MS4A15 drives ferroptosis resistance through calcium-restricted lipid remodeling

Ferroptosis is an iron-dependent form of cell death driven by biochemical processes that promote oxidation within the lipid compartment. Calcium (Ca2+) is a signaling molecule in diverse cellular processes such as migration, neurotransmission, and cell death. Here, we uncover a crucial link between ferroptosis and Ca2+ through the identification of the novel tetraspanin MS4A15. MS4A15 localizes to the endoplasmic reticulum, where it blocks ferroptosis by depleting luminal Ca2+ stores and reprogramming membrane phospholipids to ferroptosis-resistant species. Specifically, prolonged Ca2+ depletion inhibits lipid elongation and desaturation, driving lipid droplet dispersion and formation of shorter, more saturated ether lipids that protect phospholipids from ferroptotic reactive species. We further demonstrate that increasing luminal Ca2+ levels can preferentially sensitize refractory cancer cell lines. In summary, MS4A15 regulation of anti-ferroptotic lipid reservoirs provides a key resistance mechanism that is distinct from antioxidant and lipid detoxification pathways. Manipulating Ca2+ homeostasis offers a compelling strategy to balance cellular lipids and cell survival in ferroptosis-associated diseases.


47
Ferroptosis is a type of oxidative cell death induced by glutathione (GSH) deprivation 48 or uncontrolled reactive oxygen species (ROS). During ferroptosis, polyunsaturated 49 phospholipid peroxides induced by reactive iron accumulate to lethal levels, resulting in 50 membrane lapse 1 . The selenoenzyme glutathione peroxidase 4 (GPX4) is a central 51 enzyme protecting lipids from oxidative species that uses GSH as an essential cofactor 52 to convert lipid hydroperoxides to lipid alcohols 2, 3 . Loss of GPX4 activity and 53 deprivation of GSH both lead to lipoxygenase activation in a process closely linked to 54 inflammation 4, 5 . Lipoxygenases oxidize polyunsaturated fatty acids (PUFAs) to 55 generate metabolites which additionally promote calcium (Ca 2+ ) influx for the final, 56 catastrophic phase of cell death 6 . 57 Calcium is a store-operated signal transduction molecule controlling diverse cellular 58 processes such as growth and migration. It is intricately linked to cancer and the 59 pathogenesis of degenerative diseases, which feature imbalanced metabolism and 60 excessive ROS 7-9 . The endoplasmic reticulum (ER) is the main intracellular Ca 2+ 61 storage site and plays a key role in the maintenance of Ca 2+ homeostasis and regulation 62 of protein, lipid, and glucose metabolism. In response to extrinsic stimuli, inositol 1,4,5-63 trisphosphate (IP3) and ryanodine receptors release Ca 2+ from the ER to the cytosol, 64 whereas the sarco/endoplasmic reticulum Ca 2+ -ATPase (SERCA) pumps Ca 2+ against 65 the gradient to maintain a concentration difference between the ER lumen and the 66 cytosol at rest. 67 Previous studies have shown that ER Ca 2+ homeostasis is critical for adipogenesis and 68 lipid storage 10,11 . Altering Ca 2+ balance can regulate activity of key enzymes in de novo 69 lipogenesis, including fatty acid synthase (FAS) and stearoyl-CoA desaturase 1 (Scd1), 70 or, induce lipolysis 12,13 . IP3 receptor (IP3R) mutants have conserved pathways of 71 energy metabolism, with higher serum triglycerides and free fatty acids in mice 14 and 72 persistent luminal Ca 2+ depletion circumvents synthesis of ferroptosis-sensitive 99 substrates in human cancer cell lines. This is the first report directly linking modulation 100 of ER Ca 2+ homeostasis to lipid remodeling and ferroptosis sensitivity. 101 cancer cells treated with IKE (Fig. 1F). However, due to absent MS4A15 expression in 127 cell lines (1345 of 1375 have ≤1 TPM; Fig. 1G) 37,38 , siRNA knockdown cells were not 128 more sensitive to ferroptotic challenge ( Supplementary Fig. 1D). We further noted that 129 despite high expression in primary adenocarcinomas, MS4A15 is lost in cultured lung 130 cancer cell lines in a direct relationship to cell adhesion markers (Supplementary Fig. 131 1E). A defective cell migration phenotype is thus consistent with decreased 132 metastasis/increased survival of lung cancer patients with high MS4A15-expressing 133 tumors (Supplementary Fig. 1F,G). 134 Together, these results show that MS4A15 is linked to cell migration and can robustly 135 protect against ferroptosis. MS4A15 protein is increased following ferroptosis induction, 136 suggesting its presence is instrumental to survival ( Supplementary Fig. 1H). Notably, 137 this resistance is accomplished without substantially affecting regulators of ferroptosis 1 138 ( Supplementary Fig. 1I). 139

