A lipid atlas of human and mouse immune cells provides insights into ferroptosis susceptibility

Abstract

The cellular lipidome comprises thousands of unique lipid species. Here, using mass spectrometry-based targeted lipidomics, we characterize the lipid landscape of human and mouse immune cells (www.cellularlipidatlas.com). Using this resource, we show that immune cells have unique lipidomic signatures and that processes such as activation, maturation and development impact immune cell lipid composition. To demonstrate the potential of this resource to provide insights into immune cell biology, we determine how a cell-specific lipid trait—differences in the abundance of polyunsaturated fatty acid-containing glycerophospholipids (PUFA-PLs)—influences immune cell biology. First, we show that differences in PUFA-PL content underpin the differential susceptibility of immune cells to ferroptosis. Second, we show that low PUFA-PL content promotes resistance to ferroptosis in activated neutrophils. In summary, we show that the lipid landscape is a defining feature of immune cell identity and that cell-specific lipid phenotypes underpin aspects of immune cell physiology.

Main

Lipids are essential cellular components with roles in multiple aspects of cell biology¹. A fundamental property of lipids is their capacity to self-associate and form bilayers. This feature enables the formation of membranes, allowing cells to segregate their internal constituents from the external environment and, in eukaryotes, to form subcellular organelles². The varied biophysical properties of lipids enable membranes to alter their shape, thickness and order¹. These properties are essential for fission, fusion, budding, tubulation and the formation of lipid rafts¹. These membrane features underlie essential cellular processes, such as cell division, mitochondrial dynamics, endocytosis, protein export and cell signalling^3,4,5,6. Lipids also have a multitude of biochemical roles, including being harbingers of cell death (for example, glycerophospholserine (PS) exposure in apoptosis⁷ and polyunsaturated fatty acids (PUFAs) in ferroptosis⁸). Lipids act as signalling molecules and contribute to post-translational modifications (for example, myristylation and palmitoylation) that influence protein localization, activity and stability⁹. Finally, lipids are substrates in energy metabolism and heat production¹⁰. The ability of lipids to function in such wide-ranging aspects of cellular biology is due to both the chemical complexity of the lipidome, with ~100,000 theoretical lipid species, and the compositional diversity of the lipidome, which varies between tissues, cell types, organelles and membrane leaflets (the inner versus outer bilayer)¹. The importance of maintaining lipid homeostasis for cellular and organismal fitness is underscored by the many diseases that arise as a result of dysregulated lipid metabolism.

While thousands of chemically diverse lipid species are present within cells, the structural differences between many lipid species are relatively small (for example, the position and number of double bonds, the types of chemical linkage used to attach aliphatic chains to backbones, and acyl chain length). While considerable progress has been made in both describing lipid structural diversity and determining functions of particular lipid classes and species, we currently have a limited understanding of how lipid diversity, and the compositional variance in the cellular lipidome it encodes, affects cellular biology in general, and also how it may contribute to cell-specific functionality. In this regard, a limitation of our current knowledge is how lipid composition varies between cell types. The value of such information is highlighted by recent efforts describing the metabolic landscape of 928 cancer cell lines, where lipidome compositional variance between cell lines led to the identification of therapeutically exploitable lipid profiles¹¹.

While compositional differences in tissue lipidomes have been described in mice¹², parallel lipidomic analysis in human tissues is challenging. Furthermore, cellular heterogeneity within and between tissues limits the interpretation of lipid composition data obtained from whole tissues. The mammalian immune system comprises a multitude of highly specialized cells with diverse, and often unique, functional properties, and therefore provides an opportunity to explore how lipid composition may vary between cell types with cell-specific effector functions. Indeed, previous studies have described compositional variance in the lipidomes of a limited number of immune cell types¹³.

We hypothesized that the cellular lipidome would be a defining feature of immune cell identity and that differences in lipid composition between immune cells would contribute to cell-specific functionality. In this Resource, we have used mass spectrometry-based lipidomics to comprehensively describe the cellular lipidome of the human and mouse immune systems. Our immune cell lipid atlas comprises >500 individual lipid species measured in 16 human and eight murine mature immune cell types and is freely available to the community (www.cellularlipidatlas.com). Ferroptosis is a form of regulated cell death that involves the lethal lipid peroxidation of cell membrane lipids by oxygen-containing free radicals⁸. Ferroptosis is distinct from other forms of cell death and plays an important role in numerous disease conditions, including (neuro)degenerative disorders and cancer⁸. Using our immune cell lipid atlas, we show that cell-specific lipidomic phenotypes govern the susceptibility of immune cells to ferroptotic cell death.

Results

A lipidomic cell atlas of the human immune system

To test the hypothesis that the cellular lipidome is a defining feature of human immune cells, we isolated 16 distinct mature immune cell populations, representing all of the major immune cell lineages and their subsets, from the peripheral blood of 14 healthy donors. We then characterized their cellular lipidomes using ultra-high-performance liquid chromatography (HPLC) coupled with electrospray ionization and tandem mass spectrometry (MS/MS) (Fig. 1a). We quantified 524 individual lipid species, 505 of which were significantly different between at least two cell types, belonging to multiple distinct lipid classes, including diacyl, ether and vinyl–ether glycerophospholipids (PLs) (glycerophosphocholine (PC), glycerophosphoethanolamine (PE), PS, glycerophosphoinositol (PI), glycerophosphoglycerol (PG) and glycerophosphate (PA)), sphingolipids (SLs) (sphingomyelin (SM), ceramide, ganglioside and hexosylceramide (Hex-ceramide)), lyso-PLs and cholesterol (COH). In the present study, we were specifically interested in membrane lipids; therefore, neutral lipids (for example, triacylglycerol, diacylglycerol and COH ester species), which are abundant components of lipid droplets, were not analysed. Visualization of the global lipidome using t-distributed stochastic neighbour embedding (t-SNE) showed that the global cellular lipidome is an identifying feature of the cells of the human immune system (Fig. 1b).

Fig. 1: A lipidomic cell atlas of the human immune system.

a, Workflow schematic. b, t-SNE map displaying the global cellular lipidome of the human immune system. The dots represent individual human donors for each cell type. c–g, Breakdown of lipid features in human immune cells by class (c), phospholipid headgroup type (d), type of Sn1 chemical linkage in PC (e) and PE (f) and PL alkyl/acyl composition (g). h, Volcano plot of the lipid changes in CD4⁺ T_Naive cells and CD4⁺ T_EM cells. i–l, Changes in PE (i), PA (j), PC (k) and PS (l) in CD4⁺ T_Naive cells and CD4⁺ T_EM cells. In c–g and i–l, the data are presented as means + s.d. In c–g, statistically significant differences in specific lipid features between cell types were determined by one-way ANOVA with Tukey’s honestly significant difference (HSD) test after FDR correction (5%; Benjamini–Hochberg). P values for all pairwise comparisons are shown in Supplementary Table 1. In h–l, statistically significant differences were determined using two-tailed unpaired Student’s t-test. In c–l, the n values denoting the number of individual human donors for each cell type are as follows: 14 for naive B cells, CD4⁺ T_CM cells, CD8⁺ T_Naive cells, classical monocytes, intermediate monocytes and basophils; 13 for memory B cells, CD4⁺ T_Naive cells, CD4⁺ T_EM cells, CD8⁺ T_EM cells, non-classical monocytes and eosinophils; 12 for CD8⁺ T_CM cells, CD56_dim NK cells and CD56_bright NK cells; and 11 for neutrophils.

Source data

Human immune cells have a distinct phospholipid composition

To understand the primary drivers of the global lipidome differences within the human immune system, we examined how various chemical features of lipids vary between immune cells. Exploring the major lipid classes in mammalian cell membranes (that is, PLs, SLs, lyso-PLs (PLs containing a single acyl chain) and COH) revealed a generally similar profile across all cells (Fig. 1c; significance values for all human lipid atlas data are shown in Supplementary Table 1). However, we noted that the proportions of SL and COH were higher in myeloid cells, notably in neutrophils and eosinophils, with a corresponding decrease in the proportion of PL (Fig. 1c).

PLs are the most abundant lipid class present in cell membranes² (Fig. 1c) and are defined by their variable head group (for example, phosphocholine, phosphoethanolamine, phosphoserine and phosphoinositol) attached at the Sn3 position of the PL glycerol backbone. Overall, differences in the proportions of PL classes between immune cell types were unremarkable (Fig. 1d). Neutrophils had an increased proportion of PS—a finding that is consistent with the role of PS as a marker of apoptotic cells, as well as the very high number of neutrophils that must be efficiently cleared via efferocytosis by bone marrow-resident macrophages¹⁴. The proportion of PG was markedly higher in T and B lymphocytes compared with natural killer (NK) cells and myeloid cells (Fig. 1d and Extended Data Fig. 1a). PG is a critical substrate in the synthesis of cardiolipins and a major component of the mitochondrial inner membrane, and may thus be expected to be present in lower proportions in myeloid cells, particularly granulocytes, where mitochondrial abundance is typically low.

Myeloid cells are enriched in ether-linked phospholipids

Chemical linkages attaching fatty acyl chains and fatty alcohols (ester versus ether– or vinyl–ether linkages) to the sn1 position of the PL glycerol backbone also drive PL diversity. Therefore, we assessed the proportions of diacyl PLs, ether PLs (PL(O)) and vinyl–ether PLs (PL(P)) (also referred to as plasmalogens) within total PC and PE, the major PL classes within which ether PLs are enriched. A striking difference was observed in PC and PE composition (Fig. 1e,f). Diacyl PC accounted for ~90% of total PC in lymphoid cells, but was decreased to ~25–70% in myeloid cells (Fig. 1e). Similarly, the proportion of diacyl PE, which accounted for ~70% of total PE in lymphoid cells, was decreased to ~25–50% in myeloid cells (Fig. 1f). Accordingly, the proportions of PC(O), PC(P), PE(O) and PE(P) were significantly higher in myeloid cells than in lymphoid cells (Fig.1e,f and Extended Data Fig. 1b,c). These differences in diacyl, ether- and vinyl–ether composition within PE and PC are a major driver of the variance observed in the global lipidome between immune cell types.

