Abstract

   CoQ[10] (Coenzyme Q[10]) is an essential fat-soluble metabolite that
   plays a key role in cellular metabolism. A less-known function of
   CoQ[10] is whether it may act as a plasma membrane-stabilizing agent
   and whether this property can affect cancer development and
   progression. Here, we show that CoQ[10] and its biosynthetic enzyme
   UBIAD1 play a critical role in plasmamembrane mechanical properties
   that are of interest for breast cancer (BC) progression and treatment.
   CoQ[10] and UBIAD1 increase membrane fluidity leading to increased cell
   stiffness in BC. Furthermore, CoQ[10] and UBIAD1 states impair ECM
   (extracellular matrix)-mediated oncogenic signaling and reduce
   ferroptosis resistance in BC settings. Analyses on human patients and
   mouse models reveal that UBIAD1 loss is associated with BC development
   and progression and UBIAD1 expression in BC limits CTCs (circulating
   tumor cells) survival and lung metastasis formation. Overall, this
   study reveals that CoQ[10] and UBIAD1 can be further investigated to
   develop therapeutic interventions to treat BC patients with poor
   prognosis.

   Subject terms: Breast cancer, Mechanisms of disease, Cell death
     __________________________________________________________________

   The roles of CoQ10 and producing enzymes in cancer have not been well
   explored. Here, the authors identify that the CoQ10-biosynthetic enzyme
   UBIAD1 is a tumor suppressor in breast cancer. CoQ10 induced
   alterations in plasma membrane mechanical properties lead to increased
   cellular stiffness and impaired tumorigenic signaling, as well as
   enhanced sensitivity to ferroptosis.

Introduction

   CoQ[10] (Coenzyme Q10, also known as ubiquinone) represents an
   essential fat-soluble compound present in the plasma and inner
   membranes of most eukaryotic cells^[50]1,[51]2. CoQ[10] consists of a
   benzoquinone ring and a lipophilic isoprenoid side chain. CoQ[10] plays
   a role in several aspects of cellular metabolism, including
   mitochondrial respiration and shielding membrane lipids from oxidative
   stress^[52]3. A less-known function of CoQ[10] is its capacity to act
   as a membrane-stabilizing agent that alters the mechanical properties
   of phospholipid membranes. Indeed, the role of CoQ[10] in plasma
   membrane homeostasis influences cholesterol-poor membranes such as in
   mitochondria and aerobic gram-negative bacteria^[53]4,[54]5. However,
   whether increased plasma membrane CoQ[10] levels affect
   cholesterol-dependent signaling in cholesterol-rich membranes and
   pathological settings is poorly understood. Such CoQ[10] studies would
   be critical also from a clinical point of view since genetic mutations,
   disease, and aging can affect Q[10] levels in the cells and
   consequently result in serious health issues. How and whether CoQ[10]
   contents can affect cancer development and progression has thus far
   been poorly investigated^[55]6,[56]7.

   The biosynthesis of CoQ[10] involves distinct steps: production of the
   aromatic group that forms the quinone head, production of the isoprene
   tail, attachment of the quinone head to the isoprene tail, followed
   with slight biochemical modifications that lead to the formation of the
   final CoQ[10] product^[57]8,[58]9. Several metabolic pathways
   contribute to CoQ[10] biosynthesis by providing precursors that
   generate CoQ[10] molecules: the quinone head derives from tyrosine
   catabolism in higher eukaryotic cells, and the isoprene tail derives
   from acetyl-CoA in the mevalonate pathway^[59]10. There are different
   rate-limiting steps to synthesizing CoQ[10]. The first one is the
   availability of isopentenyl pyrophosphate (IPP) since the production of
   the quinone head comes from an abundant source of tyrosine, and the
   second is the condensation of the isoprenoid tail to the quinone group
   mediated by UbiA enzymes^[60]11. In mammals, two UbiA enzymes act as
   polyprenyl-4-hydroxybenzoate transferase, called COQ2 and UBIAD1,
   located in the mitochondria and Golgi/ER apparatus. Indeed,
   mitochondria and Golgi/ER apparatus are proposed sites for CoQ[10]
   synthesis in mammals^[61]12,[62]13.

   Ferroptosis is an iron-dependent form of programmed cell death mediated
   by the accumulation of lipid reactive oxygen species (ROS), inducing
   oxidative damage and disrupting antioxidant defense balance^[63]14.
   Cancer cells are more prone to undergo ferroptosis, as they have a more
   vigorous metabolism and accumulate more ROS in comparison to normal
   cells^[64]15. Recent studies reveal that ferroptosis induction holds
   great promise for cancer therapy, especially for therapeutic-resistant
   tumors or those with high metastasizing tendencies^[65]15. However,
   cancer cells can also achieve lower susceptibility to ferroptosis
   through additional mechanisms like the glutathione (GSH) system,
   CoQ[10]system, and thioredoxin (TXN) system detoxify lipid
   hydroperoxides and suppress ferroptosis^[66]16. Induction of
   ferroptosis is an explorable therapeutic strategy as triple-negative
   breast cancer (TNBC) is an iron-rich and lipid-rich
   tumor^[67]17,[68]18. Recently it has been shown that TNBCs display
   ferroptosis heterogeneity, and the luminal androgen receptor (LAR)
   subtype of TNBC shows hypersensitive to ferroptosis inducers proposing
   that the combination of GPX4 inhibitors and immune checkpoint
   inhibitors could represent a treatment strategy for LAR TNBCs^[69]19.
   Resistance to ferroptosis is also associated with the development of
   some refractory cancers. Ferroptosis is a process regulated by multiple
   metabolic pathways, but a clear picture of ferroptosis in TNBC remains
   poorly characterized. In this context we investigate whether UBIAD1 and
   CoQ[10] affect the survival and aggressiveness of the TNBC subtype that
   strongly relies on survival and invasion signaling and ferroptosis
   protection for the progression of the disease. Overall, we wonder
   whether UBIAD1 and CoQ[10] dysfunction cause the acquisition of
   aggressive phenotypes in BC being the function of the mevalonate
   pathway intermediates in BC far from being completely understood.

   In this work, we discover a function for CoQ[10] and UBIAD1 in
   controlling membrane fluidity properties that is associated with a
   reduced metastatic potential and ferroptosis resistance in BC settings
   and patients.

Results

CoQ[10] regulates membrane fluidity and cell stiffness in BC

   CoQ[10] (or ubiquinone) is a potent lipid-soluble antioxidant enriched
   not only in mitochondria but also in plasma membranes^[70]2,[71]20.
   Whether plasma membrane (PM) CoQ[10] levels might have
   redox-independent functions such as altering cell membrane lipid
   homeostasis and acting as a membrane-stabilizing agent in pathological
   settings remains not characterized. To this purpose, we treated the
   TNBC cell lines MDA-MB-231 with different formulations of CoQ[10]
   (CoQ[10] and CoQ[10]H[2]) and we evaluated changes in PM
   cholesterol-rich lipid rafts, microdomains that are known to regulate
   signal transduction and spatial organization of the PM. CoQ[10]
   treatment of TNBC cells leads to a significant increase of CoQ[10]
   contents in the membrane of these cells (Supplementary Fig. [72]1a, b)
   without affecting overall survival. By using the cholera toxin
   B-subunit staining we detected a decrease in lipid rafts microdomains
   upon CoQ[10] treatments (Fig. [73]1a, b). These effects are independent
   of the redox state of CoQ[10] and are evident also in a different TNBC
   cell line such as MDA-MB-436 (Supplementary Fig. [74]1c).

Fig. 1. CoQ[10] alters BC cell stiffness and impairs AKT2 activation.

   [75]Fig. 1
   [76]Open in a new tab

   a, b Representative images of lipid rafts (CT-B) in MDA-MB-231 cells
   treated with 20 μM CoQ[10] (a) or 20 μM CoQ[10]H[2] (b). Scale bar,
   25 μm. DAPI and CT-B/lipid rafts. Integrated density normalized to cell
   number, expressed as fold change over Di-Methyl-Formamide (a) or
   Placebo (b). Mean ± SEM from n = 5 for (a); n = 4 for b independent
   experiments, analyzed with a two-tailed t-test. c Stiffness of
   MDA-MB-231 cell after 20 μM CoQ[10] or CoQ[10]H[2] treatment. Each dot
   is the average of 3 measurements per cell from 2 independent
   experiments (n = 3 for Not-Treated, n = 8 for CoQ[10] and CoQ[10]H[2]
   at 7 days, n = 7 for others). P-values from two-tailed t-test. d
   Western blot for AKT1, AKT2, ERK1/2, and GSK3β phosphorylation to
   assess PI3K-AKT, MEK/ERK, and Wnt pathway activation. VINCULIN as
   control. Data from independent replicates (n = 3). e Chemical
   structures of CoQ compounds and antioxidants, including
   N-Acetyl-L-Cysteine (NAC) and Idebenone. f Stiffness of MDA-MB-231 cell
   after 24-h treatment with 20 μM CoQ[10], CoQ[6], CoQ[2], Idebenone, or
   10 mM NAC. Vehicles: Di-Methyl-Formamide or H[2]O. Each dot is the
   average of 3 measurements per cell from 3 independent experiments
   (n = 6 for H[2]O, n = 15 for CoQ[2], n = 14 for others). Mean ± SEM,
   adjusted p-values from one-way ANOVA and Tukey’s test. g Stiffness in
   MCF10A and breast cancer cell lines after 1 week of 20 μM CoQ[10]
   treatment. Each dot is the average of 3 measurements per cell from 2
   independent experiments (n = 8 for MDA-MB-231 CoQ[10], n = 7 for
   others). Mean ± SEM, p-values from two-tailed t-test. h Stiffness of
   MDA-MB-231 cells treated with 20 μM CoQ[10], ROCK inhibitor Y27632
   (20 μM), MLCK inhibitor ML7 (20 μM), or their combination. Each dot is
   the average of 3 measurements per cell from 2 independent experiments
   (n = 7). Mean ± SEM, adjusted p-values from one-way ANOVA and Tukey’s
   test. i Invasion assay of MDA-MB-231 cells through Matrigel-coated
   membranes, treated for 24 h with Di-Methyl-Formamide, 20 μM CoQ[10],
   Y27632 (20 μM) + ML7 (20 μM), or their combination. Data are percentage
   of invading cells relative to Vehicle (100%). Each dot is the average
   of two wells from 3 independent experiments. Mean ± SEM, adjusted
   p-values from one-way ANOVA and Tukey’s test. Source data are provided
   in Source data file.

   It has been reported that plasma membrane cholesterol depletion rises
   cellular stiffness by increasing cortical rigidity^[77]21–[78]23. To
   examine whether increased membrane CoQ[10] level influences BC cell
   mechanical properties, we directly measured single-cell cortical
   stiffness using Atomic Force Microscopy (AFM)^[79]24. We found that
   CoQ[10]-supplemented BC cells exhibited higher cortical stiffness,
   compared with the untreated or control cells (Fig. [80]1c). This is
   true for both formulations of CoQ[10] and independent of the timing of
   treatments.

   Next, we evaluated whether such alteration of PM homeostasis induced by
   CoQ[10] treatment may lead to changes in the way how cells regulate
   signal transduction^[81]25. Different signaling pathways related to BC
   development were analyzed (Fig. [82]1d). While most of them remained
   unchanged, specific inhibition of the PI3K/AKT pathway has been
   detected as the impairment of the phosphorylation status of the AKT2
   isoform. Interestingly, this specific isoform has been previously
   associated with worse breast cancer patient outcomes^[83]26.
   Considering the biochemical properties of membrane-bound CoQ[10], these
   data suggest that treatment with CoQ[10] affects PM homeostasis and
   PI3K/AKT oncogenic signaling in BC cells.

   To support our hypothesis regarding the unique dual functionality of
   CoQ[10]—combining antioxidant properties through the redox cycling of
   its quinone ring and mechano-properties associated with the PM through
   its polyprenyl tail (isoprenoid moiety)—we conducted a comprehensive
   investigation utilizing AFM. Our experimental approach involved the
   examination of diverse molecules, including structural analogs of
   CoQ[10] featuring a shortened isoprenoid moiety, such as CoQ[6] and
   CoQ[2], alongside counterparts possessing antioxidant function but
   lacking the structural features of CoQ[10], such as idebenone and NAC.

   This analysis aims to contribute valuable insights into the
   multifaceted roles of CoQ[10], providing a better understanding of its
   potential applications in both antioxidant and mechano-property
   domains. Interestingly, we observed that antioxidant compounds such as
   idebenone and NAC do not alter cellular stiffness. CoQ[2] which
   contains a short side chain of 2 isoprenoid units also does not alter
   cell stiffness, while CoQ[6] which contains a side chain of 6
   isoprenoid units significantly promotes stiffness although not at the
   same levels of CoQ[10] (Fig. [84]1e, f). Therefore, these data showed
   that it is conceivable to separate the lipid antioxidant role of
   CoQ[10](linked to the redox ubiquinone ring) from its plasma membrane
   mechanical role (linked to the polyprenyl tail).

   Next, we analyzed whether CoQ[10] can equally exert its
   mechano-properties in different BC cell lines (Fig. [85]1g). CoQ[10]
   does increase stiffness in MCF7, MDA-MB-468, and MDA-MB-231, but not in
   non-tumorigenic MCF10A suggesting that the effect of CoQ[10] is
   possibly related to the tumorigenic state of breast cancer cells or
   rather a different PM composition. To further confirm the ability of
   CoQ[10] to regulate cellular stiffness, we treated MDA-MB-231 cells
   with CoQ[10] in combination with the ROCK inhibitor Y27632 and the MLCK
   inhibitor ML7 known to make BC softer^[86]27 (Fig. [87]1h). Here we
   show that Y27632 + ML7 inhibitors promote softness in MDA-MB-231 cells
   as expected. However, in presence of CoQ[10], the two inhibitors fail
   to promote cellular softness and the cells present the same stiffness
   as with CoQ[10] treatment alone.

   Moreover, we examined whether CoQ[10] could interfere with BC
   proliferation and/or alter invasion properties (Fig. [88]1i and
   Supplementary Fig. [89]1d). CoQ[10] treatment does not alter cell
   proliferation of MDA-MB-231, while significantly impairs their ability
   to invade Matrigel-coated membranes. On the other hand, in the presence
   of the ROCK inhibitor Y27632 and the MLCK inhibitor ML7, cell
   invasiveness increases, but such effect is significantly reduced in the
   presence of CoQ[10] (Fig. [90]1i).

   Finally, we tested whether CoQ[10] treatment could alter cellular
   morphology by examining structures containing Paxillin and FAK proteins
   as crucial molecular components that play a central role in connecting
   ECM signaling with actin filaments. They regulate the dynamics of the
   cytoskeleton and cellular morphology during cell invasion^[91]28.
   MDA-MB-231 cells treated with CoQ[10] show different cellular
   protrusions compared to controls. Notably, these alterations manifest
   as multiple blebs at the plasma membrane where FAK and Paxillin are
   localized. In contrast, control cells exhibit a more elongated
   morphology with visible lamellipodia, characteristic of
   mesenchymal-like cells (Supplementary Fig. [92]1e–h). To confirm that
   CoQ[10] acts by altering cortical contractility, we examined
   cytoskeletal changes in cells that have been treated with the ROCK
   inhibitor Y27632 and the MLCK inhibitor ML7 (Supplementary
   Fig. [93]1i–k). Y27632 + ML7-treated cells showed pronounced
   lamellipodia but no blebs and, interestingly, such morphological
   features were significantly rescued by the co-treatment with CoQ[10],
   confirming that CoQ[10] acts by altering cortical contractility.

   Altogether, these findings indicate an unforeseen role for CoQ[10] in
   regulating PM mechano-properties, cellular morphology and PI3K/AKT
   signaling pathways in BC cells.

