Deep phenotyping of the lipidomic response in COVID-19 and non-COVID-19 sepsis

Abstract

Background: Lipids may influence cellular penetrance by viral pathogens and the immune response that they evoke. We deeply phenotyped the lipidomic response to SARs-CoV-2 and compared that with infection with other pathogens in patients admitted with acute respiratory distress syndrome to an intensive care unit (ICU).

Methods: Mass spectrometry was used to characterise lipids and relate them to proteins, peripheral cell immunotypes and disease severity.

Results: Circulating phospholipases (sPLA2, cPLA2 (PLA2G4A) and PLA2G2D) were elevated on admission in all ICU groups. Cyclooxygenase, lipoxygenase and epoxygenase products of arachidonic acid (AA) were elevated in all ICU groups compared with controls. sPLA2 predicted severity in COVID-19 and correlated with TxA2, LTE4 and the isoprostane, iPF2α-III, while PLA2G2D correlated with LTE4. The elevation in PGD2, like PGI2 and 12-HETE, exhibited relative specificity for COVID-19 and correlated with sPLA2 and the interleukin-13 receptor to drive lymphopenia, a marker of disease severity. Pro-inflammatory eicosanoids remained correlated with severity in COVID-19 28 days after admission. Amongst non-COVID ICU patients, elevations in 5- and 15-HETE and 9- and 13-HODE reflected viral rather than bacterial disease. Linoleic acid (LA) binds directly to SARS-CoV-2 and both LA and its di-HOME products reflected disease severity in COVID-19. In healthy marines, these lipids rose with seroconversion. Eicosanoids linked variably to the peripheral cellular immune response. PGE2, TxA2 and LTE4 correlated with T cell activation, as did PGD2 with non-B non-T cell activation. In COVID-19, LPS stimulated peripheral blood mononuclear cell PGF2α correlated with memory T cells, dendritic and NK cells while LA and DiHOMEs correlated with exhausted T cells. Three high abundance lipids - ChoE 18:3, LPC-O-16:0 and PC-O-30:0 - were altered specifically in COVID. LPC-O-16:0 was strongly correlated with T helper follicular cell activation and all three negatively correlated with multi-omic inflammatory pathways and disease severity.

Conclusions: A broad based lipidomic storm is a predictor of poor prognosis in ARDS. Alterations in sPLA2, PGD2 and 12-HETE and the high abundance lipids, ChoE 18:3, LPC-O-16:0 and PC-O-30:0 exhibit relative specificity for COVID-19 amongst such patients and correlate with the inflammatory response to link to disease severity.

FIGURE 1

Eicosanoids and disease severity. Comparison of eicosanoid profiles and Spearman correlation with severity scores and organ damage indicators between COVID‐19 patients, non‐COVID patients and healthy controls. (A) Heat maps illustrate the effect size of mean differences of eicosanoids and linoleic acid metabolites in urine, and (B) plasma, when comparing each patient group with healthy controls. The values within each box represent the Cliff's delta, indicating the magnitude of the differences. The significance of these differences was determined using non‐parametric Mann–Whitney tests and significant (p < .05) differences are indicated in bold. Additionally, the Spearman correlation coefficient (ρ) between eicosanoids and different severity scores and organ damage indicators in (C) urine and (D) plasma are presented. The colour scale represents the strength and direction of the correlation. Warmer colours indicate positive correlations and cooler colours indicate negative correlations. The p values were calculated and significant (p < .05) correlations are indicated in bold.

FIGURE 2

Selected eicosanoids in each severity group. The figure shows the normalised concentrations of urinary TxM (A), PGEM (B), PGDM (C) and concentrations of plasma LTE₄ (D), 12‐HETE (E), 12¹³‐DiHOME (F) in each severity group, with error bars indicating standard error of the mean. The statistical significance was determined using Kruskal–Wallis test and marked as follows: *p < .05, **p < .01, ***p < .001, ****p < .0001.

FIGURE 3

Correlation between eicosanoid concentration and the frequency of immune cell subtypes. Heat maps show the Spearman correlation coefficient (ρ) between eicosanoid concentrations and the frequency of immune cell subtypes in (A) urine and (B) plasma. The colour scale represents the strength and direction of the correlation, with warmer colours indicating positive correlations and cooler colours indicating negative correlations. The p value was calculated and marked in the heat maps.