Alkyl/acyl chain composition varies across immune cells

Another major driver of PL diversity is the composition of the alkyl/acyl chains, which typically vary from 16 to 24 carbon atoms in length and contain zero to six double bonds. Such differences are functionally important as they influence the biophysical and biochemical properties of the membrane¹. Saturated 16:0 and 18:0 chains constituted a large proportion (~50%) of the overall PL pool, which was relatively consistent between immune cell types (Fig. 1g). The proportions of monounsaturated chains, primarily 16:1 and 18:1, were significantly altered between cell types. The proportion of 16:1 was lower in myeloid cells compared with lymphoid cells (Fig. 1g; ~2 versus 5%, respectively), while the proportion of 18:1 was markedly higher in granulocytes (neutrophils (39%), eosinophils (22%) and basophils (27%)) compared with all other cell types (10–15% in lymphoid cells and 14–19% in monocytes) (Fig. 1g). With regards to polyunsaturated species, the proportions of 20:4 and 22:6 within the overall PL pool were decreased in granulocytes relative to the other cell types (Fig. 1g). To provide a more detailed breakdown of PL composition, we calculated alkyl/acyl chain composition within the major PL classes (PC, PE, PI and PS) (Extended Data Fig. 2). These analyses highlight the remarkable diversity in alkyl/acyl composition between PL classes, demonstrating that differences in global PL alkyl/acyl chain composition between immune cell types are attributable to compositional differences within specific PL classes. For example, the higher overall proportion of 18:1 in granulocytes is largely driven by an increase within PE and PS (Extended Data Fig. 2d,h). Moreover, within granulocytes (most notably neutrophils), the decreased overall proportion of 20:4 is largely due to differences in PC and PE (Extended Data Fig. 2a,d), while the decreased overall proportion of 22:6 is largely due to differences in PE(P) and PS (Extended Data Fig. 2f,g). Indeed, the proportions of 20:4 within PC and PE display remarkable variance between immune cell types (for example, comprising ~12% (PC) and ~30% (PE) of the total acyl chain content within lymphoid cells, but only ~1% (PC) and ~4% (PE) in neutrophils; Extended Data Fig. 1g,h). A similar degree of variance was observed for the proportions of 22:6 within PE(P) and PS (Extended Data Fig. 1i,j). Collectively, the above analysis reveals that differences in PL composition are a defining feature of the human immune cell lipidome. Specifically, differences in the usage of chemical linkages (that is, ester, ether and vinyl–ether) and alkyl/acyl chain diversity are major drivers of lipidome variance between immune cell types.

Human immune cells have similar SL composition

SLs are major constituents of cell membranes and play critical roles in cell biology¹⁵. Compositional diversity within SLs occurs as a result of differences in the sphingoid base, the N-acyl chain and the status of the headgroup (for example, SM, glucosyl-ceramides and ceramide-1-phosphate). First, we assessed SL diversity based on the absence (ceramide, dihydroceramide or deoxy-ceramide) or presence (Hex-ceramide (one, two or three sugar moieties), ganglioside or SM) of different SL head groups. While several significant differences between cell types were observed, very few lineage or cell subclass effects were observed in SL head group diversity, with two primary exceptions: (1) striking increases and decreases in the proportions of Hex2-ceramide (Hex2Cer) and SM, respectively, in neutrophils; and (2) increased proportions of ceramide and Hex-ceramide and a decrease in the proportion of SM in both naive and memory B cells (Extended Data Fig. 3a). With regards to the length of the N-acyl chain within ceramides, very few lineage or cell subclass effects were observed. A notable exception occurred within naive and particularly memory B cells, which had an increased proportion of the shorter 16:0 and 18:0 N-acyl chains compared with all other cell types (Extended Data Fig. 3b). Finally, we examined chemical diversity within the sphingoid base of ceramides. In most SLs, the palmitate (16:0)-derived d18:1 sphingoid base is the most prevalent. However, other fatty acids can form atypical sphingoid bases (that is, myristate (14:0) and stearate (18:0)), which form the corresponding d16:1 and d20:1 sphingoid bases^15,16. Finally, an atypical d18:2 sphingoid base can be formed by desaturation of the 18:1 sphingoid base by fatty acid desaturase 3 (FADS3)¹⁷. As expected, d18:1 was the primary sphingoid base in all cell types (Extended Data Fig. 3c). While little difference between cell types was observed, we noted that CD56^bright NK cells were enriched in d20:1 sphingoid bases, whereas eosinophils had elevated d18:2 sphingoid bases (Extended Data Fig. 3c). Collectively, these data show that while chemical diversity within SL is not a major driver of global lipidome variance between the cells of the immune system, specific SL phenotypes were observable in neutrophils, B cells and CD56^bright NK cells, which may impact their function.

Activation or maturation remodels the immune cell lipidome

Next, we used the human lipid atlas to examine how activation (CD4⁺ T_Naive cells versus CD4⁺ T effector memory cells (T_EM cells)) and maturation (classical versus non-classical monocytes) influence cellular lipid composition. First, exploring T cell activation, we noted a striking enrichment of PLs containing 22:4 PUFAs in CD4⁺ T_EM cells (Fig. 1h). 22:4 PUFAs are produced by a two-carbon extension to 20:4(n-6) PUFAs by elongation of very-long-chain fatty acid 5 (ELOVL5). Indeed, there was an inverse relationship between the proportions of 20:4 and 22:4 for several PL classes (Fig. 1i,j). Notably, we observed additional examples of PL acyl chain remodelling, indicative of fatty acid elongation in CD4⁺ T_EM cells (Fig. 1k,l). Consistently, using RNA sequencing data from the Human Protein Atlas, the expression of ELOVL1 and 5 family members was increased in CD4⁺ T_EM cells (Extended Data Fig. 4a).

To examine how cell maturation impacts the immune cell lipidome, we compared CD14⁺CD16⁻ classical versus CD16⁺CD14^dim non-classical monocytes¹⁸. In total, 201 out of 524 lipid species were significantly differentially abundant between classical and non-classical monocytes (Extended Data Fig. 4b). Prominent among these changes was a shift in SL composition (Extended Data Fig. 4c–f); non-classical monocytes were enriched in glycosphingolipids such as gangliosides and Hex1Cer, with proportionally lower SM levels. Furthermore, a shift towards shorter N-linked fatty acyl chains was observed in non-classical monocytes (Extended Data Fig. 4g–j). Collectively, these data show that the lipid composition of immune cells undergoes remodelling in response to cellular activation and maturation.

A lipidomic cell atlas of the mouse immune system

To determine whether the lipidomic traits observed in human immune cells were present within murine immune cells, we created a lipidomic atlas of the murine immune system. Eight distinct immune cell types were isolated by fluorescence-activated cell sorting (FACS) from the blood of C57Bl6/J mice (male; 4–6 weeks of age; fed a standard chow diet) and the cellular lipid composition was determined by mass spectrometry (Fig. 2a). Of the 524 measured lipid species, 400 were significantly different between at least two cell types. Visualization of the global lipidomes of the different murine immune cell types using t-SNE revealed a clear distinction between myeloid and lymphoid cells, as well as between the myeloid cell types (Fig. 2b).

Fig. 2: A lipidomic cell atlas of the mouse immune system.

a, Workflow schematic. b, t-SNE map displaying the global cellular lipidome of the mouse immune system. The dots represent individual mice for each cell type. c–g, Breakdown of lipid features in murine immune cells by class (c), phospholipid headgroup type (d), type of Sn1 chemical linkage in PC (e) and PE (f), and PL alkyl/acyl composition (g). In c–g, the data represent as means + s.d. The n values denoting the number of mice for each cell type are as follows: 8 for B cells and CD8⁺ T cells; 9 for CD4⁺ T cells, eosinophils, Ly6C^lo monocytes and NK cells; and 10 for Ly6C^hi monocytes and neutrophils. P values for all pairwise comparisons are shown in Supplementary Table 2. h–j, Pearson correlation coefficients of lipid abundance between human and mouse neutrophils (h; n = 11 individual human donors and 10 mice), T cells (i; n = 13 individual human donors and 9 mice) and B cells (j; n = 13 individual human donors and 8 mice). In each panel, statistically significant differences in specific lipid features between cell types were determined by one-way ANOVA with Tukey’s HSD test after FDR correction (5%; Benjamini–Hochberg). Schematic in a created with BioRender.com.

Source data

Similar to the human immune cells, there were no marked cell lineage or cell type distinctions in the overall proportions of PL, SL, lyso-PL and COH or the proportions of the major PL classes (Fig. 2c,d; significance values for all mouse lipid atlas data are shown in Supplementary Table 2). However, as in the human dataset, the proportion of PG was significantly decreased in neutrophils and eosinophils relative to other cell types (Extended Data Fig. 1d). Changes in the proportions of diacyl PC/PE, ether PC(O)/PE(O) and vinyl–ether PC(P)/PE(P) within PC and PE classes paralleled those observed in the equivalent human immune cells (Fig. 2e,f and Extended Data Fig. 1e,f). Thus, the utilization of distinct sn1 chemical linkages (ester, ether and vinyl–ether bonds) between cell types (primarily myeloid versus lymphoid) is a conserved feature of the human and mouse immune systems.

Next, we examined the diversity in global PL alkyl/acyl composition between murine immune cells. While numerous significant differences were present between specific cell types, no clear lineage distinctions were observed (Fig. 2g), with the exception of 22:6, the proportion of which was markedly lower in myeloid relative to lymphoid cells. Analysis of alkyl/acyl composition within PC, PE, PI and PS classes revealed numerous lineage and cell subtype-specific effects (Extended Data Fig. 2i–p). First, the proportion of 20:4 within PC (Extended Data Figs. 1k and 2i) and PE (Extended Data Figs. 1l and 2l) was markedly decreased in neutrophils and eosinophils, and to a lesser extent in monocytes. The decrease in the proportion of 22:6 in myeloid cells was driven by differences within PE(P) (Extended Data Figs. 1m and 2n) and PS (Extended Data Figs. 1n and 2p). Finally, the proportion of 18:1 within PS was significantly higher in neutrophils and eosinophils (Extended Data Fig. 2p). This profile of changes is similar to that observed within the human immune system.

Mouse immune cells have similar SL composition

As in the human immune system, variance in SL chemical diversity between immune cells is limited. Indeed, the only clear lineage distinction was a decrease in the proportion of Hex-ceramide and an increase in the proportion of SM in granulocytic cells relative to all other cell types (Extended Data Fig. 3d). Very few significant differences between cell types were observed in ceramide N-acyl chain composition or sphingoid base composition (Extended Data Fig. 3e,f). Furthermore, two of the most striking SL effects seen in the human immune system—(1) an increased proportion of Hex2Cer in neutrophils; and (2) an increased proportion of the 20:1 sphingoid base in CD56^bright NK cells—were not observed in mouse immune cells.