The CoQ[10] biosynthetic enzyme, UBIAD1, is associated with breast cancer
(BC) patient outcome

   Membrane-bound CoQ[10] is synthesized by the UBIAD1 enzyme that is
   specifically localized in the Golgi compartment^[94]29,[95]30.
   Therefore, we investigate the function of the UBIAD1 enzyme during BC
   tumorigenesis. At first, we analyzed the expression of the CoQ[10]
   biosynthetic enzyme UBIAD1 in human patients with BC. We interrogated
   clinical and genomic data of ~2000 BC patients with well-annotated
   clinicopathological information and long-term follow-up, comprised in
   the METABRIC (Molecular Taxonomy of Breast Cancer Consortium)
   dataset^[96]31. Copy number alterations (CNA) analysis of this database
   revealed that, compared to the normal tissue, breast tumors show
   frequent copy number aberrations at the level of the UBIAD1 locus,
   consisting in most cases in deep or shallow deletions, with only a
   minority of cases (~1%) showing a copy gain of UBIAD1 (Fig. [97]2a). In
   the analysis of the METABRIC cohort, UBIAD1 deletion appears to be
   highly significantly enriched in Luminal B, HER2+ and Basal vs. Luminal
   A patients (Fig. [98]2b). To confirm the possible alterations of UBIAD1
   expression in BC patients, we performed immunohistochemistry (IHC)
   analyses on a large retrospective consecutive cohort of ~2000 BC
   patients with long-term complete follow-up (median 14.1 years),
   available in a tissue microarray^[99]32,[100]33. The availability of
   matched normal mammary tissue samples for a subset of patients also
   enabled us to compare UBIAD1 expression in normal vs. tumor tissue.
   UBIAD1 appeared uniformly and proficiently expressed in the normal
   mammary gland, with comparatively higher expression levels in the
   luminal compared to the basal epithelial layer, arguing for a
   functional role of UBIAD1 in homeostatic conditions (Fig. [101]2c).
   This observation strikingly contrasted with the heterogeneous pattern
   of UBIAD1 staining across breast tumors, with a vast majority of tumors
   displaying a remarkable downregulation or even total loss of UBIAD1
   expression, compared to the normal tissue (Fig. [102]2c). Then, we
   categorized patients of the cohort in three groups based on UBIAD1
   protein levels in the primary tumor (see also Materials and Methods):
   UBIAD1[High], tumors with UBIAD1 protein levels undistinguishable from
   normal (9.2% of patients); UBIAD1[Intemediate], tumors with residual
   UBIAD1 expression (67.0% of patients); UBIAD1[Low], tumors with almost
   complete or total loss of UBIAD1 (23.8% of patients). Remarkably, we
   found clinical significance, with UBIAD1[Low] BC patients showing a
   significantly shorter overall survival (OS) compared to UBIAD1[High]
   patients (HR[High vs. Low] = 0.57, CI, 0.34–0.94; P = 0.03)
   (Fig. [103]2d). The behavior of UBIAD1 as a marker of poor prognosis
   was independent of all the standard clinicopathological parameters, as
   assessed in a multivariable analysis adjusted for estrogen and
   progesterone receptor expression, nodal status, tumor size,
   proliferative index, age, and grade. We also found that UBIAD1[Low]
   tumors, compared to UBIAD1[High] tumors, are associated with a more
   malignant clinicopathological status, as witnessed by an increase in
   tumor size, poorer degree of differentiation, loss of estrogen and
   progesterone receptor, and higher proliferation rate (Supplementary
   Table [104]1). While no significant associations were detected for
   nodal status, HER2 expression and Age (Supplementary Table [105]2).
   Consistently, UBIAD1[Low] tumors also significantly correlated with
   triple-negative BC (TNBC) (OR[Low vs. High] = 4.08, p = 0.0012), which
   represents the most biologically and clinically aggressive BC subtype
   (Fig. [106]2e). Similar evidence was obtained for basal tumors when
   examining the UBIAD1 mRNA expression levels in the METABRIC cohort
   (Supplementary Table [107]3).

Fig. 2. Loss of the CoQ[10]-biosynthetic enzyme UBIAD1 is associated to a
worse prognosis in BC patients.

   [108]Fig. 2
   [109]Open in a new tab

   a Distribution of UBIAD1 copy number alterations (CNA) in the METABRIC
   cohort of breast cancer patients (see Methods). Percentages of patients
   with no CNA (Normal), amplifications (Amplified), or deep and shallow
   deletions (Deleted) are indicated. b Association between UBIAD1 status
   (Normal vs. Deleted) and PAM50 BC subtypes in the METABRIC cohort.
   HER2 +, HER2-positive. OR, odds ratio; CI, confidence intervals.
   P-value from two-sided Fisher’s exact test, with Luminal A subtype as
   the reference (ref.). The number and percentage of patients in each
   group are shown. c Representative images showing heterogeneous UBIAD1
   expression in a cohort 1767 breast samples available as TMA (1717 tumor
   plus 50 normal mammary gland samples) by IHC. Red and black arrowheads:
   luminal and basal cells, respectively. Tumors are classified as High,
   Intermediate, or Low based on UBIAD1 intensity scores. Scale bars,
   100 µm. Quantitative analysis of UBIAD1 expression by IHC in normal and
   tumor tissues, with the number and percentage of samples in each UBIAD1
   category (High, Intermediate, Low) shown. d Kaplan-Meier analysis of
   overall survival in the TMA cohort, stratified by UBIAD1 status as in
   (c). *HR, multivariate hazard ratio for UBIAD1High vs. UBIAD1Low
   groups. HR is calculated with 95% CI using multivariate Cox
   proportional hazards regression, adjusted for ER/PGR, pT, pN, Ki-67,
   HER2, age, and grade. CI and P (p-value) are indicated. P-value from
   two-sided Wald test. e Association between UBIAD1 status (High, Low)
   and BC molecular subtypes. HER2 +, HER2-positive; TNBC,
   triple-negative. OR, odds ratio; CI, confidence intervals. P-value from
   Fisher’s exact test with Luminal A subtype as reference (ref.). The
   number and percentage of patients in each group are shown. P-value from
   two-sided Fisher’s exact test. f Western blot analysis of UBIAD1
   protein expression in human BC cell lines, grouped by molecular
   subtype. HER2+, HER2-positive; TN, triple-negative. Densitometric
   analysis of UBIAD1 protein expression with each dot representing n = 4
   independent seedings and treatments. Data normalized over ACTIN and
   expressed relative to the non-tumorigenic breast line MCF10A.
   Mean ± SEM. P-value calculated using one-sample (two-tailed) t-test to
   compare the mean of each sample with 1.0. ns, not significant. Source
   data are provided as a Source data file.

   In agreement with patients, we obtained similar evidence by analyzing
   UBIAD1 expression both at mRNA and protein levels in a panel of human
   BC cell lines. We found that UBIAD1 expression is reduced particularly
   in TNBC lines (Fig. [110]2f and Supplementary Fig. [111]2a), especially
   in the most aggressive MDA-MB-231, MDA-MB-436, and Hs578T lines.
   Interestingly, also the levels of total CoQ[10] are significantly
   higher in MCF10A cells compared to other tumorigenic cells lines
   (Supplementary Fig. [112]2b).

   To support the role of UBIAD1 in breast tissues, we evaluated the
   estrogen-mediated regulation of UBIAD1. Thus, we treated with the
   Estrogen receptor-inhibitor 4-OH-Tamoxifen (4OHT) three ERα-positive
   lines (Supplementary Fig. [113]2c) and we evaluated the changes in the
   expression of the ERα-targets genes such as BCL2, MYC and CCND1 and
   UBIAD1. We detected a reduction in the expression of all these
   ER-target genes as well as of UBIAD1 (Supplementary Fig. [114]2d, e).
   To confirm this regulation, we treated the ERα-positive lines with
   17β-Estradiol to activate the ER-mediated transcription. Accordingly,
   we detect an increase of UBIAD1 expression upon estrogen treatment
   (Supplementary Fig. [115]2f). These data show the regulation of UBIAD1
   expression through ERα signaling and support the involvement of the
   CoQ[10]-producing enzyme UBIAD1 in BC.

   In summary, we propose a function of the CoQ[10]-biosynthetic enzyme
   UBIAD1 as a tumor suppressor gene in mammary tumorigenesis, with an
   intrinsic clinical value as an independent biomarker for the prognostic
   stratification of BC patients.

Ubiad1 genetic loss impairs CoQ[9] synthesis and promotes tumor development
in mouse BC models

   To validate the functional role of UBIAD1 and CoQ[10] in BC
   development, we generated a Ubiad1-deficient mouse model by disrupting
   exon1 through the insertion of a β-Gal/Neo Cassette (Supplementary
   Fig. [116]3a). Ubiad1^−/− mice were not viable (Supplementary
   Fig. [117]3b) starting to die around E10.5. Protein lysates from E9.5
   embryos showed a strong reduction in UBIAD1 protein level compared to
   both Ubiad1^+/− and Ubiad1^+/+ littermates at the same embryonic stage
   (Supplementary Fig. [118]3c). We observed that almost 40% of Ubiad1^+/−
   embryos at E10.5/E11.5 show a hemorrhagic phenotype with cranial and
   heart bleeding, while the rest of the embryos show normal appearance
   (Supplementary Fig. [119]3d). These results suggested that part of
   Ubiad1^+/− embryos die during embryogenesis due to cardiovascular
   failure and explain the reduced number of Ubiad1^+/− pups observed at
   birth. Similar results were previously obtained in the zebrafish
   model^[120]30. Next, we measured by MS the level of CoQ[9] in whole
   Ubiad1^+/− embryos and we found a reduced amount of CoQ[9] in
   Ubiad1^+/− embryos compared to Ubiad1^+/+ siblings (Supplementary
   Fig. [121]3e). To test whether CoQ[9] is the limiting factor
   responsible for the embryonic lethality of Ubiad1 KO mice, we injected
   CoQ[10] or Vitamin K[2] in Ubiad1^+/− pregnant females and analyzed the
   number of pups for each genotype at P0. We found that CoQ[10], but not
   Vitamin K[2], was able to increase the percentage of Ubiad1^+/− embryos
   at P0, while both failed to rescue embryonic lethality of Ubiad1^−/−
   genotype (Supplementary Fig. [122]3f). Moreover, to validate the
   contribution of enhanced oxidative stress in determining the
   cardiovascular system failure observed in CoQ[9]-deficient embryos, we
   showed that injection of the antioxidant NAC rescues the embryonic
   lethality at similar extent as CoQ[10] treatments (Supplementary
   Fig. [123]3f).

   Next, we crossed Ubiad1^+/− mice with two mouse genetic models of BC:
   (i) MMTV-NeuT, where tumor formation is driven by the overexpression of
   an active form of the HER2 oncogene^[124]34; (ii) MMTV-PyMT, in which
   the oncogenic polyomavirus middle T antigen (PyMT) expressed under the
   control of the mammary gland specific promoter MMTV, and drives the
   spontaneous development of multifocal mammary adenocarcinoma^[125]35.
   In both models, Ubiad1^+/− mice, compared to Ubiad1^+/+ controls,
   showed a significantly accelerated tumorigenesis, with a mean tumor
   formation interval of 31 weeks instead of 37 weeks for MMTV-NeuT mice
   and 15 weeks instead of 16.5 weeks for MMTV-PyMT mice (Fig. [126]2a, b
   and Supplementary Fig. [127]4a, b). Moreover, we found an accelerated
   growth rate in Ubiad1^+/− vs. Ubiad1^+/+ control mice by monitoring
   tumor growth kinetics after the tumors had reached a palpable size
   (Fig. [128]3c and Supplementary Fig. [129]4c). Accordingly, we detected
   an increased number of proliferating Ki67+ cells in MMTV-PyMT;
   Ubiad1^+/− tumors compared to controls (Supplementary Fig. [130]4d). At
   the sacrifice, we observed an increase in the number of primary tumors
   developed per mouse and the total tumor weight per mouse in Ubiad1^+/−
   compared to Ubiad1^+/+ control for both models (Fig. [131]3d, e, and
   Supplementary Fig. [132]4e, f). We then analyzed Ubiad1 expression in
   tumors spontaneously generated in MMTV-NeuT; Ubiad1^+/− and MMTV-PyMT;
   Ubiad1^+/− mice and we found a significant reduction of Ubiad1 mRNA
   expression in Ubiad1^+/−-derived tumors compared to controls,
   supporting the association between the loss of Ubiad1 and increased
   tumorigenicity (Fig. [133]3f and Supplementary Fig. [134]4g). We did
   also measure the level of CoQ[9] in these tumors showing a reduction in
   Ubiad1^+/− mice (Fig. [135]3g).

Fig. 3. Deletion of Ubiad1 gene accelerates BC development in mouse models.

   [136]Fig. 3
   [137]Open in a new tab

   a Generation of MMTV-NeuT BC model with exon 1 disruption of the Ubiad1
   gene. b Tumor-free survival from birth to tumor detection. Dashed lines
   show 95% Confidence Interval (CI). Group mean survival is reported.
   Analysis used the Gehan-Breslow-Wilcoxon test (1 degree-of-freedom). c
   Tumor growth from first palpable tumor (week 0). Data for n = 11
   MMTV-NeuT;Ubiad1^+/+ and n = 9 MMTV-NeuT;Ubiad1^+/− mice shown as
   mean ± SEM. Adjusted p-value from two-way ANOVA and Sidak’s test. d
   Tumors from two MMTV-NeuT;Ubiad1^+/+ and two MMTV-NeuT;Ubiad1^+/− mice.
   Number of primary tumors per mouse (n = 9 mice) shown as mean ± SEM.
   P-value from a two-tailed t-test. e Total tumor weight per n = 7
   mice/group shown as mean ± SEM. P-value from a two-tailed t-test. f
   Ubiad1 mRNA levels in n = 15 MMTV-NeuT;Ubiad1^+/+ and n = 10
   MMTV-NeuT;Ubiad1^+/− tumors. Mean ± SEM. P-value from a two-tailed
   t-test. g CoQ[9] levels in tumor lysates from n = 4 mice/group. Data
   shown as mean ± SEM. P-value from a two-tailed t-test. h CD31 staining
   in tumors. Each dot represents a tumor from n = 4 MMTV-NeuT;Ubiad1^+/+
   and n = 5 MMTV-NeuT;Ubiad1^+/− mice. Scale bar, 100 µm. Micro-vessel
   density and area shown as mean ± SEM. P-value from a two-tailed t-test.
   i CD45 staining in tumors. Each dot represents a tumor from n = 5
   mice/group. Scale bar, 100 µm. CD45+ cells shown as mean ± SEM. P-value
   from a two-tailed t-test. j Generation strategy for the KBP BC model
   with exon 2 deletion of Ubiad1 in basal mammary gland cells. k Whole
   mount mammary gland staining to visualize lesions in heterozygous
   (Ubiad1^f/+) or homozygous (Ubiad1^f/f) Ubiad1 knockout KBP females.
   Red dashed lines highlight lesions. Each dot represents a mouse (n = 6
   KBP;Ubiad1f/+ and n = 4 KBP;Ubiad1f/f). Data shown as mean ± SEM.
   P-value from a two-tailed t-test. LN, Lymph node. l Ubiad1 mRNA levels
   in n = 4 KBP, n = 5 KBP;Ubiad1^f/+, and n = 4 KBP;Ubiad1^f/f tumors.
   Mean ± SEM. Adjusted p-value from one-way ANOVA and Tukey’s test. m
   Kaplan-Meier analysis of survival in METABRIC BC patients by UBIAD1
   status (522 for UBIAD1 deleted and 1363 for UBIAD1 normal. *HR,
   multivariate hazard ratio adjusted for clinicopathological parameters.
   HR and 95% Confidence Interval from Cox regression after adjustment. CI
   and P-value indicated. P-value from two-sided Wald test without
   adjustments for multiple comparisons. Source data are provided in
   Source data file.

   Considering the essential role of UBIAD1 in endothelial cells^[138]30
   and the frequency of the hemorrhagic phenotypes observed in Ubiad1^+/−
   embryos, we investigated the presence of blood vessels inside the tumor
   by analyzing vessel density and mean vessel area (Fig. [139]3h and
   Supplementary Fig. [140]4h). The analysis revealed no significant
   differences in both parameters between Ubiad1^+/+ and Ubiad1^+/− mice.
   In parallel, we analyzed immune cell infiltration (CD45+ cells) in
   these tumors with no significant difference between Ubiad1^+/− vs
   Ubiad1^+/+ animals (Fig. [141]3i and Supplementary Fig. [142]4i)
   arguing in favor of a tumor cell-autonomous function of UBIAD1.

   To better check the tissue-specific function of Ubiad1 in breast cancer
   onset, we generated a flox allele model of Ubiad1^f/f mice, carrying
   the second exon of Ubiad1 gene flanked by two lox-P sites
   (Supplementary Fig. [143]4j). We then crossed these Tg animals with the
   BC model K14-Cre; Brca1^f/f; Tp53^f/f (Fig. [144]3j^[145]36); in which
   deletion of Tp53, Brca1, and Ubiad1 occurs concomitantly and
   specifically in the basal cells of the mammary gland. We verified the
   specific deletion of exon 2 only in the mammary gland and in the breast
   cancer cells of KBP; Ubiad1^f/+ and KBP; Ubiad1^f/f mice by PCR on the
   genomic DNA (Supplementary Fig. [146]4j, k). In line with our previous
   results, deep deletion of the Ubiad1 gene in KBP; Ubiad1^f/f mice leads
   to the development of larger neoplastic areas in the mammary glands,
   compared to KBP; Ubiad1^f/+ littermates (Fig. [147]3k). We then
   confirmed that Ubiad1 expression was absent in KBP; Ubiad1^f/f tumor
   compared to KBP; Ubiad1^+/+ and KBP; Ubiad1^Δ/+ conditions
   (Fig. [148]3l). Interestingly, using the METABRIC dataset we confirmed
   these findings in humans where deep deletion of UBIAD1 gene in BC
   patients is associated with reduced overall survival (Fig. [149]3m).
   Moreover, as previously demonstrated (Fig. [150]2), absence of UBIAD1
   is more frequent in TN and HER2 + BC subtypes, that are known to be the
   most clinically aggressive (Fig. [151]2e).

