{"paper_id":"edf2db87-c99b-4bd7-9e57-18dc39c9880c","body_text":"Riboflavin metabolism shapes FSP1-driven ferroptosis resistance \n \nVera Skafar1, Izadora de Souza 1, Ancely Ferreira dos Santos 1, Florencio Porto Freitas 1, \nZhiyi Chen1, Merce Donate1, Palina Nepachalovich2, Biplab Ghosh3,4, Juliane Tschuck5, \nApoorva Mathur6, Ariane Ferreira Nunes Alves 6, Jannik Buhr 7, Camilo Aponte -\nSantamaría7, Werner Schmitz 8, Martin Eilers 8, Jessalyn Ubellacker 9, Ulrich Elling 10, \nHellmut G Augustin3,4, Kamyar Hadian5, Svenja Meierjohann11, Betina Proneth12, Marcus \nConrad12,13, Maria Fedorova2, Hamed Alborzinia14,15, José Pedro Friedmann Angeli1,16*. \n \n1- Rudolf Virchow Zentrum (RVZ), Center for Integrative and Translational Bioimaging, \nUniversity of Würzburg, Würzburg, Germany \n2- Center of Membrane Biochemistry and Lipid Research, University Hospital and Faculty of \nMedicine Carl Gustav Carus of TU Dresden, Dresden, Germany \n3- European Center for Angioscience, Medical Faculty Mannheim, Heidelberg University, \nHeidelberg, Germany \n4- Division of Vascular Oncology and Metastasis, German Cancer Research Center \nHeidelberg (DKFZ–ZMBH Alliance), Heidelberg, Germany \n5- Research Unit Signaling and Translation, Helmholtz Zentrum München, Neuherberg, \nGermany \n6- Institute of Chemistry, Technische Universität Berlin, Berlin, Germany \n7- Heidelberg Institute for Theoretical Studies, Heidelberg, Germany \n8- Department of Biochemistry and Molecular Biology, Theodor Boveri Institute, Biocenter, \nUniversity of Würzburg, Würzburg, Germany \n9- Department of Molecular Metabolism, Harvard T.H.Chan School of Public Health, Boston, \nMa, USA. \n10- Institute of Molecular Biotechnology of the Austrian Academy of Science (IMBA), Vienna, \nAustria. \n11- Department of Pathology, University of Würzburg, Würzburg, Germany. \n12- Institute of Metabolism and Cell Death, Helmholtz Zentrum München (HMGU), \nNeuherberg, Germany.  \n13- Translational Redox Biology, Technical University of Munich, TUM Natural School of \nSciences, Garching, Germany \n14- Heidelberg Institute for Stem Cell Technology and Experimental Medicine (HI -STEM \nGmbH), Heidelberg, Germany. \n15- Division of Stem Cells and Cancer, German Cancer Research Center (DKFZ), Heidelberg, \nGermany. \n16- Comprehensive Cancer Center Mainfranken, University Hospital Würzburg, Würzburg, \nGermany. \n \n*Corresponding author: pedro.angeli@uni-wuerzburg.de \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nAbstract \nMembrane protection against oxidative insults is achieved by the concerted action of glutathione \nperoxidase 4 (GPX4) and endogenous lipophilic antioxidants such as ubiquinone and vitamin E. \nDeficiencies in these protective systems lead to an increased propensity to phosphol ipid \nperoxidation and ferroptosis. More recently, ferroptosis suppressor protein 1 (FSP1) was identified \nas a critical ferroptosis inhibitor acting via regeneration of membrane-embedded antioxidants. Yet, \nregulators of FSP1 are largely un characterised, and their identification is essential for \nunderstanding the mechanisms buffering phospholipid peroxidation and ferroptosis. Here, we \nconducted a focused CRISPR -Cas9 screen to uncover factors influencing FSP1 function, \nidentifying riboflavin (vitamin B ₂) as a new modulator of ferroptosis sensitivity. We demonstrate \nthat riboflavin, unlike other vitamins that act as radical -trapping antioxidants, supports FSP1 \nstability and the recycling of lipid -soluble antioxidants, thereby mitigating phospho lipid \nperoxidation. Furthermore, we show that the riboflavin antimetabolite roseoflavin markedly \nimpairs FSP1 function and sensitises cancer cells to ferroptosis. Thus, we uncover a direct and \nactionable role for riboflavin in maintaining membrane integrity by promoting membrane tolerance \nto lipid peroxidation. Our findings provide a rational strateg y to modulate the FSP1 -antioxidant \nrecycling pathway and underscore the therapeutic potential of targeting riboflavin metabolism, \nwith implications for understa nding the interaction of nutrients and their contributions to a cell’s \nantioxidant capacity. \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nMain text \nCellular membranes are crucial structural components, acting as dynamic barriers. Membrane \nlipid constituents are nevertheless vulnerable to oxidative damage, a process known as lipid \nperoxidation (LPO) 1. LPO disrupts membrane integrity and executes ferroptosis, a type of \nregulated cell death implicated in a growing number of (patho)physiological processes, including \ncancer, neurodegeneration, and ischemia-reperfusion injury2.  \nFerroptosis is predominantly inhibited by glutathione peroxidase 4 (GPX4) 3,4, which reduces \nphospholipid hydroperoxides using glutathione. Additionally, radical -trapping antioxidants like \nubiquinone (CoQ10), vitamin E, and vitamin K are crucial in  suppressing propagation of lipid \nperoxidation5. More recently, ferroptosis suppressor protein 1 (FSP1) emerged as a key player in \nresisting ferroptosis 6,7. Unlike GPX4, FSP1 regenerates quinone-like antioxidants such as \nubiquinone and vitamin K8 using NAD(P)H as an electron donor, thereby halting the radical chain \nreaction required for ferroptosis execution . Although FSP1's enzymatic mechanism is well \nunderstood, actionable factors regulating its function  are largely uncharacterised . Identifying \nthese regulators is vital for comprehending how cells maintain membrane redox homeostasis and \nwithstand stress -induced ferroptosis, potentially revealing new therapeutic strategies for \npathological conditions in which ferroptosis plays a significant role. \nHere we used a CRISPR-Cas9 screen to identify regulators of FSP1 function, revealing riboflavin \n(vitamin B2) to be a modulator of ferroptosis sensitivity. Riboflavin, best known as a precursor for \nflavin adenine dinucleotide (FAD) and flavin mononucleotide (FMN), is essential for redox biology. \nWe show that riboflavin directly supports the stability and activity of FSP1, enhancing its capacity \nto recycle lipid-soluble antioxidants and mitigate phospholipid peroxidation. Unlike other vitamins \nthat act as direct radical-trapping antioxidants, riboflavin uniquely facilitates enzymatic recycling \nand is positioned upstream in the cascade, promoting ferroptosis resistance and preserving \nmembrane integrity. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nBy revealing this novel role for riboflavin in regulating FSP1 and lipophilic antioxidant recycling, \nour findings expand our understanding of ferroptosis regulation and provide a framework for \ntherapeutic strategies targeting the FSP1 -antioxidant axis in c ancer and other ferroptosis -\nassociated diseases. \n \nFocused CRISPR-based screen uncovers new regulators of FSP1 \nWe previously demonstrated that cells lacking GPX4 can survive and proliferate indefinitely when \nFSP1 activity is robust, either through naturally elevated FSP1 expression or enforced \noverexpression6 (Fig. 1a). Building on this observation, we generated FSP1-dependent HT1080 \ncells where  GPX4 was deleted in an  FSP1-overexpressing background  (HT1080GPX4KO/FSP1OE) \n(Fig. 1b). HT1080GPX4KO/FSP1OE cells readily undergo cell death upon FSP1 inhibition, which can \nbe rescued by co-treatment with the ferroptosis inhibitor liproxstatin-1 (Lip1) (Fig. 1a, c). \nWe reasoned that this cellular model could be combined with CRISPR-based genetic screening \nto identify factors contributing to FSP1 function. Therefore, we performed a focused  CRISPR \nscreen (targeting approximately 3000 potential druggable genes)  in this line, comparing \nconditions with and without Lip1. We posited that genetic perturbations impairing FSP1 \nexpression or activity would induce ferroptosis, which Lip1 should rescue (Fig. 1d)6. Through this \napproach, we identified several genes whose loss significantly impacted FSP1 -dependent \nferroptosis resistance, with the two top hits being stearoyl-CoA desaturase (SCD1) and riboflavin \nkinase (RFK) (Fig. 1e and Extended Table 1).  Loss of SCD1 is known to increase ferroptosis \nsensitivity by raising the PUFA/MUFA ratio 9. Consistent with this, pharmacological inhibition of \nSCD1 readily triggered ferroptosis in HT1080GPX4KO/FSP1OE cells (Extended Data Fig. 1a, b), likely \ndue to an elevated pool of oxidisable substrates overwhelming FSP1’s protective capacity. These \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nfindings suggest that cells with a high PUFA/MUFA ratio derive limited benefit from FSP1 activity. \nHowever, SCD1 depletion did not appear to directly impair FSP1 function. \n  \nFigure 1. Identification of factors supporting FSP1 function. a, Schematic representation of the FSP1-\ndependent model used to identify factors supporting FSP1 function. Upper left: the primary protective \nsystem against lipid peroxidation  (LPO) is the enzyme GPX4, complemented by FSP1. Lower left: cells \ntreated with an FSP1 inhibitor (iFSP1) can rely on GPX4 activity to survive. Upper right:  FSP1-dependent \nHT1080 cells  (HT1080GPX4KO/FSP1OE). In HT1080 cells, as in many other s, knocking out GPX4 induces \nferroptosis due to insufficient endogenous FSP1 levels to compensate for GPX4 loss. However, cell survival \ncan be rescued by overexpressing FSP1. Lower right: Upon pharmacological inhibition of  FSP1 (iFSP1 \ntreatment), cells undergo ferroptosis as they solel y rely on FSP1 function for survival. b, Immunoblot (IB) \nanalysis of ACSL4, FSP1, GPX4, and vinculin in HT1080 parental and HT1080GPX4KO/FSP1OE cells. c, Dose-\ndependent toxicity of an FSP1 inhibitor (iFSP1) in HT1080 parental and HT1080GPX4KO/FSP1OE cell lines. \nCell viability was monitored using Alamar blue after 24 h of treatment. Where indicated, cells were treated \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nwith the ferroptosis inhibitor Lip-1 (500 nM). d, Schematic representation of the screening strategy used to \nidentify novel factors supporting FSP1 function in the previously-described FSP1-dependent cellular model \n(see Fig. 1a). HT1080GPX4KO/FSP1OE cells were transduced with a gRNA library targeting approximately 3000 \ngenes and selected over 7 days in the presence of the ferroptosis inhibitor Lip -1 (500 nM). Subsequently, \ncells proliferate with or without Lip-1 supplementation for additional 14 days. e, Graphical representation of \nthe results from two independently performed CRISPR screens; plot depicts the score calculated using the \nMaGeCK package (x-axis) and the –Log false discovery rate (y-axis). Riboflavin kinase (RFK) and stearoyl-\nCoA desaturase-1 (SCD1) were identified as potentially robust candidates to modulate FSP1 function. f, \nImmunoblot (IB) analysis of RFK, FSP1, and vinculin in A375 cells transduced with either a non -targeting \ncontrol (NT) or three different RFK -targeting sgRNAs. g, Dose-dependent toxicity of RSL3 and ML210 in \nA375 cells transduced with either a non -targeting control (NT) or three different RFK -targeting sgRNAs. \nCell viability was monitored after 72 h of treatment.   \n  \nGiven these results, we focused on RFK and its role in regulating FSP1. RFK is a key enzyme,  \nphosphorylating riboflavin  to generate flavin mononucleotide (FMN) , a central step in the \nproduction of flavin adenine dinucleotide (FAD) 10—a co -factor essential for the activity of \nflavoproteins, including FSP1 11. While FSP1’s dependence on FAD implies  a direct link to RFK, \nthe extent to which RFK deficiency affects FSP1 levels and ferroptosis resistance is unknown. \nUsing A375 cells, in which FSP1 confers strong protection against GPX4 inhibitors (GPX4i) such \nas RSL3 and ML210, we found that CRISPR -mediated deletion of RFK leads to a robust loss of \nRFK expression and a decrease in FSP1 protein levels (Fig. 1f). This reduction renders RFK -\ndeficient cells highly sensitive to GPX4i, highlighting a previously uncharacterised dependency of \nferroptosis resistance on RFK (Fig. 1g). We obtained similar results using HT1080GPX4KO/FSP1OE cells, \nwhere we find that the absence of RFK impairs viability and induces ferroptosis (Extend Data 1 \nc-f). Our findings establish RFK as an actionable upstream regulator of FSP1 functionality and \nferroptosis resistance. \n \nFAD deficiency disrupts FSP1 function and promotes ferroptosis susceptibility \nInterestingly, we observed that increased sensitivity to GPX4 inhibitors (GPX4i) in cells \ntransduced with sgRNA targeting RFK was lost over time. FAD biosynthesis is a multi -step \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nprocess beginning with riboflavin uptake via members of the SLC52A family of solute carriers \n(SLC52A1, SLC52A2, and SLC52A3). Once inside the cell, riboflavin is phosphorylated by RFK \nto generate FMN, which is subsequently adenylated by flavin adenine dinu cleotide synthetase 1 \n(FLAD1) to produce FAD (Fig. 2a). Consistently, analysis of the DepMap database \n(www.depmap.org) reveals that RFK loss is poorly tolerated by most cell types (Fig. 2b). This \nintolerance was evident in our cultures, where unedited cells rapidly outcompeted RFK -deficient \ncells (Extended Data, Fig. 2a). These challenges impeded further experiments and prompted us \nto explore whether other enzymes generating the flavin co-factors could be targeted. \nNotably, unlike RFK, loss of FLAD1 —the enzyme responsible for the final step in FAD \nbiosynthesis—appears to be better tolerated by cells, likely because essential FMN-dependent \nproteins remain functional.  Therefore, we knocked out FLAD1 in A375 cells  and \nHT1080GPX4KO/FSP1OE cells and single clones thereof (Fig. 2c and Extended Data, Fig. 2b -g), which \nallowed us to establish an isogenic pair  of FLAD1-proficient and -deficient cell lines  (Fig 2c,d). \nUsing this model, we find that loss of FLAD1 leads to markedly increased sensitivity to GPX4 \ninhibition (Fig. 2e)  as well as to lipid peroxidation (Fig. 2f). We saw similar effects in the \nHT1080GPX4KO/FSP1OE model (Extended Data, Fig. 2h, j). \nLoss of flavin cofactors impairs the stability of the flavoproteome and exerts a broad metabolic \neffect, including impacting lipid metabolism 12. Therefore, we aimed to determine if the effect —\nloss of redox homeostasis—is FSP1-specific or more general. A whole proteomic analysis of the \nFLAD1 isogenic pair confirmed a reduced abundance of flavoproteins. Yet interestingly, FSP1 \nand NQO1 are the most strongly depleted flavoproteins in FLAD1 -deficient cells ( Fig. 2g), an \neffect independent of mRNA abundance (Extended Data, Fig. 2k). Building on this observation, \nwe validated the effect in other cell lines, showing that genetic loss of FLAD1 leads to a substantial \nloss of FSP1 and is accompanied by increased sensitivity to ferroptosis  Extended Data Fig. 2l, \nm).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nFigure 2. FAD deficiency disrupts FSP1 function and promotes ferroptosis susceptibility. a, \nSchematic representation of flavin mononucleotide (FMN) and flavin adenine dinucleotide (FAD) \nbiosynthesis from riboflavin (RbF). Riboflavin is first phosphorylated to form FMN in a reaction catalyzed by \nriboflavin kinase (RFK, encoded by RFK). FMN is then adenylated to form FAD in a reaction catalyzed by \nFAD synthase (FADS, encoded by FLAD1). b, CRISPR dependency scores of RFK, FLAD1, and SLC52A1-\n3 perturbations across a panel of human cancer cell lines (https://depmap.org/portal/, version 23Q2). c, \nImmunoblot (IB) analysis of FLAD1, RFK, and vinculin in A375 parental, A375 FLAD1KO single clone 1 (C1) \nand A375 FLAD1KO C1 cells stably overexpressing either an empty vector (mock) or Flag-FLAD1 (addback). \nd, Relative quantification of FAD in A375 parental and A375 FLAD1KO C1 cells stably overexpressing either \nan empty vector (mock) or Flag-FLAD1. e, Dose-dependent toxicity of RSL3 and ML210 in A375 parental, \nA375 FLAD1KO single clone 1 (C1) and A375 FLAD1KO C1 cells stably overexpressing either an empty vector \n(mock) or Flag -FLAD1. Cell viability was monitored using Alamar blue after 72 h of treatment. f, Lipid \nperoxidation evaluated by C11 -BODIPY 581/591 staining of A375 parental, A375 FLAD1 KO single clone 1 \n(C1) and A375 FLAD1KO C1 cells stably overexpressing either an empty vector (mock) or Flag-FLAD1. Cells \nwere treated with DMSO, RSL3  (200 nM), or RSL3  (200 nM) + Lip-1 (500 nM) for 6 h. g, Volcano plot of \ndifferentially-expressed proteins between A375 FLAD1 KO C1 and A375 FLAD1 KO C1 cells stably \noverexpressing Flag -FLAD1. Quantified proteins are plotted based on their fold change (FC: FLAD1 KO \nC1/FLAD1KO C1 Flag-FLAD1OE). h, Dose-dependent toxicity of RSL3 in A375 parental and A375 FLAD1 KO \ncells in the absence or presence of an FSP1 inhibitor (iFSP1 2 µM). Cell viability was monitored using after \n72 h of treatment. i, Dose-dependent toxicity of RSL3 and ML210 in A375 parental, A375 FLAD1 KO single \nclone 1 (C1) and A375 FLAD1KO C1 transduced with either a non-targeting control (NT) or a FSP1-targeting \nsgRNAs (FSP1KO). Cell viability was monitored after 72 h of treatment. j, Immunoblot (IB) analysis of ACSL4, \nFSP1, GPX4, and vinculin in A375 parental, A375 FLAD1 KO C1, A375 FLAD1 KO C1 cells stably \noverexpressing an empty vector (mock) or Flag-FLAD1 and A375 FLAD1 KO C1 transduced with either a \nnon-targeting control (NT) or a FSP1 -targeting sgRNAs (FSP1 KO). k, Lipid peroxidation evaluated by C11-\nBODIPY 581/591 staining of A375 parental, A375 FLAD1KO single clone 1 (C1), and A375 FLAD1KO C1 cells \ntransduced with either a non -targeting control (NT) or a FSP1 -targeting sgRNAs (FSP1 KO). Cells were \ntreated with DMSO, RSL3 (200 nM), or RSL3 (200 nM) + Lip-1 (500 nM) for 6 h. \n \nMolecular dynamic simulations of FSP1 with and without FAD were carried where the RMSD \nanalysis shows that the absence of FAD increases protein backbone instability ( Extended Data, \nFig. 2n). RMSF profiles reveal more significant residue fluctuations, especially in residues 282 -\n300, which interact with FAD ( Extended Data, Fig. 2 o) following our previous report 13. Overall, \nthese findings establish that FAD is essential not only for activity but also for FSP1 stability.  \nBuilding on this observation, we validated the effect in other cell lines, showing that genetic loss \nof FLAD1 leads to a s ubstantial loss of FSP1 and is accompanied by increased sensitivity to \nferroptosis Extended Data Fig. 2 l, m). Using genetic and pharmacological approaches, we find \nthat loss or inhibition of FSP1 in FLAD1 -deficient cells causes no further sensitisation to lipid \nperoxidation and ferroptosis induced by GPX4i (Fig 2h-k and Extended Data Fig. 2p). Lastly, loss \nof FLAD1 appears to sensitise cells to GPX4i specifically (Fig 2h-k): other ferroptosis inducers, \nsuch as Erastin and BSO, or other tested cytotoxic agents encompa ssing a diverse variety of \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nmodes of action (Extended Data Fig. 2p, q) exert no enhanced effects in the absence of FLAD1. \nThese findings establish FLAD1 as a key regulator of FSP1 activity and ferroptosis sensitivity. \n \nRiboflavin availability as a central determinant of ferroptosis resistance \nHaving established a functional link between intracellular flavin metabolism and ferroptosis \nresistance via FSP1, we hypothesised that the availability of riboflavin —the main precursor for \nFAD—can influence ferroptosis sensitivity. To test this, we examined proteomic changes in cells \ncultured under riboflavin -deficient conditions (Fig. 3a and Extended Fig. 3a -c). As in FLAD1 -\ndeficient cells, we observe an overall loss of flavoproteins. However, the changes appear more \nextensive, presumably because FMN-containing proteins are also affected  Extended Fig. 3d-e). \nNotably, FSP1 emerged again as the most downregulated protein in riboflavin-deficient conditions \n(Fig. 3a, b), underscoring the critical role of FAD in FSP1 stability. Riboflavin withdrawal is also \naccompanied by a characteristic upregulation of NRF2 targ et genes (e.g., AKR1Cs, TXNRD1, \nGCLM).  \nCulturing cells without riboflavin for 72 hours markedly increases their susceptibility to lipid \nperoxidation (Fig. 3c  and Extended Fig. 3 f) and enhances cell death upon GPX4 inhibition \n(GPX4i) (Fig. 3d and Extended Fig. 3g). We could rule out the sensitisation to ferroptosis upon \nriboflavin withdrawal  by a  direct antioxidant role of riboflavin (Extended Data Fig. 3h ). These \neffects are not limited to A375 cells: three additional cancer cell lines from different tissue origins \nexhibit similar responses (Fig. 3b, d). Consistent with an FSP1-dependent mechanism, combining \nGPX4i with an FSP1 inhibitor strongly sensitised cells under riboflavin-replete conditions but had \nnegligible effects in the absence of riboflavin (Fig. 3d). We further corroborated these effects in \nFSP1-deficient A375 and MDA-MB-231 cell lines. Upon GPX4 inhibition, only cells cultured with \nriboflavin displayed further ferroptosis sensitisation following FSP1 loss (Fig. 3e, f). These results \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nwere recapitulated in HT1080GPX4KO/FSP1OE and A375 cells where genetic deletion of the main \nriboflavin transporter, namely SLC52A2, disrupted FSP1 function and sensitised cells to GPX4i \n(Extended Data Fig. 4a -l). These findings  indicate that, under riboflavin -poor conditions, cells \nbecome ferroptosis-sensitive primarily through an FSP1-dependent pathway. \n \nFigure 3. Riboflavin availability as a central determinant of ferroptosis resistance. a, Volcano plot of \nquantified proteins showing their change in A375 parental cells cultured in riboflavin -deficient medium for \n96 h. Proteins are plotted based on their fold change (FC: riboflavin deficient/normal ). b, Immunoblot (IB) \nanalysis of FSP1, GPX4, and vinculin in HT1080, A375, MDA -MB-231, and A549 parental cell lines after \n96 h of growth in riboflavin -deficient medium or medium supplemented with 1 µM riboflavin. c, Lipid \nperoxidation evaluated by C11 -BODIPY 581/591 staining of A375 parental cell li ne cultured for 72 h in \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nriboflavin-deficient medium or medium supplemented with 1 µM riboflavin and after treatment with DMSO, \nRSL3 (200 nM), or RSL3  (200 nM) + Lip-1 (500 nM) for 6 h. d, Dose-dependent toxicity of RSL3 in the \nabsence or presence of an FSP1 inhibitor (iFSP1 3 µM) in HT1080, A375, MDA-MB-231, and A549 parental \ncell lines pre-cultured in riboflavin-deficient medium or medium supplemented with 1 µM riboflavin for 48 h. \nCell viability was monitored using Alamar blue after 96 h of treatment. e, Dose-dependent toxicity of RSL3 \nin the absence or presence of an FSP1 inhibitor (iFSP1 3 µM) in A375 and MDA-MB-231 cells transduced \nwith either a non -targeting control (NT) or an FSP1 -targeting sgRNAs pre -cultured in riboflavin -deficient \nmedium or medium supplemented with 1 µM riboflavin for 48 h. Cell viability was monitored after 96 h of \ntreatment. f, Immunoblot (IB) analysis of ACSL4, FSP1, NQO1, and vinculin in A375 and MDA -MB-231 \ncells transduced with either a non-targeting control (NT) or an FSP1-targeting sgRNAs cultured in riboflavin-\ndeficient medium or medium supplemented with 1 µM riboflavin for 96 h. g, Epilipidomics analysis of A375 \nparental cells pre-cultured in riboflavin-deficient medium or medium supplemented with 1 µM riboflavin for \n72 h after treatment with DMSO, RSL3 (200 nM), or RSL3 (200 nM) + Lip-1 (500 nM) for 6 h.  \n \nTo directly demonstrate that riboflavin supports protection against LPO, we analy sed the \nepilipidome of cells treated with GPX4i under riboflavin-replete and riboflavin-deficient conditions \n(Fig. 3g). Our  detailed analysis revealed a rapid and specific accumulation of oxidi sed \nphosphatidylethanolamine (PE) species 14 upon GPX4 inhibition in riboflavin -starved cells. \nCritically, treatment with Lip -1 fully reverses the formation of oxidised lipids . Together, these \nresults establish a direct link between riboflavin and membrane antioxidant capacity. \nGiven our results under riboflavin-depleted conditions, we asked whether moderate variations in \nriboflavin concentration would similarly affect sensitivity to GPX4i. Notably, human plasma \nriboflavin concentrations typically range from 10 to 20 nM15 , while standard cell culture media like \nRPMI and DMEM contain around 500 and 1000 nM, respectively. More advanced media, such as \nPlasmax16 and HPLM17, also include significantly higher riboflavin levels at 300 and 500 nM, \nrespectively. We found that physiological riboflavin levels (≤20 nM) markedly reduce FSP1 \nexpression, whereas concentrations above 100 nM are sufficient to stabilise FSP1 and confer \nferroptosis resistance (Extended Data, Fig. 4j).  Altogether, our studies demonstrate that riboflavin \navailability is a central determinant of membrane repair capacity and determines FSP1 antioxidant \nrecycling capacity. \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nRiboflavin analogues disrupt FSP1 activity and promote ferroptosis \nOur results suggest that pharmacologically targeting riboflavin metabolism, to disrupt its FSP1 -\nprotective branch, would sensitise cancer cells to ferroptosis. No selective inhibitors for any of the \nproteins involved in riboflavin uptake (SLC52A2) or its action toward FMN (RFK) and FAD \n(FLAD1) have been described. Nonetheless, bacteria from the genus Streptomyces (e.g . \ndavaonensis and cinnabarinus) produce the riboflavin antimetabolite roseoflavin that exerts \nantimicrobial activity by binding to and disrupting riboflavin riboswitches. Roseoflavin has shown \npromise as a broad-spectrum antibiotic, but few studies have explored it as an anticancer agent. \nIn eukaryotic cells, reseoflavin is thought to follow a similar metabolic route as riboflavin —being \ntransported, phosphorylated, and adenylated (Fig. 4a) 18. The dimethylamino group on C8 of the \nisoalloxazine ring leads to the formation of an altered flavin cofactor that is believed to disrupt \nnormal flavin-mediated electron transfer reactions. \nUsing our HT1080GPX4KO/FSP1OE cells, we investigated whether roseoflavin  influences FSP1 -\nmediated ferroptosis protection. Notably, at physiologically relevant levels of riboflavin, roseoflavin \ninduced ferroptosis within the single-digit nanomolar range (Fig. 4b). Moreover, roseoflavin affects \nthe response to GPX4i only in FSP1-expressing cells (Fig. 4c and Extended Data Fig. 5 a, b); no \nsensitisation was detected in FSP1-deficient cells (Fig. 4c and Extended Data Fig. 5 c), indicating \nthat the effect of roseoflavin is specific. Additionally, roseoflavin can restore FSP1 levels when \ncells are treated under low and physiological riboflavin conditions (Fig 4e), suggesting that the  \neffect is likely on-target. Finally, the effect of roseoflavin was broadly reproduced in a larger panel \nof cell lines (Fig 4f), where treatment can restore FSP1 levels (Fig 4e). These actions reflect the \nability of roseoflavin adenine dinucleotide (roFAD), like FAD, to stabilise FSP1; however, the \nmodification of the isoalloxazine group destroys its oxidoreductase function. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nAltogether, these results demonstrate that FSP1 activity can be effectively inhibited by riboflavin \nantimetabolites, offering additional opportunities to target ferroptosis protective mechanisms in \ncancer cells. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nFigure 4. The riboflavin analogue roseoflavin disrupts FSP1 activity and promote s ferroptosis. a, \nSchematic representation of the metabolism of riboflavin (RbF) and its analog roseoflavin (RoF) by \nriboflavin kinase (encoded by RFK) and FAD synthase (encoded by FLAD1). b, Dose-dependent toxicity of \nroseoflavin (RoF) and iFSP1 in the absence or presence of Lip -1 (500 nM) in HT1080GPX4KO/FSP1OE. Cell \nviability was monitored using Alamar blue after 48 h of treatment. Cells were cultured in low -riboflavin \nmedium (20 nM). c, Dose-dependent toxicity of the GPX4 inhibitor ML210 in A375 cells transduced with \neither a non -targeting control (NT) or a FSP1 -targeting sgRNAs (FSP1 KO) that were pre -treated with \nroseoflavin (RoF, 1, 3, and 10 nM) for 48 h. Cell viability was monitored after 48 h of ML210 treatment. Cells \nwere cultured in low -riboflavin medium (20 nM). d, Immunoblot analysis (IB) of FSP1, NQO1, ACSL4, \nGPX4, and vinculin treated with roseoflavin (RoF 0, 20, 100, and 1000 nM) for 96 h in the absence or \npresence of riboflavin (RbF 20 nM). e, Dose-dependent toxicity of the GPX4 inhibitor ML210 in HT1080, \nMDA-MB-436, PC-9, and H460 parental cell lines pre -treated with roseoflavin (RoF, 1, 3, and 10 nM) for \n48 h. Cell viability was monitored after 48 h of ML210 treatment. f, Immunoblot (IB) analysis of FSP1, NQO1, \nand vinculin in HT1080, MDA-MB-436, PC9, and H460 parental cell lines treated with roseoflavin (RoF, 20 \nnM) for 48 h in the absence or presence of riboflavin (RbF 20 nM). g, Riboflavin metabolism supports FSP1 \nfunction and promotes ferroptosis resistance. This is therefore a process that riboflavin analogs, such as \nroseoflavin can modulate.  \n \nDiscussion \nHere we reveal a critical role for riboflavin in governing ferroptosis via FSP1. Previous work \nestablished FSP1 as an essential safeguard against lipid peroxidation, complementing the activity \nof GPX4; however, the upstream factors regulating FSP1 functionality were largely unknown. By \nsystematically dissecting FAD biosynthesis, we demonstrate that multiple nodes in the riboflavin–\nFMN–FAD pathway directly influence FSP1’s stability and enzymatic capacity to recycle lipophilic \nantioxidants. In particular, we  highlight FLAD1 as a key enzyme in the terminal step of FAD \nproduction and reveal that its depletion specifically compromises FSP1, with broad implications \nfor ferroptosis sensitivity and cell viability. These findings enhance our understanding of how cells \ndepend on riboflavin metabolism to preserve membrane integrity under oxidative stress. \nCrucially, we observed that standard tissue culture media supply riboflavin at concentrations far \nexceeding physiological plasma levels, potentially obscuring the impact of riboflavin availability \non ferroptosis. Culturing cells at physiological or deficie nt riboflavin levels revealed pronounced \nchanges in FSP1 expression and ferroptosis outcomes, underscoring how even moderate \nfluctuations in riboflavin can alter redox balance. This insight raises important considerations for \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nmechanistic studies and potential clinical contexts in which dietary riboflavin or altered metabolism \nmay drive vulnerability to ferroptosis-related diseases19. \nInterestingly, a parallel can be drawn between the regulation of FSP1 by riboflavin and the \nregulation of GPX4 by selenium. It is well -established that selenium availability dictates GPX4 \nprotein translation, with selenium metabolism pathways attracting significant interest as potential \nmodulators of ferroptosis 20-22. Similarly, our findings suggest that riboflavin —by serving as a \nprecursor for FAD —determines the abundance and activity of FSP1 independently of mRNA \nlevels. Importantly, both GPX4 and FSP1 highlight how two crucial regulators of ferroptosis cannot \nbe understood solely by analy sing the abundance of their mRNA. Instead, the availability of \nspecific micronutrients—selenium for GPX4 and riboflavin for FSP1 —is required for their proper \ntranslation and functionality. This establishes a broader principle where micronutrients and their \nmetabolic pathways modulate protein functionality and ferroptosis susceptibility, suggesting \nsignificant new opportunities for therapeutic intervention. \nBeyond elucidating a role for endogenous riboflavin metabolism, we highlight the value of \nriboflavin antimetabolites, such as roseoflavin, as potent modulators of FSP1 function. Owing to \nits structural similarity to riboflavin, roseoflavin likely shares efficient tissue distribution and \ndemonstrates activity in the nanomolar range  in physiologically relevant riboflavin \nconcentrations—substantially lower than current FSP1 inhibitors. An additional advantage of \nroseoflavin lies in its uptake and metabolism, w hich depend on SLC52A2, RFK, and FLAD1. \nMutations in these proteins that could affect resistance would also compromise riboflavin uptake \nand metabolism, ultimately preventing the production of essential cofactors like FAD and FMN , \nas recently exemplified in Plasmodium falciparum23. This dual impact makes it unlikely for cells to \nselectively develop resistance to roseoflavin without severely impairing  riboflavin metabolisation \nand FAD production . In contrast, specific inhibitors of FSP1 are more prone to resistance \nmechanisms, such as mutations in FSP1 itself or activation of compensatory pathways, as these \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\ndo not disrupt the broader metabolic network. Roseoflavin’s reliance on essential and conserved \nmetabolic pathways could thus provide a significant therapeutic advantage. \nAltogether, our results show  that intracellular flavin metabolism is pivotal to  FSP1’s protective \nfunction against phospholipid peroxidation, operating independently of GPX4. This framework \nconstitutes a previously underappreciated approach for enhancing ferroptosis in cancer cells and \nother contexts where FSP1 supports survival. Moreover, our work unveils riboflavin’s role in the \nFSP1-driven recycling of lipophilic antioxidants, offering fundamental insight into  the complex \ninteraction of nutrients with im portant implications for understanding  inconsistent outcomes in \npreclinical and clinical studies of antioxidant therapies 24-26. \n \nAcknowledgements \nJ.P.F.A. acknowledges the support of the Junior Group Leader program of the Rudolf Virchow \nCenter, University of Würzburg and additional support from the Deutsche \nForschungsgemeinschaft (DFG), FR 3746/3 -1, FR 3746/5 -1, FR 3746/6 -1, CRC205 (INST \n269/886-1), and TRR 387/1 (514894665); the EU -H2020 (ERC -Consolidator, DeciFERR); the \nDeutsche Jose Carreras Leukämie Stiftung (DJCLS 01 R/2022) ; and The São Paulo Research \nFoundation (2023/04397-4). Work in the Fedorova lab is supported by ‘‘Sonderzuweisung zur \nUnterstützung profilbestimmender Struktureinheiten’’ by the SMWK to TUD, TG70 by Sächsische \nAufbaubank and SMWK, the measure is co-financed with tax funds based on the budget passed \nby the Saxon state parliament (to M.F.), Deutsche Forschungsgemeinschaft (FE 1236/5 -1, FE \n1236/8-1 to M.F.), and Bundesministerium für Bildung und Forschung (031L0315A, DEEP_HCC \nand 01EJ2205A, FERROPath to M.F.).  