Immune cell lipidomes are similar between mice and humans

Pearson correlation coefficients were performed to examine the relationship between the lipidomes of mouse and human immune cells. This revealed a strong correlation between overall lipid abundance in human and mouse immune cells (neutrophils, r = 0.78; T cells, r = 0.80; B cells, r = 0.83). Several lipid classes were selectively enriched or depleted within a specific immune cell type. For example, the abundances of Hex2Cer and Hex3Cer were higher in human relative to mouse neutrophils (Fig. 2h); ceramides and PE(P) were higher and lower, respectively, in human T cells (Fig. 2i); and Hex1Cer and GM3 gangliosides were higher in human B cells (Fig. 2j). Finally, we noted that the lipid abundance relationship between human and mouse immune cells was stronger for PLs (neutrophils, r = 0.88; T cells, r = 0.83; B cells, r = 0.86) than for SLs (neutrophils, r = 0.56; T cells, r = 0.68; B cells, r = 0.71), indicating that PL metabolism is more highly conserved than SL metabolism between mice and humans.

Lipid phenotypes emerge during myelopoiesis

Myelopoiesis is the developmental process that gives rise to mature myeloid cells. Myeloid cells are derived from multipotent stem cells known as haematopoietic stem cells. Haematopoietic stem cells undergo division and commitment to evermore lineage-committed cell types, until ultimately producing mature myeloid cells (for example, neutrophils, eosinophils or monocytes). One of the most striking features of mouse and human immune cell lipid composition was the finding that myeloid cells are enriched in ether PLs (Figs. 1e,f and 2e,f). To examine the ontogeny of these observations, we assessed how ether lipid content changed during myelopoiesis. To this end, we performed lipidomic analysis on mature myeloid cells (neutrophils, eosinophils and Ly6C^hi monocytes), committed neutrophil precursors (immature neutrophils), unipotent progenitors (pre-neutrophils and common monocyte progenitors), bipotent progenitors (common myeloid progenitors and granulocyte–macrophage progenitors) and haematopoietic stem and progenitor cells (lineage marker-negative Sca1⁺ cKit⁺ cells) isolated from the bone marrow of 8- to 10-week-old C57BL6/J mice. We show that the abundances of ether- (O) and vinyl–ether (P) PC (Fig. 3a) and PE (Fig. 3b) progressively increase during myeloid cell development, while diacyl PC and PE abundance decrease, collectively causing a marked shift in overall PC and PE composition. These changes were particularly marked during neutrophil maturation. Another major feature of mature myeloid cells was a decreased content of 20:4 and 22:6 within PC, PE(P) and PS (Extended Data Fig. 2) relative to lymphoid cells. To provide insight into the origin of these differences, we examined changes in the abundance of 20:4 within PC and 22:6 within PE(P) during myelopoiesis. Notably, the level of 20:4 within PC progressively decreased during neutrophil development, reaching its nadir in mature neutrophils (Fig. 3c). While not as pronounced as that observed in neutrophils, a similar effect was seen in eosinophils and monocytes (Fig. 3c). In contrast, the level of 22:6 within PE(P) was relatively unchanged during myelopoiesis: while a significant increase in 22:6 within PE(P) was observed in common myeloid progenitors, their levels were not different between any of the other cell types (Fig. 3d). Collectively, these findings demonstrate that the lower levels of 20:4 within PC in mature myeloid cells, especially neutrophils, are due to remodelling of lipid composition during myelopoiesis. To gain further insight into these lipidomic changes, we analysed the expression of genes encoding proteins involved in ether lipid and PUFA metabolism during neutrophil development¹⁹. The expression of Fads1, Fads2 and Elovl5, which encode proteins that mediate the desaturation and elongation of PUFAs, as well as Acsl4, which encodes a protein that converts 20:4 fatty acids to their active form for subsequent metabolism, was markedly downregulated during neutrophil development (Fig. 3e). Of note, while the expression of some genes encoding proteins involved in ether lipid synthesis was upregulated, others involved in ether lipid catabolism were downregulated (Fig. 3f). In contrast, Gnpat, Agps and Dhrs7b (also known as Pexrap)—genes that encode peroxisome-localized proteins involved in the early steps of ether lipid synthesis—were markedly downregulated during neutrophil development (Fig. 3f). Collectively, these data provide preliminary evidence that transcriptional reprogramming during neutrophil development may contribute to alterations in lipid composition.

Fig. 3: Diacyl and ether PL proportion changes during myelopoiesis.

a,b, Hierarchical representation of the changes in PC composition (a) and PE composition (b) during myelopoiesis. Total PC and PE are the sum of the diacyl, ether (O) and vinyl–ether (P) forms. The pie charts show the compositional changes of diacyl, ether (O) and vinyl–ether (P) PC and PE lipids as a percentage of the total PC and total PE levels, respectively, and the data were analysed by one-way ANOVA with Tukey’s HSD test. c,d, Concentration (bars) and proportion (lines) of 20:4 within PC (c) and 22:6 within PE(P) (d) in each cell type isolated from the bone marrow of C57BL6/J mice. The data are presented as means ± s.e.m. and were analysed by one-way ANOVA with Tukey’s HSD test. In a–d, n values denoting the number of mice for each cell type are as follows: 9 for common myeloid progenitors (CMPs), mature neutrophils (Mat-N) and eosinophils (Eo); and 10 for haematopoietic stem and progenitor cells (lineage marker-negative Sca1⁺ cKit⁺ cells (LSKs)), granulocyte–macrophage progenitors (GMPs), pre-neutrophils (Pre-N), immature neutrophils (Imm-N), common monocyte progenitors (cMoPs) and monocytes (Mo). e,f, Expression of genes (z score normalized) encoding PUFA metabolism enzymes (e) and ether lipid metabolism enzymes (f), determined in the indicated cell types (RNA sequencing data from ref. ¹⁹). Also shown are schematics of the plasmalogen (e) and PUFA (f) synthesis pathways. Schematics in e and f created with BioRender.com. AdA, adrenic acid; AGK, alkylglycerol kinase; AGPAT, 1-acyl-sn-glycerol-3-phosphate acyltransferase; AGPS, alkylglycerone phosphate synthase; CEPT, choline/ethanolamine phosphotransferase; DG, diacylglycerol; DGLA, dihomo-gamma-linoleic acid; DHRS7B, dehydrogenase/reductase 7b; EDA, eicosadienoic acid; FAR1/2, fatty acyl-CoA reductase; G3P, glycerol-3-phosphate; GLA, gamma-linoleic acid; GNPAT, glyceronephosphate O-acyltransferase; LA, linoleic acid; LPC, lyso-glycerophosphocoline; LPE, lyso-glycerophosphoethanolamine; LPIN, Lipin1; PLA2, phospholipase A2; TMEM, transmembrane protein.

Source data

Lymphoid cells have a higher proportion of PUFA-PLs

One of the most striking features of immune cell lipid composition was marked variance in the proportions of PUFAs between myeloid (low) and lymphoid (high) cells, particularly highly unsaturated 20:4 and 22:6 PUFAs (Extended Data Figs. 1 and 2). PUFA-PLs are susceptible to iron-dependent lipid peroxidation, which can result in a form of cell death known as ferroptosis⁸. PE(18:0/20:4) is a key substrate that triggers ferroptosis²⁰. PE(18:0_20:4) (note that PE(18:0/20:4) nomenclature is used when the position of the acyl chains is known, while PE(18:0_20:4) nomenclature is used when the identity of the acyl chains is known, but their specific positioning at either Sn1 or Sn2 is not known; the latter is the case for most PLs) displayed considerable variance between immune cells, being highest in T cell subsets and lowest in neutrophils (Extended Data Fig. 1o,p). It is now appreciated that overall cellular PUFA abundance dictates susceptibility to ferroptosis²¹. To provide a more holistic view of how differences in PUFA-PL abundance may impact immune cell susceptibility to ferroptosis, we calculated an estimated cellular phospholipid peroxidation index (CPI) based on the known relative hydrogen atom transfer propagation rate constants of fatty acids^22,23. Of note, the estimated cellular lipid peroxidation index does not consider many factors that influence the susceptibility of membrane lipids to undergoing peroxidation (for example, cumulative antioxidant defences, endogenous Fe²⁺ levels and the distribution of PUFA-PLs in the membrane). Nonetheless, the CPI does reveal potential differences in the intrinsic susceptibility of the lipidome of distinct immune cell types to lipid peroxidation. In both human and mouse immune cells, lymphoid cells had the highest CPI, followed by monocytes and granulocytes (Fig. 4a,b).

Fig. 4: Differences in PUFA-PL composition provide a basis for the differential susceptibility of immune cells to ferroptosis.

a,b, CPIs of human (a) and mouse (b) immune cells. The data are presented as a box and whiskers, where the lower and upper limits of the box correspond to the 25th and 75th percentiles, respectively, the line within the box is the median and the whiskers extend to the minimum and maximum values. In a, the numbers (n) of individual human donors are as reported in Fig. 1. In b, the numbers (n) of independent mice are as reported in Fig. 2. c, Cell viability of FACS-sorted neutrophils, monocytes, T cells and B cells treated with the indicated doses of ML210 for 24 h (n = 4 mice). d,e, Liperfluo fluorescence (d) and cell viability (e) in bone marrow neutrophils, monocytes, T cells and B cells treated with ML210 at the indicated doses for either 6 h (left) or 24 h (right) (n = 6 mice). f,g, Liperfluo fluorescence (f) and cell viability (g) in T and B cells treated with dimethylsulfoxide (DMSO) vehicle (Veh), ML210 (1 μM) or ML210 in combination with either α-tocopherol (α-TOH; 200 μM) or ferrostatin-1 (Fer-1; 1 μM) for 24 h. The data were analysed by one-way ANOVA with Tukey’s HSD test (n = 6 mice). h–k, Volcano plots showing the changes in PL species in FACS-sorted T cells (h), B cells (i), monocytes (j) and neutrophils (k) treated with ML210 (1 μM) or DMSO Veh for 24 h. Coloured dots indicate the PL species that were significantly different (one-way ANOVA with Tukey’s HSD test) after FDR correction (Benjamini–Hochberg) (n = 8 mice). l, Cell viability of bone marrow neutrophils treated with the indicated doses of ML210 alone (that is, with DMSO Veh) or in combination with methotrexate (1.5 μM), ferroptosis suppressor protein 1 inhibitor (iFSP1; 3 μM) or methotrexate + iFSP1 for 24 h (n = 6 mice with the exception of methotrexate + iFSP1 + ML210 (20 μM), where n = 5 mice). The data were analysed by one-way ANOVA with Tukey’s HSD test. The data in c–g and l are presented as means ± s.e.m.

