign indicates direction and strength of correlation in control/mild COVID-19, and number that follows indicates number of lipid pairs in the global networks exhibiting this pattern of change. For instance, red line +/++ 1 in the upper legend of the global networks indicates that correlation between two connected lipid pairs was positive (+) in controls, and the correlation became even more strongly positive (++) in mild COVID-19 patients, as defined by statistically significant (p < 0.05) increase in correlation coefficients between the lipid pair across the two conditions. A total of 1 lipid pair connected by red lines in the global network displayed this pattern of change (+/++). Blue line +/− : positive in controls → negative in mild COVID-19. Teal line +/0: positive in controls → insignificant in mild COVID-19. Gold line ++/+: strongly positive in controls → weaker positive in mild COVID-19. Purple line 0/−: insignificant correlation in controls → negative correlation in mild COVID-19. Gray line 0/+: insignificant correlation in controls → positive correlation in mild COVID-19. Four modules (I–IV) of biological interest were circled and expanded for better visual clarity. (I) Module with hub PS 34:1 connected to numerous PEs by teal lines, indicating PS-PE positive correlations in healthy controls were lost in mild COVID-19. (II) Module with hub BMP 38:5(18:1/20:4) connected to CEs by blue lines, indicating BMP-CE correlations became negative in mild COVID-19. (III) Module with GM3 d18:0/25:0 as hub connected to several PSs by blue and purple lines, indicating GM3-PS correlations became negative in mild COVID-19. (IV) Module with LysoPC 16:1 as the hub connected to numerous PUFA-PEs by blue lines, indicating lysoPC-PUFA-PE correlations changed from positive in healthy controls to negative in mild COVID-19. PS, phosphatidylserines; PE, phosphatidylethanolamines; BMP, bis(monoacylglycero)phosphates; CE, cholesteryl esters; GM3, monosiaolodihexosyl gangliosides; PUFA-PE, polyunsaturated PEs.

PUFA-PE Accumulation Was Associated with Reductions in PSs in Mild COVID-19
Module I comprises PS 34:1 as the hub connected to numerous PUFA-PEs, i.e., PE 40:4(20:0/20:4) and PE 40:5(20:1/20:4), by green lines (+/0), indicating that the positive association between PS and PUFA-PEs in healthy controls was lost in mild COVID-19. PUFA-PEs were elevated while PSs were reduced in mild COVID-19 relative to controls (Figures 2 and S4). These suggest that PS synthase, which catalyzes the production of PSs from PEs (Han, 2016), may be compromised upon early infection, resulting in a loss of correlation between these lipids.

Correlation between BMPs and CEs Was Altered in Mild COVID-19
Module II comprises BMP 38:5 (18:1/20:4) as the hub connected to numerous cholesteryl esters (CEs) by blue lines (+/−). BMPs represent a class of structurally unique phospholipids enriched in MVBs implicated in cellular cholesterol homeostasis (Hullin-Matsuda et al., 2009). BMPs also exhibit cell-type-specific compositions in their esterified fatty acyl chains, with pulmonary alveolar macrophages exhibiting distinct enrichment in n-6 fatty acids (i.e., linoleic acid 18:2 and arachidonic acid 20:4) (Cochran et al., 1987; Mason et al., 1972). Among the various BMPs analyzed in our study, only BMP 38:5(18:1/20:4) and BMP 38:6(18:2/20:4) were specifically reduced in mild COVID-19 relative to healthy controls (Figure S4), suggesting that these reductions might be specific to pulmonary alveolar macrophages upon infection. BMPs were shown to influence cellular export of cholesterol by controlling cholesterol storage capacity of endosomes (Chevallier et al., 2008), which may explain the positive correlation between BMP 38:5(18:1/20:4) and several CEs in healthy controls. The changes in correlations in mild COVID-19 (i.e., +/− and +/0) suggested that this process might be perturbed, in agreement with the observed reductions in circulating apo-A1 with COVID-19 (Figure S1C). Of interest, we noticed that BMPs enriched in alveolar macrophages, i.e., BMP 38:5(18:1/20:4) (p = 0.0054) and BMP 38:6(18:2/20:4) (p = 0.0004), increased longitudinally with saturated CEs (CE 14:0, p = 0.0017; CE 16:0, p = 0.0092) from hospital admission to discharge in a small subset of patients who recovered from COVID-19 (Figure S2).

Negative Correlation between GM3s and PSs in COVID-19 Pathogenesis
Module III consists of GM3 d18:0/25:0 as the central hub connected to several PSs by purple (0/−) and blue lines (+/−). The connection between PSs and GM3s, which are not directly connected by endogenous lipid biosynthetic pathways, suggested that they might partake in common biological processes in a manner similar to BMPs and CEs as described above. In our study, while PUFA-PSs such as PS 38:4 and PS 40:4 were reduced in mild cases, numerous PS species including PS 40:4, PS 34:1, and PS 36:2 were significantly elevated (p < 0.05) in severe compared to moderate cases (Figure S4). Plasma PSs were also increased (p = 0.0041) during recovery from COVID-19 (Figure S2).

