Glycosaminoglycan-driven lipoprotein uptake protects tumours from ferroptosis

Abstract

Lipids are essential components of cancer cells due to their structural and signalling roles¹. To meet metabolic demands, many cancers take up extracellular lipids^2-5; however, how these lipids contribute to cancer growth and progression remains poorly understood. Here, using functional genetic screens, we identify uptake of lipoproteins-the primary mechanism for lipid transport in circulation-as a key determinant of ferroptosis sensitivity in cancer. Lipoprotein supplementation robustly inhibits ferroptosis across diverse cancer types, primarily through the delivery of α-tocopherol (α-toc), the most abundant form of vitamin E in human lipoproteins. Mechanistically, cancer cells take up lipoproteins through a pathway dependent on sulfated glycosaminoglycans (GAGs) linked to cell-surface proteoglycans. Disrupting GAG biosynthesis or acutely degrading surface GAGs reduces lipoprotein uptake, sensitizes cancer cells to ferroptosis and impairs tumour growth in mice. Notably, human clear cell renal cell carcinomas-a lipid-rich malignancy-exhibit elevated levels of chondroitin sulfate and increased lipoprotein-derived α-toc compared with normal kidney tissue. Together, our study establishes lipoprotein uptake as a critical anti-ferroptotic mechanism in cancer and implicates GAG biosynthesis as a therapeutic target.

Extended Data Figure 1.. The lipid fraction of lipoproteins renders GPX4 dispensable in cancer cells.

a, b. Cellular uptake of DiI-labeled LDL (a) or HDL (b) in HeLa cells, quantified by median PE fluorescence intensity (left) and shown in representative histograms (right). Cells treated with DiI-LDL or DiI-HDL (blue) are compared to untreated controls (grey). c. Differential gene scores in HeLa cells cultured in lipoprotein-depleted serum (LPDS) with or without lipoprotein supplementation. Anti-ferroptotic genes and lipid metabolism genes are highlighted. d. Individual sgRNA scores (log₂) for GPX4 in the CRISPR screen in HeLa cells under indicated conditions. e, f. Immunoblot analysis of GPX4 in human (e) or mouse (f) isogenic cell lines with sgControl or GPX4/Gpx4 knockout (KO). GAPDH is included as loading control. g. Proliferation (log₂ doublings, 5 days) of indicated GPX4-deficient cell lines in vitro either untreated or supplemented with cholesterol (5 μg/mL), LDL (50 μg/mL), HDL (50 μg/mL) or Fer-1 (1 μM). h. Uncropped co-autoxidation of STY-BODIPY (1μM) and liposomal egg phosphatidylcholine with cholesterol, initiated by DTUN (0.2 mM), and inhibited by 50 μg/mL LDL, HDL, reconstituted HDL with Tristearin (rHDL-TG(18:0)₃), or PMC (4 μM). i. Experimental approach to replace the non-polar lipid core of native lipoproteins. Native lipoproteins are first stripped of their non-polar core, and then reconstituted with tristearin (TG(18:0)₃). j. Proliferation (log₂ doublings, 5 days) of indicated GPX4-deficient cell lines in vitro either untreated or supplemented with cholesterol (5 μg/mL), combined native LDL and HDL (50 μg/mL), combined tristearin-reconstituted LDL and HDL (rLDL+rHDL, 50 μg/mL), or Fer-1 (1 μM). a, b, g, j: Bars represent mean ± s.d.; a, b, g, j: n=3 biological replicates. Statistics by two-sided unpaired t-tests as indicated or compared to untreated cells (a, b, g, j). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 2.. The antioxidant effect of lipoproteins is independent of MUFAs and vitamin K.

