Cutting Edge: Severe SARS-CoV-2 Infection in Humans Is Defined by a Shift in the Serum Lipidome, Resulting in Dysregulation of Eicosanoid Immune Mediators

Results and Discussion

To measure differences in LM profiles between moderate and severe cases of COVID-19 and gain mechanistic insights into how these changes may drive disease severity, we used serum draws from 19 healthy patients (healthy), 18 hospitalized COVID-19 patients with respiratory symptoms who did not require ICU admission (moderate), and 20 patients with respiratory symptoms of a severity that required ICU admission (severe). Serum draws from healthy patients without known comorbidities were collected prior to the COVID-19 pandemic, to avoid any contribution of asymptomatic or convalescent cases in this dataset. The demographics, preexisting conditions, and treatment details of these patients are indicated in Supplemental Table I. To ensure recovery of both lipids and polar metabolites a modified chloroform extraction was used and measurements were made using a series of targeted liquid chromatography tandem mass spectrometry (LC-MS/MS) methods providing high-confidence feature identification (20, 22, 23).

Changes in primarily polar metabolites, including methanol-soluble polar lipids, among COVID-19 patient cohorts from China, Italy, and France have been reported (14–17). In agreement with those studies, we observed a dysregulation of amino acids and lactate and dysregulation of nucleotide catabolic products such as xanthine, hypoxanthine, and urate (Supplemental Fig. 1A) (14). Among lipid species, separation of disease from healthy and severe from moderate disease was marked by changes in sphingomyelins, lysophospholipids, cholesterol esters and multiple classes of glycerophospholipids in agreement with previous studies (Fig. 1A, 1B, Supplemental Fig. 1B–M) (14–17).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Mobilization of plasmalogen-derived PUFAs correlates with the disease severity in COVID-19. (A) Supervised partial least-squares discriminant analysis (PLSDA) of the healthy, moderate, and severe disease groups and (B) the corresponding feature loading plot. (C–F) Heatmap of the autoscaled mean intensity of each patient group for significantly varied lipids containing C20:4 (C), C20:5 (D), C22:5 (E), and C22:6 (F) between one or more groups adjusted for a false discovery rate (FDR) of 10% because of multiple comparisons. Color scale is consistent for (C)–(F).

To pursue our hypothesis that dysregulation of PUFA-containing lipids and immune-regulatory LMs may contribute to the inflammatory progression of severe COVID-19, we specifically interrogated patterns of LM precursors among the portion of lipids that significantly varied between any two groups in the cohort. Disease groups contained lowered levels of PUFA-containing phosphatidylcholine, PUFA-containing phosphatidylserine, and PUFA-containing phosphatidylethanolamine-plasmalogen and increased levels of PUFA free fatty acids as well as PUFA-containing phosphatidylethanolamine, PUFA-containing lysophospholipids, and PUFA-containing triacylglycerols compared with healthy controls (Fig. 1C–F). These PUFA patterns were exacerbated in severe patients compared with moderate patients. Of these PUFA-containing lipids, plasmalogen was of particular interest as a primary pool of PUFAs for the generation of LMs in both immune and structural cells (24, 25). This shift in PUFA pools strongly suggested differential mobilization of PUFAs for the generation of LMs between moderate and severe COVID-19.

To assess LMs directly, we targeted 67 LMs species using LC-MS/MS with comparison with standards and available spectral libraries (https://serhanlab.bwh.harvard.edu/). Principle component analysis of the LM lipidome showed separation of infected and healthy cohorts and between moderate and severe patients (Fig. 2A). Nearly all LMs measured were positively correlated with infection and moderate and severe disease were characterized by unique profiles of LMs (Fig. 2B, Supplemental Fig. 1N–P). Of particular interest were LMs present at lower levels in the severe group compared with the moderate group as these molecules behave counter to the pattern of increased immune activity expected in the severe disease. Specifically, moderate disease was characterized by significantly higher levels of the proresolving LM resolvin (Rv) E3 and a trend toward increased presence of the PG family members, particularly PGE2 (p = 0.051), PGD2 (p = 0.220) and PGF2a (p = 0.242). In contrast, severe disease was characterized by a significant increase in free PUFAs levels, monohydroxylated species, and AA-derived dihydroxylated species (Supplemental Fig. 1P). Multiple DHA- and AA-derived proresolving LMs also trend upward in severe disease (RvD3, p = 0.063) but do not reach significance.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

A unique milieu of LMs defines moderate and severe COVID-19 disease. (A) Unsupervised principle component analysis (PCA) of autoscaled LMs and (B) corresponding feature loading plot. (C–G) Heatmaps of the autoscaled mean for each patient group across molecules synthesized by ALOX5 (C), ALOX12 (D), ALOX15 (E), COX (F), or CYP (G). Color scale is consistent across (C)–(G).

LMs are generated by a single or a series of oxygenase mediated conversions of the parent PUFA. To examine the potential contribution of each oxygenase enzyme to the severe disease phenotype, we grouped LMs according to synthesis pathway (Fig. 2C–G). Several LMs are shared between multiple enzyme groups as they require sequential stereospecific hydroxylations by multiple enzymes. This grouping revealed that moderate disease was characterized by higher levels of LMs that require cyclooxygenase (COX) activity as well as certain EPA products of ALOX12, whereas severe disease was characterized by LMs that require activity of ALOX5 and cytochrome p450 (CYP) enzymes. This is in good agreement with previous observations from influenza, which associated symptom severity with ALOX5 activity (26).

