Ferroptotic cell death triggered by conjugated linolenic acids is mediated by ACSL1

Ferroptosis is associated with lipid hydroperoxides generated by the oxidation of polyunsaturated acyl chains. Lipid hydroperoxides are reduced by glutathione peroxidase 4 (GPX4) and GPX4 inhibitors induce ferroptosis. However, the therapeutic potential of triggering ferroptosis in cancer cells with polyunsaturated fatty acids is unknown. Here, we identify conjugated linoleates including α-eleostearic acid (αESA) as ferroptosis inducers. αESA does not alter GPX4 activity but is incorporated into cellular lipids and promotes lipid peroxidation and cell death in diverse cancer cell types. αESA-triggered death is mediated by acyl-CoA synthetase long-chain isoform 1, which promotes αESA incorporation into neutral lipids including triacylglycerols. Interfering with triacylglycerol biosynthesis suppresses ferroptosis triggered by αESA but not by GPX4 inhibition. Oral administration of tung oil, naturally rich in αESA, to mice limits tumor growth and metastasis with transcriptional changes consistent with ferroptosis. Overall, these findings illuminate a potential approach to ferroptosis, complementary to GPX4 inhibition.

Ferroptosis is a form of non-apoptotic, iron-dependent, regulated cell death linked to lipid hydroperoxides, a form of reactive oxygen species (ROS)1,2,3,4,5,6,7. Interest in the therapeutic potential of inducing ferroptosis in cancer cells has led to a major research focus on inhibiting cellular antioxidant pathways that oppose ferroptosis. GPX4 is the only peroxidase known to efficiently reduce esterified, hydroperoxy fatty acids into unreactive alcohols8. Recently, ferroptosis suppressor protein 1 (FSP1) was identified as a second ferroptosis suppression mechanism through its recycling of coenzyme Q10, a radical-trapping antioxidant9,10. Small molecules that trigger ferroptosis identified thus far include agents that inhibit GPX4 directly, molecules that deplete the GPX4 cofactor glutathione, and compounds that oxidize iron1,9,11,12. Lipid hydroperoxides are generated by non-enzymatic oxidation or by lipoxygenases or cytochrome P450 oxidoreductase13,14,15. Hydroperoxides disrupt membrane architecture, produce reactive aldehydes, and drive cell death16. In contrast to therapeutic targeting of antioxidant pathways, we have pursued a complementary strategy for ferroptosis induction based on enhancing the production of lipid hydroperoxides.

Hydroperoxides can spread by a free radical-mediated chain reaction leading to oxidation of adjacent polyunsaturated acyl chains, and consequently higher levels of this species increase vulnerability to peroxide propagation. Indeed, supplementing cells with arachidonic acid sensitizes them to ferroptosis triggered by GPX4 inhibitors17,18. Similarly, clear-cell carcinomas are rendered vulnerable to GPX4 inhibitors due to HIF-2α-dependent polyunsaturated lipid accumulation19. Conversely, increasing the relative proportion of monounsaturated acyl chains at the expense of polyunsaturated lipids suppresses sensitivity to ferroptosis20,21. Thus, increasing polyunsaturated lipids in cancer cells could enhance their vulnerability to ferroptosis. Importantly, this approach could exploit the propensity of cancer cells to scavenge fatty acids from their environment22.

Here, we report that linolenic fatty acids with conjugated double bonds, produced by certain plant species, can induce ferroptosis in diverse cancer cells as a single agent. Conjugated linoleate triggers ferroptosis by a mechanism distinct from canonical ferroptosis inducers. Cell death is mediated by acyl-CoA synthetase long-chain family member 1 (ACSL1), implicating this isoform in ferroptosis. Furthermore, oral administration of tung tree seed oil, naturally rich in the conjugated linoleate α-eleostearic acid (αESA), inhibits tumorigenesis and metastasis in a murine breast cancer xenograft model. αESA metabolites are detected in tumors of treated mice and are associated with the expression of a ferroptotic gene signature. These results introduce a distinct class of ferroptosis inducers and offer insights into the molecular basis of ferroptotic sensitivity. The tractability of these dietary, pro-ferroptotic fatty acids addresses the current lack of effective GPX4 inhibitors for use in vivo and suggests an opportunity to exploit a metabolic liability across cancer subtypes.