MS4A15 regulates Ca 2+ -mediated ferroptosis 167
In light of these observations we examined Ca 2+ signaling in Ms4a15 OE cells. 168 Extracellular stimuli such as EGF can trigger phospholipase C (PLC) to generate IP3, 169 which stimulates cytosolic Ca 2+ release or MAPK/PKC to mediate cellular response 30, 170 42, 43 . We observed in Ms4a15 OE cells that phospho-ERK levels show a slight 171 concentration-dependent sensitization to EGF stimulation ( Supplementary Fig. 3A). 172 However, STAT3 and AKT were unchanged, arguing against parallel activation of 173 signaling pathways. 174 We therefore directly measured Ca 2+ response using the fluorescent sensor GCaMP6s. 175 In Ca 2+ free medium, bradykinin activates its GPCR, releasing Ca 2+ from ER stores 176 ( Fig. 2D). In Ms4a15 OE cells stimulated with bradykinin, however, the Ca 2+ response 177 was strikingly reduced (Fig. 2E). Re-addition of CaCl2 induced robust transients in 178 control cells but a limited response in Ms4a15 OE cells, suggesting the inactivation of 179 SOCE. The permeant ionophore A23187 corroborated a potent decrease in total Ca 2+ 180 released from Ms4a15 OE internal stores (Fig. 2E). 181 This profile is similar to that of cells treated with thapsigargin (Tgn), a potent inhibitor of 182 SERCAs that supply the lumen with Ca 2+ (Fig. 2D,E). Remarkably, Tgn disruption of 183 ER Ca 2+ import in control cells showed diminished lipid peroxidation corresponding to 184 treatment duration (Fig. 2F). Whereas simultaneous application of Tgn with RSL3 did 185 not affect resistance, 7 and 14 days pretreatment comprehensively protected cells. Pre-186 treatment with Tgn abolished bradykinin and ionophore-induced store release, but 187 increased Ca 2+ uptake from the extracellular milieu (Fig. 2E,F). This shows that while 188 cytosolic Ca 2+ levels in Tgn-treated cells may be partially rebalanced, Ms4a15 OE cells 189 are refractory to uptake. 190 We next investigated if Ms4a15 OE resistance was due to ER-Ca 2+ depletion or SOCE-191 related effects. Inhibition of SOCE Ca 2+ import by CoCl2 as well as forced influx via 192 ionophore did not markedly affect Ms4a15 OE cell sensitivity (Supplementary Fig. 193 3B,C). In addition, rapid uptake store-operated membrane channel (Orai) expression 194 was virtually unchanged, consistent with unchanged ferroptosis sensitivity upon  inhibition with BAPTA-AM ( Supplementary Fig. 3D Fig. 3C; ESI-, Supplementary Fig. 4C). The data show the 226 vast majority of modulated lipids are glycerophospholipids (GP), followed by several 227 free fatty acid (FA) species (Fig. 3D). 228 Ms4a15 OE delivered a different free FA profile compared to control cells. Significant 229 increases of the main saturated FAs, palmitic (C16:0) and stearic (C18:0) acid, were 230 observed while PUFA fatty acids such as arachidonic (20:4, AA), andrenic (22:4), 231 eicosapentaenoic acid (20:5, EPA), docasapentaenoic acid (22:5, DPA), and 232 doxosahexaenoic acid (22:6, DHA) were decreased. Tgn long cells shared a similar albeit 233 less robust profile than Ms4a15 OE, possibly due to the abbreviated treatment 234 (Fig. 3D). Ether lipids may consist of alkyl-ether or vinyl-ether moieties, with a double bond 250 proximal to the oxygen, termed plasmalogens (Fig. 3F). MS 2 cannot differentiate 251 between isomeric alkyl-ether and vinyl-ether, thus we verified MUFA plasmalogens as 252 the main species in Ms4a15 OE cells by acidic hydrolysis (Fig. 3H). Co-elution of a 253 plasmalogen and an isomeric saturated ether was seen for several species, while many 254 upregulated ethers were entirely plasmalogens (Supplementary Table 2). Consistently, 255 Ms4a15 knockout MF cells show a decrease in the same ether species and MUFA-256 GPs, however, these lipids were mostly unaffected in knockdown Calu-1 and HT-1080, 257 in agreement with unchanged viability for these cell lines ( Supplementary Fig. 5A  We therefore compared all affected lipid species by global non-supervised principal 269 component analysis (PCA), resulting in group separation with minimal convergence 270 ( Fig. 4B). This suggests that ferroptosis is classically initiated in cells but peroxidation 271 degrades additional lipid species in Ms4a15 OE cells. We therefore investigated their 272 origin with respect to dysregulated lipids found in the Ms4a15 OE pool. 273 We found that RSL3-treatment depleted the same lipids increased that are elevated in 274 Ms4a15 OE cells (Fig. 3E,3G,4C). We therefore examined if significantly upregulated 275 and highly abundant lipids are preferred targets of RSL3 (Fig. 4D), however, the pattern 276 is independent of initial concentration. In Ms4a15 OE, RSL3 treatment extensively 277 modifies most ether-lipids and MUFA-containing GPs, rather than single or highly 278