Source data

PUFA-PL content dictates immune cell ferroptosis sensitivity

We hypothesized that lymphoid cells, which are enriched in PUFA-PLs, would be more sensitive to lipid peroxidation and ferroptosis than myeloid cells. Therefore, we examined lipid peroxidation status and ferroptosis susceptibility in murine bone marrow immune cells treated with ML210, a highly specific inhibitor of glutathione peroxidase 4 (GPX4), the primary pathway by which cells limit lipid peroxidation to prevent ferroptosis⁸. A schematic of the various compounds used to induce and inhibit ferroptosis is shown in Extended Data Fig. 5a. ML210 treatment of both FACS-purified (Fig. 4c) and unsorted bone marrow immune cells caused marked lipid peroxidation (Fig. 4d) and cell death (Fig. 4e) at 6 and 24 h in T and B cells, but not in Ly6C^hi monocytes or neutrophils. These effects were reversed by co-treatment with radical trapping antioxidants, confirming that cells were dying via ferroptosis (Fig. 4f,g). Similar results were obtained when we used a second GPX4 inhibitor, RSL3 (Extended Data Fig. 5b–g), and erastin (Extended Data Fig. 5h–l), an indirect inhibitor of GPX4 that induces ferroptosis via inhibition of the cystine–glutamate antiporter system xc⁻. Lipid peroxidation causes PUFA fragmentation, resulting in a decrease in the abundance of PUFA-PLs. ML210 treatment decreased the abundance of numerous PUFA-PLs in T and B cells (Fig. 4h,i). This included canonical executioners of ferroptosis PE(18:0_20:4) and PE(18:0_22:4), as well as PUFA-containing ether PLs, PCs and PSs. These changes were reversed by co-treatment with the ferroptosis inhibitor ferrostatin-1 (Fer-1) (Extended Data Fig. 6a,b,d,e). We also observed a decrease in several PUFA-PLs in Ly6C^hi monocytes; however, these changes were not reversed by Fer-1 (Fig. 4j and Extended Data Fig. 6c,f). ML210 treatment did not significantly alter the abundance of any PUFA-PLs in neutrophils, consistent with the resistance of neutrophils to ferroptosis (Fig. 4k and Extended Data Fig. 6g).

In addition to GPX4, several other pathways prevent excessive lipid peroxidation and ferroptosis. As such, the importance of altered lipid composition (that is, decreased PUFA levels) relative to the potential involvement of other ferroptosis-suppressing pathways (Extended Data Fig. 5a) in mediating myeloid cell resistance to ferroptosis is unclear. Accordingly, we next determined whether neutrophil and Ly6C^hi monocyte resistance to lipid peroxidation and ferroptosis following GPX4 inhibition was due to the involvement of alternative ferroptosis-suppressing pathways—specifically, the ferroptosis-suppressing protein 1 (FSP1), GTP cyclohydrolase 1/dihydrofolate reductase and dihydroorotate dehydrogenase (DHODH) pathways^24,25,26,27. Remarkably, inhibition of these pathways (alone or in combination), together with GPX4 inhibition, had no effect on ferroptotic cell death in neutrophils (Fig. 4l). Ly6C^hi monocytes showed a partial susceptibility to ferroptosis when GPX4 and FSP1 were co-inhibited (Extended Data Fig. 5m–o). Finally, while inhibition of the DHODH pathway triggered cell death in both neutrophils and monocytes, this was not due to ferroptosis, instead appearing to be largely necroptotic in nature (Extended Data Fig. 5p–u). Collectively, these data suggest that differences in the abundances of PUFA-PLs are a major determinant of immune cell susceptibility to lipid peroxidation and ferroptosis.

Human immune cells have distinct ferroptosis susceptibility

Next, we examined whether the differential susceptibility of murine lymphoid and myeloid cells to ferroptosis was observed in human immune cells. Purified T cells and neutrophils isolated from peripheral blood samples of healthy volunteers were treated with ML210, RSL3 or erastin. T cells displayed a loss of viability and increased lipid peroxidation following either ML210, RSL3 or erastin treatment, which was reversed by Fer-1 (Fig. 5a–d). In contrast, human neutrophils displayed very little change in viability and lipid peroxidation following ML210, RSL3 or erastin treatment (Fig. 5e–h). These findings demonstrate that the differential susceptibility of lymphoid and myeloid cells to lipid peroxidation and ferroptosis is a conserved feature of the mouse and human immune systems.

Fig. 5: Human neutrophils and T cells are differentially susceptible to ferroptosis.

a–h, Cell viability (left) and Liperfluo fluorescence (right) in CD3⁺ T cells (a–d) and neutrophils (e–h) treated with ML210 (a, b, e and f), RSL3 (c and g) or erastin (d and h) at the indicated doses in the presence or absence of Fer-1 (1 μM) for 72 h. a,e, Representative DAPI and Liperfluo staining in T cells (a) and neutrophils (e). All data are shown as means ± s.e.m. In b, c, f and g, n = 6 individual human donors. In d and h, n = 5 individual human donors.

Source data

Decreasing PUFA-PLs protects T cells from ferroptosis

Our data suggest that the PUFA content of immune cells governs their susceptibility to ferroptosis. We hypothesized that altering endogenous PUFA levels would change ferroptosis susceptibility. To decrease the proportion of PUFA-PLs within T cells, we first supplemented T cells with either the monounsaturated fatty acid (MUFA) oleic acid (OA; 18:1) or the PL PE(18:0/18:1). Both oleate and PE(18:0/18:1) shifted the T cell PL composition towards a MUFA-enriched profile, concomitantly decreasing the proportion of PLs containing four or more carbon–carbon double bonds (for example, 20:4-containing PLs) and decreasing the CPI (Fig. 6a–c,e–g and Extended Data Fig. 7a,b). Importantly, supplementation with either OA or PE(18:0/18:1) protected T cells from lipid peroxidation and ferroptosis triggered by GPX4 inhibition (Fig. 6d,h). To further confirm the importance of PUFA-PL content in dictating T cell susceptibility to ferroptosis, we generated Acsl4^−/− mice. Acyl-coenyzme synthetase long-chain family member 4 (ACSL4) catalyses the addition of CoA to PUFAs, which is essential for the biosynthesis of PUFA-PLs. The deletion of ACSL4 leads to decreased PUFA-PLs and resistance to ferroptosis in many cancer cell lines^20,28, although its role in mediating ferroptosis susceptibility in primary cells is less well established. Deletion of Acsl4 from T cells led to widespread alterations to PL content (Fig. 6i–k and Extended Data Fig. 7c), characterized by a substantial decrease in the proportion of PLs containing 20:4, 22:4, 22:5 and 22:6 PUFAs. These changes were offset by an increase in the proportion of PLs containing 18:1 MUFAs and 18:2 PUFAs, such that overall PL levels were not different between wild-type (WT) and Acsl4^−/− T cells. These changes in PL acyl chain composition led to increased resistance to ferroptosis and a decrease in lipid peroxidation (Fig. 6l). These findings are consistent with a previous study²⁹.

Fig. 6: Decreasing the PUFA-PL content alters T cell sensitivity to lipid peroxidation and ferroptosis.

a,e,i, Percentage of acyl chain length in purified WT T cells following treatment with oleate (200 μM) (a) or PE(18:0/18:1) (20 μM) (e) and in Acsl4^−/− T cells (i). b,f,j, Heatmaps (z score normalized) and P values showing example PL species in WT T cells following oleate treatment (b) or PE(18:0/18:1) treatment (f) and in Acsl4^−/− T cells (j). c,g,k, Calculated CPIs in purified WT T cells following treatment with oleate (c) or PE(18:0/18:1) (g) and in Acsl4^−/− T cells (k). d,h,l, Cell viability (left and middle) and Liperfluo fluorescence (right) in purified WT T cells (d and h) or purified WT versus Acsl4^−/− T cells (l) treated with the indicated doses of ML210 for 24 h. Cells in d and h were pre-treated for 16 h with either oleate (d) or PE(18:0/18:1) (h). In d and h, BSA (d) and ethanol (h) were used as controls. The middle graphs display the data highlighted in purple to the left. The data were analysed by two-tailed unpaired Student’s t-test (a, c, e, g, i and k) or two-way ANOVA (d, h and l) with Tukey’s HSD test. The data are presented as means ± s.e.m. (n = 6 mice (a–e, h and the ethanol group in g), n = 5 mice (PE(18:0/18:1) group in g) and n = 4 mice (i–l).

Source data

Increasing PUFA-PLs triggers ferroptosis in neutrophils

To determine whether increasing endogenous PUFA levels sensitized neutrophils to ferroptosis, we supplemented neutrophils with arachidonic acid (20:4) and docosahexaenoic acid (22:6) (AA + DHA). Remarkably, in the absence of ML210, AA + DHA supplementation caused dose-dependent cell death and significantly increased lipid peroxidation in neutrophils (Fig. 7a,b). These effects were largely reversed by co-treatment of neutrophils with Fer-1, but not inhibitors of apoptosis or necroptosis, indicating bona fide ferroptosis (Fig. 7a–d). Lipidomic analysis of AA + DHA-treated neutrophils revealed a striking profile of PL changes (Fig. 7e and Extended Data Fig. 8a–f). AA + DHA treatment increased the abundance of 20:4- and 22:6-containing PLs, most notably within PC and PC(O) PL classes (Extended Data Fig. 8c,e). The levels of adrenic acid (22:4)-containing PLs were increased in PC, PE and PE(P) PL classes (Extended Data Fig. 8d), which was probably the result of elongation of supplemented 20:4. All of these effects were largely prevented by Fer-1 (Extended Data Fig. 8). Higher doses of AA + DHA supplementation also increased the levels of several lyso-PL species (Fig. 7e), consistent with previous lipidomic analyses of ferroptotic cells^30,31. Furthermore, we found that AA + DHA supplementation dramatically increased the levels of multiple oxidized PC species (Fig. 7e). Collectively, these findings demonstrate that AA + DHA supplementation increases the abundance of PLs containing 20:4, 22:4 and 22:6, leading to lipid peroxidation and ferroptotic cell death.