Glutathione depletion triggers ferroptosis in triple-negative breast cancer cells

Several triple-negative breast cancer cell lines are killed by the glutathione biosynthesis inhibitor buthionine sulfoximine (BSO) (Fig. 1a and ref. 23). However, the specific cellular pool of ROS responsible for cell death is unclear. Glutathione depletion in some contexts induces ferroptosis24, which is operationally defined by three criteria; its association with lipid peroxidation, the ability to suppress cell death with lipophilic antioxidants like ferrostatin-1 (fer-1), and iron-dependence1. Consistent with ferroptosis, cytotoxicity in BSO-treated BT-549 cells was blocked by fer-1 but not by Z-VAD-FMK, an inhibitor of apoptosis (Fig. 1a, b). Similarly, iron chelators deferoxamine and deferiprone suppressed BSO-mediated cell death (Fig. 1c, Supplementary Fig. 1a). Furthermore, BSO treatment resulted in the accumulation of lipid peroxidation products and this was suppressed by fer-1 (Fig. 1d, e). Thus, glutathione depletion triggers death consistent with ferroptosis in BT-549 cells and identifies lipid hydroperoxides as the lethal ROS species underlying glutathione addiction in these cells.

Sensitivity to ferroptosis is associated with the accumulation of polyunsaturated lipids

Fer-1 suppressed BSO toxicity in half of the triple-negative breast cancer (TNBC) cell lines tested, consistent with ferroptosis (Fig. 1f, red bars). These cell lines were also generally more sensitive to ML162, a small molecule inhibitor of GPX425 (Welch’s t-test, p = 0.01, Fig. 1g, Supplementary Fig. 1b). Thus, fer-1-suppressible cell death is a common response to glutathione depletion in a subset of TNBC cells and is associated with a greater dependence on GPX4 activity. This vulnerability was not due to decreased GPX4 expression (Supplementary Fig. 1c). We used ANOVA of our prior metabolomic data for these cell lines23to compare levels of individual metabolites between TNBC cell lines in which fer-1 rescued BSO-mediated cell death and those in which it did not (Fig. 1f), and found that two of the three most significantly differentially accumulated metabolites were linoleate (18:2)-substituted phosphatidylcholines (Fig. 1h, i). These membrane phospholipids were more enriched in cell lines that undergo fer-1-suppressible death. This finding suggested the hypothesis that polyunsaturated lipid enrichment may increase dependence on GPX4 and vulnerability to ferroptosis (Fig. 1g).

αESA triggers cancer-selective ferroptosis

We postulated that elevating polyunsaturated lipid levels might sensitize TNBC cells to ferroptosis. We tested a variety of fatty acids and identified α-eleostearic acid [(9Z,11E,13E)-octadeca-9,11,13-trienoic acid; αESA] (Fig. 2a) as a fatty acid that enhanced cell death in combination with BSO (Fig. 2b). αESA is abundant in certain plants (e.g. bitter melon and tung tree)26 and was previously shown to suppress the growth of breast cancer cells in a manner reversible by antioxidants27. Others have suggested that αESA triggers apoptotic cell death28,29,30,31,32,33 although a role for ferroptosis was not investigated. Unexpectedly, we found that αESA triggered cell death as a single agent and this death was suppressed by fer-1 (Fig. 2b), the iron chelator deferoxamine (Fig. 2c), and by the lipophilic antioxidant vitamin E (Supplementary Fig. 2a), and was associated with an increase in lipid peroxidation products that could be suppressed by fer-1 (Fig. 2d). Neither Z-VAD-FMK nor the necroptosis inhibitor nec-1s blocked αESA-induced cell death (Supplementary Fig. 2b, c). Together, these findings show that αESA induces ferroptosis as a single agent and distinguishes αESA from arachidonic and adrenic fatty acids, which sensitize cells to GPX4 inhibition but do not induce ferroptosis13,17.

The non-transformed MCF-10A (Fig. 2e) and MCF-12A (Supplementary Fig. 2d) breast epithelial cell lines were resistant to death by αESA. By contrast, all TNBC cell lines were susceptible and, in all cases, death was prevented by fer-1 (Fig. 2b, f). Notably, even cell lines that did not undergo ferroptosis in response to glutathione depletion (Fig. 1f) were sensitive to αESA, pointing to a distinct mechanism of ferroptosis induction. Time-lapse microscopy of αESA-treated BT549 cells (Supplementary Movie 1) showed that during the latter part of 24 h of treatment, otherwise healthy appearing cells undergo a sudden death, morphologically similar to cell death by the canonical ferroptosis inducer ML162 (Supplementary Movie 2) and distinct from apoptosis (Supplementary Movie 3).