We also want  to acknowledge the German Cancer \nResearch Center (DKFZ) Proteomics Core Facility. M.C. acknowledges support from DFG (CO \n291/7-1, the Priority Program SPP 2306 [CO 291/9-1, #461385412; CO 291/10-1, #461507177], \nand the CRC TRR 353 [CO 291/11 -1; #471011418]); the European Research Council (ERC) \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nunder the European Union’s Horizon 2020 research and innovation programme (grant agreement \nno. GA 884754) and the German Federal Ministry of Education and Research (BMBF) \nFERROPATH (01EJ2205B) to M.C. and B.P.  Illustrations in Fig. 2a and 4g were created in \nBioRender. Skafar, V. (2025) https://BioRender.com/j20e353). \n \n \nAuthor Contributions \nV.S., and  I.S., performed most experiments with support from  A.F.S., F.P.F. and M.D.  Z.C. \nperformed and analysed the CRISPR screen . P.N. and M.F performed epilipidomics analysis .. \nW.S. performed the analysis and interpretation of flavin quantification using HPLC -MS. B. G. \ncontributed with the statistical analysis of proteomics data . S.M. designed and assisted with RT-\nqPCR analysis. A.M, A.F.N.A., J. B., C.A-S., carried molecular dynamic simulations to determine \nthe impact of FAD loss in FSP1 stability. H.G.A., S.M., U. E., M.E., J. T., K. H., A. F. N. A., J. B., \nC.A., R.G.B., B.P., M.C., M.F., and H. A. contributed with reagent, critical information and/or \nplatforms. J.P.F.A. initiated and supervised the project. All authors contributed with discussion and \ndata interpretation and read and agreed on the paper's content. \n \nConflict of Interest \nM.C. and B.P. are co-founders and shareholders of ROSCUE Therapeutics GmbH.  \n \n \nReference \n \n1 Xu, L., Davis, T. A. & Porter, N. A. Rate constants for peroxidation of polyunsaturated fatty \nacids and sterols in solution and in liposomes. J Am Chem Soc 131, 13037-13044 (2009). \nhttps://doi.org/10.1021/ja9029076 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n2 Stockwell, B. R. Ferroptosis turns 10: Emerging mechanisms, physiological functions, and \ntherapeutic applications. Cell 185, 2401 -2421 (2022). \nhttps://doi.org/10.1016/j.cell.2022.06.003 \n3 Yang, W. S. et al. Regulation of ferroptotic cancer cell death by GPX4. Cell 156, 317-331 \n(2014). https://doi.org/10.1016/j.cell.2013.12.010 \n4 Friedmann Angeli, J. P. et al. Inactivation of the ferroptosis regulator Gpx4 triggers acute \nrenal failure in mice. Nat Cell Biol 16, 1180-1191 (2014). https://doi.org/10.1038/ncb3064 \n5 Angeli, J. P. F., Shah, R., Pratt, D. A. & Conrad, M. Ferroptosis Inhibition: Mechanisms and \nOpportunities. Trends Pharmacol Sci  38, 489 -498 (2017). \nhttps://doi.org/10.1016/j.tips.2017.02.005 \n6 Doll, S. et al. FSP1 is a glutathione-independent ferroptosis suppressor. Nature 575, 693-\n698 (2019). https://doi.org/10.1038/s41586-019-1707-0 \n7 Bersuker, K.  et al.  The CoQ oxidoreductase FSP1 acts parallel to GPX4 to inhibit \nferroptosis. Nature 575, 688-692 (2019). https://doi.org/10.1038/s41586-019-1705-2 \n8 Mishima, E.  et al.  A non-canonical vitamin K cycle is a potent ferroptosis suppressor. \nNature 608, 778-783 (2022). https://doi.org/10.1038/s41586-022-05022-3 \n9 Peck, B. et al. Inhibition of fatty acid desaturation is detrimental to cancer cell survival in \nmetabolically compromised environments. Cancer Metab  4, 6 (2016). \nhttps://doi.org/10.1186/s40170-016-0146-8 \n10 Barile, M., Giancaspero, T. A., Leone, P., Galluccio, M. & Indiveri, C. Riboflavin transport \nand metabolism in humans. J Inherit Metab Dis  39, 545 -557 (2016). \nhttps://doi.org/10.1007/s10545-016-9950-0 \n11 Lienhart, W. D., Gudipati, V. & Macheroux, P. The human flavoproteome. Arch Biochem \nBiophys 535, 150-162 (2013). https://doi.org/10.1016/j.abb.2013.02.015 \n12 Martinez-Limon, A., Calloni, G., Ernst, R. & Vabulas, R. M. Flavin dependency undermines \nproteome stability, lipid metabolism and cellular proliferation during vitamin B2 deficiency. \nCell Death Dis 11, 725 (2020). https://doi.org/10.1038/s41419-020-02929-5 \n13 Nakamura, T. et al. Integrated chemical and genetic screens unveil FSP1 mechanisms of \nferroptosis regulation. Nat Struct Mol Biol  30, 1806 -1815 (2023). \nhttps://doi.org/10.1038/s41594-023-01136-y \n14 Kagan, V. E. et al. Oxidized arachidonic and adrenic PEs navigate cells to ferroptosis. Nat \nChem Biol 13, 81-90 (2017). https://doi.org/10.1038/nchembio.2238 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n15 Tan, A. et al.  Plasma riboflavin concentration as novel indicator for vitamin -B2 status \nassessment: suggested cutoffs and its association with vitamin -B6 status in women. P \nNutr Soc 79, E658-E658 (2020). https://doi.org/10.1017/S0029665120006072 \n16 Vande Voorde, J.  et al.  Improving the metabolic fidelity of cancer models with a \nphysiological cell culture medium. Sci Adv  5, eaau7314 (2019). \nhttps://doi.org/10.1126/sciadv.aau7314 \n17 Cantor, J. R.  et al. Physiologic Medium Rewires Cellular Metabolism and Reveals Uric \nAcid as an Endogenous Inhibitor of UMP Synthase. Cell 169, 258 -272 e217 (2017). \nhttps://doi.org/10.1016/j.cell.2017.03.023 \n18 Pedrolli, D. B.  et al. The antibiotics roseoflavin and 8 -demethyl-8-amino-riboflavin from \nStreptomyces davawensis are metabolized by human flavokinase and human FAD \nsynthetase. Biochem Pharmacol  82, 1853 -1859 (2011). \nhttps://doi.org/10.1016/j.bcp.2011.08.029 \n19 McNulty, H., Pentieva, K. & Ward, M. Causes and Clinical Sequelae of Riboflavin \nDeficiency. Annu Rev Nutr  43, 101 -122 (2023). https://doi.org/10.1146/annurev -nutr-\n061121-084407 \n20 Alborzinia, H. et al. LRP8-mediated selenocysteine uptake is a targetable vulnerability in \nMYCN-amplified neuroblastoma. EMBO Mol Med , e18014 (2023). \nhttps://doi.org/10.15252/emmm.202318014 \n21 Dos Santos, A. F., Fazeli, G., Xavier da Silva, T. N. & Friedmann Angeli, J. P. Ferroptosis: \nmechanisms and implications for cancer development and therapy response. Trends Cell \nBiol (2023). https://doi.org/10.1016/j.tcb.2023.04.005 \n22 Chen, Z.  et al.  PRDX6 contributes to selenocysteine metabolism and ferroptosis \nresistance. Mol Cell  84, 4645 -4659 e4649 (2024). \nhttps://doi.org/10.1016/j.molcel.2024.10.027 \n23 Hemasa, A., Spry, C., Mack, M. & Saliba, K. J. Mutation of the Plasmodium falciparum \nFlavokinase Confers Resistance to Roseoflavin and 8-Aminoriboflavin. ACS Infect Dis 10, \n2939-2949 (2024). https://doi.org/10.1021/acsinfecdis.4c00289 \n24 Bjelakovic, G., Nikolova, D. & Gluud, C. Antioxidant supplements to prevent mortality. \nJAMA 310, 1178-1179 (2013). https://doi.org/10.1001/jama.2013.277028 \n25 Kontush, A., Finckh, B., Karten, B., Kohlschutter, A. & Beisiegel, U. Antioxidant and \nprooxidant activity of alpha -tocopherol in human plasma and low density lipoprotein. J \nLipid Res 37, 1436-1448 (1996).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n26 Rimm, E. B. et al. Vitamin E consumption and the risk of coronary heart disease in men. \nN Engl J Med 328, 1450-1456 (1993). https://doi.org/10.1056/NEJM199305203282004 \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \nExtended Figures \n \nExtended Figure 1. Identification of factors supporting FSP1 function. a , Dose-dependent toxicity of \nthe SCD1 inhibitor CAY10566 in the absence or presence of the ferroptosis inhibitor Lip -1 (500 nM) in \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nHT1080GPX4KO/FSP1OE cells. Cell viability was monitored using Alamar blue after 144 h of treatment. b, Dose-\ndependent toxicity of the SCD1 inhibitor CAY10566 in the absence or presence of Lip-1 (500 nM) in HT1080 \nGPX4KO/FSP1OE cells. Cell viability was monitored by crystal violet staining after 6 days of treatment. c, \nDose-dependent toxicity of RSL3 in the presence of iFSP1 (3 µM) and/or Lip -1 (500 nM) in A375 cells \ntransduced with either a non -targeting control (NT) or three different RFK -targeting sgRNAs. Cell viability \nwas monitored using Alamar blue after 72 h of treatment. d, Cell viability of HT1080GPX4KO/FSP1OE cells \ntransduced with either a non-targeting control (NT) or three different RFK-targeting sgRNAs in the absence \nor presence of Lip-1 (500 nM). Cell viability was monitored using Alamar blue after 96 h. e, Immunoblot (IB) \nanalysis of RFK, FSP1 and vinculin in HT1080GPX4KO/FSP1OE cells transduced with either a non -targeting \ncontrol (NT) or three different RFK -targeting sgRNAs. f, Growth curves showing the percentage  of \nconfluence over time for HT1080GPX4KO/FSP1OE cells transduced with either a non -targeting control (NT) or \nan RFK-targeting sgRNAs in the absence or presence of Lip -1 (500 nM). Data are presented as mean ± \nSEM. The error was calculated per image. g, Cell viability of HT1080GPX4KO/FSP1OE cells transduced with \neither a non-targeting control (NT) or an RFK -targeting sgRNAs in the absence or presence of Lip -1 (500 \nnM). Cell viability was monitored by crystal violet staining after 96 h. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nExtended Figure 2. FAD deficiency disrupts FSP1 function and promotes ferroptosis susceptibility. \na, Immunoblot (IB) analysis of RFK, FSP1 and vinculin in A375 cells transduced with either a non-targeting \n(NT) or RFK -targeting sgRNAs collected at an early passage or after two weeks of passaging (later \npassage). b, Immunoblot (IB) analysis of FSP1, ACSL4 and ꞵ -actin in HT1080 GPX4 KO/FSP1OE cells \ntransduced with either a non-targeting control (NT) or three different FLAD1-targeting sgRNAs. c, Relative \nquantification of FAD in HT1080 GPX4KO/FSP1OE cells transduced with either a non-targeting control (NT) \nor three different FLAD1 -targeting sgRNAs. d, Cell viability of HT1080 GPX4 KO/FSP1OE cells transduced \nwith either a non -targeting control (NT) or three different FLAD1 -targeting sgRNAs in the absence or \npresence of Lip -1 (500 nM). Cell viability was monitored using Alamar blue after 96 h. e, R elative \nquantification of riboflavin (RoF), flavin mononucleotide (FMN) and flavin adenine dinucleotide (FAD) in \nHT1080 GPX4KO/FSP1OE cells transduced with either a non -targeting control (NT) or an FLAD1 -targeting \nsgRNAs (FLAD1 sgRNA1). f, Genotyping of wild -type (WT) and FLAD1 KO cells. The scheme shows the \nFLAD1 gene, indicating exons and the target site of FLAD1 sgRNA1 (red) in exon 7. The PCR products \nspanning the FLAD1 sgRNA1 cut site in WT and FLAD1KO (single clone, C1) from HT1080 GPX4KO/FSP1OE \nand A375 cells were resolved on a 1% agarose gel. g, Sanger sequencing chromatograms of PCR products \nconfirmed the presence of wild type (WT) and mutant alleles in the FLAD1 locus in HT1080 GPX4KO/FSP1OE \nand A375 cells. h, Cell viability of HT1080 GPX4KO/FSP1OE non target (NT), FLAD1KO single clone 1 (C1), \nand FLAD1KO C1 cells stably overexpressing either an empty vector (mock) or Flag -FLAD1 (addback) in \nthe absence or presence of Lip -1 (500 nM). Cell viability was monitored using Alamar blue after 24 h. i, \nRelative quantification of FAD in HT1080 GPX4 KO/FSP1OE cells “parental”, FLAD1KO C1 and FLAD1KO C1 \ncells stably overexpressing either an empty vector (mock) or Flag -FLAD1 (addback). j, Immunoblot (IB) \nanalysis of ACSL4, FSP1, Flag -tag and β-actin in HT1080 GPX4 KO/FSP1OE “parental” and FLAD1 KO C1 \ncells stably overexpressing an empty vector (mock) or Flag-FLAD1 (addback). k, Relative mRNA levels of \nFSP1 measured by quantitative RT-PCR in A375 parental, FLAD1KO single clone 1 (C1) and FLAD1 KO C1 \noverexpressing either an empty vector (mock) or Flag -FLAD1 (addback). l, Immunoblot (IB) analysis of \nFSP1, GPX4 and vinculin in HT1080, A375, MDA -MB-231 and A549 cells transduced with either a non -\ntargeting (NT) or FLAD1 -targeting sgRNA (FLAD1 sg RNA1). m, Dose-dependent toxicity of RSL3 in the \nabsence or presence of iFSP1 (3 µM) in HT1080 cells transduced with either a non -targeting (NT) or \nFLAD1-targeting sgRNA (FLAD1 sgRNA1). n, Root mean square deviation (RMSD) of FSP1 backbone in \nFSP1-NAD-FAD and FSP1 -NAD complexes for three 500  ns-MD simulations.  