a. Proliferation (log₂ doublings, 5 days) of indicated GPX4_KO cell lines in vitro either untreated or supplemented with Fer-1 (1 μM), α-toc (10 μM), vitamin K2 (10 μM), or OA (250 μM). b. Proliferation (log₂ doublings, 5 days) of indicated GPX4-deficient cell lines in vitro left untreated or supplemented with α-toc, vitamin D3, or 7-DHC at indicated concentrations. c. Proliferation (log₂ doublings, 5 days) of A-498 cells in vitro either treated with ferroptosis inducers ML162 (left) or erastin (right), with or without α-toc, vitamin D3, or 7-DHC at indicated concentrations. d. Cellular lipid oxidation ratio using BODIPY-C11 in A-498 cells untreated or treated with ML162 (125 nM) ± LDL, HDL, α-toc (10 μM), vitamin K2 (10 μM), vitamin D3 (10 μM), CoQ10 (10 μM), OA (250 μM) or Fer-1 (1 μM). e. Cellular lipid oxidation ratio using BODIPY-C11 in A-498 cells ± LDL, HDL, α-toc (10 μM), vitamin K2 (10 μM), OA (250 μM) or Fer-1 (1 μM). f. Cellular lipid oxidation ratio using BODIPY-C11 in Karpas299 cells untreated or treated with ML162 (125 nM) ± LDL, HDL, α-toc (10 μM), vitamin K2 (10 μM), vitamin D3 (10 μM), vitamin A (10 μM), CoQ10 (10 μM), OA (250 μM) or Fer-1 (1 μM). g. Schematic of genes required for cellular utilization of antioxidant lipids carried by lipoproteins. h. Immunoblot of ACSL3 and AIFM2 in the indicated cell lines transduced with a sgControl, sgACSL3 or sgAIFM2. GAPDH is the loading control. i, j. Cellular lipid oxidation ratio using BODIPY-C11 in Karpas299 (left) or A-498 (right) cell lines transduced with sgACSL3 (i) or sgAIFM2 (j) versus sgControl under GPX4 inhibition (ML162: 75 nM for Karpas299, 175 nM for A-498) and HDL supplementation. For all LDL and HDL experiments, 50 ug/mL was used. a-c, Bars represent mean ± s.d.; d-f, i, j, bars represent median; a-f, i, j, n=3 biological replicates. Statistics by two-sided unpaired t-tests as indicated or compared to untreated cells (a-c, e) or ML162 treated cells (d, f). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 3.. α-Toc is the primary lipid driving lipoprotein-mediated antioxidant protection.

a. Rank of significantly altered lipid metabolites (detected in negative mode) in B16 cells treated with HDL versus LDL. b. LC/MS analysis of eight vitamin E isoforms in HeLa cells supplemented with human LDL and HDL (50 μg/mL each), shown as fold change relative to lipoprotein-depleted cells. c. LC/MS analysis of eight vitamin E isoforms in HeLa cells treated with LDL or HDL (50 μg/mL), shown as fold change relative to lipoprotein-depleted cells. d. Relative metabolite peak area for eight vitamin E isoforms in plasma from mice fed a control or Vitamin E-deficient diet. e. Proliferation (log₂ doublings, 5 days) of Karpas299 in vitro untreated or treated with ML162 (125 nM) ± 50 ug/mL of human HDL, HDL isolated from mice fed a vitamin E-deficient diet, HDL isolated from mice fed a vitamin E-sufficient diet, or Fer-1 (1 μM). f. Representative tumours of Karpas299, HeLa, and B16 cells implanted subcutaneously in mice fed a control or vitamin E-deficient diet. g. Quantification of circulating plasma triglycerides (TGs, mg/dL) and HDL (mg/dL) in mice fed a control or a vitamin E-deficient diet. h. Schematic depicting generation of Ttpa-liver specific knockout mice to evaluate depletion of α-toc in driving tumour growth in mice. i. Immunoblot of TTPA in the liver of mice injected with liver-specific AAV8-sgControl or AAV8-sgTtpa. Ponceau staining is used as loading control. j. Plasma α-toc levels from Cas9 knock-in mice injected with liver-specific AAV8-sgControl or AAV8-sgTtpa 4 weeks after AAV injection. k. Quantification of 4-HNE H-scores in B16 tumours grown in control or liver-specific Ttpa knockout mice. l. Tumour weights from control and liver-specific Ttpa knockout mice implanted with B16 cells. m. Mass spectrometry analysis of vitamin E isoforms α-tocotrienol and δ-tocotrienol in plasma from mice fed a standard mouse diet or a Vitamin E-sufficient diet. Data is presented as relative metabolite peak area. b-e, g, k, m: Bars represent mean ± s.d.; j, l:Boxes represent median, first/third quartiles; whiskers are range. b, c, e, n=3 biological replicates; d: n=4 biological replicates; g: n=7 biological replicates; j: n=5 biological replicates; k: n=14 representative fields; l: n=10–12 biological replicates; m: n=3–4 biological replicates. Statistics by: i) two-sided unpaired t-tests as indicated or compared to untreated cells (b, c), control-diet mice (d, g), or to ML162-treated cells (e); or ii) by one-way ANOVA followed by a Kruskal–Wallis nonparametric test (a). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 4.. Loss of UGDH sensitizes cancer cells to ferroptosis in vitro.