αESA is incorporated into diverse cellular lipids

To examine how αESA is metabolized, we conducted a lipidomic analysis of MDA-MB-231 cells treated with αESA for three hours, prior to overt cell death. Lipids containing acyl chains with 18 carbons and three double bonds (18:3), consistent with αESA, were rare in untreated cells (Fig. 2g, blue bars), but were found in diverse lipids in αESA-treated cells, including phospholipids and neutral lipids (green bars; Supplementary Data 1). Fer-1 co-treatment did not dramatically alter the spectrum of lipid classes that incorporated 18:3 acyl chains (compare green & purple bars). The relative abundance of di- and triacylglycerol lipids increased ~2-fold in αESA and αESA + fer-1-treated cells compared to untreated controls, consistent with αESA being incorporated into storage lipids (Fig. 2h). αESA-induced lipidomic changes were not downstream consequences of ferroptosis because they were not replicated by ML162 treatment (orange bars in Fig. 2g, h).

Structure-activity analysis of conjugated PUFAs

To characterize structural features of αESA required for ferroptosis, we screened related conjugated and unconjugated polyunsaturated fatty acids for death induction. αESA possesses conjugated double bonds at carbons 9, 11, and 13 with cis, trans, trans stereochemistry, respectively (Fig. 2a). We examined cis-trans stereoisomers of αESA and found that isomerization at position 9 (β−eleostearic acid) or at both 9 and 13 (catalpic acid) retained cell death activity while isomerization at position 13 alone (punicic acid) led to a ~4-fold reduction in potency (Fig. 3a). Jacaric acid (conjugated 18:3, cis-8, trans-10, cis-12) showed the most potent cell killing (1.8 µM IC50), demonstrating that shifting double bond positioning while maintaining the sequential cis, trans, cis stereochemistry of punicic acid, did not disrupt activity. α−calendic acid (conjugated 18:3, trans-8, trans-10, cis-12), a stereoisomer of jacaric acid, was similarly potent. Thus, cell killing is not strictly dependent on the precise positioning or stereochemistry of double bonds within the aliphatic chain. Conjugated linoleic acid (18:2), which shares similar position and stereochemistry of its two double bonds with αESA, was a poor inducer of cell death, however, suggesting that the third conjugated double bond is required for activity. Furthermore, all tested non-conjugated polyunsaturated fatty acids (PUFAs) were much less potent single-agent inducers of cell death (Supplementary Fig. 2e) including arachidonate, which enhances ferroptosis in some contexts17,18. Cell killing by arachidonate could not be rescued by fer-1, liproxstatin or deferoxamine, highlighting an important distinction between the activity of conjugated and non-conjugated PUFAs (Supplementary Fig. 2f). By contrast, jacaric- and catalpic acid-induced death was suppressed by fer-1 (Supplementary Fig. 2g). Similar results were found in BT-549 cells (Supplementary Fig. 2h), demonstrating the conserved ability of conjugated 18:3 fatty acids to trigger ferroptosis.

To validate the relative potency of these fatty acids as inducers of cell death, we profiled their effects on viability across 100 human cancer cell lines at 6 doses using the PRISM screening platform34. Thunor35 was used to fit dose-response curves and the area over the curve (Activity Area) was calculated to reflect the potency of cell killing35. A histogram reflecting the range of Activity Areas across the cell line panel for each fatty acid is presented as a violin plot in Fig. 3b. Jacaric and catalpic acid were consistently most toxic to the panel, followed by α-calendic acid, αESA and βESA. Whether the differential potency of these conjugated linolenic acids is related to their metabolism by specific enzymes or due to differences in solubility or other physicochemical properties is an important open question. Punicic acid and arachidonic acid exhibited substantially less anti-proliferative activity and the remaining control fatty acids were largely inactive. These results support a unique death-inducing activity of conjugated linolenic acids.

The tissue of origin of cell lines did not appear to correlate with αESA sensitivity or resistance. However, the sensitivity of individual cell lines to one conjugated linolenate was correlated with sensitivity to the others, but not with control fatty acids (Fig. 3c). Taken together these results show that conjugated linolenic acids exhibit broad toxicity against multiple cancer cell lines of distinct tissue origin. Furthermore, sensitivity or resistance to conjugated linolenic acids is intrinsic to each cell line.