o, R oot mean square \nfluctuation (RMSF) of FSP1 residues in FSP1-NAD-FAD and FSP1-NAD complexes. The absence of FAD \nleads to more instability in the protein structu re. p, Dose -dependent toxicity of the indicated cytotoxic \ncompounds in A375 parental, A375 FLAD1 KO single clone 1 (C1) and A375 FLAD1 KO C1 cells stably \noverexpressing an empty vector (mock) or Flag -FLAD1 (addback). q, Dose -dependent toxicity of the \nindicated cytotoxic compounds in A375 parental, A375 FLAD1KO single clone (C1) and A375 FLAD1KO C1 \ntransduced with either a non-targeting control (NT) or a FSP1-targeting sgRNAs (FSP1KO). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nExtended Figure 3. Riboflavin availability as a central determinant of ferroptosis resistance. a, \nImmunoblot (IB) analysis of FSP1, GPX4 and vinculin in HT1080, A375, MDA -MB-231 and 549 parental \ncell lines after 24, 48 and 144 h of growth in riboflavin -deficient medium or supplemented with 1 µM of \nriboflavin. b, Volcano plots of quantified proteins showing their change in A375 parental cells cultured in \nriboflavin-deficient medium for 24, 48 and 144 h. Proteins are plotted based on their fold change (FC: \nriboflavin deficient/normal). c, Heatmaps of quantified proteins showing their change in A375 parental cells \ncultured in riboflavin-deficient medium for 24, 48, 96 and 144 h (FC: riboflavin deficient/normal). d, Volcano \nplot of quantified flavoproteins showing their change in A375 parental cells cultured in riboflavin -deficient \nmedium for 96 h (FC: riboflavin deficient/normal). e, Heatmap of quantified flavoproteins showing their \nchange in A375 parental cells cultured in riboflavin -deficient medium for 96 h (FC: riboflavin \ndeficient/normal). f, Lipid peroxidation evaluated by C11 -BODIPY 581/591 staining of A375 parental cell \nline cultured for 72 h in riboflavin -deficient medium or supplemented with 1 µM of riboflavin and after \ntreatment with DMSO, RSL3 (200 nM) or RSL3 (200 nM) + Lip -1 (500 nM)  for 6 h. g, Dose-dependent \ntoxicity of RSL3 in the presence of Lip -1 (500 nM) and iFSP1 (3 µM, when indicated) in HT1080, A375, \nMDA-MB-231 and A549 parental cell lines cultured in riboflavin -deficient medium or supplemented with 1 \nµM of riboflavin for 48 h. Cell viability was monitored using Alamar blue after 96 h of treatment. h, \nAbsorbance at 517 nm corresponding to the radical initiator 2,2 -diphenyl-1-picrylhydrazyl (DPPH) co -\nincubated with ferrostatin-1 (Fer-1), riboflavin (RbF) or roseoflavin (RoF) (n = 3). Data plotted are mean ± \nSD. Statistical significance was determined by one-way ANOVA followed by Dunnett´s multiple comparison \ntest.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nExtended Figure 4. Riboflavin availability as a central determinant of ferroptosis resistance.  a, \nTranscript expression levels of the riboflavin transporters SLC52A1, SLC52A2 and SLC52A3 for a panel of \nhuman cancer cell lines from Dependency Map public 23Q2 dataset (https://depmap.org/portal/, version \n23Q2). b, Cell viability of HT1080 GPX4KO/FSP1OE cells transduced with either a non-targeting control (NT) \nor four different SLC52A2 -targeting sgRNAs in the absence or presence of Lip -1 (500 nM). Cell viability \nwas monitored using Alamar blue after 96 h. c, Immunoblot (IB) analysis of FSP1 and ꞵ -actin in HT1080 \nGPX4KO/FSP1OE “parental”, non-target control (NT), SLC52A2 KO single clone 1 (C1) and SLC52A2 KO C1 \ncells stably overexpressing an empty vector (mock) or SLC52A2 (addback). d, Cell viability of HT1080 \nGPX4KO/FSP1OE cells non target control (NT), SLC52A2 KO single clone 1 (C1) and SLC52A2 KO C1 cells \nstably overexpressing an empty vector (mock) or SLC52A2 (addback). e, Genotyping of wild-type (WT) and \nSLC52A2KO cells. The scheme shows the SLC52A2 gene, indicating exons and the target site of SLC52A2 \nsgRNA2 (red) in exon 3. f, Sanger sequencing chromatograms of PCR products confirmed the presence \nof WT and mutant alleles in the SLC52A2 locus in HT1080 GPX4KO/FSP1OE and A375 cells. g, Schematic \nrepresentation of the sequencing results obtained from the PCR product (in blue) covering the edited region \n(in red) in comparison with the wild-type product. h, Immunoblot analysis (IB) of FSP1, GPX4 and vinculin \nin A375 parental, non-target control (NT), SLC52A2 KO single clone 1 (C1) and SLC52A2 KO single clone 1 \n(C1) stably overexpressing an empty vector (mock) or SLC52A2 (addback). i, Dose-dependent toxicity of \nRSL3 in A375 non -target control (NT ), SLC52A2 KO single clone 1 (C1) and SLC52A2 KO C1 stably \noverexpressing an empty vector (mock) or SLC52A2 (addback).  j, Immunoblot analysis (IB) of ACSL4, \nFSP1, NQO1, GPX4 and vinculin in A375, A549, HT1080 and H460 parental cell lines cultured in medium \nsupplemented with either standard FBS or dialyzed FBS and different concentrations of riboflavin (RbF).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n \nExtended Figure 5. The riboflavin analogue roseoflavin  disrupts FSP1 activity and promotes \nferroptosis. a, Dose-dependent toxicity of ML210 in A375 parental cell line pre -treated with roseoflavin \n(RoF, 0, 1, 3 and 10 nM) for 48 h. Cell viability was monitored using Alamar blue after 48 h of treatment with \nML210. Cells were cultured in low-riboflavin medium (20 nM). b, Immunoblot (IB) analysis of FSP1, NQO1 \nand vinculin in A375 parental cell line treated with roseoflavin (RoF, 20 nM) for 96 h in the absence or \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\npresence of riboflavin (RbF 20 nM). c, Dose-dependent toxicity of ML210 in A375, HT1080, MDA-MB-436, \nPC-9 and H460 parental cell lines pre-treated with roseoflavin (RoF, 10 nM) and/or iFSP1 (3 µM) for 48 h. \nCell viability was monitored using Alamar blue after 48 h of treatment. Cells were cultured in low-riboflavin \nmedium (20 nM).  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nMethods \nChemicals \nLip-1 (Sigma, cat. no. SML1414), riboflavin (Sigma, cat. no. R9504), roseoflavin (Sigma, cat. no. \nSML1583), RSL3 (Sigma, SML2234), ML210 (Sigma, SML0521), CAY10566 (MedChemExoress, \ncat. no. HY -15823), erastin (Selleckchem, cat. no. S724203), L -Buthionine-sulfoximine (BSO, \nSigma cat. no. B25159), TRi -1 (MedChemExoress, cat. no. HY -125006), etoposide \n(MedChemExoress, cat. no. HY -13629), PLX4032 (Selleckchem, cat. no. S1267), auranofin \n(Sigma, cat. no. A6733), camptothecin (MedChemExoress, cat. no. HY-16560), protamine sulfate \n(Sigma, cat. no. P3369), bortezomib (PS -341, MedChemExpress, cat. no. HY-10227) and C11-\nBODIPY (581/591) (Invitrogen, cat. no. D3861) were used in this study.  \nCell lines \nHuman cancer cell lines HT1080, A375, MDA -MB-231, MDA -MB-436, A549 and H460 were \npurchased from ATCC. LOX-IMVI cells were obtained from NCI/NIH (National Cancer Institute, \nNational Institutes of Health, USA). Cells were cultured in high-glucose DMEM GlutaMax (Gibco, \ncat. no. 31966-021, 4.5 g/L D-glucose) supplemented with 10% fetal bovine serum (FBS, Gibco, \nref. A5256701) and 1% penicillin -streptomycin (Gibco, cat. no. 15140122). GPX4, FSP1, RFK, \nFLAD1, SLC52A2 knockout cells were cultured in the presence o f Lip -1 (500 nM) for \nmaintenance. All cells were cultured at 37 °C with 5% CO2 and routinely tested for Mycoplasma \ncontamination in the laboratory.  \nLentivirus production and transduction \nHEK293T cells were used to produce replication -incompetent lentiviral particles. A third -\ngeneration lentiviral packaging system consisting of transfer plasmids, envelope plasmid (pCMV-\nVSV-G) and packaging plasmids (pRSV_Rev and HIV -1 GAG/Pol) was used. Bri efly, 700,000 \ncells per well were seeded on 6 -well plates and cultured overnight. The next day, the cells were \nco-transfected with transfer, envelope and packaging plasmids (3 µg DNA/well in a proportion \n1:1:1:1) using transfection reagent (X -tremeGENE HP reagent (Roche, cat. no. 06366236001)). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nViral particle -containing cell culture supernatants were harvested 48 hours after transfection, \nfiltered through a 0.45 µm membrane and used for lentiviral transduction. For transduction of the \ncell line of interest, 100,000 cells were seeded in a 6 -well plate in medium supplemented with \nprotamine sulfate (8 µg/mL) and Lip -1 (500 nM) and directly incubated with the lentivirus -\ncontaining supernatant for 48 hours. Transduced cells were selected by replacing the cell culture \nmedium with fresh medium containing  appropriate antibiotics, such as puromycin (1 µg/mL), \nblasticidin (10 µg/mL) or G418 (1 mg/mL) until non-transduced cells died. \nCRISPR-cas9 screen \nHT1080 GPX4KO FSP1-FlagOE stably expressing cas9 were transduced with a lentiviral human \nCRISPR knockout library (VBC). This library contains 15,062 sgRNAs targeting 3007 “druggable \ngenes” (5 sgRNAs per gene). 26 million cells were transduced to achieve 50 0x coverage with \n1000 mg of total DNA in the presence of Lip-1 (500 nM). 48 h after infection, cells were selected \nwith G418 (1 mg/mL) for 7 days. Once the selection was done, cells were harvested (day 0) and \nthe remaining cells were split into two conditi ons (DMSO and Lip-1 500 nM) and maintained for \nan additional 14 days. After these 2 weeks of passaging, cells were harvested again (day 14) and \ngenomic DNA was isolated from cell pellets (Qiagen, cat. no. 69504), followed by PCR \namplification and next -generation sequencing (Illumina MiniSeq High -Output, 25M reads). The \nmapping of raw sequencing reads to the reference library, computing of enrichment scores and \np-values were processed using the MaGeCK pipeline. MaGeCKFlute was used to visualize hits \nand associated pathways. \nCRISPR-cas9 mediated gene knockout \nsgRNAs were selected using the VBC score (https://www.vbc -score.org/). Guides were cloned \nusing annealed oligonucleotides (Eurofins genomics) with specific overhangs complementary to \nBsmBI-digested lentiCRISPRv2 -puro or lentiCRISPRv2 -blast (Addgene, cat. n o. 98290 and \n98293, respectively) and selected for 4 -7 days. Knockout efficiency was validated by \nimmunoblotting (when antibodies were available) and sequencing of genomic DNA. In the case \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nof the FLAD1 gene, the knockout was also confirmed by measuring FAD levels. Cells were used \nas pools unless stated otherwise. The sequences of the guides used in the study are provided \nbelow.  \nhGPX4_sgRNA1 - GAAGCGCTACGGACCCATGG \nhFSP1_sgRNA1 - GGTGCAGAGAATCACCAGGT \nhRFK_sgRNA1 - GAGATGTCCATAAGATGG \nhRFK_sgRNA2 - ACTTGACCCCGGCAGAAGTA \nhRFK_sgRNA3 - CTATGGGGAAATCCTCAATG \nhFLAD1_sgRNA1 - AGTCGGGAGAATACCGTG \nhFLAD1_sgRNA2 - GATCTGTGCCATAATGC \nhFLAD1_sgRNA3 - GAGTAGGGGTCAGTCCGG \nhSLC52A2_sgRNA1 - GCCAGGAAGCAGGCCAG \nhSLC52A2_sgRNA2 - TAAGCAGGAAAAGCTCTGCA \nhSLC52A2_sgRNA3 - CTGGCTGCCACCTTCACGT \nhSLC52A2_sgRNA4 - GTGGGTCAGCACCGGAC \nStable protein overexpression \nHT1080 and A375 cells overexpressing human FSP1, FLAD1 and SLC52A2 proteins were \ngenerated by transduction using lentivirus particles as described above. For cloning FSP1, FLAD1 \nand SLC52A2 expression vectors, the sequences (CCDS7297.1, CCDS1078.1 and \nCCDS6423.1, respectively) were codon -optimized (IDT Codon Optimization Tool) and \nsynthesized by IDT as gBlocksTM and cloned using ClonExpress II One Step Cloning Kit \n(Vazyme, cat. no. C112) into p442-IRES-blast or p442-IRES-neo vectors. Afterward, the plasmids \nwere delivered into the cell of interest using lentivirus particles as described above.  \nCell viability assays \nCells were seeded at 1000 cells/well on 96 -well plates and after 4 -6 hours treated with the \nfollowing compounds: RSL3, ML210, iFSP1, Lip -1, erastin, BSO, TRi -1, etoposide, PLX -4032, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nauranofin, camptothecin, bortezomib and roseoflavin. Cell viability was assessed 72 -96 hours \nafter treatment using the Alamar Blue method as an indicator of viable cells. Alamar Blue solution \nwas made by dissolving 1 g of resazurin sodium salt in 100 mL of PBS and filtered through a 0.22 \nµm membrane. The working solution was made fresh by adding 80 µL of the stock solution to \nevery 10 mL of DMEM medium. After 3 hours of incubation, viability was estimated by measuring \nfluorescence using a 540/35 excitation filter and a 590/20 emission on a Spark® microplate reader \n(Tecan, Zürich, Switzerland).  \nLive-Cell Imaging and Proliferation analysis \nCell proliferation was monitored using the IncuCyte® Live -Cell Analysis System S3/SX1 \n(Sartorius). The plate was placed in the IncuCyte incubator, and phase -contrast images were \nacquired every 6 hours for a total duration of 5 -7 days. Image acquisition was performed at 10x \nmagnification, and confluence was determined using the IncuCyte integrated analysis software. \nCell lysis and immunoblotting \nCells were lysed in RIPA buffer containing protease inhibitor cocktail (Roche, cat. no. \n11697498001) and centrifuged at 20,000g for 30 minutes at 4 °C. Protein concentration was \nnormalized by the bicinchoninic acid (BCA) assay (Thermo Fisher, cat. no. 23235), and samples \nwere heated at 70 or 90 °C for 10 minutes in the presence of SDS -containing loading buffer. \nProteins (20 µg per lane) were resolved on 12% SDS-PAGE gels and subsequently electroblotted \nonto nitrocellulose or PVDF membranes. The membranes we re blocked in non -fat milk 5% in \nTBT-T (20 mM Tris-HCl, 150 mM NaCl and 0.1% Tween-20) for 1 hour at RT. Subsequently were \nincubated overnight at 4 °C with primary antibodies diluted with 5% milk or bovine serum albumin \n(BSA) in TBS-T against FSP1 (1:10, a ntibody raised against recombinant human FSP1 protein, \nclone 6D8 -11, developed in Helmholtz Zentrum München), GPX4 (1:1,000, Abcam, cat. no. \nab125066 or 1:500, Proteintech, cat. no. 67763 -1-Ig), ACSL4 (1:500, Santa Cruz, cat. no. sc -\n271800), FLAD1 (1:250, Santa Cruz, cat. no. sc -376819), RFK (1:250, Santa Cruz, cat. no. sc -\n398830 or 1:500, Abbexa cat. no. abx124688), NQO1 (1:5,000, Santa Cruz, cat. no. sc -32793), \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nFlag-tag (1:1,000, Sigma-Aldrich, cat. no. F3165), β-actin (1:5,000, Sigma-Aldrich, cat. no. A5441) \nand vinculin (1:1,000, Santa Cruz, sc-73614). The next day, membranes were washed with TBS-\nT and incubated with horseradish peroxidase -labeled secondary ant ibodies (1:3,000, Cell \nSignaling, cat. no. 7074, 7076 and 7077) diluted in non -fat milk 5% in TBS -T for 2 hours at RT. \nFinally, membranes were washed, and antibody -antigen complexes were detected by \nchemiluminescence using enhanced chemiluminescence substr ate (BioRad, cat. no. 107 -5061) \non Amersham ImageQuant 800 (Cytiva). \nGenotyping \nFLAD1 and SLC52A2 single-clone knockout cells were confirmed through genotyping.  \nGenomic DNA from WT and knockout cell pellets was extracted by proteinase K digestion. \nSubsequently, the sgRNA -targeting regions were amplified by PCR using primers flanking the \nsgRNA cut -site and analyzed via Sanger sequencing to identify WT and knockout alleles \nsequences. Indels were characterized using CRISP-ID web-based application1. To further confirm \nthe precise editing events at the FLAD1 locus in the knockout cells, the PCR product was cloned \ninto a pJET1.2 cloning vector using CloneJET PCR Cloning Kit (Thermo Fisher, cat. no. K1231), \nfollowed by PCR amplification of 5 colonies and Sanger sequencing analysis. The sequences of \nprimers used are provided below.  \nFLAD1_Fw - CTCCACCCTTGCACTAGAGG \nFLAD1_Rv - CCCCTTAGAGTGAGCACAGC \nSLC52A2_Fw - GACTTCCTTGAGCGTTTTCCC \nSLC52A2_Rv - GGGGAACATAGCAACAGCG \nPreparation of riboflavin-deficient medium \nThe DMEM medium without riboflavin was purchased from PAN biotech (cat. no. P04-03584) and \nsupplemented with GlutaMax (Gibco, cat. no. 35050-061), 10% dialyzed fetal bovine serum (FBS, \nGibco, ref. A5256701), 1% penicillin-streptomycin (Gibco, cat. no. 15140122), 15 mg/L phenol red \n(Sigma, cat. no. P0290) and riboflavin (1 µM, unless noted otherwise) as needed for comparison. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nRiboflavin stock solution (20 mg/L) was freshly prepared by dissolving riboflavin (Sigma, cat. no. \nR9504) in PBS and sterilizing through a 0.22 µm membrane filter.  \nDialysis of Fetal Bovine Serum \nFetal bovine serum (FBS) was dialyzed to remove small molecules, including riboflavin, using \ntubing cellulose membranes (Sigma, cat. no. D9527) with a 14 kDa molecular weight cutoff. \nDialysis was performed against PBS at a 1:10 (v/v) ratio, with the buffer replaced every 24 h over \nfive consecutive cycles under constant magnetic stirring at 4 °C. The dialyzed serum was \nsubsequently filtered through a 0.22 µm membrane for sterility, aseptically transferred into sterile \ncontainers, aliquoted, and stored at -20 °C until further use. \nCell viability assays in riboflavin-deficient medium \nHT1080, A375, MDA -MB-231 and A549 parental cell lines were pre -incubated in riboflavin -\ndeficient medium for 48 hours, seeded at 1,000 cells/well in 96 -well plates and after 4 -6 hours \ntreated with RSL3 (0 -1000 nM) in the absence or presence of iFSP1 (3 µM) and/or Lip-1 (500 \nnM). Cell viability was assessed after 96 hours of treatment using the Alamar Blue method. \nTo evaluate the effect of roseoflavin, HT1080, A375, MDA -MB-231, A549, MDA-MB-436, H460, \nLOX-IMVI and PC -9 parental cell lines were cultured in low -riboflavin medium (20 nM). 3,000 \ncells/well were seeded in 96-well plates and after 4 – 6 hours pre-treated with roseoflavin (0, 1, 3 \nand 10 nM) for 48 hours. Following pre-treatment with roseoflavin, cells were treated with ML210 \n(0-10 µM) or iFSP1 (3 µM) for an additional 48 hours, after which cell viability was assessed using \nthe Alamar Blue method.  \nRiboflavin, FMN and FAD measurements  \nWater soluble metabolite measurements were made by liquid chromatography -mass \nspectrometry (LC/MS) analysis. For sample preparation, 1 million cells were harvested, washed \nwith PBS and rapidly frozen with liquid nitrogen.  \nWater-soluble metabolites were extracted with 500 µL of ice -cold MeOH/H2O (80/20, v/v) \ncontaining 0.01 μM lamivudine and 1 µM each of D2-glucose, D4-succinate, D5-glycine and 15N-\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nglutamate (Sigma-Aldrich). After centrifugation of the resulting supernatants were evaporated in \na rotary evaporator (Savant, Thermo Fisher Scientific, Waltham, USA). Dry sample extracts were \nredissolved in 150 μL 5 mM NH4OAc in CH3CN/H2O (50/50, v/v). 20 µL supernatant was \ntransferred to LC-vials. Metabolites were analyzed by LC-MS using the following settings: 3 μL of \neach sample was applied to a XBridge Premier BEH Amide (2.5 μm particles, 100 × 2.1 mm) \nUPLC-column (Waters, Dublin, Ireland). Metabolites were separated with Solvent A, consisting of \n5 mM NH4OAc in CH3CN/H2O (40/60, v/v) and solvent B consisting of 5 mM NH4OAc in \nCH3CN/H2O (95/5, v/v) at a flow rate of 200 µL/min at 45 °C by LC using a DIONEX Ultimate \n3000 UHPLC system (Thermo Fisher Scientific, Bremen, Germany). A linear gradient starting after \n2 min with 100 % solvent B decreasing to 10% solvent B within 23 min, followed by 16 min 10% \nsolvent B and a linear increase to 100% solvent B in 2 min was applied. Recalibration of the \ncolumn was achieved by a 7-minute pre-run with 100% solvent B before each injection.  \nAll MS analyses were performed on a high -resolution Q Exactive mass spectrometer equipped \nwith a HESI probe (Thermo Fisher Scientific, Bremen, Germany) in alternating positive - and \nnegative- full MS mode with a scan range of 69.0 -1000 m/z at 70K resolution  and the following \nESI source parameters: sheath gas: 30, auxiliary gas: 1, sweep gas: 0, aux gas heater \ntemperature: 120 °C, spray voltage: 3 kV, capillary temperature: 320 °C, S-lens RF level: 50. XIC \ngeneration and signal quantitation was performed usin g TraceFinder™ V 5.1 (Thermo Fisher \nScientific, Bremen, Germany) integrating peaks which corresponded to the calculated \nmonoisotopic metabolite masses (MIM +/ - H+ ± 3 mMU). All analyses were performed in three \nindependent biological replicates. \nRNA extraction, cDNA synthesis and RT-qPCR \nIsolation of total RNA from cell pellets was performed using TRIzol reagent (Invitrogen, cat. no. \n15596026) according to the manufacturer instructions. cDNA synthesis was done using HiScript \nIII 1st Strand cDNA Synthesis Kit (Vazyme, cat. no. R312) and hexamer primers in accordance to \nthe manufacturer’s protocol. RT -qPCR was performed and analysed with a Mastercycler ep \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nrealplex (Eppendorf) or CFX Connect (Biorad) using SYBR Green reagent. Gene expression was \nnormalized to ACTB as a housekeeping gene using the ∆∆ct method. \nThe qPCR primer pairs are indicated below.  \nhFSP1_1_Fw - GACTCCTTCCACCACAATGTGG \nhFSP1_1_Rv - CAGCACCATCTGGTTCTTCAGG  \nhFSP1_2_Fw - CCGCTATCCAGGCCTATGAG \nhFSP1_2_Rv - AATCTCTGCTGCCATCTCCA \nLipid peroxidation assays using C11-BODIPY (581/591) \n50,000 cells per well were seeded on 6-well plates one day prior to the experiment. The next day, \ncells were treated with the following conditions (i) DMSO, (ii) RSL3 200 nM and (iii) RSL3 200 nM \n+ Lip-1 500 nM for 6 hours. After treatment, cells were washed with PBS and incubated with C11-\nBODIPY (1 µM) for 30 minutes at 37 °C before they were harvested by trypsinization. \nSubsequently, cells were resuspended in 400 µL of PBS supplemented with 2% FBS followed by \nanalysis using a flow cytometer (FACS Canto II , BD Biosciences). Data was collected from the \nFITC detector (for the oxidized form of BODIPY) with a 502LP and 530/30 BP filter and from the \nPE detector (for the reduced form of BODIPY) with 556 LP and 585/42 BP filter. At least 10,000 \nevents were analyzed per sample. Data was analyzed using FlowJo Software. The ratio FITC/PE \n(oxidized/reduced ratio) was calculated as follows: (median FITC-A fluorescence – median FITC-\nA fluorescence of unstained samples)/(median PE-A fluorescence – median PE-A fluorescence of \nunstained samples). \nFor riboflavin deprivation experiments, cells were pre-incubated in riboflavin-deficient medium or \nmedium supplemented with riboflavin (1 µM) for 72 hours before undergoing the treatments \ndescribed above.  \nEpilipidomics analysis \n500,000 A375 parental cells were seeded on 15-cm dishes in a standard DMEM medium. After 4 \nhours, cells were divided into two conditions: (i) riboflavin-deficient medium and (ii) supplemented \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nwith riboflavin (1 µM) medium. Cells were maintained under these conditions for 72 hours. \nFollowing the 72 -hour incubation, cells were treated with one of the following conditions for 6 \nhours: (i) DMSO, (ii) RSL3 200 nM, and (iii) RSL3 200 nM + Lip -1 500 nM. After treatment, cells \nwere harvested and washed with PBS. The cell pellets were snap -frozen in liquid nitrogen and \nstored at -80 °C until further analysis.  \nLipids were extracted according to Folch protocol 2. All solvents contained 1 μg/mL butylated \nhydroxytoluene and were cooled on ice before use. Briefly, cell pellets (3 -5 x106 cells) collected \nin PBS were washed, centrifuged, and resuspended in 50 μL of H2O in 2 mL Eppendorf tubes. 5 \nµL of SPLASH® LIPIDOMIX® (Avanti Polar Lipids Inc., Alabaster, AL, USA) were added, tubes \nwere vortexed for 10 s and left on ice for 15 min. 365 μL ice-cold MeOH was added, the samples \nwere vortexed for 10 s, then 740 μL of ice-cold CHCl3 were added, vortexed for 10 s, and the \nsamples were incubated for 1 h in a rotary shaker (40 rpm) at 4 °C. Phase separation was induced \nby adding 225 μL of H2O, the samples were vortexed for 10 s and centrifuged at 2000g for 10 \nmin. 770 μL of the lower phase was transferred to 1.5 mL Eppendorf tubes and dried in a vacuum \nconcentrator at 40 mbar and 20 °C. Meanwhile, lipids from the remaining upper phase were re -\nextracted by adding 400 μL of CHCl3:MeOH 2:1 (v/v), vortexed for 10 s, 100 μL H2O was added \nto promote phase separation, vortexed for 10 s, centrifuged at 2000g for 10 min, and 380 μL of \nthe lower phase was transferred to the 1.5 mL Eppendorf tube containing the first extract portion \nand the sample was continued to dry in the vacuum concentrator.  \nThe dried lipid extracts were reconstituted in 125 μL of i-PrOH, centrifuged at 10000g for 5 min, \nand 120 μL were transferred into vials with glass inserts for LC -MS analysis. The upper phase \nremained after extraction was dried in the vacuum concentrator a t 20 mbar and 20 °C and used \nfor total protein quantification. Group-specific samples (gQC) were prepared by pooling 40 μL of \nthe corresponding individual samples, total quality control (tQC) samples were prepared by mixing \n40 μL of gQC samples. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nReversed phase liquid chromatography (RPLC) was carried out on a Vanquish Horizon (Thermo \nFisher Scientific, Bremen, Germany) equipped with an Accucore C30 column (150 x 2.1 mm; 2.6 \nµm, 150 Å, Thermo Fisher Scientific, Bremen, Germany). Lipids were separat ed by gradient \nelution with solvent A (MeOH/H2O, 1:1, v/v) and B (i -PrOH/MeCN/H2O, 85:15:5, v/v/v) both \ncontaining 5 mM HCOONH4 and 0.1% (v/v) HCOOH. Separation was performed at 50 °C with a \nflow rate of 0.3 mL/min using the following gradient: 0-20 min – 10 to 86 % B (curve 4), 20-22 min \n– 86 to 95 % B (curve 5), 22 -26 min – 95 % isocratic, 26 -26.1 min – 95 to 10 % B (curve 5) \nfollowed by 5 min re-equilibration at 10% B.  \nMass spectrometry was performed on Thermo Scientific Orbitrap Exploris 240 (Thermo Fisher \nScientific) equipped with a heated electrospray ionization (HESI) source with an EASY-IC unit for \nlock mass correction and operated with the following global HESI par ameters: sheath gas 40 \narbitrary units, auxiliary gas 10 arbitrary units, sweep gas 1 arbitrary units, spray voltage 3.5 kV \n(positive mode) or 2.5 kV (negative mode), ion transfer tube temperature 300 °C, vaporizer \ntemperature 370 °C, S-lens RF level 35%, EASY-IC lock mass correction was set to RunStart. \nFor identification of oxidized lipids, 6 μL of gQC samples were applied onto the LC column and \nMS data was recorded in ionization polarity switch mode (0-22.5 min – negative, 22.5-39.9 min – \npositive) with the instrument operating in semi-targeted data dependent acquisition (stDDA) mode \nwith 6 MS/MS scans per cycle. Full scans (MS1) had the following settings: Orbitrap resolution \n60000 at m/z 200, scan range m/z 500 -980 (negative mode) and 480 -1000 (positive mode), \nabsolute AGC value set to standard, maximu m injection time set to auto, 1 microscan. StDDA \nMS2 scans had the following settings: 1.5 m/z precursor selection isolation window, Orbitrap \nresolution 30000 at m/z 200, stepped higher-energy collisional dissociation (HCD) at normalized \ncollision energies (nCE) of 22-32-43% (negative mode) or 32 -43-54% (positive mode), absolute \nAGC value 1.105, maximum injection time 200 ms, 2 microscans. The following filters were \napplied prior to MS2 scans: dynamic exclusion after 5 times if occuring within 6 s, exclusio n \nduration 5 s, mass tolerance ±5 ppm, isotope exclusion; allowed charge state 1; targeted mass \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\ninclusion using the mass list of in silico predicted oxidized lipids (667 individual m/z values split \ninto 3 LC-MS runs), inclusion mass tolerance ±5 ppm. Data was acquired in profile mode.  \nOxidized lipids were identified using LPPtiger2 software (Fedorova Lab) 3,4. For relative \nquantification of oxidized lipids identified in stDDA runs, retention time-scheduled parallel reaction \nmonitoring (PRM) in ionization polarity switch mode (0 -22 min – negative, 22-40 min – positive) \nat the resolution of 15000 at m/z 200, absolute AGC value of 1.105 and a maximum injection time \nof 100 ms. The isolation window for precursor selection was 1.5 m/z. HCD nCE for every target \nwas chosen based on optimization PRM runs of gQC samples (fixed nCE from 15 to 40%). Data \nwas acquired in profile mode.  PRM data were processed in Skyline v. 24.1.0.199 (MacCoss Lab)5 \nconsidering fragment anions of oxidized fatty acyl chains as quantifier. The obtained peak areas \nwere normalized by appropriate lipid species from SPLASH® LIPIDOMIX® Mass Spec Standard \n(Avanti), e.g. by LPC(18:1(d7)), LPE(18:1(d7)), PC(15:0/18:1(d7)), or PE(15:0/18:1(d7)), and \nprotein concentration measured for the corresponding sample. Normalized peak areas were \nfurther log -transformed and autoscaled in MetaboAnalyst online platform \n(https://www.metaboanalyst.ca, Xia Lab)6. \nProteomic analysis \nA375 FLAD1KO clone 1 (C1) and FLAD1KO C1 Flag-FLAD1OE (addback) cells were seeded at \na density of 1 million cells on 10 -cm dishes. After 24 hours, cells were harvested, washed once \nwith PBS, snap-frozen in liquid nitrogen and stored at -80 °C until further processing.  \nIn a separate experiment, A375 parental cells were cultured in DMEM medium either deficient in \nriboflavin or supplemented with 1 µM of riboflavin. Cells were seeded at varying densities 500,000, \n250,000, 60,000 and 15,000 cells per 10-cm dish to collect at different time points: 24, 48, 96 and \n144 hours, respectively. At each time point, 1 million cells were harvested, washed with PBS, \nsnap-frozen in liquid nitrogen and stored at -80 °C until further analysis. \nFor proteomics analysis, cell pellets were lysed in RIPA buffer supplemented with protease and \nphosphatase inhibitor cocktails (Roche, cat. no. 11697498001 and 4906845001, respectively) and \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n1% (v/v) benzonase (Merck, cat. no. E1014). Lysates were sonicated with 3 cycles of 10 seconds \non/30 seconds off at 40% amplitude (Bioruptor Plus). Removal of residual cell debris was made \nby centrifugation at 17,000g for 2 hours at 4 °C. The supernatants containing soluble proteins \nwere collected, and protein concentrations were determined using the BCA assay (Thermo Fisher, \ncat. no. 23235). Protein concentrations were normalized to 1 µg/µL using water. Normalized \nprotein samples were aliquoted and stored at -80 °C until further analysis.  \nTryptic digestion of proteins (Input: 10 µg) was performed using an AssayMAP Bravo liquid \nhandling system (Agilent technologies) running the autoSP3 protocol according to Müller et al.7. \nThe resulting peptides were vacuum dried and stored at -20 °C until LC-MS/MS analysis. \nThe LC-MS/MS analysis was carried out on an Ultimate 3000 UPLC system directly connected to \nan Orbitrap Exploris 480 mass spectrometer (both Thermo Fisher Scientific) for a total of 120 min \ninjecting an equivalent of 1 µg of peptide. Peptides were online d esalted on a trapping cartridge \n(Acclaim PepMap300 C18, 5 µm, 300 Å wide pore; Thermo Fisher Scientific) for 3 min using 30 \nµl/min flow of 0.05% TFA in water. The analytical multistep gradient (300 nl/min) was performed \nusing a nanoEase MZ Peptide analytical column (300 Å, 1.7 µm, 75 µm x 200 mm, Waters) using \nsolvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in acetonitrile). For 102 min \nthe concentration of B was linearly ramped from 4% to 30%, followed by a quick ramp to 78%, \nafter two minutes the concentration of B was lowered to 2% and a 10 min equilibration step \nappended. Eluting peptides were analyzed in the mass spectrometer using data independent \nacquisition (DIA) mode. A full scan at 120k resolution (380 -1400 m/z, 300% AGC targe t, 45 ms \nmaxIT) was followed by 47 MS2 windows covering the mass range from 400 -1000 m/z with \nvariable width and overlapping by 1 Da (30k resolution, AGC target 1000%, maxIT 54 ms, 28% \nHCD collision energy). Each sample was followed by a wash run (40 min) to minimize carry-over \nbetween samples. Instrument performance throughout the course of the measurement was \nmonitored by regular (approx. one per 48 hours) injections of a standard sample and an in-house \nshiny application.