   In summary, these in vivo data provide genetic evidence that loss of
   the CoQ[9/10]-biosynthetic enzyme UBIAD1 cooperates with the
   oncogene-driven tumorigenesis to increase spontaneous breast tumor
   formation, pointing to a fundamental tumor suppressor function of
   UBIAD1 and possibly CoQ[9/10], in BC development.

UBIAD1 expression limits metastasis formation in aggressive BC lines

   To better characterize the role of CoQ[10] and UBIAD1 during BC
   development and progression, we took advantage of the metastatic human
   BC cell line MDA-MB-231 that showed a remarkable reduction of UBIAD1
   expression. We then (re)expressed UBIAD1 (UBIAD1^OE) in MDA-MB-231
   cells (Supplementary Fig. [152]5a) and measured the level of different
   UBIAD1-derived metabolites through HPLC (Supplementary Fig. [153]5b).
   CoQ[10] amount was increased while free and esterified cholesterol did
   not change. Vitamin K[2] was not detectable by MS in both control and
   UBIAD1^OE cells (data not shown). Interestingly, UBIAD1 expression
   stabilizes HMGCR (Supplementary Fig. [154]5c), the rate-limiting enzyme
   of the mevalonate pathway, suggesting a metabolic response due to the
   CoQ[10] increase. Then, we measured whether UBIAD1 expression might
   alter the tumorigenic properties of MDA-MB-231. While UBIAD1^OE cells
   do not show alteration in cell proliferation rate, they do show
   impairment in cell invasiveness compared to Scramble control cells
   (Supplementary Fig. [155]5d, e, respectively). Moreover, expression of
   UBIAD1 reduced the ability of MDA-MB-231 cells to grow and form
   colonies in anchorage-free conditions (Supplementary Fig. [156]5f).
   These results indicated a decreased cell ability to escape anoikis,
   which is an essential process also during the dissemination of
   circulating tumor cells (CTCs)^[157]37,[158]38. Consistently,
   restoration of UBIAD1 expression in MDA-MB-436 has similar effects
   (Supplementary Fig. [159]5g, h).

   Next, we tested the in vivo tumorigenic ability of UBIAD1-expressing
   MDA-MB-231 cells using xenografts assays in immunodeficient
   mice^[160]39. After labeling these cells with Luciferase-eGFP, we
   injected UBIAD1^OE or control cells in the fat pad of the fourth
   mammary gland of NSG mice (Fig. [161]4a, b). We measured tumor growth
   weekly, and we found a decreased growth rate of UBIAD1^OE compared to
   control cells (Fig. [162]4b). Moreover, we evaluated the number of
   circulating tumor cells (CTCs) and the metastasis formation at
   different sites (Fig. [163]4c–f). The number of CTCs sorted from the
   whole blood is dramatically reduced in mice injected intracardiacally
   with UBIAD1^OE cells compared to controls (Fig. [164]4d), suggesting a
   decreased ability of UBIAD1^OE cells to survive in the bloodstream.
   Moreover, the lung metastasis formed by UBIAD1^OE cells was reduced
   both by analyzing the whole organs (Fig. [165]4e) or sorting the GFP+
   cells in dissociated lungs (Fig. [166]4f). No significant differences
   were detected in the other whole organs analyzed. We confirmed through
   qPCR that UBIAD1 expression was maintained high in the UBIAD1^OE cells
   sorted from the primary tumor, blood, and different metastatic sites
   (Supplementary Fig. [167]5i). In line with our experimental evidence,
   we found that UBIAD1 expression is specifically turned off in
   MDA-MB-231-derived CTCs and lung metastasis but not in metastasis
   formed in other sites (i.e., brain and liver), supporting the idea that
   low level of UBIAD1 could be beneficial specifically for the survival
   of CTCs and the growth of pulmonary metastasis (Fig. [168]4g, h).

Fig. 4. UBIAD1 expression increases CoQ[10] level and reduces BC
dissemination.

   [169]Fig. 4
   [170]Open in a new tab

   a, b Tumor growth in NSG tumor-bearing mice engrafted in the mammary
   fat pad with Scramble and UBIAD1-overexpressing MDA-MB-231 cells. Data
   are mean ± SEM, with n = 8 mice per group from two independent
   experiments. Adjusted p-value from two-way ANOVA and Sidak’s test. c–f
   Scramble and UBIAD1-overexpressing MDA-MB-231 cells injected
   intracardially into NSG mice. d–f Data shown as box plots with median,
   25th, and 75th percentiles and with whiskers at minimum and maximum
   values. P-value from two-tailed unpaired t-test. d Percentage of
   Circulating Tumor Cells (CTCs) in whole blood. Each dot represents a
   mouse (n = 10 MDA-MB-231 Scr and n = 7 MDA-MB-231 UBIAD1^OE). Gating:
   eGFP+ cells from Single_cell_2 population. e eGFP signal of
   MDA-MB-231-derived metastasis in different organs. Each dot represents
   a mouse. f Percentage of eGFP+ cells in dissociated lungs. Each dot
   represents a mouse (n = 10 MDA-MB-231 Scr and n = 8 MDA-MB-231
   UBIAD1^OE). g, h UBIAD1 mRNA and protein expression in eGFP+ cells from
   whole blood, organs, or primary tumors of MDA-MB-231 Scr-injected mice.
   Each dot represents a single organ pooled from 3 independent
   experiments. Data are mean ± SEM. Adjusted p-value from one-way ANOVA
   and Sidak’s test. i–m Generation and characterization of Scramble
   (PyMT) and UBIAD1-overexpressing (PyMT^UBIAD1) ex vivo cultures from
   MMTV-PyMT-derived tumors. i UBIAD1 protein expression. j Total CoQ[9]
   in mouse ex vivo cultures. Each dot represents an independent seeding
   (n = 6). Data shown as mean ± SEM. P-value from two-tailed unpaired
   t-test. k Cholesterol esters (CE) and free cholesterol levels in mouse
   ex vivo cultures. Each dot represents an independent seeding (n = 6).
   Data shown as mean ± SEM. P-value from two-tailed unpaired t-test. l
   Invading mouse ex vivo PyMT and PyMT^UBIAD1 cells. Data expressed as
   percentage of invading cells compared to PyMT (set at 100%). Each dot
   represents an independent seeding (n = 3). Data shown as mean ± SEM.
   P-value from two-tailed unpaired t-test. m PyMT tumorspheres.
   Quantification of total tumorspheres and total area relative to PyMT
   cells. Each dot represents an independent seeding (n = 4 for number,
   n = 5 for area). Data shown as mean ± SEM. P-value from two-tailed
   unpaired t-test. Source data are provided in Source data file.

   To verify if our findings might be extended to other aggressive BC
   lines independently by UBIAD1 basal expression, we performed similar
   experiments using the JIMT-1 cell line, a HER2+ human breast cancer
   cell line with high UBIAD1 expression (Supplementary Fig. [171]5j, k).
   Thus, we overexpressed UBIAD1 in JIMT-1 and injected them in the fat
   pad of the fourth mammary gland of NSG mice or into the bloodstream
   through intracardiac injection (Supplementary Fig. [172]5l–r).
   Similarly, to MDA-MB-231, we found that increased UBIAD1 expression led
   to a decreased primary tumor growth (Supplementary Fig. [173]5m, n).
   Moreover, we found a reduction in the number of CTCs isolated from mice
   blood at 35 days post-intracardiac injection of JIMT-1 UBIAD1^OE cells
   and in the colonization of different distal sites, like lung, brain,
   and bone (Supplementary Fig. [174]5o–r)^[175]40.

   To validate the ability of UBIAD1 expression to impair BC
   tumorigenesis, we took advantage of MMTV-PyMT tumors. We isolated
   primary PyMT tumor cells and we (re)expressed UBIAD1 by lentiviral
   infections (Fig. [176]4i). We confirm that UBIAD1 expression increases
   CoQ[10] without affecting free cholesterol or CE levels (Fig. [177]4j,
   k). Then, we tested their invasion properties on Matrigel-coated
   membranes as well as their capacity to generate tumorspheres
   (Fig. [178]4l, m). In agreement with previous results in human BC cell
   lines, UBIAD1 expression in ex vivo PyMT cells impaired both their
   invasion properties and cancer stem cell-like features evaluated in the
   tumorsphere assay.

   Altogether, our data showed a role of the metabolic enzyme UBIAD1 in
   decreasing BC primary tumor growth and metastasis formation with
   specificities related to the metastatic site.

CoQ[10] and UBIAD1 expression impairs ECM-mediated signaling and AKT2 pathway
in BC cells

   To investigate the molecular mechanisms driven by UBIAD1 in BC, we
   performed RNA-seq analyses on control and UBIAD1^OE MDA-MB-231 cells
   (Fig. [179]5a–c, Supplementary Tables [180]4–[181]5, and Supplementary
   Fig. [182]6a). KEGG pathway and Gene Ontology (GO) enrichment analyses
   identified the PI3K-AKT pathway as one of the most affected oncogenic
   signaling pathways, and the Extracellular matrix organization as the
   most significant biological process affected. Several genes that have
   been pointed out as oncogenes in breast cancer were shown to be
   downregulated in UBIAD1^OE cells, i.e., LAMA5, IL6R, SRC, CHUK/IKKα,
   ITGA6, and LPAR1^[183]41–[184]46, and likely oncosuppressor genes were
   upregulated, i.e., CST3 and TINAGL1^[185]47,[186]48 (Fig. [187]5c and
   Supplementary Tables [188]4–[189]5, and Supplementary Fig. [190]6a).

Fig. 5. UBIAD1 expression impairs AKT oncogenic pathway and alter
cytoskeletal stiffness of BC cells.

   [191]Fig. 5
   [192]Open in a new tab

   a–c RNA-seq analysis of MDA-MB-231 Scramble (Scr) and
   UBIAD1-overexpressing (UBIAD1^OE) cells. a KEGG pathway and b Gene
   Ontology (GO) enrichment analysis on differentially expressed genes. c
   Downregulated and upregulated genes in UBIAD1^OE cells. Arrowheads:
   main genes listed in the GO analysis’ top rank. d Cell adhesion assay
   in MDA-MB-231. Data shown as ratio of attached cells over total seeded
   cells. Mean ± SEM of n = 10 for PLL, n = 6 for Collagen I, n = 7 for
   Vitronectin, n = 12 for Collagen IV, and n = 13 for Laminin independent
   seedings. P-value from two-tailed unpaired t-test. ns, not significant.
   e Phospho-FAK and phospho-AKT2 in laminin-attached cells. Densitometric
   analysis of phosphorylated protein over total protein. Each dot
   represents an independent seeding (n = 5). Data are mean ± SEM. P-value
   from two-tailed one-sample t-test comparing the mean of UBIAD1^OE with
   1.0. f Paxillin, DAPI, and Phalloidin staining. Arrows: cell adhesions
   or blebbing structures. Scale bar, 60 μm; crop, 30 μm. Results of three
   independent cell seedings, stainings, and analyses. g Lipid rafts
   (CT-B). Scale bar, 25 μm. Integrated density normalized over number of
   cells. Data as fold change over Scr. Each dot represents the average of
   five pictures for n = 5 independent seedings. P-value from two-tailed
   one-sample t-test comparing the mean of UBIAD1^OE with 1.0. h FRAP
   analysis on MDA-MB-231 cells. Data shown as relative fluorescence
   intensity over recovery time. Each dot represents a single cell (n = 25
   Scr and n = 27 UBIAD1^OE) analyzed across three independent seedings,
   stainings, and analyses. Adjusted p-value from two-way ANOVA and
   Sidak’s test. ns, not-significant. i Membrane fluidity analysis. EtOH
   as a positive control. Each dot represents an independent seeding
   (n = 5 for Scr and UBIAD1OE, and n = 3 for Scr+EtOH). Data are
   mean ± SEM. Adjusted p-value from one-way ANOVA and Tukey’s test. j
   Cellular stiffness. Each dot represents the average of 3 repeated
   measurements for a single cell (n = 6 cells/group). P-value from
   two-tailed unpaired t-test. k Percentage of invading cells compared to
   Vehicle-treated Scr (set at 100%). Scale bar, 100 µm.
   N-Acetyl-L-Cysteine (NAC). Each dot represents an independent seeding
   (n = 3). Data are mean ± SEM. Adjusted p-value from one-way ANOVA and
   Dunnett’s test. Source data are provided in Source Data file.

   The most significant downregulated gene is LM-511 (aka LAMA5). LM-511
   has been recently shown to play a crucial role in the luminal mammary
   epithelial cells highlighting its function in mammary gland remodeling
   and luminal differentiation^[193]49. LM-511 expression correlates with
   breast tumor grade and is linked to metastatic potential in
   experimental models making an interesting target of our study. Indeed,
   it has been proposed that LM511 contributes to the metastatic
   progression of breast cancer and its stemness in a mechano-dependent
   manner^[194]50,[195]51. Thus, we investigated the ability of
   MDA-MB-231^OE cells to bind different extracellular matrix components
   by the engagement of integrin receptors (Fig. [196]5d). Cell attachment
   assays revealed that compared to controls, UBIAD1^OE cells had a
   reduced capacity to bind to laminin-coated plates, while no differences
   were detected for other integrin-ligands, like Collagen I and IV,
   Vitronectin, or Fibronectin-coated surface.

   Laminin-binding integrins are recruited in BC cells to activate AKT
   through the activation of the Focal Adhesion Kinase (FAK)^[197]25.
   Thus, we led the cells to adhere on a laminin-coated surface to confirm
   that both FAK and AKT signaling were more active compared to cells in
   suspension (BSA treated-surface) or cells attached to a PLL-coated
   plate (Supplementary Fig. [198]6b). Given that, we compared the level
   of the activating phosphorylation Tyr397 on FAK and Ser474 on AKT2 in
   MDA-MB-231 vs UBIAD1^OE let adhere on a laminin-coated plate
   (Fig. [199]5e). As expected, UBIAD1 expression significantly decreased
   both FAK and AKT2 activation. We previously showed the effect of an
   extended CoQ[10] treatment on AKT pathway activation (Fig. [200]1d). We
   then decided to investigate AKT2 activation in MDA-MB-231 upon UBIAD1
   overexpression in standard culture conditions. Again, we detected a
   decrease in AKT2 phosphorylation in UBIAD1^OE cells, in line with a
   CoQ[10] overload (Supplementary Fig. [201]6c). We obtained similar
   results using PyMT^UBIAD1 ex vivo cell line being PyMT^UBIAD1 cells
   impaired in their ability to bind laminin (Supplementary Fig. [202]6d).

   To establish a functional role for AKT2 in this context, we
   overexpressed wild-type AKT2 or its constitutively active form,
   AKT2-Myr, in UBIAD1-expressing cells (Supplementary Fig. [203]6e). We
   found that AKT2 did not alter UBIAD1-overexpressing cell invasion
   ability, while the constitutively active form is able to restore the
   invasion properties of these cells. Next, we have analyzed the behavior
   of UBIAD1 mRNA levels in the METABRIC cohort upon conducting a subgroup
   analysis by stratifying patients according to the status of PIK3CA/PTEN
   alterations (Supplementary Fig. [204]6f). We observed a significantly
   worse prognosis associated with low UBIAD1 mRNA expression exclusively
   in the subgroup without PIK3^CA/PTEN alterations. These data lead to
   finding that the presence of constitutive active PI3K-AKT2 pathway
   overrides UBIAD1 functions and tumor suppressor status.

   Overall, we conclude that CoQ[10] treatment or UBIAD1 overexpression
   leads BC cells to lose their capacity to bind to laminins and engaged
   integrin-dependent FAK/AKT oncogenic signaling pathway.

   Lastly, we examined cellular protrusions and adhesions as
   integrin-containing, multi-protein structures that form mechanical
   links between intracellular actin bundles and the extracellular
   substrate in many cell types. Confocal imaging analyses of
   FAK-containing structures clearly showed alterations in membrane
   morphology in MDA-MB-231 overexpressing UBIAD1 compared to controls
   (Fig. [205]5f and Supplementary Fig. [206]6g–i). Notably, these
   alterations manifest as multiple blebs at the plasma membrane
   resembling previous conditions in MDA-MB-231 cells treated with
   CoQ[10]^[207]52 (Supplementary Fig. [208]1e–k). Similarly, we observed
   that MDA-MB-231 overexpressing UBIAD1 were less responsive to the
   cytoskeletal changes induced by the ROCK inhibitor Y27632 and the MLCK
   inhibitor ML7 (Supplementary Fig. [209]6j–l).

