Lipocalin 2 orchestrates resistance to ferroptosis via AXL

SUMMARY

Glioblastoma (GBM) remains a lethal tumor, largely due to robust mechanisms that prevent effective induction of cell death. Ferroptosis, a form of iron-dependent cell death, is a promising vulnerability in GBM. Here, we demonstrate that lipocalin-2 (LCN2) suppresses ferroptosis in GBM cells via the receptor tyrosine kinase AXL. LCN2 was elevated in GBM cells compared to lower-grade tumor and non-transformed cells, and Lcn2 knockdown impaired GBM cell fitness and growth in vitro and in vivo. Mechanistically, Lcn2 knockdown triggered ferroptosis, which was specifically rescued with ferroptosis inhibitors but not apoptosis or necroptosis inhibitors. Lcn2 knockdown reduced AXL phosphorylation, which was elevated in GBM patient tumors relative to non-tumor tissue. Notably, the combination of Lcn2 knockdown and pharmacological AXL inhibition extended survival compared to Lcn2 knockdown alone. Taken together, these data reveal a link between LCN2-mediated suppression of ferroptosis with AXL phosphorylation and support this axis as a potential therapeutic target for GBM.

Graphical Abstract

In brief

Wang et al. show that lipocalin-2 suppresses iron-dependent cell death in glioblastoma. Targeting multiple members in this pathway activates iron-dependent cell death, increases survival in mouse models, and demonstrates a potential therapy for glioblastoma.

INTRODUCTION

Glioblastoma (GBM), the most common primary malignant brain tumor in adults, remains challenging to treat despite aggressive therapies including maximal safe surgical resection with concomitant radiation and chemotherapy. The intrinsic heterogeneity of GBM contributes to its resistance to cell death, which reduces the efficacy of current treatment options.^1,2 Evasion of cell death represents a major challenge to the current standard of care, as existing therapies mainly act to induce apoptotic cell death.³ However, apoptotic resistance in GBM has been well characterized, with multiple mechanisms identified, including upregulation of the BCL-2 family of proteins, XIAP, and the TRAIL receptor.^4-8 Attempts to target these apoptotic pathways have not demonstrated clinical benefit in terms of overall survival,^9,10 and thus new methods to target other modes of cell death are critical. We have previously shown that GBM can be sensitized to lysosome-mediated cell death.¹¹ Given the inherent heterogeneity within GBM tumors, it is now evident that several other cell death mechanisms remain largely unexplored in GBM and warrant further examination.

Ferroptosis is an iron-dependent mode of cell death driven by lipid peroxidation and is mechanistically and morphologically distinct from other forms of cell death.^12-15 Although first conceptualized in 2008 following the discovery of erastin,¹⁴ ferroptosis is now recognized as one of the most conserved forms of cell death, with ferroptosis-like cell death observed in plants, protozoa, and fungi.^16-18 Its importance has recently been recognized across multiple cancers, including breast cancer, hepatocellular cancer, and pancreatic ductal adenocarcinoma,^19,20 as well as neurodegenerative diseases such as Alzheimer’s, Huntington’s, and Parkinson’s disease.²¹ Most recently, ferroptosis has been shown to propagate in waves to trigger limb development, a role previously attributed primarily to apoptosis.²² As the hallmark of ferroptosis is iron-dependent lipid peroxidation, the brain—and GBM in particular—provide an environment well-suited for these conditions. The brain is a lipid-rich organ based on its myelin content, and neuronal membranes also contain a high abundance of polyunsaturated fatty acids (PUFAs), the lipids that execute ferroptosis upon peroxidation through reactive oxygen species (ROS), iron, or PUFA accumulation.²³ GBM exhibits an enriched PUFA lipid profile. GBM metabolism has been reported to be critically regulated by the oxidation of PUFAs,^24,25 and upregulation of PUFA populations has also been shown to be a GBM resistance mechanism to radiation treatment.²⁶ Additionally, the brain has high iron and oxygen demands due to high energy requirements, and iron imbalance has been implicated in many brain pathological states, including GBM.^27-29 These features position ferroptosis as an attractive candidate for exploration in GBM.^2,28-32