FIGURE 4

(A) Changes in the lipidomic profile of PBMC supernatant from COVID‐19 patients compared with healthy volunteers. The heat map shows the effect size of differences of levels of arachidonic acid and linoleic acid metabolites, corrected by cell number. The values within each box represent the Cliff's delta, indicating the magnitude of the differences. The significance of these differences was determined using non‐parametric Mann–Whitney tests and significant (p < .05) differences are marked with * in the heat map. (B and C) Correlation between eicosanoid concentrations and the frequency of immune cell subtypes in PBMCs are shown. The heat map shows Spearman's correlation coefficient (ρ) between lipid metabolites and the frequency of immune cell subtypes in PBMCs from healthy controls (B) and COVID‐19 patients (C). p Values were calculated and significant (p < .05) correlations are marked with * in the heat maps. The colour scale reports the Spearman's correlation coefficient ρ. Warmer colours indicate positive correlations, while cooler colours indicate negative correlations.

FIGURE 5

High abundance lipidomic response to COVID‐19 infection. Targeted positive and negative mode lipidomics data were assessed to evaluate signatures of COVID‐19 specific alterations. OPLS‐DA scores plot from (A) positive mode and (B) negative mode lipid data. (C) Lipids specific to COVID‐19 assessed using Kruskal–Wallis test and pairwise Mann–Whitney test (Kruskal–Wallis FDR < .2, p (Mann–Whitney) < .05 between COVID/non‐COVID and COVID/control; >.05 between non‐COVID/control. Median value of each lipid for each group is presented as heatmap. (D) Boxplots showing three lipids that are COVID specific as well as significantly different (FDR < 0.2) between moderate (ordinal score > 4) and severe (ordinal score < 4) COVID subjects (#p < .1, *p < .05, **p < .01, ***p < .001, ns = not significant). Inset density plots depict the distribution of respective lipids across the moderate and severe COVID groups. Dotted line represents the median values.

FIGURE 6

Highly abundant lipids restrain inflammation. LPC‐O‐16:0 is associated with the inflammatory response (A–C) and ChoE‐18:3 is specific to severe COVID (D–F). Levels of LPC‐O‐16:0 from COVID subjects were used for correlation analysis with protein and immunotype data to identify biological pathways associated with the lipid. (A) Top functionally enriched pathways (FDR < 0.001, for all significantly enriched pathways (FDR < 0.05), see Table S6) associated with the protein sets that are significantly associated with LPC‐O‐16:0 (Spearman's rho < −0.4 and p < .05) and vary with COVID severity (FDR < 0.05). The proteins are listed in Table S5. (B) Functional enrichment network of proteins (described in Table S5) significantly associated to LPC‐O‐16:0 (Spearman's rho < −0.4 and p < .05). (C) Immune cell types significantly correlated to LPC‐O‐16:0 (Spearman's rho < −0.4 or >0.4 and p < .05). The points are coloured by −log (p value of correlation) and sized by −log (q value between moderate and severe COVID (q_sev)). (D) ChoE‐18:3 is the only cholesteryl ester molecule that is depleted in the severe COVID group compared with moderate COVID, but elevated in the non‐COVID subjects. (E) Functional enrichment network of proteins significantly associated (Spearman's rho < −0.4 or >0.4 and p < .05) to ChoE‐18:3. (F) ChoE‐18:3 is significantly, and exclusively, associated to various B cell populations (Spearman's rho < −.4 or >.4 and p < .05).

FIGURE 7

Visualisation of COVID‐19 correlation network. (A) UMAP projection of features measured in COVID‐19 cohort. Each feature is represented by a circle whose size is negatively proportional to log10 of the p value testing difference of the feature in moderate versus severe COVID‐19. Hence, bigger circles correspond to features that differ more significantly between moderate and severe cases of COVID‐19. Correlations of sPLA2 with other features are represented by lines, red lines correspond to positive correlations, with Spearman's correlation coefficient > 0.4, blue lines correspond to negative correlations, with Spearman's correlation coefficient < −0.4. Named features are further discussed in this paper. (B) Region in close proximity to host response cluster 1 containing sPLA₂. Features printed in black show significant (p < .05) correlation with sPLA2, either with Spearman's correlation coefficient > 0.4 (red lines) or <−0.4 (blue lines). (C) Correlation subnetwork for three high abundance lipids of interest. Red lines denote pairs of features with Spearman's correlation coefficient >0.4, blue lines denote pairs of features with Spearman's correlation coefficient < −0.4. (D) Region in close proximity to host response cluster 1 containing urinary 12‐HETE and several other urinary eicosanoids. Correlations of 12‐HETE with other features are represented by red (Spearman's correlation coefficient > 0.4) or blue (Spearman's correlation coefficient < −0.4) lines. Compared with sPLA2, 12‐HETE exhibits lower number of strong correlations with other features. However, the positioning of 12‐HETE in the network is based on all correlations among all pairs of features, so its proximity to other features, such as IL‐12B, TNFB and SIT1 could be of interest.