   It has been reported that the oncogenic PI3K/AKT pathway sustains
   cancer cell survival, invasion, and metastasis formation via activation
   of integrin receptors localized on the plasma membrane^[210]53. Among
   the integrin analyzed, laminin-binding integrins require a proper PM
   lipid raft assembly to recruit the transduction machinery and activate
   the downstream pathways^[211]54. Considering the ability of CoQ[10]
   treatment to increase membrane fluidity, reduce lipid rafts contents
   and increase cortical stiffness (Fig. [212]1a–c), we analyzed whether a
   similar phenotype was induced by UBIAD1^OE. Indeed, we detected a
   decrease of lipid rafts in UBIAD1^OE cells (Fig. [213]5g). In line with
   this evidence, using the fluorescent probe Dil-C16 to perform a
   Fluorescence Recovery After Photobleaching (FRAP) analysis, we found an
   increased recovery of fluorescence in UBIAD1^OE MDA-MB-231 cells
   compared to controls. Indeed Dil-C16 has more affinity for disordered
   cholesterol-poor membrane domains (where it can freely diffuse) than
   for cholesterol-enriched lipid rafts (Fig. [214]5h^[215]55). Knowing
   that alteration of PM lipid raft structures is associated to changes in
   membrane fluidity^[216]56, we tested the effect of UBIAD1
   overexpression on membrane fluidity of MDA-MB-231 cells. The analysis
   showed an increased membrane fluidity of UBIAD1^OE cells compared with
   control conditions that are similar to what was observed after
   treatment with EtOH, used as a positive control (Fig. [217]5i).

   Modifications in membrane fluidity are important also in the context of
   the organization of the actin cytoskeleton that can control the
   activity of receptors exposed on cell surface^[218]25,[219]56. To
   confirm whether this is the case, we analyzed the single-cell cortical
   stiffness of MDA-MB-231 cells upon UBIAD1 overexpression. Using AFM, we
   showed that UBIAD1^OE cells are stiffer than control cells
   (Fig. [220]5j) as for CoQ[10] treatment (Fig. [221]1c) suggesting that
   BC adapts to increased plasma membrane fluidity, impairment of
   cholesterol-rich lipid rafts with increasing cortical cytoskeletal
   stiffness.

   Interestingly, the inhibitory effect of UBIAD1 expression on blocking
   BC invasion is attenuated either by inducing softness using
   Y27632 + ML7 inhibitors or by supplying antioxidants such as NAC
   (Fig. [222]5k). This suggests that the increased stiffness and
   potential alteration of redox homeostasis caused by UBIAD1-synthesized
   CoQ[10] are likely responsible for decreasing tumorigenic activities in
   these cells.

   Taken together our data demonstrate that UBIAD1 and CoQ[10] alter
   plasma membrane mechano-properties that in turn impair ECM-mediated
   oncogenic AKT pathway activation during BC progression and decrease
   their tumorigenic properties.

UBIAD1 and CoQ[10] enhance BC sensitivity to ferroptosis inducers via FSP1

   Modification of PM lipid properties is associated with the alteration
   of cell mechanical properties and signaling. However, whether it might
   also modulate cell sensitivity to lipid peroxidation (e.g., ferroptosis
   and lipid-peroxidation-triggered death upon treatment with ferroptosis
   inducers) is currently poorly understood. Resistance to ferroptosis has
   been recently associated with the development of some refractory
   cancers, in particular BC^[223]19. Thus, we treated MDA-MB-231
   UBIAD1^OE cells with the lipid peroxidation inducers Erastin and RSL3
   and we evaluated the cell sensitivity to lipid peroxidation. We found
   that in vehicle-treated cells, the level of Malondialdehyde (MDA), a
   bioproduct derived from lipid peroxidation, was not affected by UBIAD1
   overexpression, while UBIAD1^OE cells showed high redox stress upon
   Erastin treatment compared to controls (Fig. [224]6a). Similar evidence
   was obtained evaluating the oxidative status of BODIPY C11, a
   fluorescent lipid peroxidation probe that shifts the emission spectrum
   when oxidized. Indeed, UBIAD1^OE showed a higher level of oxidized
   BODIPY compared to control cells after RSL3 treatment (Fig. [225]6b).
   Interestingly, we found that while treatment with the ROCK inhibitor
   Y27632 and the MLCK inhibitor ML7 tends to protect cells from RSL3,
   lipid peroxidation is significantly higher in MDA-MB-231 UBIAD1^OE
   cells treated with the same inhibitors. This suggests that changes in
   PM cortical properties might influence ferroptosis sensitivities
   (Fig. [226]6c).

Fig. 6. UBIAD1 affects ferroptosis resistance in BC cells.

   [227]Fig. 6
   [228]Open in a new tab

   a Malondialdehyde (MDA) staining in MDA-MB-231 cells treated with
   Vehicle (DMSO) or 2 μM Erastin for 24 h. Scale bar, 100 μm. MDA signal
   intensity normalized over cell number. Mean ± SEM of three independent
   experiments. Adjusted p-value from one-way ANOVA and Sidak’s test. b
   Lipid peroxidation in cells treated with vehicle (DMSO) or 0.45 μM RSL3
   for 24 h. Data expressed as fold change over vehicle-treated cells.
   Each dot represents an independent experiment (n = 6 for
   Vehicle-treated Scramble (Scr), n = 6 for RSL3-treated Scr and
   Scr+Y2736 + ML7, and n = 3 for RSL3-treated UBIAD1^OE and
   UBIAD1^OE + Y2736 + ML7). Mean ± SEM. Adjusted p-value from one-way
   ANOVA followed by Sidak’s test. c Lipid peroxidation in cells treated
   with vehicle (DMSO), 0.45 μM RSL3 for 12 h, in combination with 20 μM
   Y27632 and 20 μM ML7 inhibitors. Gating: P1-Live_Cells-FITC/PE. Data
   expressed as fold change over vehicle-treated cells. Each dot
   represents an independent experiment (n = 5 for Vehicle-treated and
   n = 6 for RSL3-treated). Mean ± SEM. Adjusted p-value from one-way
   ANOVA and Tukey’s test. d, e Cell viability assay with Erastin and RSL3
   (d), Doxorubicin (e), and Menadione (e). Cells treated for 24 h.
   Mean ± SEM of n = 5 independent replicates. P-value from two-way ANOVA.
   f Cell viability assay with RSL3 in MMTV-PyMT ex vivo cultures.
   Mean ± SEM of n = 3 independent replicates. P-value calculated using
   two-way ANOVA. EC50 reported. g Cell viability assay with FIN56 in
   PyMT;Ubiad1^+/+ and PyMT;Ubiad1^+/− cells. Mean ± SEM of n = 5
   independent replicates. P-value from two-way ANOVA. EC50 reported. h
   GPX4 protein in ex vivo cells treated with vehicle (DMSO) or 50 nM RSL3
   for 24 h. ACTIN used as loading control. Each column represents an
   independent replicate (n = 3). i Cell viability assay in cells were
   treated for 24 h with 0.45 μM RSL3 alone, 20 μM CoQ[10] alone or in
   combination. Each dot represents an independent seeding, treatment and
   analysis (n = 8 for DMF + DMSO (Vehicle) and CoQ[10] only, and n = 13
   for RSL3 only and CoQ[10] + RSL3). Mean ± SEM. Adjusted p-value from
   one-way ANOVA and Sidak’s test. j FSP1 protein with DMF (Vehicle) or
   RSL3 0.45 μM for 24 h with or without CoQ[10] 20 μM. VINCULIN used as
   loading control. Each dot represents an independent experiment (n = 4).
   Mean ± SEM. P-value from two-tailed unpaired t-test. Source data are
   provided in Source data file.

   We then evaluated cell viability at increasing concentrations of
   Erastin and RSL3 and we found that MDA-MB-231 UBIAD1^OE cells showed
   reduced viability at lower concentrations of both drugs (Fig. [229]6d).
   Similar evidence was obtained for the two TNBC cell lines MDA-MB-436
   and Hs578T (Supplementary Figs. [230]5g and [231]7a–c). Such effect was
   specific for ferroptosis-inducers since cell sensitivity to general
   oxidative stress inducers (e.g., Menadione) or chemotherapeutic drug
   treatment (e.g., Doxorubicin) was not affected by UBIAD1 expression
   (Fig. [232]6e and Supplementary Fig. [233]7b, c). In line with these
   results, we found that also ex vivo PyMT^UBIAD1 cells were more
   sensitive to RSL3 than controls (Fig. [234]6f). To strengthen our
   results, we tested also tumor-derived cells from MMTV-PyMT-Ubiad1^+/+
   and MMTV-PyMT-Ubiad1^+/− mice. We then treated with increasing
   concentrations of the ferroptosis inducers FIN56 (Fig. [235]6g) and
   found that MMTV-PyMT-Ubiad1^+/− cancer cells with low UBIAD1 levels
   (Supplementary Fig. [236]4g) were more resistant to ferroptosis. GPX4,
   the central regulator that drives ferroptosis resistance, was analyzed
   in PyMT cells after treatment with RSL3 (Fig. [237]6h) and we found
   that UBIAD1 expression in PyMT cancer cells decreases importantly GPX4
   levels and, therefore, their resistance to ferroptosis. Similar
   outcomes were obtained for MDA-MB-231 cells where both Erastin and RLS3
   promote GPX4 level increase in control cells but not in
   UBIAD1-overexpressing cells (Supplementary Fig. [238]7d). Furthermore,
   we quantified Acsl4 mRNA expression in PyMT control and PyMT^UBIAD1
   cells. Indeed, ACSL4 functions to catalyze the ligation of CoA to free
   PUFAs to generate PUFA-CoA that predispose for ferroptosis onset. ACSL4
   expression is tuned down by activation of integrin-mediated signaling,
   especially in cells detached by the ECM^[239]57. Interestingly, we
   found that in detached PyMT^UBIAD1 cells, Acsl4 mRNA expression is
   significantly increased compared to control cells (Supplementary
   Fig. [240]7e). The same results were achieved in tumorspheres obtained
   by PyMT^UBIAD1 cells (Supplementary Fig. [241]7f). Similarly, ACSL4
   mRNA expression increases more in detached MDA-MB-231 UBIAD1^OE cells
   compared to Scr (Supplementary Fig. [242]7g). These data suggest that
   UBIAD1 levels can modulate the sensitivity of BC cells to ferroptosis
   inducers through different anti-ferroptosis mechanisms^[243]58.

   Lastly, we tested whether UBIAD1 increases ferroptosis sensitivity in
   BC cells through its biosynthetic product CoQ[10]. Therefore, we
   treated MDA-MB-231 cells with CoQ[10] with or without the ferroptosis
   inducers RSL3 and tested cell viability (Fig. [244]6i). While CoQ[10]
   treatment alone does not affect MDA-MB-231 cell viability, the
   co-treatment with RSL3 significantly enhanced cell death. To explain
   these results, we investigate FSP1 expression in MDA-MB-231 compared to
   other BC cell lines since the oxidized form CoQ[10] needs to be reduced
   by the NADH-dependent oxidoreductases FSP1 to CoQ[10]H[2], and only
   CoQ[10]H[2] acts as a radical-trapping antioxidant to suppress
   ferroptosis^[245]59. We examined FSP1 expression by Western blot
   analysis in different BC cell lines and scored that MDA-MB-231 and
   other TNBC lines express very low amounts of FSP1 compared to, for
   instance, the luminal BC cell line MCF7 (Supplementary Fig. [246]7h).
   We therefore tested whether FSP1 expression correlates with ferroptosis
   sensitivity of BC cells by comparing RSL3-mediated lipid peroxidation
   between MCF7 and MDA-MB-231 (Supplementary Fig. [247]7i). MDA-MB-231
   are more sensitive to RSL3 compared MCF7. This biological tendency was
   evident also analyzing cell viability after RSL3 treatment with or
   without CoQ[10]; indeed, CoQ[10] treatment made MDA-MB-231 (but not
   MCF7) significantly more sensitive to RSL3 (Supplementary
   Fig. [248]7j). Next, to explain the increased sensitivity of the
   CoQ[10]-treated cells for ferroptosis inducers we analyzed FSP1 levels
   in CoQ[10]-treated cells after ferroptosis inductions (Fig. [249]6j).
   RSL3 stimulation reduces FSP1 protein levels only in CoQ[10] treated
   cells suggesting the possibility that during ferroptosis CoQ10 act by
   impairing the stability of FSP1.

   Overall, we propose that treatment with CoQ[10] or UBIAD1 expression
   sensitizes breast cancer cells to ferroptosis (and consequently reduces
   ferroptosis resistance) by altering PM mechanical properties and,
   possibly, the effectiveness of the FSP1 shield system.

Discussion

   While it is known that CoQ[10] acts as a redox-cycling agent in
   cellular membrane^[250]59,[251]60, its previously hypothesized role as
   a membrane-stabilizing agent remains elusive^[252]5. Here we
   demonstrate that CoQ[10] can alter membrane mechanical properties by
   increasing PM fluidity and possibly cortical stiffness in BC. Such
   biological changes can be achieved either by expressing
   Golgi/ER-localized producing enzyme UBIAD1 or treating cells with the
   metabolite itself, an effect that is independent of its redox state.
   Such CoQ[10]-mediated changes in PM homeostasis leads to an impairment
   of lipid rafts content and ECM-mediated signaling in BC. Interestingly,
   CoQ[10]-dependent impairment of integrin signaling has been previously
   described during atherosclerosis^[253]61. Aggressive BC cells, such as
   TNBC cells, display higher levels of cholesterol-rich lipid
   rafts^[254]62. Therefore, it is conceivable that UBIAD1 and CoQ[10]
   affect the survival and aggressiveness of the TNBC subtype that
   strongly relies on lipid raft integrity and integrin/ECM-mediated
   signal transduction for invasion/survival and ferroptosis
   protection^[255]63,[256]64. Mechanistically, we foresee that such
   effect depends on the ability of CoQ[10] to intercalate within the
   lipid bilayer and altering cholesterol-dependent PM rigidity and
   cortical cell softness vs. stiffness^[257]23,[258]65,[259]66.
   Cholesterol is known for its stabilizing effect in biological lipid
   membranes, therefore, it is tempting to speculate that CoQ[10] might
   increase the fluidity state of cellular PM by directly competing with
   the function of cholesterol in membrane stabilization, causing opposite
   effects (cholesterol→rigidity vs. CoQ[10]→fluidity). Alternatively,
   CoQ[10] might exert a membrane-destabilizing effect in certain types of
   biological membranes, independently of cholesterol^[260]67. We can
   speculate that these membrane alterations might be responsible for the
   spatial organization of membrane-associated contractile proteins,
   potentially altering their function and interactions with the plasma
   membrane. This could interfere with the effect of inhibitors of MLCK
   and ROCK such as ML7 and Y27632 limiting their effect on cellular
   softness. Regardless, the sum of results of this study clearly point to
   a direct role for the UBIAD1/CoQ[10] axis in membrane homeostasis, with
   its dysfunction being causal to the acquisition of aggressive
   phenotypes in BC, thus highlighting alterations in cancer cell membrane
   fluidity and mechano-properties as potential therapeutically actionable
   targets for specific treatments in BC. Supporting this view, clinical
   trials showed a decrease of both plasma and tumor tissue levels of
   CoQ[10] in breast cancer patients, suggesting a statistically inverse
   relationship between the plasma CoQ[10] level and breast cancer
   prognosis^[261]68,[262]69. Of note, reduced plasma concentrations of
   CoQ[10] have also been reported for other types of human cancers, such
   as melanoma, lymphoma and lung cancer^[263]70,[264]71 and correlated,
   in particular in one study in melanoma patients, with increased risk of
   metastasis^[265]72.

   Another peculiar aspect of our studies is the correlation of UBIAD1 and
   CoQ[10] expression with the regulation of ferroptosis resistance during
   BC spreading and CTCs survival. It has been recently shown that
   melanoma cells exit the primary site of tumor growth through the
   bloodstream; the oxidative stress that they encounter makes them prone
   to undergo a type of cell death called ferroptosis that in theory
   limits the formation of distant metastases. However, melanoma cancer
   cells develop specific mechanisms to counteract ferroptosis and survive
   in the bloodstream leading to metastasis^[266]73. Our data show that
   UBIAD1 expression limits significantly CTCs and metastasis in xenograft
   models of BC and that this is possibly related to its ability to
   enhance sensitivity to ferroptosis. We were able to recapitulate these
   findings using CoQ[10] treatments suggesting that an increased
   concentration of plasma CoQ[10] could be therapeutic for BC patients.
   Mechanistically, we demonstrate that altering PM function (e.g.,
   cortical actin, lipid raft contents) CoQ[10] and UBIAD1 impair
   laminin/α6 integrin activation and oncogenic functions in BC. This also
   limits integrin-dependent ferroptosis protection function. These data
   are consistent with previous data showing that laminin (e.g., LAMA5)
   and integrin (e.g., α6β4) are essential to protect BC cells by
   sustaining GPX4 expression^[267]51,[268]57. This would be of particular
   importance for the luminal androgen receptor (LAR) subtypes of TNBC
   characterized by an upregulation of the GPX4 ferroptosis defense
   mechanism^[269]19.