a. Gene essentiality plot showing genes scores in untreated (x-axis) and ML210-treated (GPX4 inhibition) Karpas299 cells. b. Differential gene scores in Karpas299 cells ML210-treated CRISPR screen. Negative scores represent genes whose loss potentiates ML210 toxicity; positive scores represent genes whose loss provides resistance to ML210. GAG biosynthesis, canonical anti-ferroptotic, and lipid metabolism genes are highlighted. c. Differential gene scores in Karpas299 cells for DiI-LDL uptake CRISPR screen. Negative scores represent genes whose loss reduce cellular DiI-LDL uptake. GAG biosynthesis and canonical lipoprotein uptake genes are highlighted. d. Individual sgRNA scores for UGDH in untreated and ML210-treated conditions from Karpas299 screen. e. Individual sgRNA scores for UGDH in high and low DiI-LDL uptake populations from Karpas299 screen. f. Immunoblot of UGDH in the indicated isogenic cell lines transduced with sgUGDH and expressing or not a sgRNA-resistant UGDH cDNA. GAPDH included as a loading control. g. Proliferation (log₂ doublings, 5 days) of indicated HeLa (left) or Caki-2 (right) cell lines transduced with sgControl or sgUGDH under the indicated concentrations of ML162. h. Cellular lipid oxidation ratio using BODIPY-C11 in A-498 (left) or Karpas299 (right) UGDH_KO cell lines expressing a sgRNA resistant UGDH cDNA under GPX4 inhibition (ML162: 200 nM for Karpas299, 250 nM for A-498) and Fer-1 supplementation (1 μM). i. Proliferation (log₂ doublings, 5 days) of indicated UGDH-deficient cell lines expressing a sgRNA resistant UGDH cDNA or an empty vector under the indicated concentrations of erastin. g, i: Bars represent mean ± s.d.; h: bars represent median; g-i: n=3 biological replicates. Statistics by two-sided unpaired t-tests compared to untreated empty vector transduced cells (g-i). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 5.. UGDH expression is a major determinant of LDL and HDL uptake in cancer cells.

a. Proliferation (log₂ doublings, 5 days) of indicated UGDH-deficient cell lines expressing a sgRNA resistant UGDH cDNA or an empty vector under treatment with the indicated concentrations of ferrous ammonium citrate (FAC). b. Immunoblot of TFRC (top) and IRP1 (bottom) in the indicated UGDH_KO cell lines expressing a sgRNA resistant UGDH cDNA or an empty vector treated or not with FAC (0.1 mg/mL). GAPDH included as a loading control. c. Immunoblot of GPX4 (left), SLC7A11 (center), and AIFM2 (right) in the indicated UGDH_KO cell lines expressing a sgRNA resistant UGDH cDNA or an empty vector. GAPDH included as a loading control. d. Histograms of DiI-LDL uptake in Karpas299 UGDH_KO cells expressing sgRNA-resistant UGDH cDNA or vector control. e, f. Cellular uptake of DiI-labeled LDL (e) or HDL (f) in the indicated Karpas299 (left), A-498 (center) and HeLa (right) UGDH_KO cells expressing a sgRNA resistant UGDH cDNA or an empty vector, quantified by median PE fluorescence intensity. g. Mass spectrometry analysis of isotope labeled ([3-¹³C]Cholesterol, left) and unlabeled (right) cholesterol in A-498 UGDH_KO cells expressing a sgRNA resistant UGDH cDNA or an empty vector cultured in the presence of native LDL (grey) or ¹³C-Cholesterol-labeled LDL for 4 hours (dark blue) and 8 hours (light blue). Data is presented as fold of the control LDL condition. a, e-g, Bars represent mean ± s.d.; a, e-g, n=3 biological replicates. Statistics by two-sided unpaired t-tests compared to empty vector transduced cells (a, e-g). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 6.. GAG sulfation and xylose synthesis are essential for lipoprotein-mediated ferroptosis resistance in lymphoma cells.