The role of GPX4 in αESA-induced ferroptosis

Next, we sought mechanistic insights into ferroptosis induction by αESA. One possibility is that αESA triggers ferroptosis by inhibiting GPX4 directly or indirectly by depleting glutathione. Unlike BSO, treatment with αESA did not significantly affect total cellular glutathione levels despite decreasing cell viability (Fig. 4a). Direct GPX4 inhibitors have been shown to reduce GPX4 protein levels17. We confirmed this for ML162 and RSL3 but found that αESA or the negative control erastin had no effect on GPX4 levels (Fig. 4b), suggesting that αESA does not directly inhibit GPX4. As an alternative approach to test if αESA inhibits GPX4, we compared the kinetics of ferroptosis in MDA-MB-231 cells treated with αESA, ML162, and RSL3 alone and in combination. Combining both GPX4 inhibitors at 500 nM triggered fer-1-suppressible cell death with similar kinetics to either GPX4 inhibitor alone, indicating this was a saturating dose of inhibitor (Fig. 4c). In contrast, cell viability was lost significantly more rapidly when cells were co-treated with 500 nM ML162 and 50 µM αESA, a dose of αESA that took longer to induce cell death than either of the GPX4 inhibitors alone. Consistent with ferroptosis, cell death could be rescued by fer-1 (green in Fig. 4c). The ability of αESA to enhance the kinetics of cell death when GPX4 is maximally inhibited shows that αESA and GPX4 inhibitors trigger ferroptosis by distinct mechanisms.

αESA enhanced the toxicity of GPX4 inhibitors in multiple cell lines (Fig. 4d and Supplementary Fig. 3), suggesting that the reduction of hydroperoxides by GPX4 opposes αESA-induced ferroptosis. Consistent with this model, transient knockdown of GPX4 by siRNA also enhanced cell killing by αESA (Fig. 4e). Finally, we generated MDA-MB-231 and BT549-derived cell lines stably overexpressing GPX4 or eGFP as control (Fig. 4f) and found that cell lines overexpressing GPX4 were resistant to ML162-triggered ferroptosis (Fig. 4g) and αESA-induced ferroptosis, though the resistance to ferroptosis triggered by αESA was less robust in BT-549 cells (Fig. 4h).

FSP1 mediates a second pathway for hydroperoxide neutralization that operates in parallel to GPX4. By reducing coenzyme Q10, FSP1 maintains an active pool of this radical-trapping lipophilic antioxidant9,10. As expected, iFSP1, a small-molecule inhibitor of FSP19, enhanced cell killing by ML162 in both BT-549 and MDA-MB-231 cells (Fig. 4i). By contrast, iFSP1 had no effect on αESA toxicity in BT-549 cells, but enhanced cell death in MDA-MB-231 cells (Fig. 4j), highlighting a further distinction between ferroptosis induced by αESA and canonical ferroptosis inducers.

αESA-induced ferroptosis is ACSL1-dependent

The first step in esterification of fatty acids into lipids is their conjugation to coenzyme A (CoA), catalyzed by acyl-CoA fatty ligases. There are five long chain-specific isoforms (ACSLs) that are distinguished by fatty acid specificity and subcellular localization36. ACSL4 preferentially esterifies certain PUFAs including arachidonic acid and loss of ACSL4 protects cells from ferroptosis induced by direct or indirect inhibition of GPX4 (Fig. 5a)37,38,39. αESA, however, was similarly toxic to ACSL4-deficient Pfa1 cells and control parental cells (Fig. 5b), supporting a difference in the mechanism of action of αESA and canonical ferroptosis inducers. αESA-induced cell death was rescued by fer-1 in both cell lines, consistent with ferroptosis (Supplementary Fig. 4a). To identify ACSL isoforms that contribute to αESA-induced ferroptosis, we individually knocked down expression of each of the five isoforms in BT-549 and MDA-MB-231 cells by RNAi (Supplementary Fig. 4b, c). As expected, ML162 toxicity was suppressed by ACSL4 depletion in both cell lines (Fig. 5c and Supplementary Fig. 4d–f). αESA toxicity, however, was significantly suppressed by knockdown of ACSL1, though other ACSL isoforms may also contribute.

We used CRISPR/Cas9 to generate multiple clonal BT-549 cell lines deficient in ACSL1 (Fig. 5d). ACSL1 loss reduced αESA sensitivity compared to controls (Fig. 5e) and increased sensitivity to ML162 (Fig. 5f). Similar results were obtained using an independent guide RNA (Supplementary Fig. 4g, h). Re-expressing CRISPR-resistant ACSL1 in ACSL1-deficient cells restored αESA sensitivity (Fig. 5g), and increasing ACSL1 expression in control BT-549 cells further enhanced αESA toxicity (Fig. 5h). These results demonstrate that ACSL1 promotes αESA-triggered cell death. Furthermore, they highlight a mechanistic distinction between αESA and canonical ferroptosis inducers and expand the spectrum of ACSL isoforms involved in ferroptosis.