   Some questions remain unanswered such as why we can detect UBIAD1
   effects on GPX4 levels only under a ferroptosis-inducing condition, but
   not under normal basal situations. It has been shown that cancer cells
   use the mevalonate pathway (MVA) and its intermediates, such IPP and
   DMAPP, to support selenoproteins and, thus, GPX4
   synthesis^[270]74,[271]75. We could therefore speculate that when
   UBIAD1 is low in TNBC, MVA intermediates are used to support the
   DMAPP-tRNASelCys-GPX4 branch while when UBIAD1 is high MVA
   intermediates are needed to support CoQ[10] synthesis explaining the
   reduced levels of GPX4 and higher sensibility to ferroptosis in UBIAD1
   ^high condition (Fig. [272]7). It would be interesting to metabolically
   test the hypothesis whether ferroptosis inducers heavily engaged the
   DMAPP-tRNASelCys-GPX4 branch response that would explain the higher
   sensitivity to RSL3 of cells expressing UBIAD1, being the metabolic
   mevalonate pathway of the cells fully rewired to synthetize CoQ[10]
   instead of GPX4. In agreement with this hypothesis, it has been shown
   that soft conditions in BC increase the mevalonate IDI gene required to
   synthesize DMAPP to support selenoprotein/GPX4 synthesis^[273]27. FSP1
   is also an alternative suppressor of ferroptosis. Interestingly, It has
   been shown that myristoylation of FSP1 mediates the recruitment of this
   protein to the plasma membrane and leads to ferroptosis
   inhibition^[274]60,[275]76. Therefore, we can not exclude that by
   reducing the number of lipids rafts and altering PM cortical properties
   UBIAD1/CoQ[10] also act to reduce the stability of FSP1 making cells,
   with already low levels of FSP1, more sensitive to ferroptosis.

Fig. 7. CoQ[10] and UBIAD1 expression impair BC spreading and survival via
modification of plasma membrane-dependent signaling.

   [276]Fig. 7
   [277]Open in a new tab

   Expression of the CoQ[10]-biosynthetic enzyme UBIAD1 is reduced in
   highly aggressive BC, especially during the metastatic cascade.
   Moreover, the ability of colonize specific distant sites is associated
   to specific UBIAD1 expression levels. High levels of CoQ[10] or
   expression of UBIAD1 in BC alters lipid raft plasma membrane fluidity
   that leads to increase cellular stiffness, reduced integrin activity
   and associated oncogenic AKT2 signaling. Furthermore, BC cells with
   high UBIAD1/CoQ10 levels show increased sensitivity to ferroptosis and
   lipid peroxidation possibly by decreasing GPX4 synthesis and FSP1
   levels contributing to the impaired blood circulating cancer cell
   survival and reduced lung colonization. The images were generated using
   Smart Servier Medical Art under license CC BY4.0.

   In conclusion, our study shows that UBIAD1 loss as well as reduced
   CoQ[10] plasma membrane levels promote BC development and metastatic
   spreading and, possibly contribute to BC resistance to
   ferroptosis^[278]19. Therapeutic interventions able to stabilize UBIAD1
   expression and/or increase CoQ[10] levels (e.g., based on chronic
   CoQ[10] supplementations) would act to increase the fluidity of the
   plasma membrane of tumor cells/CTCs and counteract ECM-mediated
   oncogenic signaling, representing an interesting therapeutic level of
   action in BC patients with aggressive disease course.

Methods

Ethics statement

   All procedures involving animals for research purposes were approved by
   the Organismo preposto al benessere degli animali (O.P.B.A.) of the
   University of Padua and by the Italian Ministry of Health (Permit
   Numbers: 336/2019-PR).

Mouse model

   We generated constitutive Ubiad1 Knockout mice by cloning a
   neomycin-resistance cassette into the ATG start codon in the first exon
   of Ubiad1, transfected by electroporation in mouse ES cells. Southern
   blotting identified 5 clones that carried the targeted allele, 2 of
   them were injected into C57BL/6J mouse blastocysts. We obtained
   heterozygous Ubiad1^+/− mice using chimeras. Transgenic male C57BL/6J
   mice expressing the polyoma virus middle T antigen under the control of
   the mouse mammary tumor virus (MMTV) long terminal repeat (LTR)
   promoter (MMTV-PyMT, The Jackson Laboratory JAX #022974) were crossed
   with the Ubiad1^+/− females to generate MMTV-PyMT; Ubiad1^+/+ and
   MMTV-PyMT; Ubiad1^+/− females. We monitored the female mice from birth
   to week 28. Similarly, transgenic male C57BL/6J mice expressing an
   activated rat c-neu oncogene (Erbb2) under the control of the mouse
   mammary tumor virus (MMTV) long terminal repeat (LTR) promoter
   (MMTV-NeuT/Erbb2^[279]77), were crossed with the Ubiad1^+/− females to
   generate MMTV-NeuT; Ubiad1^+/+ and NeuT; Ubiad1^+/− females. We
   monitored the female mice from birth to week 50. We obtained MMTV-NeuT
   mice in BALB/c background and were backcrossed in C57BL/6J for more
   than 10 generations before proceeding with the experiments. Ubiad1^f/f
   mice carrying the second exon of Ubiad1 flanked by two Lox-P sites were
   custom-made from Cyagen in C57BL/6J background. C57BL/6J K14-Cre; Brca1
   ^f/f; Tp53 ^f/f mice, alias KBP^[280]36, were kindly obtained from
   Prof. Bouwman at The Netherlands Cancer Institute. We crossed KBP mice
   with Ubiad1^f/f mice to generate the KBP; Ubiad1^f/f used in this
   study. Only males K14-Cre-positive were used for breeding^[281]78,
   while only females were used in the experiments. 6–8-week-old NOD/SCID
   common γ chain knockout (NSG; Charles River) female mice were used for
   xenografts experiments.

   Mouse genotyping began with whole mouse embryos or from a small piece
   of the ear that was digested for 16 h at 56 °C with 0.2 mg/ml
   Proteinase K (Thermo Fisher Cat. #EO0492). After heat inactivation of
   the Proteinase K, samples were used for genotyping with the following
   primers: Ubiad1^+/− FW 5′-CACGGGGCCGGCAATTTGGTCAAG-3′, RV
   5′-GGGGGTCTCCAGCCCCACAGTAAGAGA-3′ and Neo-Cassette
   5′-CAGCTCGGGCTCGACTAGAG CTTGC-3′ (400 bp KO allele and 500 bp WT
   allele); MMTV-PyMT IMR-15 5′- CAAATGTTGCTTGTCTGGTG-3′, IMR-16
   5′-GTCAGTCGAGTGCACAGTTT-3′, IMR-384 5′-GGAAGCAAGTACTTCACAAGGG-3′ and
   IMR-385 5′-GGAAAGTCACTAGGAGCAGGG-3′ (200 bp internal CTRL and 556 bp
   PyMT allele); MMTV-NeuT/Erbb2 FW 5′-ATCGGTGATGTCGGCGATAT-3′, RV
   5′-GTAACACAGGCAGATGTAGGA-3′, β-cas-FW 5′-GATGTGCTCCAGGCTAAAGTT-3′ and
   β-cas-RV 5′-AGAAACGGAATGTTGTGGAGT-3′ (525 bp internal CTRL and 225 bp
   NeuT allele); Ubiad1^f/f LoxP1_FW 5′-CTCCTGAAGACCATTGTCACCTGTC-3′,
   LoxP1_RV 5′-GAACACAGAGCCACAGTAACAAACAC-3′, LoxP2_FW
   5′-CACAAGGAGCAAGCAAGCCAGTA-3′ and LoxP2_RV
   5′-GGGAGGATAGAGGGAAATGAGGTTA-3′; K14-Cre
   FW5′-CGATGCAACGAGTGATGAGGTTC-3′, RV5′-GCACGTTCACCGGCATCAAC-3′,
   oIMR8744_Int_ctrl5′-CAAATGTTGCTTGTCTGGTG-3′ and
   oIMR8745_Int_ctrl5′-GTCAGTCGAGTGCACAGTTT-3′; Brca1^f/f LoxP1_FW
   5′-TATCACCACTGAATCTCTACCG-3′, LoxP1_RV 5′-GACCTCAAACTCTGAGATCCAC-3′,
   LoxP2_FW 5′-TATTCTTACTTCGTGGCACATC-3′ and LoxP2_RV
   5′-TCCATAGCATCTCCTTCTAAAC-3′; Tp53^f/f LoxP1_FW
   5′-CACAAAAACAGGTTAAACCCAG-3′, LoxP1_RV 5′-AGCACATAGGAGGCAGAGAC-3′,
   LoxP2_FW 5′-AAGGGGTATGAGGGACAAGG-3′ and LoxP2_RV
   5′-GAAGACAGAAAAGGGGAGGG-3′. HOT FIREPol® Blend Master Mix (Solis
   BioDyne) was used for genotyping and T[annealing] was set at 58 °C for
   all the reactions, except for Ubiad1^+/− in which T[annealing] was set
   at 62 °C.

   Tissue-specific deletion of exon 2 of Ubiad1 was verified using the
   following primers: Ubiad1^f/f LoxP1_FW 5′-CTCCTGAAGACCATTGTCACCTGTC-3′,
   LoxP2_FW 5′-CACAAGGAGCAAGCAAGCCAGTA-3′ and LoxP2_RV
   5′-GGGAGGATAGAGGGAAATGAGGTTA-3′. HOT FIREPol® Blend Master Mix (Solis
   BioDyne) was used, T[annealing] was set at 63 °C and time of extension
   was set at 20 s.

   Mice were housed under conventional conditions in individual cages in
   the controlled room at 22 °C and 12 h light/dark cycle with ad libitum
   access to food (Mucedola S.r.l., 4RF21 standard diet) and water and
   were regularly monitored for weight and food consumption. The maximal
   permitted tumor volume of 4000 mm³ per single tumor has not been
   exceeded. Only female mice were used for the experiments as the gold
   standard model to study breast cancer.

Primary cells

   Ex vivo primary BC PyMT (alias PyMT; Ubiad1^+/+) or PyMT; Ubiad1^+/−
   cells were isolated from mammary gland tumors of MMTV-PyMT (alias
   MMTV-PyMT; Ubiad1^+/+) or MMTV-PyMT;Ubiad1^+/− transgenic C57BL/6J
   female mice, respectively. The collected tumor was placed on ice in
   DMEM/F12 medium (Thermo Fisher Scientific Cat. #31331028) and processed
   in less than 1 h. A piece of 1 g was dissected from the tumor, cut into
   small pieces (<4 mm) and dissociated using Tumor Dissociation Kit
   (Miltenyi Biotec Cat. #130-096-730) according to manufacturer’s
   instructions (program for soft tumors). The suspension was applied to a
   70 μm and a 40 μm cell strainer, then centrifuged at 300 × g for
   10 min. Cells were resuspended in DMEM/F12 GlutaMAX™ medium containing
   10% FBS (Carlo Erba Reagents Cat. #FA30WS1810500), 5 ng/ml EGF
   (PeproTech Cat. #AF10015), 1 µg/ml Hydrocortisone (Sigma Cat. #H0888),
   100 ng/ml Cholera Toxin (Sigma Cat. #C8052), 5 µg/ml Insulin (Sigma
   Cat. #I6634), 100 µg/ml penicillin and 100 µg/mL streptomycin (Thermo
   Fisher Scientific Cat. #15140122) and placed in culture. The medium was
   changed regularly and any fibroblasts were removed by several passages
   of differential trypsinization^[282]79. Then, PyMT cells were
   maintained in DMEM/F12 GlutaMAX™ medium supplemented with 10% FBS,
   100 µg/ml penicillin and 100 µg/mL streptomycin in a 37 °C incubator
   with a humidified atmosphere of 5% CO[2]. The cells were tested for
   mycoplasma infection using N GARDE Mycoplasma PCR Reagent Set
   (EuroClone Cat. #EMK090020) prior to use for experiments.

Cell lines

   MCF7 (ATCC Cat. #HTB-22™), T-47D (ATCC Cat. #HTB-133™), BT474 (ATCC
   Cat. #HTB-20™), HCC202 (ATCC Cat. #CRL-2316™), HCC1143 (ATCC Cat.
   #CRL-2321™), MDA-MB-438 (ATCC Cat. #HTB-132™), MDA-MB-436 (ATCC Cat.
   #HTB-130™), MDA-MB-231 (ATCC Cat. #HTB-26™), Hs578T (ATCC Cat.
   #HTB-126™), JIMT-1 (Leibniz Institute DSMZ-German Collection of
   Microorganisms and Cell Cultures, Cat. #ACC-589) and HEK293T (ATCC Cat.
   #CRL-11268) cells were cultured in DMEM, high glucose, GlutaMAX™
   Supplement, pyruvate (Thermo Fisher Scientific Cat. #31966047)
   containing 10% FBS (Carlo Erba Reagents Cat. #FA30WS1810500) and
   100 µg/ml penicillin and 100 µg/mL streptomycin (Thermo Fisher
   Scientific Cat. #15140122). MCF10A (ATCC Cat. #CRL-10317™) cells were
   cultured in DMEM/F12 GlutaMAX™ medium (Thermo Fisher Scientific Cat.
   #31331028) supplemented with 5% Horse Serum (Sigma Cat. #H1138),
   20 ng/ml human EGF (PeproTech Cat. #AF10015), 0.5 µg/ml Hydrocortisone
   (Sigma Cat. #H0888), 100 ng/ml Cholera Toxin (Sigma Cat. #C8052),
   10 µg/ml Insulin (Sigma Cat. #I6634), 100 µg/ml penicillin and
   100 µg/mL streptomycin. Cells were maintained in a 37 °C incubator with
   humidified atmosphere of 5% CO[2]. All cell lines were regularly tested
   for mycoplasma infection using N GARDE Mycoplasma PCR Reagent Set
   (EuroClone Cat. #EMK090020).

Public dataset analysis

   Publicly available Copy Number Alteration (CNA), UBIAD1 mRNA level,
   PIK3CA/PTEN mutational status, associated clinicopathological, and
   survival data of 1904 breast cancer patients enrolled in the METABRIC
   cohort were retrieved from cBioPortal
   ([283]http://www.cbioportal.org/).

   Patients were stratified according to UBIAD1 CNA in Normal (without
   CNA), Amplified (with amplifications), and Deleted (with deep or
   shallow deletions).

   The 19 patients carrying UBIAD1 amplifications were not included in the
   survival analysis as they represented a rare subpopulation not of
   primary interest.

   Inferred ER and HER2 status from RNA expression status were used for
   the molecular classification of breast cancer to compensate for the
   lack of IHC information in a relevant fraction of patients.

Human breast cancer immunohistochemical analysis

   IHC analysis was performed on formalin-fixed paraffin-embedded tissue
   microarrays (TMA) of breast cancer specimens from a retrospective
   consecutive cohort of ~2400 BC patients enrolled at the European
   Institute of Oncology, Milan, from 1997 to 2000, with complete
   long-term follow-up (~15 years)^[284]80. Expression of UBIAD1 was
   assessed using rabbit anti-human UBIAD1 custom-made from ProteoGenix
   (peptide sequence: cys-FRSQAFNKLPQRTAK), which was used at final
   concentration of 0.17 µg/ml following an antigen retrieval procedure in
   Citrate pH 6.0. Immunocomplexes were visualized by the EnVision+ HRP
   Rabbit (DAB+) kit (DAKO Cat. #K4007), acquired with the Aperio
   ScanScope system (Leica Biosystems). Informed consent was obtained from
   all subjects. For correlation analysis with clinical and pathological
   parameters, patients were stratified based on the intensity of UBIAD1
   staining, as follows: UBIAD1[Low] (IHC score <2), UBIAD1[Intermediate]
   (IHC score >2 and <3) and UBIAD1[High] (IHC score ≥3).