Fig. 7: Increasing the PUFA-PL content triggers NOX2-dependent lipid peroxidation and ferroptosis in neutrophils.

a,b, Cell viability (a) and Liperfluo fluorescence (b) in purified neutrophils treated with an AA + DHA mixture (0 (that is, with BSA Veh), 50, 100, 200 or 400 μM) in the presence of Fer-1 (1 μM) or Veh (DMSO) for 24 h (n = 6 mice). c,d, Cell viability (c) and Liperfluo fluorescence (d) in purified neutrophils treated with an AA + DHA mixture (200 μM (left) or 400 μM (right)) in the presence of Fer-1 (1 μM) or vehicle (DMSO) for 24 h or following pre-treatment with z-VAD (25 μM) or Nec-1s (10 μM) for 1 h (n = 6 mice in two independent experiments). e, Heatmap (z score normalized) showing example PL, lyso-PL and oxidized PL species in neutrophils treated with AA + DHA in the presence or absence of Fer-1 (1 μM). f,g, Cell viability (f) and Liperfluo fluorescence (g) in purified neutrophils from WT and Nox2^−/− mice treated with an AA + DHA mixture (0 (that is, with BSA vehicle), 50, 100, 200 or 400 μM) for 24 h (n = 6 mice). h,i, Liperfluo fluorescence (h) and cell viability (i) in purified neutrophils from WT and Nox2^−/− mice treated as indicated with PMA (50 nM), ML210 (5 μM) and Fer-1 (1 μM) for 6 h (n = 6 mice). j, Cell viability in purified neutrophils supplemented with either AA + DHA (50 μM) or vehicle (BSA) for 16 h before being treated as indicated with PMA (50 nM), ML210 (5 μM) and Fer-1 (1 μM) for 6 h (n = 8 mice). The data were analysed by either one-way ANOVA (c and d) or two-way ANOVA (a, b and f–j) with Tukey’s HSD test. The data are presented as means ± s.e.m. NS, not significant.

Source data

In considering the source of the reactive oxygen species (ROS) that trigger lipid peroxidation in neutrophils treated with AA + DHA, we turned our attention to the NADPH oxidase family of enzymes (NOX1–5 and DUOX1 and 2), specifically NADPH oxidase 2 (NOX2), which is highly expressed in neutrophils. We hypothesized that NOX2-derived ROS trigger lipid peroxidation and ferroptosis in neutrophils treated with AA + DHA. Indeed, neutrophils from NOX2 knockout (KO) mice (Extended Data Fig. 8g) were protected from lipid peroxidation and ferroptosis following AA + DHA treatment (Fig. 7f,g). Following activation, NOX2 rapidly produces large quantities of superoxide, which plays a critical role in neutrophil-mediated immunity. A notable feature of NOX2 activation is that in addition to producing large amounts of ROS, the reaction also consumes large amounts of NADPH. NADPH is a crucial co-factor in regenerating reduced glutathione, which is required for GPX4 function. Thus, we hypothesized that the activation of NOX2 would create conditions within the cell that promote lipid peroxidation and ferroptosis. Neutrophils treated with phorbol 12-myristate 13-acetate (PMA), a potent NOX2 activator, had increased lipid peroxidation and a slight decrease in cell viability (Fig. 7h,i). Notably, inhibition of GPX4 in PMA-stimulated neutrophils caused a marked loss of cell viability and further increased lipid peroxidation (Fig. 7h,i). Importantly, these effects were reversed by Fer-1 and NOX2 deletion, confirming that the cell death observed was ferroptosis and the dependence on NOX2 (Fig. 7h,i). Thus, while unactivated neutrophils do not appear to use GPX4 to prevent excessive lipid peroxidation and ferroptosis, activated neutrophils rely on GPX4 to minimize self-inflicted lipid damage caused by heightened ROS production. Finally, we hypothesized that increasing the PUFA-PL content of neutrophils would exacerbate PMA-induced ferroptosis. In this experiment, we supplemented neutrophils with a dose (50 μM) of AA + DHA that caused minimal cell death (Fig. 7a) but still effectively increased the PUFA-PL content (Extended Data Fig. 8a–e) for 16 h before stimulating with PMA. Once again, neutrophils treated with PMA alone displayed only a modest loss of viability, which was enhanced by concomitant GPX4 inhibition and reversed by Fer-1 (Fig. 7j). Notably, PMA treatment induced a more pronounced loss of cell viability in neutrophils supplemented with AA + DHA (Fig. 7j). ML210 treatment did not further decrease cell viability (Fig. 7e). Collectively, our data demonstrate that a low PUFA-PL content is essential for both resting and activated neutrophils to minimize lipid peroxidation and ferroptosis driven by NOX2-dependent ROS production.

Discussion

Many features define immune cell identity, from physical parameters such as cell size, nuclear morphology and granule content to molecular features such as the transcriptome and protein surface marker expression. Importantly, such differences underlie the unique functional properties of distinct immune cell types. The major motivation for the current work was the idea that the lipid landscape of the cells of the immune system is also a defining feature of immune cell identity and that differences in lipid composition contribute to cell-specific functionality. To address this idea, we created a cellular lipid atlas of the immune system. This resource, which comprises >500 individual lipid species measured in immune cell types from different developmental lineages and with distinct roles in host immunity, in both mice and humans, demonstrates that the lipid composition of immune cells is distinct. Furthermore, we show that physiological and pathological processes such as cell maturation and activation alter immune cell lipid composition, while key lipid features are acquired during immune cell development. By analysing the major chemical features that define lipid identity and function (for example, PL headgroup, acyl chain diversity and chemical linkages), we have identified unique, immune cell-specific lipid profiles and provided evidence that these differences have important biological consequences. Specifically, using the lipid atlas, we find that differences in the abundance of PUFA-PLs provide a potential molecular basis for the previously described susceptibility of T cells to ferroptosis, as well as the resistance of myeloid cells, including neutrophils and monocytes, to ferroptosis^{29,32,33,34,35}.

One of the most striking differences in lipid composition we identified from the lipid atlas was the roughly two- to threefold enrichment of ether lipids in myeloid cells, in particular granulocytes, compared with lymphoid cells. Furthermore, the high level of ether lipids in mature myeloid cells is the result of lipidome remodelling during myelopoiesis. These changes are particularly pronounced during neutrophil development, from ~90% diacyl PC to ~90% ether or vinyl–ether PC and from ~50% diacyl PE to ~90% ether or vinyl–ether PE in lineage marker-negative Sca1⁺ cKit⁺ cells and mature neutrophils, respectively. While an increase in the proportion of ether lipids in myeloid cells, particularly neutrophils, has been appreciated for several decades^36,37, their functional importance is not well understood. Ether PLs are distinguished from conventional diacyl PLs by the presence of an ether or vinyl–ether bond used to attach the sn1 alkyl chain, as opposed to an ester linkage in diacyl PLs^38,39. This relatively modest structural difference imparts ether PLs with unique biophysical and biochemical roles, including altering membrane properties, in cell signalling, and as antioxidants^38,39. Previous studies have shown that activated neutrophils and eosinophils produce chlorinating and brominating reactive species that attack the vinyl–ether bond to produce 2-chlorohexadecal and 2-bromohexadecanal, respectively, that can act as phagocyte chemoattractants^40,41,42. In neutrophils, ether lipids may influence membrane lipid composition to limit endoplasmic reticulum stress-induced apoptosis and neutropenia⁴³; however, it has also been reported that mice deficient in ether lipids have normal neutrophil numbers⁴⁴. Our data support the idea that ether lipids have critical roles in the development and function of neutrophils and probably within other granulocytes and myeloid cells, the elucidation of which will be an important future goal.

Having determined that different immune cell types have specific lipid signatures, we wanted to establish that such differences confer cell-specific functional properties; for several reasons, we focused on PUFA-PLs. First, in both mice and humans, the proportions of 20:4 and 22:6 within PC, PE and PS were markedly lower in mature myeloid cells, most notably in neutrophils. Due to the presence of multiple abstractable H atoms, PUFAs are highly susceptible to damage by oxygen-containing radicals²³. This feature of PUFAs is central to their involvement in ferroptosis⁸. While ferroptosis has primarily been studied in cancer cells, among non-cancer cells, the susceptibility of T cells to ferroptosis is quite well described. While ferroptosis does not appear to contribute to T cell development, T cells are highly susceptible to ferroptosis triggered by GPX4 deletion, leading to decreased numbers of T cells in the periphery and a failure to expand in number in response to infection³⁴. Moreover, even when GPX4 is present, T cells appear to be susceptible to lipid peroxidation and ferroptosis, as has been described for tumour-infiltrating CD8⁺ T cells^45,46 and follicular helper T cells⁴⁷, the consequences of which are impaired antitumour immunity and impaired responses to immunization, respectively. Why T cells are susceptible to lipid peroxidation, yet other immune cells (in particular myeloid cells) are resistant, was unclear. We provide a parsimonious explanation for this phenomenon, demonstrating that T cells are enriched in PUFA-PLs, notably those containing 22:6, which are highly susceptible to lipid peroxidation. Importantly, by either supplementing T cells with MUFAs or genetically deleting Acsl4, we show that biasing the PL acyl chain composition of T cells towards a MUFA-enriched profile decreased the susceptibility of T cells to lipid peroxidation and ferroptosis. In view of their propensity to undergo peroxidation, the high content of 22:6-containing PLs in T cells implies an essential role for these PLs in T cell function. Indeed, while ACSL4-deficient T cells are resistant to ferroptosis, they have impaired antitumour CD8⁺ T cell immune responses²⁹. The role of PUFA-PLs, particularly 22:6, in T cell function will be an important area for future work.

Previous studies have shown that myeloid cells, including neutrophils and monocytes, are resistant to ferroptosis triggered by GPX4 deletion³². Using our lipid atlas, we identified that neutrophils, and to a lesser extent Ly6C^hi monocytes, had markedly decreased levels of PUFA-PLs, particularly those containing the highly peroxidizable 20:4 and 22:6. Given the cardinal role of PUFA-PLs in ferroptosis, these observations provide an explanation for the resistance of myeloid cells to lipid peroxidation and ferroptosis. Indeed, using multiple approaches, we show that the neutrophil lipidome is resistant to lipid peroxidation triggered by GPX4 inhibition. GPX4’s unique ability to reduce PL-hydroperoxides to PL-alcohols places it at the centre of the cell’s defence against lipid peroxidation and ferroptosis^8,48. However, several additional pathways, in particular the DHODH²⁷, FSP1 (refs. ^24,25) and dihydrofolate reductase²⁶ pathways, help to limit ferroptosis. Although we observed a modest protective role of the FSP1 pathway in Ly6C^hi monocytes, these pathways do not appear to contribute to the resistance of neutrophils to PL peroxidation and ferroptosis. It is noteworthy that compared with neutrophils Ly6C^hi monocytes have higher levels of PUFA-PL yet are similarly resistant to ferroptosis. Indeed, a similar observation can be made about B cells, which have a very similar PUFA-PL profile to T cells, yet B cells are more resistant than T cells to ferroptosis triggered by GPX4 inhibition. These differences may be explained by differences in the profiles of PLs that undergo peroxidation in T and B cells (for example, we observed a marked decrease in ether PL only in T cells, and the peroxidation of ether lipids was recently shown to promote ferroptosis²¹) or potentially differences in the levels of known regulators of ferroptosis (for example, glutathione and iron) and the abundance and activity of GPX4 and FSP1. Alternatively, yet to be defined protective mechanisms may exist to limit ferroptosis in B cells and Ly6C^hi monocytes.