a. Schematic showing that GAG sulfation depends on PAPSS1-derived 3′-phosphoadenylylsulfate and its Golgi import via SLC35B2. b. Individual sgRNA scores (log₂) for PAPSS1 in untreated and ML210-treated conditions (left) or in high and low DiI-LDL populations (right) of the Karpas299 CRISPR screens. c. Immunoblot of PAPSS1 in Karpas299 cells transduced with a sgControl or sgPAPSS1. GAPDH included as a loading control. d. Quantification of heparan sulfate (HS) in Karpas299 cells transduced with a sgControl or sgPAPSS1. e. Quantification of total sulfate per glycosaminoglycan disaccharide in Karpas299 cells transduced with a sgControl or sgPAPSS1. f. Proliferation (log₂ doublings, 5 days) of Karpas299 cells transduced with a sgControl or sgPAPSS1 under the indicated concentrations of ML162. g. Schematic showing UGDH-driven synthesis of UDP-GlcUA and its conversion by UXS1 into UDP-xylose, the initiating sugar of the GAG linker to proteoglycans. h. Individual sgRNA scores (log₂) for UXS1 in the presence and absence of ML210 (left), or in high and low DiI-LDL populations (right) of Karpas299 CRISPR screens. i. Sanger sequencing of UXS1 gene exon 9 in Karpas299 parental cells (WT) or transduced with sgUXS1. j. Quantification of total milligrams of HS per gram of protein in Karpas299 cells transduced with a sgControl or sgUXS1. k. Proliferation (log₂ doublings, 5 days) of Karpas299 cells transduced with a sgControl or sgUXS1 under the indicated concentrations of ML162. l. Cellular uptake of DiI-LDL in Karpas299 cells transduced with a sgControl, sgUGDH, sgPAPSS1 or sgUXS1, quantified by median PE fluorescence intensity. d-f, j-l, Bars represent mean ± s.d.; d-f, j-l, n=3 biological replicates. Statistics by two-sided unpaired t-tests compared to sgControl cells (d-f, j-l). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 7.. The anti-ferroptotic effect of UGDH is essential for tumour growth.

a. Representative images of subcutaneous tumours of HeLa UGDH_KO cells expressing a sgRNA resistant UGDH cDNA or an empty vector. b. Tumour weights resulting from implantation of the indicated UGDH_KO cell lines in immunocompromised mice. c. Representative IHC images (20X magnification) of CD45, a pan-immune marker, in tumours resulting from subcutaneous implantation of Karpas299 cells expressing a sgRNA-resistant UGDH cDNA or an empty vector. Scale bars, 100 μm. d. Quantification of CD45 H-scores in tumours described in (c). e. Immunoblot of UGDH in the indicated isogenic cell lines transduced with a doxycycline-inducible shGFP or a shUGDH construct ± doxycycline treatment (doxy, 1 μg/mL) for 24 hours. GAPDH included as a loading control. f. Representative images of subcutaneous tumours of Karpas299 cells expressing a doxycycline-inducible shRNA against either GFP (control) or UGDH implanted in mice given sucrose-water with or without doxycycline (doxy, 2 g/L). g. Tumour weights resulting from implantation of Karpas299 cells expressing a doxycycline-inducible shRNA against either GFP (control) or UGDH in immunocompromised mice given sucrose-water with or without doxycycline (doxy, 2 g/L). h. Tumour weights resulting from implantation of the indicated UGDH_KO cell lines expressing a sgRNA-resistant UGDH cDNA treated with daily intraperitoneal injection of vehicle or Lip-1. i. Fold change in tumour weight relative to vehicle-treated empty vector controls for Karpas299 UGDH_KO cells expressing UGDH cDNA or an empty vector in mice treated with vehicle or Lip-1. b, g-i: Boxes represent median, first and third quartiles, and whiskers are range; b, g-i: n=10 biological replicates. Statistics by two-sided unpaired t-tests compared to empty vector transduced cells (b, d), shGFP-expressing cells (g), or vehicle-treated mice (h, i). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 8.. Contribution of GAGs and lipoprotein receptors to lipoprotein uptake and tumour growth.