Lipid elongation and desaturation mediate resistance 308
Ms4a15 OE lipids are shorter but more saturated (Fig. 3E,G). Thus, these lipids may 309 derive from de novo lipogenesis upon compromised ER-resident elongase and 310 desaturase activities. Analogously, ML239 agonizes fatty acid Δ6 desaturase 2 (FADS2) 311 activity to increase PUFA synthesis and ferroptosis sensitivity 49 . We considered that 312 supplementation with free exogenous PUFA fatty acids may overcome protective lipids. redistributed to smaller droplets in the cytosol rather than lost (Fig. 5H). Collectively, 330 these data show that depletion of ER calcium lead to qualitative changes in ferroptosis-331 sensitive lipids in concert with subcellular LD rearrangement. 332

Global Ca 2+ genes define a signature for ferroptosis 333
We speculated that changes in Ca 2+ homeostasis resulting in ferroptosis-resistant lipids 334 may contribute to resistance in different cell lines. We tested this theory by cross-335 referencing sensitivity of the 100 most RSL3-resistant and -sensitive cancer cell lines 336 from the CTRP database 49 to KEGG gene expression 37 . 337 Using unsupervised clustering of Ca 2+ genes, we observed segregation corresponding these results demonstrate that elevating ER Ca 2+ levels by blocking signals at the 367 membrane can sensitize certain ferroptosis-resistant cell lines. 368

369
In this report we define a unique mechanism for ferroptosis resistance based on the 370 discovery of MS4A15, an uncharacterized four-pass membrane protein. MS4A  and -plasmalogens abate ferroptosis-induced lipid peroxidation. Taken together, these 434 data strongly support the conclusion that MS4A15 is an independent contributor to 435 ferroptosis resistance. 436