It was previously shown that neutrophils are resistant to ferroptosis triggered by GPX4 deletion³². Our results support and extend these observations, demonstrating that even for conditions under which several major lipid peroxidation and ferroptosis-suppressing pathways are inhibited, neutrophils are almost completely resistant to lipid peroxidation and ferroptotic cell death. Importantly, we also demonstrate that increasing the PUFA content of neutrophils is sufficient to trigger ferroptosis. These findings suggest that biasing their PL composition towards a less PUFA-enriched profile, thereby minimizing peroxidative damage to the cell membrane, is a major mechanism by which neutrophils avoid ferroptotic cell death. Indeed, as described above, a major feature of neutrophil development and maturation is a decrease in the proportion of highly peroxidizable 20:4 and 22:6 PUFAs.

The resistance of neutrophils to lipid peroxidation and ferroptosis relative to other immune cell types strongly suggests that avoiding ferroptotic cell death is an important aspect of neutrophil biology. This raises the question of why neutrophils, and indeed other myeloid cell types, are so resistant to ferroptotic cell death. As a consequence of their roles in host physiology and pathology, neutrophils operate within highly oxidative environments (for example, sites of infection and tissue damage). Additionally, a critical homeostatic function of neutrophils and monocytes is to patrol the blood vasculature⁴⁹—an environment that is rich in oxygen and iron and that promotes ferroptosis⁵⁰. Finally, the production of superoxide (O⁻₂) by NADPH oxidase is a critical feature of neutrophil-mediated immunity⁵¹. O⁻₂ is rapidly converted to H₂O₂, which, in the presence of iron, is converted to HO^•, a major inducer of lipid peroxidation and ferroptosis. In support of this concept, we have demonstrated that neutrophils with an increased content of highly unsaturated PUFA-PLs are more susceptible to ferroptosis induced by NOX2 activation. Accordingly, we propose that the neutrophil lipidome is specifically adapted to guard against oxidative damage to membrane lipids and ferroptotic cell death. These adaptations would allow myeloid cells to survive and function effectively in the toxic environments in which they must operate.

While our lipid atlas of the human and mouse immune systems provides information on a comprehensive (>500 individual lipid species) set of lipids important in various aspects of cell biology, there are nonetheless limitations to our work. First, our work is not an all-encompassing characterization of the lipid composition of immune cells. Future work should investigate how neutral lipid content (for example, triacylglycerols and COH esters stored within lipid droplets) varies between immune cell types, as well as trying to work towards a fuller characterization of the total lipid complement of immune cells. Our work has initially focused on the whole cell lipidome; however, it is well known that lipid composition varies between cellular organelles. It will be fascinating to determine, for example, whether nuclear or mitochondrial lipid composition varies between immune cell types and how such differences contribute to the cell-specific requirements of these organelles. Finally, while our work has focused on the immune system in the resting state, how the lipid content of immune cells and their progenitors is altered in vivo following scenarios of activation such as inflammation or infection will be an important area to explore in the future.

In summary, we have developed a lipid atlas of the cells of the human and mouse immune system. Using this resource, we have shown that cellular lipid composition is an identifying feature of immune cells. Our analysis of the lipid features of the human and mouse immune system has identified numerous, cell-specific lipid phenotypes that should provide fruitful avenues for further research into how differences in lipid composition impact cell-specific functionality. In this regard, we have used the lipid atlas to identify differences in the abundance of PUFA-containing PL as a major mechanism by which immune cells are differentially susceptible to lipid peroxidation and ferroptosis and why this may be important for fundamental aspects of neutrophil biology.

Methods

All of the mouse experiments adhered to national guidelines on the ethical treatment and utilization of laboratory mice, receiving approval from the Alfred Research Alliance (ARA) Animal Ethics Committee in Melbourne, Australia. Nox2^−/− mice, based on a C57Bl6/J background, were accommodated at La Trobe University with procedures approved by its institutional ethics committee. All of the human studies were approved by the Human Research Ethics Committee of Alfred Health (Melbourne, Australia).

Generation of Acsl4 ^−/− mice

Acsl4^−/− mice were generated at the Melbourne Advanced Genome Editing Centre (MAGEC). Cas9 messenger RNA (20 ng μl⁻¹) and single guide RNA of the sequences TTACTTATTTCGCCGCACTG and ACTTAGGCTGAGTTCAGCGT (10 ng μl⁻¹) were injected into fertilized one-cell-stage embryos from WT C57Bl6/J mice. Founder mice were identified by genotyping using the primers 5′-ACTTCATTCCTAAAGCATGACCAC-3′ (forward) and 5′-GCCACTGCCTCAAAACATGG-3′ (reverse) to identify the Acsl4 KO allele, under the following PCR conditions: 95 °C for 2 min, then 95 °C for 30 s, 60 °C for 30 s and 72 °C for 30 s for 35 repeats and finally 72 °C for 5 min. Next-generation sequencing was used to confirm the sequence of the deleted region. Selected founder mice were backcrossed for two generations to C57Bl6/J mice to breed out any potential off-target mutations. Acsl4^+/− mice were intercrossed to generate Acsl4^−/− mice and WT (Acsl4^+/+) littermate controls. Mice were routinely genotyped from tail clip samples. Genomic DNA was extracted using the Bioline MyTaq Extract-PCR kit (BIO-21126) according to the manufacturer’s instructions. WT, homozygous Acsl4^−/− and Acsl4^+/− alleles were distinguished using three primers (5′-ACTTCATTCCTAAAGCATGACCAC-3′ (Acsl4 common forward), 5′-GCCACTGCCTCAAAACATGG-3′ (Acsl4 KO reverse) and 5′-GGTTCTTGAGATGGTTTCAGG-3′ (Acsl4 WT reverse)) under the PCR conditions described above.

Preparation of cells for the human immune cell lipid atlas

Human immune cells were obtained from buffy coats received from the Australian Red Cross (Melbourne, Australia). Buffy coats were diluted 1:5 with phosphate-buffered saline (PBS) containing 5% foetal bovine serum (FBS) and 0.5 mM ethylenediaminetetraacetic acid. Blood was layered onto a discontinuous histopaque density gradient with the densities 1.077 and 1.119 g ml⁻¹ to isolate peripheral blood mononuclear cells and granulocytes, respectively, and centrifuged for 30 min at 300g with the brakes off. The two fractions were transferred into separate tubes, washed and centrifuged for 10 min at 200g with the brakes on to remove platelets. The supernatant was removed and the pellets were lysed with 1× red blood cell (RBC) lysis buffer for 5 min. Samples were then centrifuged at 300g for 5 min and the white cell pellet was obtained.

Preparation of cells for the mouse immune cell lipid atlas

Mouse immune cells were obtained from the peripheral blood of 4- to 6-week-old, male C57Bl/6J mice fed a standard laboratory mouse diet (SF00-100; Speciality Feeds). Mice were housed at the ARA Animal Services Facility. Mice were sacrificed via CO₂ asphyxiation and blood was obtained via cardiac puncture. Samples were lysed for 15 min in 1× RBC lysis buffer, which was stopped with the addition of 1× FACS buffer (Hanks’ balanced salt solution without Ca²⁺ and Mg²⁺, containing bovine serum albumin (BSA) and 0.5 mM ethylenediaminetetraacetic acid). Samples were then centrifuged at 500g for 5 min at 4 °C and the white cell pellet was obtained.

Preparation of cells for the myelopoiesis lipid atlas and in vitro experiments

Mature myeloid cells and their progenitors were obtained from the bone marrow (BM) of 8- to 10-week-old, male C57Bl/6J mice. Bones were flushed and cells were treated with 1× RBC lysis buffer for 5 min, with the reaction stopped with an excess of FACS buffer. Bone marrow cell pellets were then obtained after a final spin (300g for 5 min).

Antibody staining and FACS

Cells were stained with antibody cocktails developed to identify cell-specific surface markers, then incubated for 30 min on ice. Antibodies were used at a 1:400 dilution unless otherwise stated. Staining was stopped with FACS buffer and cells were subsequently washed and filtered through a 35-µm strainer before sorting. Information about the antibodies and sorting panels used to purify the specific human immune cell populations, murine cell populations and mature and progenitor cells for the lipid atlas data can be found in Supplementary Table 3.

FACS was performed at the ARA Flow Cytometry Core Facility. Individual cell populations were sorted using BD FACSAria, BD FACSAria Fusion and BD Influx (BD Biosciences). All of the gating strategies were first set up based on forward scatter area versus side scatter area, forward scatter height versus forward scatter area (doublet exclusion) and side scatter area versus viability dye (viable cell isolation). Sorted event thresholds were set to 250,000 and 60,000 cells for human and murine samples, respectively, and the cells were sorted according to the expression of the specific surface markers detailed in Supplementary Figs. 1–3. Following isolation, cells were washed with PBS without Ca²⁺ and Mg²⁺ and stored at −80 °C.

Lipid extraction

Cell samples were lyophilized using either a Savant SpeedVac (Thermo Fisher Scientific) or a Scanvac CoolSafe freeze dryer (LaboGene) before extraction, then resuspended in 10 µl Milli-Q H₂O. Lipids were extracted using a modified single-phase chloroform/methanol extraction method⁵². Briefly, 200 µl chloroform:methanol (2:1) was added to each sample along with an internal standard (ITSD) mixture containing stable isotope-labelled or non-physiological lipids. In tandem, blank control samples and plasma quality controls were extracted and dispersed evenly throughout the extraction order to ensure optimal assay performance and to monitor variation that may arise from the extraction. Samples were subsequently mixed with a rotary mixer for 10 min, sonicated for 30 min at room temperature and centrifuged at 10,000g for 10 min to precipitate proteins from the lipid extracts. Supernatant containing the extracted lipids was transferred to a 96-well plate and evaporated using the SpeedVac. Once dried, extracts were reconstituted in H₂O-saturated butanol and methanol with 10 mM ammonium formate and moved to glass vials to be stored until mass spectrometry analysis. To measure oxidized phospholipid species in neutrophils, cells were washed and stored in PBS with 0.1 mM thylenetriaminepentaacetic acid and 0.1 mM butylated hydroxytoluene before lyophilization and lipid extraction.