ACSL1 regulates αESA metabolism

To elucidate the impact of ACSL1 on lipid metabolism, we conducted a lipidomic analysis of two independent ACSL1-deficient cell lines as well as an ACSL1-overexpressing line and their parental control cells (Fig. 6). Overall, the pattern of lipid enrichment or depletion in response to ACSL1 modulation was generally similar in the absence or presence of αESA (compare Fig. 6 left and middle panels). Among the differentially expressed lipids, we noted that diacylglycerols (DAGs) and triacylglycerols (TAGs) were significantly reduced in ACSL1-deficient cells compared to controls or ACSL1-overexpressing cells treated with αESA (Supplementary Data 2). These findings are consistent with prior studies linking ACSL1 to the production of these lipids40,41.

Next, we examined the impact of ACSL1 specifically on lipids bearing 18:3 acyl chains (Fig. 6, right). The mole fraction of 18:3-containing lipids as a class was modestly but significantly decreased in αESA-treated ACSL1−/− cells compared to αESA-treated controls (p = 0.006, Fisher’s combined test). This suggests that the resistance of ACSL1-deficient cells to αESA could be related to decreased αESA incorporation. Lipid subgroup-focused analysis revealed 18:3 decreases specifically in phosphatidylglycerols, DAGs, and TAGs (Supplementary Data 3). Conversely, we observed a significant increase in 18:3-bearing cholesterol esters and vinyl ether-linked phosphatidylcholine and phosphatidylethanolamine in αESA-treated ACSL1-deficient cells. A potential explanation is that αESA may also be weakly utilized by other ACSL isoforms such as ACSL4, which is known to direct substrates into cholesterol esters40. Similar and more robust differences were observed comparing ACSL1-deficient to ACSL1-overexpressing cells. They included significantly lower 18:3-bearing lipids overall (p = 0.0002, Fisher’s combined test) and particularly in DAGs (p = 1 × 10−10) and TAGs (p = 0.0002). In an independent experiment, we confirmed that transient silencing of ACSL1 expression (Fig. 7a) reduced 18:3 chain incorporation into TAG species of ESA-treated cells (Fig. 7b). Together these findings demonstrate a role for ACSL1 in the incorporation of αESA and other fatty acids into DAGs/TAGs, which could suggest a role for lipid droplets in αESA-induced ferroptosis.

αESA triggers ACSL1-dependent increases in lipid hydroperoxides

We next sought to characterize αESA-induced lipid oxidation. First, we quantified 141 mono-, di-, and tri- oxygenated lipids using liquid chromatography–mass spectrometry across seven membrane lipid classes in BT-549 cells and found the abundance of oxidized membrane lipids was significantly increased in cells treated with αESA (p = 0.02, Student’s t-test, one-tailed) (Fig. 7c). A wide variety of oxygenated lipids have been previously reported to be increased in cells undergoing ferroptosis, although a key role for phosphatidylethanolamine species bearing arachidonic or adrenic acids has been reported17. Within the seven lipid classes examined, we observed that 51 (36%) oxygenated lipid species were significantly changed by αESA and a majority of those were increased (41/51, 80%) (Fig. 7c). These results are consistent with our observation of increased lipid peroxidation products in αESA-treated cells (Fig. 2d). The classes with both the highest number and frequency of increased oxidized species were cardiolipins (17/36 47%) and phosphatidylethanolamines (18/40 45%) and included two di-oxygenated phosphatidylethanolamine species containing arachidonate residues (Fig. 7d, Supplementary Fig. 5).

Of the 41 oxygenated membrane lipids that were significantly increased following incubation with αESA, 10 (24%) were responsive to ACSL1 silencing and all were decreased (Fig. 7e; Supplementary Fig. 5c, e, g, h). Some, though not all, αESA-induced, di-oxygenated phosphatidylethanolamine lipids were decreased on ACSL1-silencing (Fig. 7d). Given the abundant incorporation of αESA in TAGs, we also examined oxidation of this lipid class. Notably, in αESA-treated but not control cells we detected oxygenated 52:4 TAG (18:1/18:3/16:0) species including mono- and di-oxygenated species (Fig. 7f, g), consistent with a prior report28. However, few other oxidized TAGs were detected (Supplementary Data 4). The abundance of these species was decreased by ACSL1-silencing, as was the parental, unoxidized 52:4 TAG precursor (Supplementary Fig. 5i), demonstrating that these oxidized TAG species are generated in an αESA and ACSL1-dependent manner.