a. Immunoblot of LDLR, SCARB1, and UGDH in HeLa cells transduced with the indicated sgRNAs. GAPDH included as a loading control. b, c. Fold change in the uptake of DiI-labeled LDL (b) and HDL (c) of the indicated HeLa cells relative to sgControl-transduced cells. d. Immunoblot of LRP8 in 786-O cells transduced with a sgControl or sgLRP8. ACTINB included as a loading control. e. Fold change in the uptake of DiI-LDL (left) and DiI-HDL (right) of the indicated 786-O cells relative to sgControl-transduced cells. f. Proliferation (log₂ doublings, 5 days) of the indicated 786-O cells under ML162 treatment. g. Immunoblot of LDLR and SCARB1 in HeLa LDLR_KO (left) and SCARB1_KO (right) cells expressing a sgRNA-resistant LDLR, SCARB1 cDNA or empty vector. ACTINB and vinculin included as loading controls. h. Fold change in the uptake of DiI-LDL and DiI-HDL by the indicated HeLa cells relative to cells expressing an empty vector. i. Proliferation (log₂ doublings, 5 days) of LDLR_KO and SCARB1_KO cells expressing the indicated cDNAs or an empty vector (grey) under the indicated concentrations of ML162. j. Representative images of tumours resulting from implantation of HeLa UGDH-KO, LDLR_KO, or SCARB1_KO cells expressing the indicated cDNAs or empty vector. k. Fold change in tumour weight relative to isogenic KO tumours for HeLa UGDH_KO, LDLR_KO, or SCARB1_KO cells expressing the indicated cDNAs or empty vector. b, c, e, f, h, i: Bars represent mean ± s.d.; k: Boxes represent median, first and third quartiles, and whiskers are range; b, c, e, f, h, i: n=3 biological replicates; k: n=10 biological replicates. Statistics by two-sided unpaired t-tests compared to sgControl-expressing cells (b, c, e, f) or empty vector transduced cells (h, i, k). For gel source data, see Supplementary Fig. 1.

Extended Data Figure 9.. The proteoglycan VCAN modestly increases resistance to ferroptosis and lipoprotein uptake in cancer cells.

a. Schematic of CRISPR screens in 786-O (ccRCC) and Karpas299 (lymphoma) cells transduced with a proteoglycan-focused sgRNA library (55 genes). 786-O cells were subjected to a proliferation-based screen in the presence or absence of the GPX4 inhibitor, ML162 (left). Karpas299 cells were subjected to flow cytometry-based cell sorting for high and low fluorescent populations after DiI-LDL treatment. b. Differential gene scores from the 786-O CRISPR screen under ML162 treatment relative to untreated cells. Negative scores indicate genes whose loss sensitizes cells to ML162. UGDH served as a positive control. c. Differential gene scores in high DiI-LDL uptake versus low uptake CRISPR screen in Karpas299 cells. Negative scores indicate genes whose loss reduces LDL uptake. UGDH served as a positive control. d. Quadrant map showing overlapping hits from both screens; genes essential in both are highlighted, with VCAN emerging as a shared hit. e. Sanger sequencing of VCAN gene exon 9 in Karpas299 parental cells (WT) or transduced with sgVCAN. f. Proliferation (log₂ doublings, 5 days) of 786-O transduced with a sgControl or sgVCAN under the indicated concentrations of ML162. g. Proliferation (log₂ doublings, 5 days) of Karpas299 cells transduced with a sgControl or sgVCAN under the indicated concentrations of ML162. h. Cellular uptake of DiI-LDL in the indicated Karpas299 cells transduced with a sgControl or sgVCAN, quantified as median PE intensity. i. Tumour weights resulting from implantation of Karpas299 cells transduced with a sgControl or sgVCAN in mice. f-h: Bars represent mean ± s.d.; i: Boxes represent median, first and third quartiles, and whiskers are range. f-h, n=3 biological replicates; i, n=10 biological replicates. Statistics by two-sided unpaired t-tests compared to sgControl-expressing cells (f-i).