Spontaneous tumor models

   2-month-old mice were monitored once a week for palpable tumors. When
   tumors became palpable, the volumes of all the mammary tumors in each
   mouse were measured once a week till week 28 for MMTV-PyMT and week 50
   for MMTV-NeuT mice. Tumor volume was measured with a caliper using the
   formula Tumor volume = (length × width^2)/2, where length indicates the
   largest tumor diameter and width indicates the perpendicular tumor
   diameter. The total tumor volume per mouse was calculated as the sum of
   the volumes of all the tumors developed from the single mouse. At the
   experimental endpoint, tumors were counted, weighed, and collected for
   histological examination or quantitative PCR/Western blot analysis.
   Total tumor weight was calculated as the sum of the weights of all the
   tumors developed from a single mouse. 4-month-old KBP mice were
   monitored once a week for palpable tumors. As soon as a tumor became
   palpable in one female of the litter (between 29 and 32 weeks of age),
   all siblings were sacrificed, the 2nd−3rd (thoracic) and the 4th
   (abdominal) mammary glands were dissected and stained for whole mammary
   gland staining. Alternatively, tumor lesions were collected and
   dissociated in TRIzol for RNA extraction or digested with Proteinase K
   for DNA extraction. Genomic DNA was purified through phenol/chloroform
   method prior to proceed with PCR.

Xenografts tumor models

   MDA-MB-231 and JIMT-1 cScr and UBIAD1^OE cells were transduced with a
   lentivirus carrying the plasmid for Firefly Luciferase and eGFP
   expression^[285]81. Only eGFP+ cells were selected with FACS Aria™ IIIu
   Cell Sorting and expanded in vitro. 6–8-week-old NSG female mice were
   anesthetized and 5 × 10^5 eGFP+ MDA-MB-231 or 4 × 10^5 JIMT-1 cells
   resuspended in 100 μL of equal amount PBS: Matrigel (Growth Factor
   Reduced Matrigel, Corning Cat. #354230) were injected into the fat pad
   of one 4th mammary gland using a syringe with a 26G needle. Tumor
   growth was monitored once a week using a caliper till 21 days
   post-injection for MDA-MB-231 and 35 days post-injection for JIMT-1
   cells. For intracardiac injection, 2 × 10^5 eGFP+ MDA-MB-231 or JIMT-1
   were resuspended in 100 μL of PBS and injected into the left ventricle
   of anesthetized NSG females using a syringe with 30G needle^[286]82.
   Metastasis formation was evaluated once a week using bioluminescence
   imaging (BLI) till 21 days post-injection for MDA-MB-231 and 35 days
   post-injection for JIMT-1 cells. Bioluminescence images were collected
   with the IVIS Spectrum Imaging System (PerkinElmer). Ten minutes before
   imaging, animals were anesthetized with isoflurane/oxygen and
   administered i.p. with 150 mg/kg of D-luciferin (PerkinElmer) in PBS.
   Ventral and dorsal images were obtained for each animal and quantified
   through the region of interest (ROI). Living Image Software
   (PerkinElmer) was used to acquire and quantify the bioluminescence
   imaging data sets. At sacrifice, mouse blood was collected via cardiac
   puncture using a syringe with 25G needle containing 50 μL of 500 U/mL
   Heparin (Sigma). After centrifugation at 1100 × g for 5 min at RT, the
   upper layer of plasma was discarded and the underlying buffy coat layer
   containing WBC and Circulating Tumor Cells was collected. The buffy
   coat was then washed with Red Bood Cell Lysis Buffer (Sigma, Cat.
   #11814389001) according to manufacturer instructions to remove any
   remaining red blood cells. WBC and CTCs were resuspended in 3%BSA in
   PBS and sorted with FACSAria™ IIIu Cell Sorting to isolate only
   CTC-eGFP+ cells. After blood collection, the primary tumor and several
   organs were collected and the eGFP signal was evaluated in the whole
   organ using Xenogen IVIS Lumina II (Perkin Elmer) or using Leica
   Stereomicroscope connected to a fluorescent lamp. Organs and tumors
   were then cut into 2 mm pieces, resuspended in 1 mg/mL Collagenase A
   (Sigma, Cat. #11088793001) in DMEM (Thermo Fisher), and dissociated at
   37 °C for 30 min. After red blood cells were lysed, cells were
   resuspended in 3%BSA in PBS and sorted with FACSAria™ IIIu Cell Sorting
   to isolate only eGFP+ cancer cells. Isolated cells were used either for
   protein extraction using home-made RIPA supplemented with 1x halt
   protease inhibitor cocktail (Roche Cat. #04693116001) and phosphatase
   inhibitor (Sigma Cat. #4906845001) or for RNA extraction. Depending on
   the number of eGFP+ cells isolated, either TRIzol™ Reagent
   (ThermoFisher Scientific Cat. #15596018) or NucleoSpin RNA XS Kit
   (Carlo Erba, Cat. #FC140902L) was used according to manufacturer
   instructions.

Mouse treatment

   N-Acetyl-L-cysteine (NAC; Sigma Cat. #A9165) and Vitamin K[2] (Sigma
   Cat. #V9378) were diluted in PBS while CoQ[9] (Sigma Cat. #C9538) was
   diluted in sunflower oil. NAC, Vitamin K[2] and CoQ[10] were injected
   at a dose of 150 mg/kg. Pregnant females were injected daily between
   E2.5 and E9.5 intraperitoneally. Pups were genotyped at P0.

CoQ[9] measurement from mouse embryos and tumors

   Samples for HPLC-MS analysis were prepared as described
   previously^[287]30. CoQ[9] levels were detected by HPLC-MS using CoQ[6]
   (Avanti Cat. #900150 O) as an internal standard. Protein concentration
   in each sample was measured by Pierce™ BCA Protein Assay (ThermoFisher
   Scientific Cat. #23225) and the quantity of CoQ[9] was normalized on
   this protein quantity.

Whole mount mammary gland staining

   Mouse mammary gland whole mount preparation and staining was performed
   as previously described^[288]83. In particular, 2nd−3rd (thoracic) and
   4th (abdominal) mammary glands from mice aged 29 to 32 weeks were fixed
   in Carnoy’s solution for 24 h, and staining was performed using a
   Carmine dye solution (Sigma, C1022) and aluminum potassium (Sigma, Cat.
   #A7167) in deionized H[2]O for 3 days. Slides were kept in Xylene while
   pictures were taken using a Leica MZ16F stereomicroscope. The area of
   neoplastic lesions was calculated with ImageJ and normalized over
   different mouse litters.

Cell viability assay

   1 × 10^4 MDA-MB-231, MCF7, MDA-MB-436, Hs578T cells/well, and 5 × 10^3
   ex vivo PyMT cells/well were seeded in 96-well plates and allowed to
   adhere for 4 h or O/N. Cells were treated for 24 h with indicated
   compounds at the indicated concentrations. In the case of combinatory
   treatment, cells were allowed to adhere for 4 h and then,
   simultaneously treated for 24 h with medium containing 20 μM CoQ[10]
   (Sigma) and 0.45 μM RSL3 (Sigma Cat. #SML2234). Erastin (Sigma Cat.
   #329600), RSL3, and Doxorubicin hydrochloride (Sigma Cat. #D2975000)
   were solved in DMSO (Sigma Cat. #D2650), Menadione (Sigma Cat. #M5625)
   in 96% ethanol and CoQ[10] in DMF (Sigma Cat. #227056). Then, the
   medium was removed and cells were washed 2 times with tap water. Then,
   50 µl of 0.5% (w/v) Crystal Violet staining solution (Sigma Cat.
   #C6158) in 20% methanol was added to each well and incubated for 20 min
   at room temperature on a bench rocker with a frequency of 20
   oscillations per minute. The plates were washed four times with tap
   water and allowed to dry. Crystal Violet was then dissolved by adding
   200 µl of methanol to each well and incubating for 20 min at room
   temperature on a bench rocker with a frequency of 20 oscillations per
   minute. Next, the absorbance at 595 nm was analyzed with a plate
   reader. Alternatively, pictures of each well were taken using a Leica
   DMI4000 or Nikon AZ100 microscope, and crystal violet-stained cells
   were counted with ImageJ software.

Cell treatments

   Cells were treated with 20 μM CoQ[10] (Sigma Cat. #C9538), CoQ[6]
   (Avanti Polar Lipids, Inc Cat. #900150), CoQ[2] (Sigma Cat. #C8081) or
   DMF (vehicle, Sigma Cat. #227056), 20 μM water-dispersible CoQ[10]H[2]
   (alias ShiroQ™, 40% Kaneka Ubiquinol water-soluble powder) or Placebo
   (vehicle), Idebenone (Tocris Bioscience Cat. #3002), Y27632 (Adooq
   Bioscience Cat. #A11001), ML7 (Tocris Bioscience Cat. #4310), NAC
   (Sigma Cat. # A9165) or equal volume of ddH[2]O. Cells were treated for
   14 days (unless specified otherwise) changing medium every two days.

   4-OHT-treated cells were seeded in complete medium in 6-well plates and
   let adhere overnight. The following day, cells were treated with 10 μM
   4-OH-Tamoxifen (Sigma Cat. #H7904) in DMSO or with Vehicle (DMSO) for
   48 h before protein and RNA collection.

   Before treatment with 17β-Estradiol, cells were seeded and grown for
   48 h in DMEM without phenol-red (Thermo Fisher Scientific, Cat.
   #A1443001) supplemented with 4.5 g/L D-glucose (Sigma, Cat. #G7021),
   1 mM Sodium Pyruvate (Thermo Fisher Scientific, Cat. #11360070), 2 mM
   L-Glutamine (Thermo Fisher Scientific, Cat. #25030024), 10% Fetal
   Bovine Serum charcoal stripped (Thermo Fisher Scientific Cat.
   #A3382101) and 100 µg/ml penicillin and 100 µg/mL streptomycin (Thermo
   Fisher Scientific Cat. #15140122). In the same medium composition,
   cells were treated for 24 h with 10 nM 17β-Estradiol (Merk, Cat.
   #2001128945) in EtOH or with Vehicle (EtOH), before proceeding with RNA
   extraction.

Cell proliferation assay

   3000 cells per well were seeded in 96-well plates and stained after
   24 h, 48 h, 72 h, 96 h, and 120 h intervals using Crystal Violet
   staining solution (Sigma Cat. #C6158). The medium was removed, and
   cells were washed twice with tap water at each interval. Then, 50 µl of
   0.5% (w/v) Crystal Violet staining solution in 20% methanol was added
   to each well and incubated for 20 min at room temperature on a bench
   rocker with a frequency of 20 oscillations per minute. The plates were
   washed four times with tap water and allowed to dry. Crystal Violet was
   then dissolved by adding 200 µl of methanol to each well and incubating
   for 20 min at room temperature on a bench rocker at a frequency of 20
   oscillations per minute. Next, we determined the value of absorbance at
   595 nm with a plate reader. For cell proliferation assay in presence of
   CoQ[10], 10,000 cells per well were seeded in 96-well plates and
   treated with DMF (Vehicle) or 20 μM CoQ[10] (Sigma Cat. #C9538). Medium
   with treatment was replaced every 48 h.

Viral transduction and generation of stable lines

   The coding sequence of the human UBIAD1 gene ([289]NM_013319.3) was
   cloned into the pLenti-CMV-Neo-DEST vector (Addgene Cat. #17392) using
   the Gateway cloning system. Recombinant lentiviruses carrying the
   pLenti-CMV-Neo-DEST-UBIAD1, pLenti-CMV-Neo-GFP-DEST (Addgene Cat.
   #19732) or pLenti-CMV-Neo-DEST-empty were produced by co-transfecting
   HEK293T cells with a mixture of plasmid DNA consisting of pMD2.G (VSV-G
   envelope; Addgene Cat. #12259), pMDLg/pRRE (Gag/Pol; Addgene Cat.
   #12251), and pRSV-Rev (Rev; Addgene Cat. #12253) using Lipofectamine™
   2000 Transfection Reagent (Life Technologies Cat. #11668019) according
   to the manufacturer’s recommendations. Supernatants containing
   lentivirus were collected, passed through 0.45-µm filters, and stored
   at −80 °C. Lentiviral particles were quantified using Lenti-X™ p24
   Rapid Titer Kit (Takara Cat. #632200). Similarly, lentiviruses carrying
   lentiMPH v2 plasmid (Addgene Ref.# 89308) and lenti_SAM_v2 plasmid
   (Addgene Ref.# 75112) containing scramble-gRNA
   (5′-GGCACTCACATCGCTACATCAGTTT-3′) or UBIAD1-gRNA
   (5′-GGCCAGAGGCGGGGAGAAGGGGTTT-3′) for the generation of CRISPR-edited
   lines were produced.

   Stable lines overexpressing UBIAD1 were generated by transducing
   MDA-MB-231, MDA-MB-436, Hs578T and ex vivo PyMT cells with 5 MOI of
   lentivirus. Transduced cells were selected with 1.5 mg/ml G418 (Gibco™
   Cat. #10131019) until the complete death of MOCK control (not
   transduced cells). All the following experiments were performed in
   culture medium without G418. Alternatively, MDA-MB-231 stable UBIAD1
   overexpressing cells were generating using CRSIPR-dCas9 technology:
   first cells were transduced with lentiMPH v2 and selected with
   100 μg/mL Hygromycin b (Santa Cruz Cat. #sc-29067) and second with
   lenti_SAM_v2_UBIAD1-gRNA or lenti_SAM_v2_scamble-gRNA and selected with
   2 μg/mL Blasticidine (Sigma, Cat. #15205)^[290]84.

   For the generation of the Luc-eGFP lentivirus, the following lentiviral
   packaging vectors were used: pMDLg/pRRE, pRSV-Rev, pMD2.VSVG, and
   pADVantage. The transfer vector #945.pCCL.sin.cPPT.SV40ployA.eGFP.-
   nCMV.hPGK.deltaLNGFR.Wpre is a self-inactivating (SIN) HIV-derived
   vector^[291]85, which carries a minCMVPGK divergent bidirectional
   promoter driving the simultaneous expression of two genes in antisense
   orientation. The transfer vectors used in this study were modified to
   express two reporter genes: the Luciferase sequence (fluc) under the
   control of the hPGK promoter, and the eGFP (enhanced Green Fluorescent
   Protein) under the control of minCMV. fluc sequence were synthetized by
   GeneArt, Life Technologies (Regensburg, Germany).

AKT2 overexpression

   Cells were transfected via reverse-transfection using Lipofectamine
   2000 (Invitrogen, Cat. #11668). Briefly, 1.5 × 10^6 cells were mixed
   with Lipofectamine and 2 μg of pcDNA3-empty, pcDNA3-AKT2-wt, and
   pcDNA3-Myr-AKT2 in Pen/Strep- and FBS-free medium and seeded in 10 cm
   petri-dish^[292]86. After 6 h, cells were washed with 1x PBS and full
   culture medium was added. 48 h-post transfection, cells were
   trypsinized and seeded for invasion assay. AKT2 overexpression was
   confirmed by Western blot analysis using anti-human AKT2 antibody
   (1:1000, GeneTex Cat. #GTX128457, RRID:AB_2750992).

Fluorescence recovery after photobleaching (FRAP)

   Cells were seeded in µ-Dish 35 mm, high (Ibidi, Cat. #81156) in
   complete medium and observed overnight. The following day, cells were
   washed with PBS and incubated with 1 μg/mL DiI-C16(3) solution
   (ThermoFisher Scientific Cat. #D-384) in PBS for 1 min at RT. After
   three washes with PBS, cells were maintained in warm HBSS (ThermoFisher
   Scientific, Cat. #14025050). Cells were kept at 37 °C using a heating
   chamber for the entire experiment. FRAP was performed using a 63×
   oil-immersion objective in a confocal microscope (Leica SP8). 60–80% of
   photobleaching of a defined area (3 × 3 μm^2) of the labeled membrane
   was obtained with four 100% 552 nm laser scans for a total duration of
   4 s. The fluorescence intensity was recorded before photobleaching and
   after taking pictures every 1 sec over a period of 3 min. After
   background normalization, the fluorescence was plotted over time to
   generate a fluorescence recovery curve. Unwanted photobleaching during
   imaging was negligible with the laser power set.

Membrane fluidity assay using lipophilic pyrene probes

   Cells were seeded in 96-well Black/Clear Flat Bottom TC-treated Imaging
   Microplates (Corning Cat. #353219) in standard medium. When reached
   80–90% confluency, cells were stained using the Membrane Fluidity Kit
   (Abcam, Cat. #ab189819) according to manufacturer’s instructions. In
   brief, a solution of 5 μM Fluorescent Lipid Reagent with 0.08% Pluronic
   F127 in Perfusion Buffer was used to stain the cells at 25 °C for 1 h
   in the dark. Cells were then washed two times with PBS. In PBS, the
   fluorescence emission between 390 nm and 510 nm (5 nm step size) was
   recorded from the bottom of the plate after an excitation at 350 nm
   (±10 nm) using a Tecan Plate reader Infinite 200 PRO. The mean of
   fluorescent intensity at 390–395 nm or 475–485 nm was considered for
   probe monomers or excimer, respectively. After background (cells
   without staining) subtraction, the ratio of excimers over monomers
   fluorescence intensity was calculated as a measure of membrane
   fluidity. Increased membrane fluidity leads to pyrene probes forming
   excimers instead of monomers^[293]87. Treatment for 48 h with 400 mM
   EtOH was used as the positive control.