Cell lines and culture conditions 438
Cell lines used in the study: Immortalized conditional Gpx4 -/-mouse embryonic 439 fibroblasts expressing Cre-ERt2 (MEF, male) 33 were previously generated 32 with the 440 CRISPR activation system 78 and a mouse Ms4a15 CRISPR guide (Supplementary 441 Table 5)  were taken with an Operetta High-Content Screening System (PerkinElmer) with a 20X 503 objective. For colony-forming assays, cells were treated with 1.25 μM RSL3 overnight, 504 then trypsinized single-cells, diluted 1:300 and seeded into 6-well plates. After 7 d 505 colonies were stained with cresyl violet and imaged. 506 Three-dimensional spheroids. MF control and Ms4a15 OE cells were seeded into the 507 GravityTRAP ULA 96-well plates (InSphero/PerkinElmer) to form 3D spheroids. 508 Interwell variations <10% were confirmed and spheres were grown for 4 days, treated 509 with 2µM RSL3 for additional 16 h and stained with PI. Spheroids were imaged directly 510 with an Operetta High-content system. Images from a single plate were acquired using 511 Brightfield and PI channels and 20x High-NA objective in wide field mode. Ten planes 512 of each sample were tracked and four replicates per cell condition were collected with 513 the same parameters and PI intensity of different cell conditions were analyzed with 514 Harmony software (PerkinElmer) using the same settings to optimize the results.

Lipid peroxidation analysis by flow cytometry 535
Cells were seeded in 6-well plates to reach 70% confluency. The next day, 0.3 µM 536 RSL3 was added for 3 h. Cells were loaded with 2 µM BODIPY 581/591 C11 (Thermo 537 Fisher Scientific) for 30 min and harvested for analysis on an Attune acoustic flow 538 cytometer (Applied Biosystems). At least 30,000 events per condition were collected 539 from the BL-1 channel (excited by 488 nm laser). Each experiment was repeated at 540 least three times independently and representative results are shown. 541 542

Intracellular calcium measurements 543
Cells containing the cytosolic calcium sensor GCaMP6s were seeded the day before in 544 10 cm dishes to reach 70% confluency. The following day, cells were treated with 545 Accutase (Sigma) and resuspended in PBS, washed twice with Ca 2+ -free buffer (NaCl 546 116 mM, KCl 5.6 mM, MgCl2 1.2 mM, NaHCO3 5 mM, NaH2PO4 1 mM, HEPES 20 mM, 547 Glucose 1 g/L). Cell pellets were resuspended in 2 mL of Ca 2+ -free buffer and were 548 analyzed with a BD FACSCanto II (Becton Dickinson). Untreated cell suspensions were 549 recorded for 2 min (approx. 2,000 events/second) to establish a baseline signal. Ca 2+ 550 release mediated by Bradykinin (Sigma) and Ionophore (Sigma) was measured for 4 551 and 6 min, respectively. After Bradykinin stimulation, 2 mM CaCl2 was added to the 552 cells and data for the uptake of Ca 2+ was collected for additional 9 min. Kinetic data 553 were created by FlowJo V10 of viable, GFP positive cells and exported for visualization 554 to GraphPrad Prism 8. All experiments were repeated at least three times. 555 556 AAPH oxidation assay using BODIPY 581/591 C11 557 Ester lipids, plasmalogens and ferrostatin (fer-1) were added into 150 µL PBS as 558 indicated to achieve 150 ppm, 150 ppm and 9 ppm, respectively. Freshly dissolved 559 1.875 µM BODIPY 581/591 C11 in 150 µL PBS and 7.5 mM 2,2'-560 Azobis ( For plasmalogen experiments, MF control cells were seeded the day before on 96-well 576 plates. The following day, cells were washed with PBS and incubated with 25 µM 577 plasmalogens in serum-free medium for 8 h. After serum starvation, 10% FBS was 578 added back and the cells were treated with RSL3 and aToc to achieve final 579 concentrations as indicated. Cell viability assay was performed as described above. 580