Liquid chromatography MS/MS

To facilitate better characterization of lipid species in immune cells, we adapted our previously reported approach⁵³, optimizing the workflow on pooled immune cell samples rather than plasma. Structural elucidation of PLs and SMs was carried out using a combination of conditions, as outlined by Huynh et al.⁵³, before the main analytical run. Tentative phospholipid acyl chain compositions for PC, PC(O), PC(P), SM, PE, PE(O), PI, PS, PG and PA were determined using pooled immune cell samples that were examined under the same gradient conditions. PE(P) composition was determined via the quantifier ion.

For the main analytical analysis, lipid extracts were analysed using an Agilent 6490 or 6495C triple quadrupole mass spectrometer coupled to an Agilent 1290 HPLC system and a ZORBAX Eclipse Plus C18 column (2.1 × 100 mm; 1.8 µm; Agilent) with the thermostat set to 45 °C. The following mass spectrometer conditions were used: gas temperature = 150 °C; gas flow rate = 17 l min⁻¹; nebulizer = 20 psi; sheath gas temperature = 200 °C; capillary voltage = 3,500 V; and sheath gas flow = 10 l min⁻¹. The final mass spectrometry analysis on each cell population was performed in positive mode with dynamic scheduled multiple reaction monitoring. The solvents comprised solvent A (50% H₂O, 30% acetonitrile and 20% isopropanol with 10 mM ammonium formate) and solvent B (1% H₂O, 9% acetonitrile and 90% isopropanol with 10 mM ammonium formate) and followed a modified 20-min gradient (Supplementary Table 4).

A wash vial comprising 1:1 butanol:methanol was used after each sample injection. To further improve chromatic peak shape for many anionic or acidic lipid species (notably, PS, PA, glycerophosphoinositol monophosphate and sphingosine-1-phosphate), an additional pre-run passivation step was performed with phosphoric acid to minimize interaction between the HPLC unit and these lipids.

While it is well reported that acyl-specific product ions in negative ioniziation mode have a variable signal response dependent on acyl compositions, previous studies have also highlighted substantial differences in phosphatidylcholine signal response with the common 184.1 product ion⁵⁴. However, it was noted that optimal electrospray ionization conditions, along with appropriate sample concenrtations, mitigated these effects⁵⁵. As we relied on headgroup-specific fragmentation to quantify phosphatidylcholines (and infer composition), we examined the signal response of phosphatidylcholine when quantifying with the phosphocholine 184.1 product ion under our conditions. We utilized a pre-mixed PC ITSD set (330825; Avanti) comprising the following: PC(17:0_14:1)-d5 (50 µg ml⁻¹), PC(17:0_16:1)-d5 (100 µg ml⁻¹), PC(17:0_18:1)-d5 (150 µg ml⁻¹), PC(17:0_20:3)-d5 (100 µg ml⁻¹) and PC(17:0_22:4)-d5 (50 µg ml⁻¹). Standards were thawed and 10 µl of each standard was dried under a stream of nitrogen and reconstituted into 100 µl 1:1 butanol:methanol with 10 mM ammonium formate, then further diluted 1:16 to final concentrations of 0.432 µM PC(17:0_14:1)-d5, 0.832 µM PC(17:0_16:1)-d5, 1.203 µM PC(17:0_18:1)-d5, 0.778 µM PC(17:0_20:3)-d5 and 0.377 µM PC(17:0_22:4)-d5. A 1 µl injection was used for analysis under the conditions described previously. Resulting areas were calculated relative to PC(17:0_16:1)-d5 (equivalent to our ITSD acyl length) and contrasted to the expected response based on the injected concentration. Using the pre-mixed deuterated PC ITSDs normalized to PC(17:0_16:1)-d5, we observed a small overestimation for shorter species (PC(17:0_14:1)-d5; 1.16-fold; Extended Data Fig. 9a,b) and a small underestimation for longer PC species (PC(17:0_22:4)-d5; 0.77-fold; Extended Data Fig. 9a,b).

Liquid chromatography MS/MS for oxidized lipid measurements

In a secondary analysis of oxidized PL species, the method was adjusted to increase the sensitivity for oxidized phospholipids. The mass spectrometer conditions (6495C; Agilent) were as follows: gas temperature = 200 °C; gas flow rate = 17 l min⁻¹; nebulizer = 20 psi; sheath gas temperature = 325 °C; capillary voltage = 3,500 V; and sheath gas flow = 10 l min⁻¹. The chromatography was further adjusted to further separate the oxidized phospholipids from co-eluting species. Oxidized species were annotated and identified by examining masses with an increasing number of oxygen atoms (16 Da increments), plus corresponding earlier elution times relative to their non-oxidized species. Chromatographically separated peaks were annotated with ‘a’ and ‘b’ notations. The majority of the oxidized species were identified within the PC, PC(O) and PC(P) subclasses and displayed a distinct 184.1 product ion. Solvent A was modified to include 5 µM medronic acid. Chromatography adjustments can be found in Supplementary Table 5.

To determine the levels of oxidized lipids in free fatty acid stocks (from unopened vials of AA and DHA), we examined the amount of lipid peroxidation from handling AA and DHA stock solutions before our in vitro experiments. Stock solutions were made to 500 mM in ethanol. The following mass spectrometer conditions were used: gas temperature = 150 °C; gas flow rate = 17 l min⁻¹; nebulizer = 20 psi; sheath gas temperature = 200 °C; capillary voltage = 3,000 V, nozzle voltage = 1,500 V; and sheath gas flow = 10 l min⁻¹. Serial injections of 1 µl were performed on each fatty acid solution (19.53, 39.06, 78.13 and 156.25 µM) diluted with 1:1 butanol:methanol. A Q1 scan was conducted over the range 150–600 m/z, with a cycle time of 800 ms in negative ionization mode. We examined ratios of AA ([M − H]⁻ = 303.2 m/z) and DHA ([M − H]⁻ = 327.2 m/z) with their corresponding –OH and –OOH species (319.2, 335.2, 343.2 and 359.2 m/z for 20:4;O, 20:4;O₂, 22:6;O and 22:6;O₂ respectively.

The acyl compositions were determined via a Q Exactive HF-X Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific) in tandem with a Vanquish HPLC running under identical gradient conditions to those described above. Collision-induced dissociation was performed in positive ionization mode for PC, PC(O), PC(P) and SM, where the sodiated adduct [M + Na]⁺ fragments corresponding to the neutral loss of the acyl chains plus trimethylamine were examined. Collision-induced dissociation of all other phospholipid species was performed in negative ionization mode, monitoring the fragments corresponding to fatty acyl chains. Acyl chain composition was determined by aligning species by their relative retention times. Isomers that were not fully determined (low abundance or substantial mix of species) are annotated as their sum composition and/or ‘a’ and ‘b’ notations if resolved chromatographically into distinct peaks.

Lipid nomenclature

The lipid names used follow guidelines set by the LIPID MAPS consortium^56,57,58. Phospholipids with detailed characteristics (that is, acyl chain composition) are annotated as ‘PC(16:0_20:4)’, with PC being the lipid class and (16:0_20:4) representing the acyl chains found on the glycerol backbone, irrespective of sn1 or sn2 position. Lipids without specific structural annotations are named based on their sum acyl chain length and degrees of saturation (for example, PC(36:4)). Isomeric lipid species separated chromatographically but incompletely annotated were designated ‘a’, ‘b’ and so on, with ‘a’ and ‘b’ representing the elution order. Owing to technical limitations, we were unable to assign acyl chains to a specific sn1 or sn2 position for the majority of PL species—only that the acyl chain was present at either the sn1 or sn2 position. In the case of ether PC and PE, the alkyl or alkenyl chains are always located at the sn1 position and the acyl chain is always at the sn2 position. In addition, we were unable to determine acyl chain composition for all PLs and the data shown therefore represent those PLs for which we were able to obtain information on structural composition. Importantly, this represents >90% of total PLs for all cell types, with the exception of neutrophils (~80%); therefore, our results probably accurately reflect global PL alkyl/acyl chain composition.

Data normalization

Analyte areas were obtained from integrating chromatograms that corresponded to a lipid of interest using the MassHunter Quantitative Analysis Software (version B9.00; Agilent). Lipid concentrations were determined using the following formula:

$$\begin{array}{l}\left\{{{\rm{Sample}}}\left(\frac{{{{\rm{area}}}}_{{{\rm{analyte}}}}}{{{{\rm{area}}}}_{{{\rm{ITSD}}}}}\right)-{{\rm{study}}}\,{{\rm{median}}}\,{{\rm{blank}}}\,\left(\frac{{{{\rm{area}}}}_{{{\rm{analyte}}}}}{{{{\rm{area}}}}_{{{\rm{ITSD}}}}}\right)\right\}\\\times {{{\rm{concentration}}}}_{{{\rm{ITSD}}}}\times {{R}}_{{{\rm{analyte}}}}\,\end{array}$$

In brief, individual analyte areas were divided by the area of the corresponding ITSDs, and the median of ITSD mixtures containing blank samples was subtracted from each analyte (background subtraction). This value was then multiplied by the ITSD concentration and the individual analyte’s response factor (R). Any values that were zeroed after background subtraction as a consequence of being less than the median value of all blank + ITSD samples were replaced with 1/10th of the minimum value for the corresponding analyte. The data were ultimately normalized to pmol or µmol of total lipidome, where the background subtracted datum for an individual lipid was divided by the sum of the total lipidome and multiplied by a factor of 10⁶ for ease of graphical representation.