Extended Data Figure 10.. Cell-surface sulfated GAGs promote lipoprotein uptake.

a, b. Total cell surface HS (a) or CS (b) measured in A-498 cells treated or not with heparinases (0.1 U/mL) or chondroitinase (0.1 U/mL). c, d. Fold change in DiI-labeled LDL (c) and HDL (d) uptake of A-498 cells upon treatment with heparinases (0.1 U/mL), chondroitinase (0.1 U/mL), or both (blue), relative to uptake of untreated cells. e. Total cell surface hyaluronic acid (HA) measured in A-498 UGDH_KO cells expressing an UGDH cDNA or empty vector. f. Total cell surface HA (left), HS (middle), and CS (right) measured in A-498 cells treated or not with hyaluronidase (5 U/mL). g. Fold change in DiI-LDL uptake of A-498 cells treated with hyaluronidase (5 U/mL) or combined heparinases/chondroitinase (0.1 U/mL) relative to uptake of untreated cells. h. Representative image of HS immunostaining in A-498 cells treated or not with combined heparinases/chondroitinase (0.1 U/mL). Nuclei stained with DAPI. Scale bars, 20 μm. i. Representative image of Laurdan staining in A-498 cells treated or not with combined heparinases/chondroitinase (0.1 U/mL). Generalized polarization (GP) values represent membrane rigidity (higher value, red) or fluidity (lower value, blue). j. Histogram of GP values from images from experiment in (i). k. Fold change in pHrodo-transferrin uptake in A-498 cells treated or not with combined heparinases/chondroitinase (0.1 U/mL). l. Histogram of DiI-LDL uptake in cell suspension of 786-O tumours resected from mice treated or not with combined heparinases/chondroitinase. m. Uptake of DiI-LDL in the tumour suspension in (l). n, o. Representative IHC images (n) (20x magnification) and H-score quantification (o) of CD45 in Karpass299 tumours injected daily with HBSS or combined enzymes (10 U/mL). Scale bars, 100 μm. a-g, k, m, o: Bars represent mean ± s.d.; a-g, k, m: n=3 biological replicates; o: n=25 representative fields. Statistical significance determined by two-sided unpaired t-tests as indicated or compared to untreated cells (a-d, f-g, k, m, o) or empty vector expressing cells (e).

Extended Data Figure 11.. Proteoglycan expression in human ccRCCs and the role of the GAG-lipoprotein axis in ferroptosis resistance.

a. Violin plot showing the relative levels of cholesterol in ccRCC patient tissues compared to paired adjacent kidney. b. Violin plot showing the concentration (ng/mL) of α-toc in the plasma of ccRCC patients and healthy donors. c. Violin plot showing the relative levels of total HS per gram of protein in ccRCC patient tissues compared to paired adjacent kidney. d. Heatmap showing expression of individual proteoglycan genes in ccRCC tumours and paired adjacent kidney. Each sample was assigned a proteoglycan score, based on their expression of all proteoglycans. Relevant proteoglycans are highlighted in blue. e. Rank of proteoglycans by fold change in gene expression in ccRCC tumours relative to paired adjacent kidney. Relevant proteoglycans are highlighted in blue. f. Fold change in DiI-LDL uptake by melanoma PDX cell suspensions treated with heparinases and chondroitinase (1 U/mL of both, dark blue) relative to untreated samples (grey). Each dot represents a different PDX. f: Bars represent mean ± s.d.; a: n=20 biological replicates; b: n=8 biological replicates; c: n=18 biological replicates; f, n=3 different melanoma PDXs. Statistical significance was determined by a two-sided unpaired t-test compared to non-cancer samples (a, b, c), or untreated cells (f).