Triglycerides contribute to αESA toxicity

ACSL1 promotes αESA accumulation in DAGs and TAGs (Fig. 6) as well as αESA toxicity (Fig. 5), suggesting a potential mechanistic link. We, therefore, examined the role of diacylglycerol acyl transferase (DGAT), which acylates DAGs to produce TAGs, in αESA toxicity. Transient silencing of DGAT1 but not DGAT2 restored viability of αESA-treated MDA-MB-231 cells (Fig. 8a, Supplementary Fig. 6a, b). Likewise, treatment of cells with the DGAT1-specific inhibitor PF-04620110 but not the DGAT2-specific inhibitor PF-06424439 suppressed αESA toxicity (Fig. 8b). Consistent with redundancy, the combination of both inhibitors produced an additive effect in MDA-MB-231 cells (Fig. 8b) and this combination also suppressed αESA-induced cell death in BT-549 cells (Fig. 8c). ML162 toxicity was not affected by this inhibitor combination (Fig. 8b, c) indicating a role for TAGs specifically in αESA-induced ferroptosis. Knockdown of adipose triglyceride lipase (ATGL; Supplementary Fig. 6c), which liberates fatty acids from TAGs, suppressed the toxicity of ML162 but not αESA in two cell lines (Supplementary Fig. 6d–g). Atglistatin, a pharmacological inhibitor of ATGL, in contrast, suppressed both αESA and ML162 toxicity (Supplementary Fig. 6h–j), suggesting the possibility that atglistatin might act as a direct radical trapping agent as has been observed for other small molecule inhibitors42.

Fatty acids and TAGs with conjugated double bonds undergo complex and poorly understood oxidative reactions distinct from that of unsaturated chains with non-conjugated double bonds43,44,45,46,47. Products of conjugated acyl chain oxidation include aldehydes, peroxides, and high molecular weight covalent polymers45,46,47,48,49. Our mass spectrometry analysis of low molecular weight lipids from αESA-treated cells revealed the formation of two distinct adducts of di-oxygenated αESA free fatty acid with DAG containing two αESA-derived chains (Fig. 8d, e). These adducts were only detected in αESA-treated cells and were significantly reduced in ACSL1-knockdown cells, demonstrating an ACSL1-dependent formation of oxidative adducts of αESA-derived acyl chains in neutral lipids. Higher-order polymers or adducts of αESA-containing TAGs were beyond the size range analyzed and thus our analysis likely underestimates the scope of TAG oxidation.

A remaining question is whether oxidative decomposition of αESA-derived acyl chains in neutral lipids could lead to the oxidation of membrane phospholipids. The seed oil of the tung tree is composed of TAGs in which ~80% of the acyl chains are esterified αESA. We, therefore, considered tung oil as a model for the αESA-rich TAGs that accumulate in αESA-treated cells (Fig. 2g). We generated an emulsion of tung oil in cell culture medium and, following sedimentation and removal of the oil layer, incubated the conditioned medium with BT-549 cells. Two hours in conditioned medium led to an increase in cellular lipid peroxidation products, which were suppressed by co-treatment with fer-1 (Fig. 8f). Incubating cells with conditioned medium for 24 h resulted in significantly decreased cell viability, which was partially rescued by fer-1 or iron chelation by deferoxamine (Fig. 8g). These findings demonstrate that αESA-rich triglycerides can produce a diffusible mediator of ferroptosis that could potentially propagate lipid peroxidation from neutral lipids to membrane lipids.

Conjugated PUFAs exhibit anti-cancer activity and promote expression of ferroptotic markers in vivo

Finally, we assessed the anti-cancer activity of tung oil in an aggressive TNBC orthotopic xenograft model. Mice with established tumors were treated by oral gavage five times per week with 100 µl tung oil. Control mice received high-oleic (18:1) safflower oil. Body weights of tung oil-treated mice did not differ significantly from controls (Supplementary Fig. 7a). However, tung oil significantly suppressed tumor growth (Fig. 9a) and endpoint tumor weight (Fig. 9b) and also substantially reduced lung metastatic invasion compared to controls (Fig. 9c, d). The addition of BSO further enhanced tumor growth suppression by αESA (Supplementary Fig. 7b), consistent with BSO enhancing αESA toxicity in vitro (Fig. 2b).