EGF signaling in cultured cells 582
MF cells were pre-seeded in 6-well plates one day before for reaching 70% confluency. 583 The culture medium was changed to serum-free medium and incubated at 37 °C for 4 h 584 starvation. Subsequently, the serum-starved MF cells were stimulated with 0-5 ng/mL 585 EGF for 10 min at 37 °C, washed with PBS and lysed for western blot analysis. 586 587

Western blotting 588
Cells were lysed for 20 min in lysis buffer (63 mM Tris-HCl, pH 6.8, 10% glycerol, 2% 589 SDS, 2.5% DTT and 1x protease inhibitor tablet (Roche)) and DNA was shredded with 590 a sonicator. After separation on a 6-12% SDS-PAGE gel according to the protein sizes, 591 proteins were transferred to PVDF membranes. After blocking with 5% non-fat milk for 592 1h at room temperature, the membranes were incubated in specific primary antibodies 593 diluted in 2.5% BSA at 4 °C overnight. The next day, membranes were incubated with 594 secondary antibodies for 2 h at room temperature. ECL prime Western blotting 595 detection reagents (Bio-Rad) were used at a ratio of 1:1 for chemiluminescence 596 detection. Each experiment presented was repeated at least three times. Primary

Confocal microscopy and immunofluorescence 606
Cells were plated at a density of 4x10 3 cells/well on 96-well plates (Perkin Elmer Cell 607 Carrier Ultra Viewer). Cells were transfected with corresponding expression constructs 608 for 24 h before 4% formaldehyde fixation. Images were taken with a laser scanning 609 confocal microscope (Olympus FluoView 1200; Olympus Corporation). Nuclei were 610 labeled with DAPI staining (blue). MS4A15 was visualized with Anti-FLAG antibody 611 (Sigma F7425; 1:500) and a secondary goat anti rabbit antibody (Cy3 Jackson Immuno 2 h at room temperature using 0.5 µg Lys-C (Wako Chemicals) and for 16 h at 37 °C 674 using 1 µg trypsin (Promega). After centrifugation (10 min at 14,000 g) the eluted 675 peptides were acidified with 0.5% TFA and stored at -20 °C. 676

LC-MS/MS analysis was performed on a Q-Exactive HF mass spectrometer (Thermo 677
Scientific) online coupled to an Ultimate 3,000 nano-RSLC (Thermo Scientific). Tryptic 678 peptides were automatically loaded on a C18 trap column (300 µm inner diameter (ID) 679 x 5 mm, Acclaim PepMap100 C18, 5 µm, 100 Å, LC Packings) at 30 µL/min flow rate 680 prior to C18 reversed phase chromatography on the analytical column (nanoEase MZ 681 HSS T3 Column, 100Å, 1.8 µm, 75 µmx250 mm, Waters) at 250 nl/min flow rate in a 682 95 min non-linear acetonitrile gradient from 3% to 40% in 0.1% formic acid. Profile 683 precursor spectra from 300 to 1,500 m/z were recorded at 60,000 resolution with an 684 automatic gain control (AGC) target of 3e6 and a maximum injection time of 50 ms. 685 TOP10 fragment spectra of charges 2 to 7 were recorded at 15,000 resolution with an 686 AGC target of 1e5, a maximum injection time of 50 ms, an isolation window of 1.6 m/z, 687 a normalized collision energy of 27 and a dynamic exclusion of 30 seconds. 688 689