There are several methods that a cell can use to protect itself against ferroptosis. The canonical ferroptosis inhibition pathway utilizes the selenoprotein glutathione peroxidase 4 (GPX4), which scavenges ROS to prevent lipid peroxidation. Iron is a major regulator of GBM pathogenesis, as it can impact both the immune response and GBM cell growth. Work from our group and others has shown that GBM cells preferentially take up iron, which is essential for their growth and the epigenetic maintenance of a pro-proliferative state.^28,29 While these studies have mainly focused on canonical iron handling proteins, such as transferrin,²⁹ ferritin,³³ and HFE,²⁸ the degree to which iron signaling, including non-canonical pathways, is involved in GBM cell growth and survival has yet to be determined.

Lipocalin-2 (LCN2) is a glycoprotein in the family of transport lipocalin proteins. It indirectly chelates ferric iron (Fe(III)) via small molecules such as bacterial siderophores or catechols.³⁴ In cancer, LCN2 has been shown to drive phenotypes such as metastasis in breast and lung cancers and growth in colon and pancreatic cancers.^35-37 LCN2 also promotes leptomeningeal infiltration of breast and lung cancers, governing an early event in metastasis to the brain.^38,39 However, the function of LCN2 in primary brain tumors, including GBM, has not been extensively studied. Based on its involvement in tumor growth and iron regulation, we hypothesized that LCN2 could mediate ferroptotic cell death in GBM cells. Our results demonstrate that LCN2 regulates ferroptosis through the expression of the selenoprotein GPX4, and this pathway can also be leveraged therapeutically by combining knockdown of Lcn2 with the AXL inhibitor bemcentinib (R428).

RESULTS

Increased LCN2 expression is associated with worse outcomes in GBM

To evaluate the correlation between lipocalins and GBM patient survival, we examined the expression of all lipocalin family members in GBM from the data portal GlioVis (z https://gliovis.bioinfo.cnio.es). Of the lipocalins expressed in the brain, LCN2 was the only family member associated with significantly decreased survival in GBM (Figure S1). LCN2 mRNA expression was also higher in GBM compared to lower-grade brain tumors (Figure 1A; Table S3) and, in GBM patients, was associated with a worse prognosis (Figure 1B). We further assessed LCN2 protein in a cohort of patient tissue samples from the Cleveland Clinic and observed a significantly higher abundance of LCN2 in GBM patients compared to non-tumor controls (Figure 1C). This elevation of LCN2 protein was also observed in conditioned media from a panel of syngeneic GBM mouse models compared to that from normal astrocytes, which had no detectable secreted LCN2 (Figure 1D). Taken together, these results suggest that LCN2 is elevated in GBM and syngeneic mouse models and that LCN2 mRNA expression negatively correlates with prognosis.

Figure 1. LCN2 is associated with worsened GBM prognosis.

(A) LCN2 mRNA expression levels in The Cancer Genome Atlas (TCGA) data compared among GBM and lower-grade brain tumors. Significance was determined by one-way ANOVA with Dunnett’s multiple comparisons test (*p < 0.05).

(B) Overall survival of IDH1 wild-type GBM patients as a function of LCN2 mRNA expression in TCGA data, stratified by high versus low LCN2 expression, as determined by log rank test (*p < 0.05). Median survival and patient numbers are shown in the graph.

(C) LCN2 abundance of flash-frozen samples measured via ELISA in human patients (Cleveland Clinic) in GBM tissue versus non-tumor tissue. Significance determined with unpaired t test (***p < 0.001). All data points are shown, with the bars indicating the range.

(D) LCN2 secretion measured in mouse glioma models with ELISA. mAstro: primary mouse astrocytes. Data are represented as mean ± SEM.