Afflicted PUFA-PC Homeostasis (Enhanced Breakdown or Abated Synthesis) in COVID-19
Module IV is represented by LysoPC 16:0 as the central hub, which became negatively correlated with several PUFA-PEs in mild COVID-19 (blue lines; +/−) relative to healthy controls. Negative correlations between PUFA-PEs and lysoPCs in mild COVID-19 may be explained considering the limiting pool of interconnecting PUFA-PCs that were reduced in mild disease versus controls (Figure 2). Enhanced production of lysoPCs via phospholipase A-mediated cleavage of PUFA-PCs would require increased methylation of PUFA-PEs to generate more PUFA-PCs via phosphatidylethanolamine N-methyltransferase (PEMT). It was demonstrated in a human cohort that PEMT preferentially utilizes PUFA-PEs as substrates, over the more saturated PEs, to selectively produce PUFA-PCs (Grothe et al., 2015).

Isolated Exosomes Displayed Increasing Enrichment in GM3s with Elevating Disease Severity
Lipid analyses were conducted on exosomes isolated from the plasma of the same cohort (n = 75). The purity of isolated exosomes was validated by the enrichment in exosome-specific protein marker ALG-2-interacting protein X (Alix) (Théry et al., 2001) (Figure S5A). In line with previous reports on exosome lipid compositions (Subra et al., 2007), the isolated exosomes displayed selective enrichment of raft-associated lipids, including free cholesterol (Cho) and SMs (Figure S5B). Interestingly, we also noted exosome-specific enrichments in 20:4-BMPs, PUFA-PSs, and several GM3s relative to plasma (Figure S5B), suggesting that these lipids might partake in exosome-specific processes. In corroboration with postulations drawn based on plasma lipid profile changes, we found that GM3s were increasingly enriched in the exosomes of COVID-19 patients of elevating disease severity (Figure 5 ). Several GM3s exhibited greater than 2-fold increases in the exosomes (p < 0.05) of COVID-19 patients relative to healthy controls (Figure 5A). Progressive increases with increasing disease severity were similarly noted for GM3s of varying acyl chain lengths, e.g., very long-chain GM3 d18:0/26:0 (p = 0.0008) and GM3 d18:1/24:1 (p = 0.0015), as well as medium-chain such as GM3 d18:1/16:0 (p = 0.0017) (Figure 5B). BMP38:5(18:1/20:4) was reduced in the exosomes of COVID-19 patients compared to controls (Figure 5A). Increasing endosomal BMP content was previously found to impede intercellular transmission of HIV particles (Chapuy-Regaud et al., 2013). PUFA-PSs such as PS 40:7 and PS 40:5 were specifically increased in the exosomes of severe compared to moderate cases (Figure 5C). While plasma changes in sphingolipids, particularly GM3s, were recapitulated in exosomes, many of the changes in phospholipids (with the exception of PSs) were not observed in isolated exosomes. These observations suggested that plasma changes in glycerophospholipids associated with COVID-19 might be attributed to other components, such as lipoproteins, in the circulation.
Figure 5 Lipid Changes in Exosomes of COVID-19 Patients
(A) Lipid changes in isolated exosomes from plasma of healthy controls (n = 25) and COVID-19 patients (n = 50). Colored dots (blue and red) in volcano plot indicate lipids that were significantly different (p < 0.05) between healthy controls and COVID-19 patients. Blue dots indicate significant lipids with fold change ≥ 2 in COVID-19 patients relative to controls listed in the blue panel, while red dots indicate significant lipids with fold change < 2 in patients relative to controls (listed in red and green panels). Lipids on the right side of vertical line at x = 0 were increased in COVID-19 patients compared to controls (blue and red panels), and lipids on the left side were decreased in COVID-19 patients compared to controls (green panel).
(B) Boxplots illustrate sum of all GM3s (p = 0.0037), and three representatives, GM3 d18:0/26:0 (p = 0.0008), GM3 d18:1/16:0 (p = 0.0017), and GM3 d18:1/24:1 (p = 0.0015), that displayed increasing trends in isolated exosomes from healthy controls to COVID-19 patients of increasing severity. Exosome lipids were expressed in nanomoles of lipids per gram of total protein (nmol/g protein).
(C) Volcano plots illustrate exosome lipids that were significantly different (p < 0.05) in pairwise comparisons indicated on top of each plot. Dots corresponding to significant lipids (p < 0.05) were colored (blue and red), and lipids with fold change ≥ 2 were colored blue and listed in the blue panel accompanying each volcano plot, while lipids with fold change < 2 were colored red.
See also Figure S5.