Initial data analysis

To identify any technical outliers within the mouse and human lipid atlas datasets, we initially inspected the datasets using a variety of unbiased analytical methods. A heatmap of log-transformed lipid proportions was created for each sample group. Samples were hierarchically clustered using complete linkage on Euclidean distance. Each heatmap was visually inspected for potential outlier samples. Concomitantly, principal component analysis (PCA) was performed on each sample group and score plots were manually examined to aid outlier identification. In addition, the distances to the origin values (distO; that is, the distances between each sample and the origin in principal component space) were calculated as a measure of sample extremeness within each PCA, using as many principal components as was necessary to capture at least 70% of the total variability in the data. Samples with the largest distO values within sample groups were flagged as potential outliers. Next, we examined the lipid data for each sample that was identified as a potential technical outlier. Only samples for which a clear technical error in the data could be identified were removed; typically, such samples would contain multiple lipid parameters that were three to four standard deviations from both the group mean (that is, within a given cell type) and all other cell group means within the atlas. For the human cell atlas, this resulted in the following n numbers from the initial 14 human donors: naive B cells (14); memory B cells (13); CD4⁺ T_Naive cells (13); CD4⁺ T central memory cells (T_CM cells) (14); CD4⁺ T_EM cells (13); CD8⁺ T_Naive cells (14); CD8⁺ T_CM cells (12); CD8⁺ T_EM cells (13); CD56^dim NK cells (12); CD56^bright NK cells (12); classical monocytes (14); intermediate monocytes (14); non-classical monocytes (13); basophils (14); eosinophils (13); and neutrophils (11). For the mouse cell atlas, this resulted in the following n numbers from the initial ten C57Bl6/J mouse donors: B cells (8); CD4⁺ T cells (9), CD8⁺ T cells (8); NK cells (9); Ly6C^lo monocytes (9); Ly6C^hi monocytes (10); eosinophils (9); and neutrophils (10).

Lipidomic data analysis

All downstream analyses were conducted in R version 3.6.2 (Comprehensive R Archive Network). PCA was performed using the FactoMineR and factoextra packages using log₁₀-transformed lipid concentrations. Data were centred and scaled before PCA. For t-SNE analysis, datasets were log transformed and z score or unit variance scaled (all lipids have a mean of 0 and a standard deviation of 1). t-SNE was used to embed the lipidomics data on to two dimensions for visualization. One-way analyses of variance (ANOVAs) followed by post-hoc testing using Tukey’s test were performed using in-built functions in R. ANOVA-generated P values were corrected for multiple comparisons using the Benjamini–Hochberg method for false discovery rate (FDR) (5%) correction. FDR-corrected P values were statistically significant when P < 0.05. Z scores were calculated by subtracting the group mean from the individual raw score and dividing by the standard deviation.

Calculation of the CPI

The propensity for a PUFA to be peroxidized is determined by the number of bis-alllyic carbons (CH₂ centres flanked by double bonds) that are the target of free radical H atom abstraction. We calculated the CPI based on the known relative H atom transfer propagation rate constants of fatty acids in solution (0.014, 1.000, 3.200, 4.000 and 5.400 for fatty acids with one, two, four, five and six double bonds, respectively^22,23); saturated fatty acids are essentially unreactive to free radical oxidation at 37 °C. Because a linear relationship exists between the H atom transfer propagation rate constants of fatty acids and the number of bis-allylic carbons, we interpolated a relative H atom transfer propagation rate constant for fatty acids with three double bonds (2.08) as this was not empirically determined in previous work. To determine the CPI, we first calculated the sum of each acyl chain within the total PL acyl chain pool (expressed as a percentage) multiplied by its corresponding relative H atom transfer propagation rate constant (for example, 4.56 (the percentage of 22:6 acyl chains within the total PL acyl chain pool in naive B cells) × 5.4 (the relative H atom transfer propagation rate constant for fatty acids with six double bonds)).

Lipidomic analysis in ML210-treated immune cells

In addition to the tibias and femurs, spleens were collected. Bone marrow and spleen cells were flushed separately through a 40-μm filter with RPMI to obtain a single-cell suspension. RBCs were lysed and cells were stained on ice with antibodies (Supplementary Table 6 and Supplementary Fig. 4) at 1:400 dilution for 30 min in the dark. After 30 min, cells were washed and centrifuged (500g for 5 min). Cell pellets were resuspended in ~1 ml RPMI and filtered (40 μm) before sorting on a BD FACSAria Fusion flow cytometer. Neutrophils, monocytes and B cells were sorted from the bone marrow and T cells from spleen. Sorted cells were centrifuged (500g for 10 min), resuspended in RPMI and seeded (200,000 cells per well) into a 96-well plate, respectively, with the respective reagents. After the experimental period, cells were collected and washed twice in 200 μl ice-cold PBS (without Ca²⁺ and Mg²⁺) and sonicated using an S-4000 sonicator (Misonix) at an amplitude of 13% for 25 s. Protein concentration was determined by Quant-iT Protein Assay Kit (Thermo Fisher Scientific) per the manufacturer’s protocol. To prepare samples for lipid extraction, the cell homogenates were stored at −80 °C for 2 h and dried down overnight in a Savant SpeedVac (Thermo Fisher Scientific). The sample order was randomized and lipids were extracted using a single-phase butanol/methanol extraction method, as described previously⁵⁹. Mass spectrometry analysis was performed as described above. Data were normalized to the total lipidome (the sum of all phospholipids, SLs, free COH and lyso-PLs) and data from individual lipid species are presented relative to the total lipidome.

Preparation of human immune cells for in vitro experiments

Peripheral blood samples were collected in lithium heparin-coated tubes from healthy male volunteers (21–50 years of age) after informed consent was provided. To lyse RBC, samples were treated with sterile 1× RBC lysis for 15 min with frequent vortexes, after which samples were washed in FACS buffer. Neutrophils were isolated using a MojoSort Whole Blood Human Neutrophil Isolation Kit (BioLegend) and T cells were isolated using a MojoSort Human CD3 T cell Isolation Kit (BioLegend). Purified human neutrophils (200,000 cells per well) and T cells (100,000 cells per well) were cultured in Human Plasma-Like Medium (Thermo Fisher Scientific) supplemented with 5% FBS in 96-well plates, respectively. Cells were treated with either ML210, RSL3 or erastin for 72 h in the absence or presence of Fer-1 (1 μM), after which lipid peroxidation and cell viability were assessed via flow cytometry.

Flow cytometry for in vitro experiments

Following treatment in 96-well plates, cells were washed, centrifuged (500g for 5 min) and stained in 100 μl FACS buffer containing either mouse or human antibodies. Antibody panel and gating strategy information can be found in Supplementary Table 6 and Supplementary Figs. 4 and 5. Upon identification of cell type, cell viability and lipid peroxidation levels were determined on an LSRII Fortessa (BD Biosciences). Data were analysed using FlowJo.

Cell viability analysis

The viability of the cultured cells was determined by 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI; BD Pharmingen), a nucleic acid stain that is excluded from viable cells. Immediately before flow cytometry analysis, cells were resuspended in FACS buffer containing 0.05 μg ml⁻¹ DAPI solution and incubated for 5 min at room temperature in the dark. After incubation, cells were immediately analysed on the flow cytometer. Upon identification of cell type, cell viability was determined using DAPI fluorescence, where DAPI⁻ cells were considered viable cells and DAPI⁺ cells were considered dead cells.

Lipid peroxidation analysis

Lipid peroxidation was assessed using the fluorescent probe Liperfluo (Dojindo). In the basal state, Liperfluo does not fluoresce, but upon reaction with lipid hydroperoxides it emits a fluorescence signal with an emission of ~510 nm (fluorescein isothiocyanate channel). Before the collection of cells, 2 μM Liperfluo was resuspended in the culture medium for the last 30 min of treatment at 37 °C, after which cells were washed with cold FACS buffer and stained for surface markers before the mean fluorescence intensity of Liperfluo was determined.

Mouse cell purification for in vitro experiments

Following RBC lysis, neutrophils were isolated from bone marrow using an EasySep Mouse Neutrophil Enrichment kit (STEMCELL Technologies) and T cells were isolated from the spleen using the MojoSort Mouse CD3 T Cell Isolation Kit (BioLegend). For experiments in which purified T cells were cultured for longer than 24 h, murine interleukin-7 (10 ng ml⁻¹; PeproTech) was supplemented in the medium to prolong survival and decrease basal cell death.

Lipid supplementation to modulate immune cell lipidomes

Lipid supplementation experiments were performed in purified cell types using enrichment kits. Before the start of the experiment, AA (20:4; Sigma–Aldrich), DHA (22:6; Sigma–Aldrich) and OA (18:1; Sigma–Aldrich) were solubilized in ethanol to a stock concentration of 1 M. Subsequently, the respective fatty acid control (ethanol), 200 μM OA and 400 μM DHA and AA were dissolved in filtered (0.2 μM Minisart filter; Sigma–Aldrich) RPMI media containing 5% FBS and 2% wt/vol BSA (Sigma–Aldrich). Fatty acid mixtures were gently rocked for 1 h at 37 °C to aid conjugation of the fatty acids to BSA. Once prepared, BSA-conjugated AA + DHA or vehicle (BSA + ethanol) mixtures were then added to respective wells containing purified bone marrow mouse neutrophils (250,000 cells per well). BSA-conjugated OA (200 μM) or vehicle (BSA + ethanol) mixtures were then added to respective wells containing purified spleen CD3⁺ T cells (250,000 cells per well). 1-stearoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (PE(18:0/18:1); Avanti) was dissolved in ethanol and sonicated in a water bath for 15 min. Isolated spleen CD3⁺ T cells (250,000 cells per well) were treated with either vehicle (ethanol) or PE(18:0/18:1) (20 μM).

Western blot analysis

Analysis of NOX2 protein expression was performed in cells isolated from bone marrow. Briefly, 250,000 bone marrow cells per mouse were directly lysed in 2× Laemmli buffer with 200 μM dithiothreitol and heated at 95 °C for 15 min. Lysates were separated on 12% sodium dodecyl sulfate polyacrylamide gel electrophoresis gels and transferred to a 0.2 μM nitrocellulose membrane. On a rotary shaker, the membranes were washed in Tris-buffered saline containing 0.1% Tween20 (TBST) (×3 for 15 min), blocked (1 g skim milk in 20 ml TBST for 1 h at room temperature) and probed (4 °C overnight) with primary antibody against NOX2. Membranes were subsequently washed and then stained with a secondary antibody (anti-rabbit IgG (H + L)–HRP conjugate; Bio-Rad). After a final series of washes in TBST (×3 for 15 min), membranes were reacted with a chemiluminescent substrate solution (Thermo Fisher Scientific) and immunoreactive bands were detected using a ChemiDOC XRS+ system (Bio-Rad) and imaged with Quantity One.

Statistics and reproducibility

All statistical testing was performed using either R or GraphPad Prism software. No statistical methods were used to predetermine sample sizes, but our sample sizes are consistent with recent publications on lipidomics and ferroptosis in Nature Cell Biology^60,61. The data distribution was assumed to be normal, but this was not formally tested. Mouse or human cell samples with insufficient viable cell yield were excluded from analysis by flow cytometry. All samples were randomized before lipid extraction. The investigators were not blinded to allocation during the experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.