Exogenous LCN2 does not impact proliferation and survival

As LCN2 is a secreted protein, we first added exogenous LCN2 to our low- and medium-secreting cell lines OL61 and KR158, with the standard LCN2 dose of 1 μg/mL as commonly reported in the literature^40-42 (Figures S2A and S2B). Surprisingly, the addition of recombinant LCN2 did not affect cell proliferation as assessed by live-cell imaging. To determine whether this was an artifact of the in vitro system, we orthotopically implanted wild-type C57BL/6 (WT) mice and global LCN2^−/− mice with cells exhibiting a range of LCN2 secretion levels: GL261 and CPA (high-secreting) and KR158 and SB28 (medium-secreting). We observed no differences in survival between WT mice or LCN2^−/− mice implanted with these tumor cells, suggesting a limited cell-intrinsic role of LCN2 in GBM (Figures S2C-S2F). To further investigate why global LCN2 deletion did not affect median survival, we harvested naive mouse brains with no tumor cell implantation and both tumor-bearing and contralateral hemispheres of brains of mice with tumors at neurological endpoint. We found that under naive conditions, LCN2 expression was absent in both WT mice and LCN2^−/− mice (Figure S2G). However, in tumor-bearing LCN2^−/− mice, both contralateral and tumor-bearing hemispheres of mice exhibited equivalent expression of LCN2, which suggested that the LCN2 detected was produced by the tumor cells and that LCN2 was mobile and spread throughout the brain after tumor cell implantation.

Lcn2 knockdown decreases cellular fitness and tumor initiation

To determine the functional relevance of LCN2 expression in GBM cells, we selected two mouse syngeneic GBM cell lines with relatively high LCN2 expression: KR158 and CPA. We used short hairpin genetic constructs to knock down Lcn2 via three nonoverlapping short hairpin RNAs and validated each knockdown with ELISA. Each knockdown targeted a different region of Lcn2 (KD1, KD2, and KD3), and we selected the two best knockdown constructs for each experiment. Lcn2 knockdown significantly decreased LCN2 secretion (Figure 2A), cell confluence via live-cell imaging (Figure 2B), and overall cell viability as assessed by CellTiter-Glo (Figure 2C). Lcn2 knockdown in KR158, which also expressed detectable levels of LCN2 protein, resulted in a similar phenotype (Figures 2D-2F). These results demonstrate that in GBM cells, Lcn2 deficiency led to decreased cellular fitness.

Figure 2. Lcn2 knockdown decreases cell fitness and increases survival.

(A) LCN2 secretion in CPA cells containing non-target construct versus knockdown constructs as determined via ELISA. Data points shown are biological replicates. Significance was determined with one-way ANOVA (*p < 0.05). Data are represented as mean ± SEM.

(B) Confluence of CPA cells containing non-target construct versus knockdown constructs as measured by IncuCyte live-cell imaging system. (One-way ANOVA, repeated measures test, ***p < 0.001, **p < 0.01). Data are represented as mean ± SEM.

(C) CPA cell viability as measured via Cell Titer-Glo on day 3, normalized to day 0. Representative data points shown are biological repeats. Significance determined by one-way ANOVA (****p < 0.0001). Data are represented as mean ± SEM.

(D) LCN2 secretion in KR158 cells containing non-target construct versus knockdown constructs as determined via ELISA. Data points shown are biological replicates. Significance was determined with one-way ANOVA (***p < 0.001 and **p < 0.01). Data are represented as mean ± SEM.

(E) Cell number of KR158 cells containing non-target construct versus knockdown constructs as measured by IncuCyte live-cell imaging system. Significance determined via one-way ANOVA repeated measures test (***p < 0.001, **p < 0.01). Data are represented as mean ± SEM.

(F) KR158 cell viability as measured via Cell Titer-Glo on day 3, normalized to day 0. Representative data points shown are biological repeats. Significance determined by one-way ANOVA (*p < 0.05). Data are represented as mean ± SEM.

(G–I) Survival analysis of CPA non-target control cells (NTCs) compared to CPA Lcn2 knockdown constructs in (G and H) C57BL/6 mice and (I) NSG mice. Significance determined by log rank test (***p < 0.001, **p < 0.01).

(J and K) Survival analysis of KR158 NTCs compared to KR158 Lcn2 knockdown constructs in (J) C57BL/6 mice and (K) NSG mice. Significance determined by log rank test (**p < 0.01 and *p < 0.05). For survival assessments, the number of mice per group and the median survival are shown on the plot. For IncuCyte studies, one representative biological replicate is shown.