To obtain evidence that the reduction in tumor growth is associated with tumor exposure to αESA, we conducted a lipidomic analysis of αESA-treated and control tumors. The two most differentially expressed lipids were DAG (34:3) and TAG (52:7), which were increased in tung oil-treated tumors by 30% and 60%, respectively (Fig. 9e). The masses of these lipids are consistent with at least one linolenate moiety, although their precise acyl chain composition was not determined. Notably, the expression of these two lipids was also increased in αESA-treated cultured cells (Fig. 9e). Thus, increased DAG (34:3) and TAG (52:7) could reflect tumor exposure to circulating αESA released from tung oil and taken up by tumor cells.

Next, we assessed whether tumors in αESA-treated mice expressed markers of ferroptosis. We conducted RNA sequencing of three tumors each from treated and control mice and compared the differentially expressed genes with a separate in vitro experiment using cultured MDA-MB-231 cells treated with either αESA, the GPX4 inhibitor ML162, or vehicle controls. Using a false discovery rate of 5% and a filter of >2-fold change in RNA expression, 124 genes were altered in tung oil-treated tumors compared to safflower oil-treated controls (Fig. 9f). Significantly, 89 genes (89/122, 73%) altered in tung oil-treated tumors were also altered in αESA-treated cells in culture. Fisher’s exact test showed significant overlap between these gene sets (p = 3.2 × 10−246) and all 89 overlapping genes were altered in the same direction. This is consistent with our hypothesis that the inhibition of tumor growth in vivo is related to the cell death caused by αESA in vitro. Importantly, there was also significant overlap in genes altered by tung oil in vivo, αESA in vitro, and by in vitro treatment with ML162 (Fig. 7f). Fisher’s exact test between any two gene sets demonstrated p-values of <7.2 × 10−61. Furthermore, all of the shared genes were altered in the same direction. These results demonstrate concordant gene signatures between tung oil treatment and αESA treatment, consistent with the hypothesis that tumors in tung oil-treated mice are exposed to αESA. Furthermore, the overlap in the signatures of αESA- and ML162-treated cells provide strong evidence that both treatments trigger a similar ferroptotic response.

All three treatments altered a common set of 23 genes (Fig. 9f) including genes previously identified as increased during ferroptosis (RHOB, SLC2A3, DDIT4, and HMOX1)50,51. In addition, the previously reported ferroptosis marker CHAC150 was commonly upregulated in both tung-oil treated tumors and αESA-treated cells. 66 genes were altered by both αESA and tung oil but not ML162, suggesting a potential source of biomarkers for αESA-induced ferroptosis.

While anti-cancer activity has been previously attributed to various conjugated linolenic acids including αESA52,53, here we reveal the mode of action as ferroptosis and highlight mechanistic differences compared to existing ferroptosis inducers. Our findings suggest that while GPX4 inhibitors promote ferroptosis by inhibiting the glutathione-dependent reduction of hydroperoxides, αESA promotes ferroptosis by enhancing hydroperoxide production. GPX4 activity reduces αESA-induced hydroperoxide accumulation but ultimately GPX4 is overwhelmed.

We found that ACSL1 mediates αESA-induced ferroptosis as well as αESA incorporation into specific lipid species including DAGs and TAGs (Figs. 5, 6 and 7b). Further supporting a role for TAGs in ferroptosis, depletion or pharmacological inhibition of DGAT activity suppressed αESA toxicity (Fig. 8a–c). We found that ACSL1, which is constitutively localized to lipid droplets54, concentrates αESA in neutral lipids compared to other lipid pools. By three hours of treatment with αESA, TAG levels increase 2-fold (Fig. 2h) with >70% containing at least one 18:3 acyl chain while no membrane phospholipid class has more than 20% (Fig. 2g). We hypothesize that neutral lipids bearing αESA-derived acyl chains undergo spontaneous or enzyme-mediated oxidation after achieving a critical concentration in lipid droplets. This concentration dependence could reflect the intermolecular nature of the rate-limiting propagation step of lipid autoxidation. The peroxy radicals that mediate the chain-reaction propagation phase of autoxidation either abstract a hydrogen atom from adjacent hydroperoxides or form intermolecular covalent bonds by adding to carbon-carbon double bonds. The rate of abstraction of a hydrogen atom adjacent to the triene system of conjugated linolenic acid is 2.5-fold faster than the corresponding rate for abstraction of the bis-allylic hydrogen in non-conjugated linolenic acid43. Furthermore, peroxyl radicals also add to the triene system of conjugated linolenic acid at an even faster rate while peroxyl radicals do not appreciably add to double bonds in non-conjugated linolenic acid. This results in an overall rate of free radical oxidation propagation that is over 8-fold faster for conjugated linolenic acid43, providing a chemical basis for the distinct ferroptosis-inducing activity of this lipid class above a threshold concentration.