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nAll sample handling (sample preparation and LC -MS/MS analysis) have been performed in a \nblock randomization order8. \nAnalysis of DIA RAW files was performed with Spectronaut (Biognosys, version \n19.1.240724.62635; Bruderer R. et al. (2015) 9 in directDIA+ (deep) library -free mode. Default \nsettings were applied with the following adaptions. Within DIA Analysis under Identification the \nPrecursor PEP Cutoff was set to 0.01, the Protein Qvalue Cutoff (Run) set to 0.01 and the Protein \nPEP Cutoff set to 0.01. In Quantification the Proteotypicity Filter was set to Only Protein Group \nSpecific, the Protein LFQ Method was set to MaxL FQ and the Quantification window was set to \nNot Synchronized (SN 17). The data was searched against the human proteome from Uniprot \n(human reference database with one protein sequence per gene, containing 20,597 unique \nentries from ninth of February 2024) and the contaminants FASTA from MaxQuant (246 unique \nentries from twenty-second of December 2022). Conditions were included in the setup. \nStatistical Analysis of Proteomics Data \nStatistical analysis was performed using MetaboAnalyst software10, with Spectronaut output files \nas input. After Spectronaut analysis, the data set comprised the identified proteins, along with \nLFQ intensities. The data were filtered to remove features mat ched with contaminant or reverse \nsequences. The missing values were replaced with LoDs (1/5 of minimum positive values of \ncorresponding variables) and data were log10 transformed and used for further statistical \nanalysis. The fold change threshold was taken to be 1.5, and RAW P value cutoff was 0.05 to get \nthe DEPs. Volcanoplot and heatmap representing the expression of top DEPs was also obtained \nfrom statistical analysis in MetaboAnalyst. Metascape tool 11 was used to obtain statistically \nenriched terms (GO/Kyoto Encyclopedia of Genes and Genomes pathway [KEGG] terms, \ncanonical pathways) for all DEPs. All analyses were performed in five independent biological \nreplicates. \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nDPPH Assay \nTo assess whether riboflavin and roseoflavin act as radical-trapping antioxidants, the compounds \nwere tested in a cell-free antioxidation assay. 10 mM compound was diluted in 1 mL 2,2-diphenyl-\n1-picrylhydrazyl (DPPH, 0.05 mM in methanol, Sigma -Aldrich) to a final concentration of 50 µM. \nSamples were rotated at room temperature for 10 minutes before they were transferred into a \nclear 96-well plate in quadruplicates. Absorbance was measured at 517 nm with an EnVision 2104 \nMultilabel plate reader (PerkinElmer). Experiment was performed in three biologic al replicates, \neach consisting of four technical replicates. DMSO was used as normalization control; Ferrostatin-\n1 was used as a positive control potent antioxidant.  \nMolecular dynamics simulations \nThe crystal structure of the human FSP1 complex (PDB ID: 8WIK) 12 was used as the initial \nstructure to prepare two systems: the FSP1-NAD complex and the FSP1-NAD-FAD complex. The \nstructure of FAD was obtained by substituting OH from C6 with H in the 6FA ligand present in \nPDB ID 8WIK. The protonation states of the residues at pH 7.4 were determined using the protein \npreparation wizard tool from Maestro (Schrödinger v.2024.2) 13,14. Ligand parameters for NAD+ \nwere obtained from the Manchester Amber parameter da tabase15. The partial charges of FAD \nwere determined using AmberTools2316 by the restrained electrostatic potential charges (RESP) \nmethod. A quantum mechanical calculation was performed to obtain the electrostatic potentials \nusing Gaussian 0917, Hartree Fock, the 6-31G* basis set and full optimi sation. The bond, angle, \ntorsion, and van der Waals parameters were generated using the general AMBER force field \n(GAFF)18 for co-factors, and the AMBER-ILDN force field19 for the protein.   \nThree independent simulations of each system, the FSP1-NAD complex and the FSP1-NAD-FAD \ncomplex, were carried out using GROMACS 2024.2 (Abraham et al. GROMACS 2024.2)20. Each \nsystem was solvated with TIP3P 21 water molecules with a margin of at least 10 Å, and Na+ and \nCl− ions were added to ensure system neutrality at an ion concentration of 150 mM. The steepest \ndescent method was used to perform energy minimization for 5000 steps on each system using \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\npositional restraints of 1000 kJ/mol/nm2 on the heavy atoms of the protein and cofactors. The \nsystems were then equilibrated with an NVT ensemble to achieve constant temperature at 300 K \nusing the velocity rescaling thermostat 22. Next, the systems were equilibrated to a constant \npressure of 1 bar using the cell rescaling barostat 23. Following equilibration, positional restraints \nwere gradually lifted from the heavy atoms of the protein and the cofactors (500,100,10 \nkJ/mol/nm2). Production runs of 500 ns were performed with a timestep of 2 fs. The temperature \ncoupling was achieved with the velocity rescaling at time constant of 0.1 ps and pressure coupling \nwas achieved with cell rescaling at compressibility of 4.5.10−5 bar−1 and a time constant of 5 ps. \nLINCS24 algorithm was used to constrain covalent H -bonds and SETTLE25 algorithm was used \nto constrain the solvent bond lengths. The particle mesh Ewald (PME) 26,27 method was used to \ncalculate the electrostatic forces with a real -space cutoff of 1.2 nm, PME orde r of four, and a \nFourier grid spacing of 1.2 Å. A cut-off of 1.2 nm was used for the calculation of the Van der Waals \ninteractions. Data analysis (root mean square deviation, root mean square fluctuation and DSSP \nanalysis) was performed using GROMACS 2024.2 (Abraham et al. GROMACS 2024.2)20. \nHuman cancer cell line datasets \nHuman cancer cell line datasets were obtained from the DepMap portal \n(https://depmap.org/portal/, version 23Q2). \nStatistical analysis  \nAll experiments (except those described otherwise in the legend) were performed independently \nat least twice. Graphs were generated using GraphPad Prism v10 (GraphPad Software) if not \nstated otherwise. \n  \nMethods References \n1 Dehairs, J., Talebi, A., Cherifi, Y. & Swinnen, J.V. CRISP-ID: decoding CRISPR mediated \nindels by Sanger sequencing. Sci Rep 6, 28973 (2016). \n2 Eggers, L.F. & Schwudke, D. Liquid Extraction: Folch. in Encyclopedia of Lipidomics (ed. \nWenk, M.R.) 1-6 (Springer Netherlands, Dordrecht, 2016). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\n3 Ni, Z., Angelidou, G., Hoffmann, R. & Fedorova, M. LPPtiger software for lipidome-specific \nprediction and identification of oxidized phospholipids from LC -MS datasets. Scientific \nReports 7, 15138 (2017). \n4 Criscuolo, A. et al. Analytical and computational workflow for in-depth analysis of oxidized \ncomplex lipids in blood plasma. Nat Commun 13, 6547 (2022). \n5 Adams, K.J. et al. Skyline for Small Molecules: A Unifying Software Package for \nQuantitative Metabolomics. J Proteome Res 19, 1447-1458 (2020). \n6 Chong, J., Wishart, D.S. & Xia, J. Using MetaboAnalyst 4.0 for Comprehensive and \nIntegrative Metabolomics Data Analysis. Curr Protoc Bioinformatics 68, e86 (2019). \n7 Muller, T. et al. Automated sample preparation with SP3 for low-input clinical proteomics. \nMol Syst Biol 16, e9111 (2020). \n8 Burger, B., Vaudel, M. & Barsnes, H. Importance of Block Randomization When Designing \nProteomics Experiments. J Proteome Res 20, 122-128 (2021). \n9 Bruderer, R. et al. Extending the limits of quantitative proteome profiling with data -\nindependent acquisition and application to acetaminophen-treated three-dimensional liver \nmicrotissues. Mol Cell Proteomics 14, 1400-10 (2015). \n10 Pang, Z. et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional \ninsights. Nucleic Acids Res 49, W388-W396 (2021). \n11 Zhou, Y. et al. Metascape provides a biologist -oriented resource for the analysis of \nsystems-level datasets. Nat Commun 10, 1523 (2019). \n12 Feng, S. et al. The crystal structure of human ferroptosis suppressive protein 1 in complex \nwith flavin adenine dinucleotide and nicotinamide adenine nucleotide.  MedComm (2020) \n5, e479 (2024). \n13 Sastry, G.M., Adzhigirey, M., Day, T., Annabhimoju, R. & Sherman, W. Protein and ligand \npreparation: parameters, protocols, and influence on virtual screening enrichments. J \nComput Aided Mol Des 27, 221-34 (2013). \n14 Schrödinger, L. Schrödinger Release 2024 -2: Protein Preparation Wizard. (Schrödinger, \nLLC, New York, NY, 2024). \n15 Ryde, U. On the role of Glu-68 in alcohol dehydrogenase. Protein Sci 4, 1124-32 (1995). \n16 Case, D.A. et al. AmberTools. J Chem Inf Model 63, 6183-6191 (2023). \n17 M. J. Frisch, G.W.T., H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, G. \nScalmani, V. Barone, G. A. Petersson, H. Nakatsuji, X. Li, M. Caricato, A. Marenich, J. \nBloino, B. G. Janesko, R. Gomperts, B. Mennucci, H. P. Hratchian, J. V. Ortiz, A. F. \nIzmaylov, J. L. Sonnenberg, D. Williams -Young, F. Ding, F. Lipparini, F. Egidi, J. Goings, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint \n\nB. Peng, A. Petrone, T. Henderson, D. Ranasinghe, V. G. Zakrzewski, J. Gao, N. Rega, \nG. Zheng, W. Liang, M. Hada, M. Ehara, K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, \nT. Nakajima, Y. Honda, O. Kitao, H. Nakai, T. Vreven, K. Throssell, J. A. Montgomery, Jr., \nJ. E. Peralta, F. Ogliaro, M. Bearpark, J. J. Heyd, E. Brothers, K. N. Kudin, V. N. \nStaroverov, T. Keith, R. Kobayashi, J. Normand, K. Raghavachari, A. Rendell, J. C. Burant, \nS. S. Iyengar, J. Tomasi, M. Cossi, J. M. Millam, M. Klene, C. Adamo, R. Cam mi, J. W. \nOchterski, R. L. Martin, K. Morokuma, O. Farkas, J. B. Foresman, and D. J. Fox. Gaussian \n09, Revision A.02. (Gaussian, Inc., Wallingford, CT, 2016). \n18 Wang, J., Wolf, R.M., Caldwell, J.W., Kollman, P.A. & Case, D.A. Development and testing \nof a general amber force field. J Comput Chem 25, 1157-74 (2004). \n19 Lindorff-Larsen, K. et al. Improved side -chain torsion potentials for the Amber ff99SB \nprotein force field. Proteins 78, 1950-8 (2010). \n20 Abraham, M.J. et al. GROMACS: High performance molecular simulations through multi-\nlevel parallelism from laptops to supercomputers. SoftwareX 1-2, 19-25 (2015). \n21 Jorgensen, W.L., Chandrasekhar, J., Madura, J.D., Impey, R.W. & Klein, M.L. Comparison \nof simple potential functions for simulating liquid water. The Journal of Chemical Physics  \n79, 926-935 (1983). \n22 Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J \nChem Phys 126, 014101 (2007). \n23 Bernetti, M. & Bussi, G. Pressure control using stochastic cell rescaling. J Chem Phys  \n153, 114107 (2020). \n24 Hess, B., Bekker, H., Berendsen, H.J.C. & Fraaije, J.G.E.M. LINCS: A linear constraint \nsolver for molecular simulations. Journal of Computational Chemistry  18, 1463 -1472 \n(1997). \n25 Miyamoto, S. & Kollman, P.A. Settle: An analytical version of the SHAKE and RATTLE \nalgorithm for rigid water models. Journal of Computational Chemistry 13, 952-962 (1992). \n26 Darden, T., York, D. & Pedersen, L. Particle mesh Ewald: An N⋅log(N) method for Ewald \nsums in large systems. The Journal of Chemical Physics 98, 10089-10092 (1993). \n27 Cheatham, T.E., III, Miller, J.L., Fox, T., Darden, T.A. & Kollman, P.A. Molecular Dynamics \nSimulations on Solvated Biomolecular Systems: The Particle Mesh Ewald Method Leads \nto Stable Trajectories of DNA, RNA, and Proteins. Journal of the American Chemical \nSociety 117, 4193-4194 (1995). \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 6, 2025. ; https://doi.org/10.1101/2025.08.05.668651doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}