To understand how LCN2 impacts tumor initiation, we orthotopically implanted non-target control versus Lcn2 knockdown cells into the brains of C57BL/6 (WT) mice. We observed the most robust increase in the median survival in Lcn2 knockdown cohorts of CPA tumors, which have the highest LCN2 secretion as compared to the controls (Figures 2G and 2H). As LCN2 has known roles in immune regulation,⁴³ we repeated this experiment in immunocompromised NOD-Scid-gamma (NSG) mice. Lcn2 knockdown cells implanted into NSG mice similarly led to increased median survival (Figure 2I), which further suggested a dominant cell-intrinsic role for LCN2 regardless of immune background. We observed consistent results using KR158 Lcn2 knockdown cells, which also led to similarly prolonged survival in vivo (Figures 2J and 2K).

To better understand the role of Lcn2 in tumor initiation, we implanted non-target cells versus Lcn2 knockdown cells in mice and collected all brain tissue at the time of neurological endpoint for the first mouse (which was a mouse implanted with non-target control cells). At this time point, we observed that qualitatively, tumor volume was smaller in mice implanted with Lcn2 knockdown constructs compared to Lcn2 non-target control constructs (Figures S3A-S3C). We then performed the same experiment and allowed tumors in each mouse to develop to neurological endpoint before harvesting tissue to determine whether tumor volume was changed at the endpoint. We did not observe a difference in tumor area (Figures S3D-S3F). This was consistent with ELISA analysis of LCN2 protein expression, which showed no difference between non-target control and Lcn2 knockdown tumors at the endpoint (Figure S3G). Furthermore, we observed a reduction in tumor cell proliferation (Figures S3H-S3J), as well as a suppression of apoptosis in mice implanted with Lcn2 knockdown tumors (Figure S3K). These data suggest that despite antibiotic selection, cells without Lcn2 knockdown may have outcompeted the knockdown cells over time, with the tumor regaining LCN2 levels that are comparable to those of non-target control tumors at the endpoint. Collectively, these findings demonstrate that LCN2 may play a role in GBM cell survival and in preclinical models in a predominantly cell-intrinsic manner.

LCN2 buffers ferroptosis in GBM

Given the reduction in cell viability with Lcn2 knockdown, we wanted to determine whether this was due to decreased cellular proliferation or increased cell death under these conditions. When seeded at equal densities, Lcn2 knockdown cells exhibited reduced cell numbers compared to non-target control cells after three days in culture (Figure 3A). To assess whether this was due to apoptosis, we performed flow cytometry on cells stained with Annexin/DRAQ7 and observed only a modest induction in the proportion of late apoptotic cells in Lcn2 knockdown cells (Figure S4A). We next examined key mediators of mitochondrial outer membrane permeabilization, including BAK, BCL-XL, and BIM.⁴⁴ If Lcn2 knockdown triggered substantial apoptosis, we would expect decreased levels of the proapoptotic proteins BIM and BAK and increased levels of the anti-apoptotic proteins BCL-XL. However, we did not observe a notable change in these protein levels between non-target control cells and Lcn2 knockdown (Figure S4B), which aligns with our earlier findings (Figures S3J-S3L). Together, these data suggest that Lcn2 knockdown did not dramatically affect apoptotic pathways.

Figure 3. LCN2 buffers ferroptosis in GBM.

(A) Cell count in CPA cells measured via trypan blue, normalized to non-target control. Significance measured by one-way ANOVA (****p < 0.0001, ***p < 0.001, and **p < 0.01). Biological replicates are shown. All data points are shown, with the median indicated.

(B) Lipid peroxidation in CPA cells measured via Image-It lipid peroxidation assay. Data represent biological replicates. Significance measured by one-way ANOVA (***p < 0.001 and **p < 0.01). All data points are shown, with the median indicated.

(C) ROS levels in CPA non-target control versus Lcn2 knockdown cells measured via CellROX Deep Red flow assay (***p < 0.001). All data points are shown, with the median indicated.

(D) Immunoblot demonstrating that GPX4 expression decreases with Lcn2 knockdown constructs.

(E) IC₅₀ of RSL3 as measured by cell viability via CellTiter-Glo assay on day 3, normalized to day 0 for CPA non-target controls versus Lcn2 knockdown. IC₅₀ is shown in the figure.