Cell adhesion assay

   Cell adhesion assay was performed to analyze integrin-mediated binding
   to different extracellular matrix components. Fibronectin (Sigma Cat.
   #ECM001), Collagen I (Roche Cat. #11179179001), Collagen IV (Sigma Cat.
   #C7521), Vitronectin (Sigma Cat. #SRP3186), Laminin from human placenta
   (Sigma Cat. #L6274), and Poly-L-lysine (PLL; Sigma Cat. #P4707) were
   solubilized according to manufacturer’s instructions and 100 μl/well
   were used to coat 96-well plates for 1 h at 37 °C. Plates were coated
   with 1.25 μg/ml Fibronectin, 0.4 μg/ml Collagen I, 0.2 μg/ml Collagen
   IV, 4 μg/ml Vitronectin, and 2.5 μg/ml Laminin. 12.5 μg/ml
   Poly-L-lysine (PLL) was used as integrin-independent attachment
   control. After coating, unspecific binding was reduced by blocking the
   plate with 2.5 mg/ml of heat-denatured BSA (Sigma Cat. #10735094001) in
   water for 2 h at 37 °C. The blocking solution was removed and the wells
   were washed twice with serum-free DMEM. Cells were detached using an
   enzyme-free buffer containing 2 mM EDTA in PBS pH 7.4 and were seeded 5
   × 10^4 cells/well in 100 μl of serum-free DMEM. Cells were allowed to
   adhere for 30 min in a 37 °C incubator with a humidified atmosphere of
   5% CO[2] before staining using Crystal Violet staining protocol.
   Briefly, cells were washed twice with tap water before adding 50 µl of
   0.5% (w/v) Crystal Violet staining solution (Sigma) in 20% methanol to
   each well. The plates were incubated for 20 min at room temperature on
   a bench rocker with a frequency of 20 oscillations per minute. Wells
   were washed four times with tap water and allowed to dry. Crystal
   Violet was then dissolved by adding 200 µl of methanol to each well and
   incubating for 20 min at room temperature on a bench rocker with a
   frequency of 20 oscillations per minute. Next, the absorbance at 595 nm
   was analyzed with a plate reader.

   Similarly, 6-well plates were coated with 7.5 μg/ml Laminin from human
   placenta (Sigma) or 7.5 μg/ml Poly-L-lysine (Sigma) and blocked as
   described above. 5 × 10^5 cells were seeded in 2 ml of serum-free
   medium and allowed to adhere for 30 min in a 37 °C incubator with a
   humidified atmosphere of 5% CO[2]. Then, samples were prepared for
   Western blotting as described in “Methods” details.

Staining of lipid rafts

   To visualize lipid rafts in living cells we used Vybrant™ Alexa Fluor™
   555 Lipid Raft Labeling Kit (Thermo Fisher Scientific Cat. #V34404)
   following the manufacturer’s instructions. Briefly, 1.5 × 10^4
   cells/well were seeded on glass coverslips placed in 24-well plates
   with DMEM GlutaMAX™ medium (Thermo Fisher Scientific) supplemented with
   10%FBS and allowed to adhere overnight. Then, the medium was removed
   and cells were incubated with 350 μl/well of CT-B conjugate in medium
   w/o FBS for 10 min at +4 °C with gentle shaking. After the incubation,
   cells were washed twice with PBS and incubated with 350 μl/well of
   anti-CT-B antibody and Hoechst 33342 (1:2000, Thermo Fisher Scientific
   Cat. #62249) in medium w/o FBS for 15 min at +4 °C with gentle shaking.
   Cells were washed three times with PBS and fixed with 4% formaldehyde
   at room temperature for 15 min. After three additional washes with PBS,
   the glass coverslips were mounted with Mowiol® 4-88 (Sigma Cat.
   #81381). Samples were visualized using Leica SP8 DLS microscopy and
   images were processed using Leica LASX software
   ([294]https://www.leica-microsystems.com/). The fluorescence intensity
   was measured with ImageJ software ([295]https://imagej.nih.gov/ij/).

Real time q-PCR

   Harvested tumors were washed in PBS and radially divided into four
   pieces. One piece was resuspended in TRIzol™ Reagent (Thermo Fisher
   Scientific Cat. #15596018) and homogenized using IKA T-10 Basic
   Homogenizer. Cells were rinsed with ice-cold PBS and either TRIzol™
   Reagent or NucleoSpin RNA XS Kit (Carlo Erba, Cat. #FC140902L) was
   directly added to the plate. After sample collection, RNA isolation was
   performed according to the manufacturer’s instructions. RNA
   concentration and purity were determined using a NanoDrop
   spectrophotometer at A260 and A280/260, respectively. RNA was reverse
   transcribed to cDNA using a High-Capacity cDNA Reverse Transcription
   Kit (Thermo Fisher Scientific Cat. #4368814). Quantitative PCR (q-PCR)
   was carried out with 5x HOT FIREPol® EvaGreen® qPCR Supermix (Solis
   Cat. #08-36-00020) on QuantStudio™ 5 Real-Time PCR System (Thermo
   Fisher Scientific). Transcripts levels were normalized to β-Actin and
   reported as relative to a reference sample. All primers used in q-PCR
   have an efficiency close to 100%, and their sequences are listed below.
   QuantStudio™ Design & Analysis Software ([296]www.thermofisher.com) was
   used for performing q-PCR data processing. Mouse Ubiad1 FW:
   5′-CACAGGAGGAATTGGATTCAAGT-3′; Mouse Ubiad1 RV:
   5′-GGATGGCGTAGATTAGAGGGAAG-3′; Mouse β-Actin FW
   5′-GTACTCTGTGTGGATCGGTGG-3′; Mouse β-Actin RV:
   5′-AAACGCAGCTCAGTAACAGTCC-3′; Human UBIAD1 FW:
   5′-CACTTGGCTCTTATCTACTTTGGA-3′; Human UBIAD1 RV:
   5′-GTCTCCCAGAGCCACGTACTTG-3′; Human β-ACTIN FW:
   5′-GATGGAGTTGAAGGTAGTTTCGT-3′; Human β-ACTIN RV:
   5′-GCGGGAAATCGTGCGTAGCATT-3′; Mouse Acsl4 FW:
   5′-GGCTGACAGAATCATGTGGTG-3′; Mouse Acsl4 RV:
   5′-GAACTGTATAACCACCTTCCTGC-3′; Human MYC FW: 5′-CTCTGAAAGGCTCTCCTTG-3′;
   Human MYC RV: 5′-CGTAGTCGAGGTCATAGTTC-3′; Human BCL2 FW:
   5′-TGGATGACTGAGTACCTGAA-3′; Human BCL2 RV: 5′-AGGAGAAATCAAACAGAGGC-3′;
   Human CCND1 FW: 5′-TCGGTGTCCTACTTCAAATG-3′; Human CCND1 RV:
   5′-CTCGCACTTCTGTTCCTC-3′.

Western blotting

   Cells were rinsed with ice-cold PBS and collected from culture plates
   by scraping. Cells were lysed in ice-cold homemade RIPA Lysis Buffer
   (20 mM Tris-HCl, pH 7.4, 200 mM NaCl[2], 2% sodium deoxycholate, 0.2%
   SDS, 2% NP40) supplemented with 1x halt protease inhibitor cocktail
   (Roche Cat. #04693116001) and phosphatase inhibitor (Sigma Cat.
   #4906845001) for 20 min, and soluble lysate fractions were clarified by
   centrifugation at 20,000 × g for 10 min at 4 + C.

   Mouse embryos were resuspended in ice-cold RIPA Lysis and Extraction
   Buffer (Life Technologies) supplemented with 1x halt protease and
   phosphatase inhibitor cocktail (Roche and Sigma), gently homogenized
   with a pestle and incubated for 30 min on ice. Soluble lysate fractions
   were clarified by centrifugation at 20,000 × g for 10 min.

   Protein concentrations were determined with Bradford (Bio-Rad Cat.
   #5000006). Twenty micrograms of protein per well were loaded in
   SDS-PAGE (4–12% pre-cast gels Invitrogen) and transferred to
   nitrocellulose membrane or 100% methanol-activated PVDF membrane. To
   check the proper transfer, membranes were stained with Ponceau
   (EuroClone Cat. #EMR194500). After Ponceau removal with several washes
   in 1x TBS-T, membranes were blocked with 5% milk in 1x TBS-T for 1 h
   and then incubated with primary antibodies overnight at 4 °C in 1% BSA.
   For anti-human UBIAD1 (Santa Cruz Biotechnology), membrane was blocked
   with 5% BSA and then incubated with primary antibody 3 h at RT.
   Following the incubation, all membranes were washed three times before
   incubation with the appropriate horseradish peroxidase (HRP)-conjugated
   anti-rabbit immunoglobulin G (IgG) (Sigma Cat. #A6154, RRID:
   AB_258284), HRP-conjugated anti-mouse IgG (Sigma Cat. #A4416, RRID:
   AB_258167). The bands were visualized and acquired using ChemiDoc™
   Imaging System (Bio-Rad). The band intensities on developed films were
   quantified using Image Lab software (Bio-Rad Laboratories;
   [297]www.bio-rad.com). Data were normalized on loading control (ACTIN
   or VINCULIN). The following primary antibodies were used: anti-human
   and mouse UBIAD1 (1:1000, Atlas Antibodies Cat. #HPA044862, RRID:
   AB_2679117), anti-human VINCULIN (1:5000, Sigma Cat. #V9131,
   RRID:AB_477629), anti-human and mouse β-ACTIN (1:5000, Sigma Cat.
   #A5316, RRID:AB_476743), anti-human p-FAK (Tyr397) (1:1000, GeneTex
   Cat. #GTX129840, RRID:AB_2886103), anti-human FAK (1:200, Santa Cruz
   Cat. #sc-558, RRID:AB_2300502), anti-human AKT(pan) (1:1000, Cell
   Signaling Cat. #4691, RRID:AB_915783), anti-human p-AKT1 (Ser473)
   (1:1000, Cell Signaling Cat. #9018, RRID:AB_2629283), anti-human p-AKT2
   (Ser474) (1:1000, Cell Signaling Cat. #8599, RRID:AB_2630347),
   anti-human GSK-3β (1:1000, Cell Signaling Cat. #9832,
   RRID:AB_10839406), anti-human p-GSK-3β (Ser9) (1:1000, Cell Signaling
   Cat. #9323, RRID:AB_2115201), anti-human ERK1/2 (1:1000, Santa Cruz
   Cat. #sc-514302, RRID:AB_2571739), anti-human p-ERK1/2 (Thr202/Tyr204)
   (1:1000, Cell Signaling Cat. #9101, RRID:AB_331646), anti-human
   FSP1/AMID (1:1000, Santa Cruz Cat. #sc-377120), anti-human GPX4
   (1:1000, Abcam Cat. #41787, RRID:AB_941790), anti-human UBIAD1 (1:250,
   Santa Cruz Biotechnology Cat. #sc-377013, clone H8, Lot. #I2313),
   anti-human HMGCR (1:500, Santa Cruz Biotechnology Cat. #sc-271595,
   clone C-1, RRID: AB_10650274) and anti-human ERα (1:500, Santa Cruz
   Biotechnology Cat. #sc-542, clone MC-20, RRID:AB_631470).

   For Western blot analysis, human cell lines and mouse ex vivo cells
   were seeded at 80% confluency and allowed to adhere for 4 h. Then,
   cells were treated for 24 h with Vehicle or 0.9 μM ERASTIN, 0.45 μM
   RSL3 for human cells and 50 nM RSL3 for mouse ex vivo cells, before
   proceeding with protein sample collection.

Invasion assay

   Cell invasion was assessed using a permeable support for a 24-well
   plate with an 8.0 µm pore size transparent PET membrane (Falcon Cat.
   #353097). The membrane was coated with 70 μl of ice-cold Growth Factor
   Reduced Matrigel (Corning Cat. #354230) diluted 1 mg/ml in DMEM w/o FBS
   and incubated overnight at 37 °C to allow the Matrigel to solidify. The
   next day, 2 × 10^5 cells were added to the upper chamber in 200 μl of
   medium without FBS and the lower chamber was filled with 650 μl of
   medium supplemented with 10%FBS to create a gradient of FBS. In the
   case of treatment with Vehicle, NAC, CoQ[10], Y27632, and ML7,
   compounds were added at the indicated concentrations to the upper
   chamber. Cells were allowed to invade for 24 h in a 37 °C incubator
   with humidified atmosphere of 5% CO[2]. The chambers were then removed,
   washed twice in PBS and fixed with 4% PFA for 2 min at RT. After two
   additional washes in PBS, the membranes were stained with 0.5% (w/v)
   Crystal Violet staining solution (Sigma) in 20% methanol for 20 min at
   RT. Finally, the excess staining solution was removed with several
   washes in PBS and the not invading cells were gently removed from the
   upper surface of the membrane using a cotton swab. Membranes were let
   to air-dry before proceeding with imaging. Invading cells were
   photographed using Leica DFC300 FX digital camera (software Leica
   Application Suite V4.13) connected to Nikon AZ100 microscope and
   counted with ImageJ software.

Soft agar assay

   12-well plates were coated with a layer of 600 µl phenol red- and
   serum-free DMEM (Thermo Fisher Scientific Cat. #11594416) containing 1%
   low-gelling temperature agarose (Sigma Cat. #A9414). Upon
   solidification, 1 × 10^4 cells/well were seeded in 400 µl phenol
   red-free DMEM containing 0.6% agarose and 4% FBS. Upon solidification,
   1 ml DMEM containing 10% FBS was added on top of the agarose layers.
   The medium was replaced once a week and after 4 weeks the medium was
   substituted with a solution containing 0.5% (w/v) Crystal Violet
   staining solution (Sigma) in 20% methanol diluted 1:100 in PBS for 2
   hrs at RT. The staining solution was removed, and the agarose layers
   were washed twice with PBS before imaging with a Leica MZ16F
   stereomicroscope. The number and area of the colonies were calculated
   with ImageJ software ([298]https://imagej.nih.gov/ij/).

Tumorsphere formation assay

   Tumorsphere formation assay was performed as previously
   described^[299]88. Cells were trypsinized and centrifuged at 200 × g
   for 5 min at RT. The supernatant was removed and cells were resuspended
   in 5 ml of tumorsphere medium: DMEM/F12 GlutaMAX™ medium (Thermo Fisher
   Scientific) containing 100 µg/ml penicillin, 100 µg/mL streptomycin,
   20 ng/ml recombinant human epidermal growth factor (EGF; PeproTech Cat.
   #AF10015), 10 ng/ml recombinant human basic fibroblast growth factor
   (bFGF; Sigma Cat. #F3685) and 1x B27 supplement (Gibco Cat. #7504044).
   The cell suspension was passed through a 40 µm cell strainer to obtain
   a single-cell suspension. Cells were counted and 2 × 10^4 cells/well in
   500 µl of tumorsphere medium were seeded in ultra-low attachment
   24-well plates (Corning Cat. #3473). After 5 days, tumorspheres were
   photographed using Leica DMI4000 microscope and counted with ImageJ
   software. Only tumorspheres greater than 100 µm were considered.

Immunofluorescence

   Tumor cryosections (10 µm-thick) were fixed in 4% PFA for 10 min,
   blocked in PBS containing 5% goat serum (Merk Millipore Cat. #S26), 1%
   BSA and 0.3 M glycine for 30 min. Sections were then incubated with
   anti-mouse CD31 (1:20, Dianova Cat. #DIA-310, RRID:AB_2631039) and
   anti-mouse Ki-67 (1:100, Cell Signaling Cat. #9129-S, RRID:AB_2687446)
   primary antibodies and APC-Cy7 anti-mouse CD45 (1:100, BD Biosciences
   Cat. #557659, RRID:AB_396774) antibody overnight at 4 °C. After three
   washes with PBS, sections were incubated for 1 h at RT with the
   corresponding Alexa Fluor conjugated secondary antibodies (5 µg/ml,
   Thermo Fisher Scientific) and 300 nM DAPI (Thermo Fisher Scientific
   Cat. #D1306) for nuclear staining. Sections were mounted and then
   analyzed using Leica SP8 DLS microscopy. For micro-vessel density and
   mean vessel area measurements, CD31 staining was evaluated in four
   randomly chosen fields per tumor sample. Quantification of Ki-67+ and
   CD45+ cells was assessed by counting all the stained cells of the tumor
   section and normalizing the number for the total tumor section area.