While we only detected two oxidized TAG species associated with αESA treatment (Fig. 7f, g), we speculate that neutral lipids containing esterified αESA undergo oxidative polymerization to form higher molecular weight polymers, as has been previously shown ex vivo45,46,47,48. Because we focused our analysis on lipids <1,200 molecular weight, very large species, and any oxygenated derivatives were not examined by our lipidomic analysis. However, our demonstration of adducts of free αESA with DAGs containing esterified αESA (Fig. 8d, e), which fall under this size cutoff, supports the possibility that larger covalent aggregates may also form.

A role for lipid droplets in potentiating lipid peroxidation is contrary to a proposed protective role for this organelle against oxidative stress55,56. This may be explained, however, by the rapid rate of free radical oxidation propagation by conjugated linolenic acids. Dietary conjugated linoleic acids have been linked to increased lipid peroxidation in humans57,58 and animal models59, and αESA exhibits even more rapid spontaneous oxidation than these species53,60. While lipid droplets may protect polyunsaturated lipids from oxidation by sequestering them from ROS species generated elsewhere, the accumulation of esterified conjugated linolenic acids, which are not naturally synthesized by mammals, in lipid droplets may promote their oxidation and polymerization there. Propagation of lipid peroxidation from lipid droplets to membrane phospholipids may be mediated by diffusion of soluble mediators (Fig. 8f, g). The formation of large covalent lipid polymers in lipid droplets may also trigger an autophagic response61 which could drive ferroptosis62.

A model in which accumulation of conjugated linolenic acyl chains in lipid droplets initiates ferroptosis may also explain the differences in ACSL isoform roles between ferroptosis triggered by αESA and by GPX4 inhibitors. GPX4 inhibitors suppress the antioxidant response, allowing natural lipid ROS species to accumulate. In many contexts these are likely enzymatically generated peroxidation products of arachidonyl and adrenyl acyl chains, whose incorporation into membrane phospholipids is mediated by ACSL438. By contrast, the ACSL1-dependence of αESA toxicity appears to be mediated at least in part by ACSL1-driven incorporation of αESA into TAGs. Overall, these findings suggest that conjugated linolenic acids trigger ferroptosis by a mechanism that is mediated by ACSL1 and support emerging evidence implicating lipid composition as a critical determinant of ferroptotic sensitivity63.

The discrepant ability of GPX4 and FSP1 to suppress αESA-triggered ferroptosis compared to ferroptosis induced by ML162 in BT549 cells (Fig. 4g–j) could also be explained by distinct origins of lipid peroxidation. We speculate that GPX4 and FSP1 may have evolved and be localized specifically to target natural sources of lipid ROS. Indeed, FSP1 is N-myristoylated and is localized to intracellular membranes and the plasma membrane in a cell type-specific manner9,10. Plasma membrane-restricted FSP1 restored RSL3-resistance to FSP1-deficient cells while FSP1 targeted to other intracellular locations did not10, suggesting that the plasma membrane is an important site of lipid radical accumulation on GPX4 inhibition. αESA, by contrast, is metabolized like other nutrient fatty acids and stored in lipid droplets where, because of its oxidizable conjugated triene system, it may act as a source of lipid ROS that is poorly targeted by endogenous protective mechanisms. This concept has clear applicability to targeting cancer cells that robustly scavenge extracellular fatty acids22.

Studies have shown that cancer cells, including renal cell carcinoma, melanoma, and glioblastoma are vulnerable to ferroptosis38,51,64,65,66,67,68. Furthermore, therapy-resistant mesenchymal cancer cells exhibit a greater reliance on GPX4 activity for survival64,69. These findings have highlighted the therapeutic potential of ferroptosis, but it remains unclear whether induction of ferroptosis can be used to kill cancer cells in vivo1. A major challenge to testing this hypothesis has been the absence of ferroptosis-inducing agents, such as GPX4 inhibitors, suitable for use in vivo70. Here, we found that treatment of mice with tung oil, a rich source of αESA, suppressed tumor growth and metastasis, and resulted in gene expression changes in tumors consistent with ferroptosis. The data suggest that GPX4 inhibitors, should these become clinically viable, may enhance the anti-tumor activity of αESA. Thus, these findings expand our understanding of the pathways that regulate ferroptosis and provide an approach for exploiting ferroptosis therapeutically.