(F) IC₅₀ of PAC-1 as measured by cell viability via CellTiter-Glo assay on day 3, normalized to day 0 for CPA non-target controls versus Lcn2 knockdown. IC₅₀ is shown in the figure.

(G) Summary of rescue matrix shown as a heatmap indicating that only ferroptosis inhibitors restored the observed cell death phenotype. ROS, reactive oxygen species; fer-1, ferrostatin-1; liprox, liproxstatin-1; nec-1, necrostatin-1; nd ZVAD, ZVAD-FMK. For CellROX analysis, one representative biological replicate is shown.

To further investigate mechanisms of cell death in GBM beyond apoptosis, we focused on ferroptosis as the brain and GBM microenvironment are well-suited to ferroptosis conditions due to their enrichment in PUFAs and high iron and oxygen demand.^25,45 Additionally, as LCN2 indirectly binds iron, we hypothesized that alterations in ferroptosis could underly the functional effects we observed upon Lcn2 knockdown.¹⁵ Ferroptosis is executed through iron-dependent lipid peroxidation, a process that can be driven by PUFAs, iron, and ROS accumulation. We therefore first measured lipid peroxidation and observed a significant increase in peroxidated lipids upon Lcn2 knockdown (Figure 3B). Increased lipid peroxidation was also observed in KR158 Lcn2 knockdown cells compared to a non-target construct (Figure S5A). As ROS generation is thought to initiate lipid peroxidation by targeting the PUFA-rich membranes of cells and organelles,^46,47 we next measured ROS levels and observed a significant increase in ROS production with Lcn2 knockdown (Figure 3C). As Lcn2 knockdown increased lipid peroxidation and elevated ROS production, we next investigated whether these changes translated into enhanced sensitivity to ferroptotic cell death in GBM cells.

As previously mentioned, GPX4 is a canonical ferroptosis inhibitor protein.^12,15 When we analyzed its expression via immunoblot, we observed that GPX4 was decreased with Lcn2 knockdown constructs compared to non-target control (Figure 3D), with a similar pattern for Gpx4 mRNA (Figure S5B). Biological replicates and KR158 immunoblots showed consistent results (Figure S5C). As GPX4 reduces lipid peroxides to lipid alcohols,^48,49 decreased GPX4 may contribute to the increase of ROS production upon Lcn2 knockdown.

Next, we assessed whether Lcn2 knockdown cells could be further sensitized to ferroptotic induction via treatment with the ferroptotic inducer RSL3 and found that Lcn2 knockdown cells were 3-fold more sensitive to RSL3 compared to control cells (Figure 3E). RSL3 is a direct inhibitor of GPX4, the major lipid peroxide neutralizing enzyme.²³ Comparatively, when we treated cells with the apoptosis inducer PAC-1, we observed no difference in IC₅₀ concentrations between non-target cells and Lcn2 knockdown cells (Figure 3F), which was replicated in KR158 (Figure S5D). These results indicated that knockdown of Lcn2 induced ferroptosis with increased lipid peroxidation and ROS generation and this effect was ferroptosis specific.