Figure 1.. Lipoprotein delivery of α-toc promotes cancer cell resistance to ferroptosis.

a. Schematic of focused CRISPR screen in HeLa cells under lipoprotein-depleted or -repleted conditions. b. Gene essentiality graph showing changes based on lipoprotein availability. c. Rank of genes by change in essentiality; anti-ferroptotic genes are highlighted (left). Diagram of the glutathione peroxidase pathway inhibiting lipid peroxidation (right). d. Proliferation (log₂ doublings, 5 days) of GPX4_KO HeLa cells treated with PBS, cholesterol (5 μg/mL), LDL, HDL, or Fer-1 (1 μM). e. Heatmap of proliferation (log₂ doublings, 5 days) of GPX4/Gpx4_KO cell lines treated with PBS, LDL, HDL, or Fer-1 (1 μM). f. Heatmap of proliferation (log₂ doublings, 5 days) in response to ML162 (GPX4 inhibitor) with or without LDL or HDL. g. Representative co-autoxidation of STY-BODIPY (1 μM) and liposomal egg phosphatidylcholine-cholesterol, initiated by DTUN (0.2 mM), and inhibited by LDL, HDL, reconstituted HDL with Tristearin (rHDL-TG(18:0)₃), or PMC (4 μM). h. Rank of significantly altered lipid metabolites (detected in positive mode) in B16 cells treated with HDL versus LDL. i. Relative peak area for α-toc in B16 cells under depletion or treatment with LDL or HDL. j. Schematic depicting dietary depletion of α-toc to evaluate its role in driving tumour growth in mice. k. Plasma α-toc levels after 4 weeks on vitamin E-sufficient (100 IU/kg) or -deficient (7 IU/kg) diets. l. Tumour weights from mice implanted with Karpas299, HeLa, or B16 cells under control or low vitamin E diets. For all LDL and HDL experiments, 50 μg/mL was used. d, i: Bars represent mean ± s.d.; l, k: Boxes represent median, first/third quartiles; whiskers are range. d-f: n=3; h-i: n=5; k: n=7; l: n= 12, 12, 9 (respectively). Statistics by two-sided unpaired t-test (d, i, k, l) or one-way ANOVA with Kruskal–Wallis post-test (h).

Figure 2.. Cancer cells require GAG biosynthesis for lipoprotein uptake, ferroptosis resistance, and tumour growth.

a. Schematic of parallel metabolism-focused CRISPR screens in Karpas299 lymphoma cells. The first screen assessed proliferation ± GPX4 inhibitor ML210 over 14 doublings (left); the second used flow cytometric sorting of top/bottom 5% fluorescent populations following DiI-LDL treatment (right). b. Rank of genes by change in essentiality under ML210 treatment; GAG biosynthesis genes (dark blue) and known anti-ferroptotic genes (light blue) are highlighted. c. Rank of genes by change in essentiality for DiI-LDL uptake; GAG biosynthesis genes (dark blue) and known lipoprotein uptake genes (orange) are shown. d. Quadrant map of overlapping genes from both screens; shared hits belong exclusively to the GAG biosynthesis pathway. e. Schematic of GAG biosynthesis from disaccharide repeats containing glucuronic acid (GlcUA); UGDH is the pathway’s rate-limiting enzyme. f. Quantification of heparan sulfate (HS) and chondroitin sulfate (CS) in UGDH_KO cell lines expressing sgRNA-resistant UGDH cDNA or empty vector. g. Proliferation (log₂ doublings, 5 days) of UGDH_KO cell lines expressing sgRNA-resistant UGDH cDNA or empty vector treated with ML162. h. Histograms of DiI-LDL (left) and DiI-HDL (right) uptake in A-498 UGDH_KO cells expressing sgRNA-resistant UGDH cDNA or empty vector. i. Representative tumours from immunocompromised mice implanted with Karpas299 or A-498 UGDH_KO cells expressing sgRNA-resistant UGDH cDNA or vector; non-engraftments marked with “X.” j. Schematic of daily intraperitoneal treatment with vehicle (grey) or liproxstatin-1 (Lip-1, dark blue) in mice implanted with UGDH_KO cells expressing sgRNA-resistant UGDH cDNA or empty vector. k. Tumour weights from mice treated as in (j). f, g: Bars represent mean ± s.d.; k: Boxes represent median, first and third quartiles, and whiskers are range. f, g: n=3; k: n=10 biological replicates. Statistics by two-sided unpaired t-test vs. empty vector (f, g) or vehicle (k) controls.