Metabolomics and proteomics 690
Briefly, 1x10 7 Ms4a15 OE and parental MF cells per replicate (n = 5) were lysed and 691 equal amounts were proteolyzed using a modified FASP procedure 81 . The proteins 692 were digested for 2 h at room temperature using 0.5 µg Lys-C (Wako Chemicals) and 693 for 16 h at 37 °C using 1 µg trypsin (Promega), eluted by centrifugation, acidified with 694 TFA and stored at -20 °C. Peptides were measured on a Q-Exactive HF mass 695 spectrometer online coupled to an Ultimate 3,000 nano-RSLC (Thermo Scientific) in 696 data-independent acquisition (DIA) mode as described previously (Lepper et al., 2018). 697 Raw files were analyzed using the Spectronaut Pulsar software (Biognosys; 82 ) with a 698 false discovery rate setting of < 1%, using an in-house mouse spectral meta library 699 generated using Proteome Discoverer 2.1 (Thermo Scientific), the Byonic search 700 engine (Protein Metrics) and the Swissprot Mouse database (release 2016_02). instrument was calibrated with a 1 ppm arginine solution. A mass error below 100 ppb 715 was achieved. Injected velocity was set to 120 µL/h. Mass lists were generated with a 716 signal-to-noise ratio (S/N) of four, exported, and combined to one data matrix by 717 applying a 1 ppm window. Ions (m/z mass/charge) were annotated using MassTRIX 718 allowing 1ppm mass tolerance. Unidentified metabolites were annotated by elemental 719 composition using mass-differences based network approach allowing 0.1ppm mass 720 tolerance 83 . 721 722

Lipid extraction and global lipidomics 723
Procedures for lipid extraction and global lipidomics profiling using UPLC-MS were 724 described previously 45 . In short, we used a two-step MTBE extraction in a cooled 725 Precellys (Bertin). The organic content was analyzed using data-dependent auto LC-726 MS² (maXis, Bruker Daltonics) coupled to an UHPLC ACQUITY (Waters) using reverse 727 phase chromatography (CORTECS UPLC C18 column, 150 mm x 2.1 mm ID 1.6 µm, 728 Waters Corporation) in both positive and negative electrospray modes. The injection 729 volume was set to 10 µL. Lipid elution was achived using 10mM ammonium formate 730 and 0.1 % formic acid in 60% acetonitrile/water mixture (A) and in 90% 731 isopropanol/acetonitrile mixture (B) as mobile phase. Quality control consisting of an 732 aliquot of each sample and pure solvent blanks were used for column equilibration. The 733 MS analysis alternated between MS and data-dependent MS n scans using dynamic 734 exclusion. Alignment, peak picking and identification as well as quality control 735 processing was done in Genedata software (Genedata Expressionist 13.5, Genedata).   The following analysis was performed by variables with the highest 100 median 810 absolute deviations (MAD). Multivariate biplot were performed to characterize the 811 variability of the data in each group using "ggplot2" 97 , "factoextra" 98 , and "ade4" 99 812 packages. 813 814 ssGSEA implementation 815 The correlations between gene expression levels were calculated by Pearson's test. 816 The 50 genes with the most significant correlation coefficients were identified from 817 whole transcriptome. The heatmap was plotted with R package "pheatmap" 100 . 818

GO_CALCIUM_ION_TRANSMEMBRANE_TRANSPORT, KEGG_CELL_ADHESION, 819
and KEGG_CALCIUM_SIGNALING_PATHWAY term lists were derived from GSEA. 820 The correlation between each term and gene expression level was calculated by 821 Pearson's test and plotted with package "ggplot2" 97 . Briefly, all tumor samples were 822 The results shown here are in whole or part based upon data generated by the TCGA 855 Research Network: https://www.cancer.gov/tcga. Calu-1 cells, (1S, 3R)-RSL3 (RSL3) and 856 imidazole ketone erastin (IKE) were kindly provided by Brent Stockwell. We also thank 857 Relative mRNA expression is shown as mean ± SD of n = 3 technical replicates of three 1262 independent experimental repetitions. Viability data are plotted as representative 1263 mean ± SD of n = 3 technical replicates for independent experiments repeated at least 1264 three times with similar outcomes. Curve statistics, p-values of two-way ANOVA, are 1265 shown for comparisons. *P <0.05, **P <0.01, ***P <0.001, ****P <0.0001. 1266 1267

Fig. 2. MS4A15 regulates calcium-mediated ferroptosis 1268
A Enrichment of MS4A15-FLAG co-immunoprecipitated proteins in HEK293T cells as 1269 determined by label-free proteomic quantification. Mean abundance ratios were 1270