Given that Lcn2 knockdown appeared to sensitize cells to ferroptosis, we then tested whether we could rescue the effect of Lcn2 knockdown and RSL3 induction. We first performed baseline viability testing of non-target control and Lcn2 knockdown cells, which recapitulated our earlier findings and showed fewer viable Lcn2 knockdown cells compared to non-target cells (Figure S5E) and chose our best two knockdown sequences for the ensuing matrix. We then generated a drug treatment matrix that combined RSL3 with inhibitors of a variety of types of cell death, with the following conditions: 1) vehicle, 2) cell death inhibitor, 3) RSL3 (ferroptosis inducer), or 4) combination of cell death inhibitor and RSL3 (Figures 3G and S5F-S5I). For each set of conditions, results were normalized to the vehicle control of each genetic condition (control or Lcn2 knockdown) to better visualize differences in viability. Treatment with cell death inhibitors (Figures 3G and S5F-S5I, condition 2) did not increase viability beyond baseline. Upon treatment with RSL3 (Figures 3G and S5F-S5I, condition 3), we once again observed differential sensitivity to RSL3 induction, as Lcn2 knockdown cells were more sensitive to ferroptosis induction. Lastly, when RSL3 was combined with ferrostatin-1 (Fer-1) or liproxstatin-1 (Liprox-1), synthetic radical-trapping antioxidants that are potent ferroptotic inhibitors,⁵⁰ we observed Lcn2 knockdown cell viability return to baseline, indicating a full rescue (Figures 3G and S5F-S5I). This effect was specific only to ferroptosis inhibitors, as when we combined RSL3 with necrostatin-1, a necroptosis inhibitor, or ZVAD-FMK, an apoptosis inhibitor, we observed no difference in cell viability. These results are summarized through a heatmap (Figure 3G), which strikingly shows that the combinations of RSL3 with ferrostatin-1 or liproxstatin-1 were the only conditions that resulted in a rescue of cell viability. This experiment was replicated with a higher dose of RSL3 to the same effect (Figures S5J-S5M).

To confirm that this reduction in viability and consecutive rescue was driven by ferroptosis and not due to the result of combining an inducer and an inhibitor of the same mode of cell death, we induced cell death in our control and Lcn2 knockdown cells with PAC-1, a selective apoptosis inducer (Figures S5N and S5O). Not only was there no differential response to PAC-1-induced cell death in our control compared to Lcn2 knockdown constructs (consistent with earlier findings from Figure 3E), but we also observed that the combination of apoptotic inhibitor ZVAD-FMK and PAC-1 did not change cell viability. Taken together, these results indicate that Lcn2 inhibits ferroptosis in GBM cells and that the effect can be further induced with the ferroptosis-specific inducer RSL3 and rescued with ferroptosis-specific inhibitors Fer-1 and Liprox-1.

We next wanted to determine whether the induction of ferroptosis following Lcn2 knockdown was specific to GPX4 inhibition or whether it could also be triggered by other classes of ferroptosis inhibitors. To test this, we treated cells with erastin, a ferroptosis inducer that operates through a different mechanism than RSL3. Although both RSL3 and erastin are ferroptosis inducers, they target different mechanisms and can induce differential responses in the cell.⁵¹ RSL3 specifically targets GPX4 by directly binding to and inhibiting the active site of GPX4,⁵² whereas erastin inhibits the cystine/glutamate antiporter xCT, which depletes intracellular cysteine and subsequently glutathione (GSH), which is upstream of GPX4 function.^53,54 Upon treatment of erastin with either Fer-1 or Liprox-1, we observed that the combination of Fer-1 and erastin or Liprox-1 and erastin could not rescue Lcn2 knockdown cells from erastin-induced cell death (Figures S6A-S6E). Indeed, when we probed for xCT expression via immunoblot, we observed no difference between non-target control cells and Lcn2 knockdown cells in both CPA and KR158 cells (Figures S6F and S6G). Taken together, these results demonstrate that the ferroptotic effects of LCN2 occur via GPX4 and that reduction of LCN2 alone results in an induction of ferroptosis in GBM cells.

Lcn2 knockdown halts Fe (III) import into the cell

As ferroptosis is an iron-dependent form of cell death, we next investigated whether Lcn2 knockdown altered iron status in GBM cells. To answer this question, once again, we orthotopically implanted non-target versus Lcn2 knockdown cells into mice and collected brain tissue at the first sign of neurological endpoint. In mice implanted with Lcn2 knockdown cells, we observed increased Prussian blue staining at the tumor periphery compared to mice implanted with non-target control cells (Figures 4A and 4B). Prussian blue stains for ferric iron (Fe(III)), which is also the form of iron that LCN2 binds indirectly.⁵⁵ When we investigated iron populations with thioglycolic acid to investigate iron pools, we observed that Lcn2 knockdown dramatically decreased Fe(III) import into the cell (Figure 4C). We then used deferoxamine (DFO) to chelate extracellular Fe(III) to determine whether we could mimic Lcn2 knockdown pharmacologically. Indeed, non-target control cells treated with DFO phenocopied the effects of Lcn2 knockdown cells with vehicle treatment (Figure 4D). When we assessed lipid peroxidation, we found that Lcn2 knockdown cells treated with DFO had further increased lipid peroxidation compared to non-target cells treated with DFO (Figures 4E and S6H). Lastly, a survey of a number of iron-handling proteins via immunoblot showed that Lcn2 knockdown disrupted iron homeostasis in the cell. Lcn2 knockdown decreased transferrin-mediated iron import, decreased ferroportin-mediated iron export, and increased iron storage through ferritin. These data suggest that Lcn2 knockdown alters iron homeostasis by decreasing the Fe(III) population in the cell, which halts iron export and increases iron storage as the cell attempts to maintain the labile iron pool balance. This shift likely promotes more ROS accumulation and lipid peroxidation to sensitize cells to ferroptosis (Figure 4G).