Elevation of ALOX5- and CYP-dependent LMs in severe COVID-19 patient sera suggested upregulation of these pathways but provided no data as to the origin of these LMs. To begin to examine potential cellular origins of these enzymes in COVID-19 patients, we interrogated a published single-cell RNA sequencing dataset of severe COVID-19 patient PBMCs (Fig. 3A) for expression of ALOX and CYP genes (21). CYP gene expression was not detected (data not shown), whereas ALOX5 expression was detected in multiple cell types (Fig. 3B). ALOX5 was significantly increased in neutrophils and trended upward in CD14 monocytes, CD16 monocytes, and developing neutrophils (a population found almost exclusively in diseased individuals) from severe COVID-19 patients compared with healthy controls (Fig. 3C). Although expression is important, ALOX enzymes are primarily regulated by subcellular location and pathways involving LM production are intercellular highlighting the importance of ALOX expressing cell numbers and not just expression levels in these pathways (27, 28). Interestingly, severe COVID-19 is characterized by elevated ALOX5 expressing monocyte/macrophage population and depletion of lymphocyte populations (21, 29, 30). The absence of CYP gene expression in peripheral blood is consistent with the primarily hepatic localization of these enzymes (31). Taken together, these data support a systemic increase in ALOX5 activity in severe COVID-19 and provide a plausible foundation for the data in this study wherein ALOX5 products separate moderate and severe disease (32).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3.

Human PBMCs from COVID-19 patients are enriched for ALOX5 expressing cells and express higher levels of ALOX5. (A) UMAP dimensionality reduction plot of a published human PBMC single-cell RNA Seq dataset (21) identifying 20 cell types. (B) UMAP depicting ALOX5 expressing cells in blue. (C) Violin plots indicating ALOX5 expression levels within specific cellular populations in healthy (blue) or COVID (red) PBMCs. *p < 0.05, **p < 0.01, determined by a Mann–Whitney U test.

COVID-19 comorbidities are known to pathologically disrupt the homeostatic lipidome and may predispose patients to the different signatures of LMs seen in this study. To examine this potential association further, we correlated the LM dataset with the severity of disease, comorbidities, and demographics (Fig. 4). Thirteen LMs were significantly (p ≤ 0.05) correlated with disease severity and several families of LMs group together in the positive and negative direction including COX2 and ALOX12 products in the negative direction and ALOX15 and CYP in the positive direction despite not all members reaching significance. Among comorbidities, body mass index (BMI) most closely reflected the LM pattern of disease severity with all but three LMs sharing a similar coefficient direction and magnitude as disease severity (Fig. 4A). Hypertension and heart disease reflected the pattern of disease severity but not to the degree of BMI. Of note, PGs (PGE2, PGD2 and PGF2a) and RvE3 were negatively correlated with disease severity, BMI, hypertension, and heart disease. Neither age, gender, nor presence of diabetes correlated well with the pattern of LM dysregulation observed in severe cases of COVID-19. This correlative analysis supports the association of the specific LM dysregulation in severe disease with known comorbidities, in particular high BMI.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4.

Correlation of comorbidities with LMs and severe COVID-19. (A) Correlation plot of spearman (BMI and age) or point biserial Pearson (disease severity, hypertension, heart disease, diabetes, and gender) correlation coefficient of each LM with disease severity or patient demographics and comorbidities. LMs are ordered by the strength of correlation with disease severity. For gender, a positive correlation is correlated with male. (B) Corresponding p values of each LM patient condition correlation displayed as −log(p).

The downward trend of PGs in severe disease and the strong negative correlation of this family of LMs with COVID-19 comorbidities suggested a role for these molecules in the differential immune response between moderate and severe disease. Products of COX, including the PGs, have been shown to play complex roles in regulating initiation and suppression of elements of inflammation in viral infection (33). In the context of COVID-19, the potentially negative effects of both inhibitors of COX and the COX product thromboxane (TXB2) in severe disease and death have been discussed, albeit inconclusively, further highlighting the complexity and multifunctional nature of this pathway (34, 35). In this study, lower levels of PGs, in the context of hospitalized COVID-19 patients, were associated with higher disease severity, but further investigation is needed to understand the mechanistic role PGs play in COVID-19 progression.

It remains unclear whether these LM patterns represent drivers of disease, arise from preexisting conditions, or COVID-19 related treatment. LMs operate as part of an expansive immune signaling network and broadly deconvoluting their role from other immune ligands will require controlled experimentation. Baseline PUFA and LM levels are known to be dysregulated in several of the comorbidities examined in this study (11–13). Thus, although the LM levels in this study reflect the COVID-19 disease state, the differences between disease severity groups is likely to be affected by different starting levels of both LMs and LM precursors. Differences in LM levels could also arise from treatment differences. Unfortunately, in this cohort, therapies were either overrepresented or underrepresented to an extent that prevented correlative analysis and these therapies remain potentially confounding factors.

Together, the results presented in this study provide the first, to our knowledge, detailed description of the behavior of LMs in COVID-19 disease progression (26, 36). We provide evidence that a systemic lipid network consisting of liberation of PUFAs from plasmalogen and their subsequent conversion to LMs, capable of modulating inflammatory responses, characterizes both the onset and severity of COVID-19. Specifically, the loss of the immune-regulatory PGs and RvE3 and the increased products of ALOX5 and CYP provides both a measure of disease severity and a potentially mechanistic understanding of the immune balance allowing for patient recovery (37). Importantly, these pathways are directly targetable with drugs previously approved for use in other inflammatory conditions and, thus, provide new therapeutic opportunities to control severe COVID-19 (2, 5).