   For cultured cell immunofluorescence, cells were seeded either on glass
   coverslips placed in a 24-wells plate or in 15 μ-Slide 8 well (Ibidi
   Cat. #80826). For the analysis of blebs and lamellipodia, cells were
   seeded after coating with 2.5 μg/ml of Laminin from human placenta
   (Sigma Cat. #L6274) and 3 h after seeding, samples were processed as
   follows. Cells were washed three times with PBS and fixed with 4%
   paraformaldehyde for 15 min at RT. Only for MDA staining, cells were
   not permeabilized and the use of detergents was avoided also in the
   following steps. The fixed cells were then permeabilized with 0.2%
   Triton X-100 in 1× PBS for 10 min at RT and blocked with 3% goat serum
   + 3% BSA + 1.5 mg/ml glycine in PBS for 1 h at room temperature. Cells
   were incubated overnight at +4 °C with anti-human MDA (1:100, Abcam
   Cat. #ab6463, RRID: AB_305484), anti-human Paxillin (1:400,
   Transduction Laboratories Cat. #[300]P13520), anti-human FAK (1:50,
   Santa Cruz Cat. #sc-558, RRID:AB_2300502) diluted in PBS + 3%BSA.
   Filamentous actin (F-actin) was stained with Phalloidin (1:300,
   Sigma-Aldrich Cat. #P1951, RRID:AB_2315148) diluted in PBS for 10 min
   at RT. After three washes with PBS, cells were incubated for 1 h at
   room temperature with the corresponding Alexa Fluor conjugated
   secondary antibodies (0.75–2 µg/ml, Thermo Fisher Scientific) in 0.2%
   BSA/PBS and 300 nM DAPI (Thermo Fisher Scientific) for nuclear
   staining. After three additional washes with PBS, coverslips were
   mounted with Mowiol® 4-88 (Sigma) and then visualized using Leica SP8
   DLS microscopy. Images were processed using Leica LASX software and the
   fluorescence intensity was measured with ImageJ software. The
   percentage of cells without any protrusions, with lamellipodia only, or
   with bleb structures was calculated using samples of cells stained for
   PAX and Phalloidin, or FAK and Phalloidin, by manually counting the
   number of cells in each group. Cells were considered to have blebs if
   two or more bleb structures were visible along the plasma membrane.

Lipid peroxidation measurement

   Lipid peroxidation was evaluated in adherent living cells using
   cholesteryl BODIPY® 576/589 C11 (Thermo Fisher Scientific Cat. #
   [301]C12680). 1.5 × 10^4 cells/well were seeded in the standard medium
   in 96-well Black/Clear Flat Bottom TC-treated Imaging Microplates
   (Corning Cat. #353219) and allowed to adhere for 4 h. Alternatively, 5
   × 10^5 cells/well were seeded in 6-well plates and allowed to adhere
   overnight. Then, cells were treated with 0.45 μM RSL3 (Sigma) or 0.1%
   DMSO (NT) for 12 or 24 h. The medium was removed and cells were
   incubated for 15 min with 3 μM BODIPY® 576/589 C11 and Hoechst 33342
   (1:2000, Thermo Fisher Scientific Cat. #62249) at 37 °C in the dark.
   The staining solution was removed and wells were gently washed three
   times with PBS. The fluorescence signal was analyzed (1) using an
   Infinite 200 PRO plate reader (Tecan) with the Hoechst Ex360/Em465,
   Oxidized BODIPY Ex485/Em535, and Reduced BODIPY Ex535/Em590 filters and
   data were expressed in Oxidized/Reduced BODIPY normalized on Hoechst
   intensity signal, or (2) by flow cytometry analysis using BD FACSCanto™
   II Cell Analyzer (BD Biosciences) with non-treated cells without Bodipy
   C11 staining used as blank and the ratio between oxidized (FITC-A) and
   reduced (PE-A) Bodipy C11 was measured (a minimum of 20,000 events were
   obtained per sample).

   To evaluate the lipid peroxidation in fixed cells, immunofluorescence
   staining using anti-MDA an antibody (Abcam) was performed as described
   in the relative section. Lipid peroxidation was induced by treating the
   cells for 24 h with 2 μM ERATIN (Sigma) or 0.1%DMSO (NT) before
   fixation.

Growth in attachment-free condition

   6-well plates were coated with 1 ml of 1% low-gelling temperature
   agarose (Sigma Cat. #A9414) in standard medium w/o FBS and the agarose
   was let solidify for 1 h at +4 °C. Then, 1 × 10^5 cells/well were
   seeded on top of the agarose layer in 2 ml of complete growth medium
   and allowed to grow in suspension for 2–6 days. Before resuspending the
   cells in TRIzol™ Reagent (Thermo Fisher Scientific Cat. #15596018), the
   cell suspension was collected, cells were pelleted by centrifugation at
   300 × g for 5 min at +4 °C and the pellet was washed three times with
   cold PBS. Then, we proceeded with RNA extraction as described.

Atomic force microscopy (AFM)

   MDA-MB-231, MCF10A, MCF7, and MDA-MB-468 cells were seeded in a
   complete growth medium in 35 mm dishes (Falcon Cat. #353001) to have
   30–40% confluency the day after. Immediately before the analysis, the
   culture medium was replaced with 3 mL of PBS, after 2 previous washes
   with PBS. AFM force indentation measurements were performed using a XE
   Bio - AFM (Park Systems, South Korea) equipped with an inverted
   microscope (Nikon Eclipse Ti).

   The force-displacement curves were acquired using PPP-CONTSCR-10
   pyramidal tips mounted on Si3N4 cantilevers with a nominal spring
   constant of ~0.2 N/m (NanoSensors, Neuchatel, Switzerland). Values for
   the cantilever spring constants were obtained by the manufacturer
   before use and were used in all calculations. Before each test, the
   sensitivity of the AFM photodetector (optical lever sensitivity) was
   calculated by measuring the slope of the force-distance curve acquired
   on a silicon standard. All experiments were performed at room
   temperature.

   Indentation curves were acquired by approaching the cell surface at a
   rate of 5 μm/s and producing an indentation with a depth of 0.5 μm. For
   each cell, at least two-force curves were collected.

   Young’s modulus was calculated by applying a fit of the Hertz model to
   each force-distance curve, assuming a Poisson ratio of 0.5.

RNA-seq analysis

   MDA-MB-231 cells stably transduced with lenti_SAM_v2_UBIAD1-gRNA or
   lenti_SAM_v2_scamble-gRNA were seeded in a complete medium to obtain
   70% confluency the day after. Cells were washed three times with
   autoclaved PBS and resuspended in TRIzol™ Reagent (Thermo Fisher
   Scientific Cat. #15596018). Four biological replicates per condition
   were collected. Total RNA was extracted according to TRIzol™ Reagent
   manufacturer’s instructions. 1 μL of DNase I (Thermo Fisher Scientific,
   Cat. #18068015) was used for every 1 μg of isolated RNA to digest any
   remaining genomic DNA. After DNase I inactivation, RNA Clean &
   Concentrator-5 kit (Euroclone, Cat. #R1015) was used to purify the RNA
   samples. RNA concentration and purity were determined using a NanoDrop
   spectrophotometer at A260 and A280/260, respectively, while integrity
   was verified by Tapestation System (AGILENT). Only samples with RIN > 8
   were used for the following RNA-seq analysis. For library preparation,
   total RNA was quantified using the Qubit 4.0 fluorimetric Assay (Thermo
   Fisher Scientific). Libraries were prepared from 125 ng of total RNA
   using the NEGEDIA Digital mRNA-seq research grade sequencing service
   (Next Generation Diagnostic srl)^[302]89 which included library
   preparation, quality assessment, and sequencing on a NovaSeq 6000
   sequencing system using a single-end, 100 cycle strategy (Illumina
   Inc.).

   The raw data were analyzed by Next Generation Diagnostic srl
   proprietary NEGEDIA Digital mRNA-seq pipeline (v2.0) which involves a
   cleaning step by quality filtering and trimming, alignment to the
   reference genome, and counting by gene^[303]90,[304]91. The raw
   expression data were normalized, analyzed, and visualized by Rosalind
   HyperScale architecture^[305]92 (OnRamp BioInformatics, Inc.) Only
   genes with −1.3 ≥ fold-change ≤ 1.3 and p-Adj ≤ 0.05 were considered
   differentially expressed and shown as a heatmap or volcano plot.

   Starting from the differentially expressed genes between Scramble and
   UBIAD1^OE a Gene Ontology (GO) and KEGG pathway enrichment analysis
   were performed using ShinyGO 0.76.3
   ([306]http://bioinformatics.sdstate.edu/go/) with FDR cutoff 0.05 and
   minimum pathway size 2. Enrichment was calculated relative to a set of
   background genes relevant to the experiment.

HPLC analysis of CoQs level

   Cells were resuspended in standard medium and seeded in a 100 mm petri
   dish to reach 70% of confluency the next day. After three washes with
   ice-cold PBS, cells were scraped and resuspended in 250 µl ice-cold
   PBS. 10 µl of cells/PBS suspension were put aside to quantify the
   protein content to normalize the quantification of CoQ[10] or CoQ[9].
   The remaining sample was immediately frozen in liquid nitrogen and
   stored at −80 °C. To extract CoQs from cells, 100 µl of cell suspension
   in PBS containing 30ug of protein was transferred to 300 µl of ice-cold
   extraction solution (ethanol/hexane 1:2) containing 0.2 µM of CoQ[9]
   (Sigma Cat. #27597) or CoQ[6] (Avanti Cat. #900150 O) as an internal
   standard for human and mouse cells, respectively, followed by
   vortexing. The upper, CoQ containing, the hexane layer was separated by
   centrifugation (5 min, 13,200 × g, +4 °C) and added to a glass mass
   spectrometry vial. The extraction procedure was repeated twice. The
   hexane was dried down under a stream of nitrogen and dried samples were
   then resuspended in 100 µl of methanol. The calibration standards were
   prepared by serial dilutions of stock solution of CoQ[10] or CoQ[9] in
   methanol within the range from 2 to 1000 nM. Internal standard (CoQ[9]
   or CoQ[6]) was spiked in each calibration standard at a concentration
   of 0.2 µM. The UHPLC-MS/MS analysis was performed on a Hybrid
   quadrupole-Orbitrap mass spectrometer (Q Exactive, Thermo Scientific)
   coupled to a UHPLC system (Ultimate 3000, Thermo Dionex) via a heated
   electrospray ionization source. Chromatographic separation of targeted
   analytes and internal standard was achieved on an Accucore C18
   100 × 2.1 (2.6 µ particle size) column with a gradient elution of 10 mM
   of ammonium formate in water (phase A) and 10 mM of ammonium formate in
   methanol/2-propanol 80:20 (phase B) at a flow rate of 250 µl/min. The
   sample injection volume was 4 µl.

   Detection and determination were performed in full MS/ddMS2 mode with
   positive electrospray ionization mode. The mass spectrometer was
   calibrated before analysis using a commercial calibration solution to
   maintain mass accuracy below 5 ppm. CoQs were detected as [M + H] + and
   [M + NH4] + adducts ions. CoQs concentration was quantified relative to
   the internal standard by an external calibration curve. Results are
   expressed as ng of CoQ[10] or CoQ[9] normalized over mg of total
   proteins.

Quantification of lipid classes

   Cells were resuspended in standard medium (10% FBS) and seeded in a
   100 mm petri dish to reach 70% of confluency the next day. After three
   washes with ice-cold PBS, cells were scraped and resuspended in 250 µl
   ice-cold PBS. 10 µl of cells/PBS suspension were put aside to quantify
   the protein content to normalize the quantification of lipid classes.
   The remaining sample was immediately frozen in liquid nitrogen and
   stored at −80 °C.

   For lipid class quantification the following internal standard was
   used: trinonanoylglycerol (Sigma Cat. #92909) and
   tripentadecanoylglycerol (Sigma Cat. #T4217) for triacylglycerols (TG),
   5-α-cholestane (Sigma Cat. #C8003) for cholesteryl esters (CE), (3α,
   5β)- cholestan-3-ol (epicoprostanol) (Steraloids Inc. Cat. #C5050-000)
   for free cholesterol (CO).

   The trimethylsilylating agents, N,
   O-Bis(trimethylsilyl)trifluoroacetamide (BSTFA) with
   trimethylchlorosilane (Cat. #15238) and HCl methanol (Cat. #90964) were
   purchased from Merk. All organic solvents were of analytical grade and
   were purchased from Merk. The fatty acid reference standard mixture
   (GLC 461) was supplied by Nuchek Prep (Cat. #GLC-461).

   Lipids were extracted from 100 µl of a sample using a
   chloroform-methanol solution by the Folch method^[307]93. Lipid
   extracts were resolved in classes by thin layer chromatography as
   previously reported^[308]94.

   The TG fraction was hydrolyzed with 2 ml of HCl-methanol (5%) and the
   resulting free fatty acids were trans-esterified and extracted with
   hexane. Hexane-containing fatty acid methyl esters (FAME) was directly
   taken for gas chromatographic analysis.

   In brief, CO and CE was eluted from the silica with 5 ml of
   chloroform-methanol 2:1. The solvent was collected and evaporated to
   dryness under a stream of nitrogen. CE were hydrolyzed by
   saponification and derived free CO was extracted with hexane. Finally,
   both dried residues containing CO were derivatized in 50 µl of BSTFA
   for 30 min at room temperature, and 1 µl of the resulting mixture was
   injected for GC-MS analysis.

   The sterols were quantified by gas chromatography-mass spectrometry
   (GC-MS). The instrument used was Agilent 6890 GC coupled with Agilent
   5973 inert MS on the single ion monitoring (SIM) mode. Samples were
   injected in splitless mode at 270 °C and separated through an HP-5ms
   column (30 m × 0.25 mm internal diameter × 0.25 µm film thickness;
   Agilent Technologies, Folsom, CA, USA). The oven temperature was held
   initially at 200 °C for 1 min, ramped to 275 °C at 10 °C/min, increased
   to 277 °C at 0.1 °C/min, and finally increased to 290 °C at 10 °C/min
   and held for 3 min. Peak identification was achieved by comparing the
   retention time and matching the height ratios of the characteristic
   ions. The internal standard method was used for quantitative analysis.

   Gas chromatography analysis was performed with an Agilent 5890 gas
   chromatograph (GC) equipped with a flame-ionization detector. FAMEs
   derived from TG were resolved with an Omegawax column (30 m × 0.25 mm
   internal diameter × 0.25 µm film thickness; Supelco) with 1 µl
   injection volume running in on-column mode. The oven temperature was
   programmed as follows: 60 °C for 3 min, increased 20 °C/min to 205 °C,
   which remained constant for 15 min. The temperature then increased
   0.4 °C/min up to 213 °C, which was maintained for 10 min and finally
   increased to 240 °C at 5.0 °C/min and held for 8 min. Peaks were
   identified to a reference standard mixture (GLC 461, Nuchek Prep). The
   internal standard method was used for quantitative analysis. Results
   are expressed as pg of TGs, CE, or CO per mg of proteins.

Quantification and statistical analysis

   Statistical analysis was performed in GraphPad Prism 8.0 software and R
   (version 4.1). The values reported in the figures represent ± SEM
   unless otherwise stated. The exact value of sample size (n) is given in
   the figure legends or represented as single dots in the figures. Three
   or more groups were analyzed with one-way ANOVA and Dunnett’s post-test
   or using Sidak’s multiple comparisons tests. Two-way ANOVA was used to
   compare data with two variables. Two-tailed unpaired t-test was used to
   compare the two groups. One-sample t-test was used to compare a group
   to a hypothetical value when values were expressed as relative to the
   control and the control was set at 1.0 or 100. Multivariate Hazard
   ratio (HR) was calculated with 95% confidence interval using
   multivariate Cox proportional hazards regression after adjustment for
   age, pT (Tumor size), pN (Regional lymph nodes), ER (Estrogen
   receptor), HER2 (Erb-B2 Receptor Tyrosine Kinase 2/HER2), tumor grade,
   patient age at diagnosis. Fisher’s exact test was used to determine
   statistical significance in contingency tables. Statistical
   significance is reported as exact p-value, adjusted-p-value or n.s.,
   when not significant (p-value or adjusted-p-value > 0.05). Additional
   statistical details can be found in the figure legends.

   The terms ‘independent experiment’ and ‘biological replicate’ present
   in the figure legends refer to experiments in which seedings,
   treatments, stainings, and analyses were conducted independently, using
   different cell preparations and working solutions.

Reporting summary

   Further information on research design is available in the [309]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [310]Supplementary Information^ (2.6MB, pdf)
   [311]Reporting Summary^ (212.2KB, pdf)
   [312]Peer Review File^ (3.5MB, pdf)

Source data

   [313]Source Data^ (4.9MB, xlsx)

Acknowledgements