Figure 3.. Cell-surface GAGs linked to proteoglycans drive lipoprotein uptake and ferroptosis resistance in tumours.

a. Schematic of the canonical proteoglycan linkage region showing a HS chain attached to a core protein via O-glycosylation of serine residues. Genes essential for proteoglycan linker formation that scored in our screens are highlighted (blue). b. Experimental strategy to enzymatically degrade cell-surface HS and CS in cultured cancer cells or dissociated xenograft tumours using heparinases and chondroitinase. c. Quantification of total cell surface HS (left, grey) and CS (right, blue) in 786-O cells before and after treatment with heparinases (0.1 U/mL) and chondroitinase (0.1 U/mL). d-e. Fold change in DiI-LDL (d) and DiI-HDL (e) uptake in 786-O cells treated with heparinases (0.1 U/mL), chondroitinase (0.1 U/mL), or both (blue), relative to uptake of untreated cells. f. Experimental strategy to assess ferroptosis sensitivity following enzymatic GAG degradation. Cells are treated with enzymes, cultured ± lipoproteins, and exposed to GPX4 inhibition, followed by BODIPY-C11 staining and flow cytometry analysis. g. Cellular lipid oxidation ratio using BODIPY-C11 in 786-O cells untreated (grey) or GAG-degraded (blue), under the indicated conditions of lipoprotein availability and GPX4 inhibition (ML162, 200 nM). Fer-1 (1 μM) is included as a positive control. h. Representative IHC images (20x magnification) of 4-hydroxy-2-nonenal (4-HNE) in 786-O tumours injected daily with HBSS or a heparinases/chondrotinase cocktail (10 U/mL). Scale bars, 100 μm. i. Quantification of 4-HNE H-scores in tumours treated as in (h). j. Representative tumour images and weights from 786-O tumours injected daily with HBSS (grey) or a heparinases/chondroitinase cocktail (10 U/mL, blue). c-e, i, j: Bars represent mean ± s.d.; g: Bars represent median; c-e, g, j: n=3 biological replicates; i: n=10 representative fields. Statistics by two-sided unpaired t-tests vs. untreated (c-e, g) or vehicle controls (i, j).

Figure 4.. GAG-mediated lipoprotein uptake in human ccRCC tissues and PDXs.

a. Primary ccRCC patient samples (paired tumour and adjacent kidney tissue) were analyzed by LC/MS for α-toc, HS, and CS quantification, and by RNA-seq. b. Violin plot showing relative levels of α-toc in ccRCC tumours compared to paired adjacent kidney tissues. c. Proliferation (log₂ doublings, 5 days) of A-498 cells in vitro in the absence or presence of HDL (50 μg/mL) isolated from either healthy donor plasma or ccRCC patient plasma, with or without treatment with ML162 (150 nM) and Fer-1 (1 μM). d. Violin plot displaying relative levels of total CS per gram of protein in ccRCC tumours compared to paired adjacent kidney tissues. e. Violin plot of proteoglycan gene expression scores derived from RNA-seq in ccRCC tumours relative to paired adjacent kidney tissue. f. Kaplan-Meier survival analysis of TCGA ccRCC patients stratified by proteoglycan score (top 25%, blue; bottom 25%, grey). g. Fold change in DiI-LDL uptake in ex vivo ccRCC PDX cell suspensions treated with a heparinases/chondrotinase cocktail (10 U/mL) versus untreated. Each dot represents one PDX. h. Schematic of an in vivo focused sgRNA competition assay using pooled control (sgControl) and UGDH-targeting (sgUGDH) sgRNAs in ccRCC PDXs implanted in the renal capsule of immunocompromised mice. i. Average guide scores (log₂) of the in vivo focused sgRNA competition assay in two ccRCC PDX tumours. Each dot represents one sgRNA. c, i: Bars represent mean ± s.d.; g: Bars represent median; b: n=20 patients; c: n=3 biological replicates; d, e: n=18 patients; f: n=133 patients per group; g: n=7 independent PDXs; i: n=4 sgRNAs per group. Statistics by: i) two-sided unpaired t-tests as indicated, compared to ML162-treated cells (c), untreated cells (g) or sgControl-transduced PDX cells (i); or ii) using a log-rank (Mantel–Cox) test (f).