Figure 4. Lcn2 knockdown halts iron (III) import into the cell.

(A and B) (A) Perls Prussian blue staining of the non-target control brain compared to (B) Lcn2 knockdown brain. Fe(III) is stained blue. Bars are shown for distance.

(C) Intracellular iron populations measured with and without thioglycolic acid for iron population ratios in CPA non-target control versus Lcn2 knockdown cells.

(D) Cell viability of CPA non-target cells and Lcn2 knockdown cells treated with either vehicle or deferoxamine (DFO). Data are represented as mean ± standard deviation.

(E) Lipid peroxidation of CPA non-target cells and Lcn2 knockdown cells treated with DFO. Results are normalized to non-target cells. Significance measured by one-way ANOVA (**p < 0.01 and *p < 0.05). All data points are shown, with the median indicated.

(F) Immunoblot demonstrating transferrin, ferroportin, and ferritin expression in CPA non-target control cells compared to Lcn2 knockdown cells.

(G) Schematic demonstrating that iron-labile pool changes with Lcn2 knockdown, which leads to an unbalanced labile iron pool (LIP), and increased radicals to affect ferroptosis in the cell.

AXL mediates LCN2 inhibition of ferroptosis

As LCN2 is a secreted protein and iron chelation therapies have historically been associated with significant toxicity in patients,⁵⁶ we wanted to therapeutically target LCN2, potentially through downstream mechanisms. Previous work has shown that receptor tyrosine kinases are overexpressed and central to GBM pathogenesis.^57,58 We therefore performed an unbiased phosphorylation array of 39 receptor tyrosine kinases comparing phosphorylated proteins in non-target control and Lcn2 knockdown cells. This revealed a significant decrease in phosphorylated AXL in Lcn2 knockdown cells (Figures 5A and S7A; Table S4). AXL is a receptor tyrosine kinase that has been shown to play a role in GBM but has not been linked to LCN2.^59-61 While AXL signaling has been shown to be important for cell proliferation, migration, and differentiation, AXL has only recently been linked to ferroptosis. In both hepatic injury⁶² and experimental autoimmune encephalomyelitis,⁶³ AXL was shown to inhibit ferroptotic cell death. We surveyed AXL phosphorylation at Y702, which is located in the intracellular kinase domain and able to auto-phosphorylate,⁶⁴ and observed a reduction in the phosphorylation of AXL in Lcn2 knockdown cells (Figures 5B and S7B). We additionally analyzed a second AXL phosphorylation site, tyrosine 703, which is also located in the activation loop in the kinase domain, and did not observe a change with Lcn2 knockdown (Figure S7C). Notably, we also observed a reduction in phosphorylated cleavage products of AXL in Lcn2 knockdown cells compared to non-target controls (Figure 5B). AXL is cleaved by ADAM10 and ADAM17 to release soluble AXL (~80 kDa),^65,66 as well as a ~55 kDa C-terminal fragment of AXL and an additional cleavage product approximately ~8 kDa smaller (AXL-ICD; Figure S7D).⁶⁷ We confirmed that the change in the phosphorylation of AXL from Lcn2 knockdown was due to the posttranslational modification, as we did not observe a difference in AXL transcript levels (Figure S7E).

Figure 5. AXL mediates LCN2 induction of ferroptosis.

(A) Unbiased phosphorylation array demonstrating that AXL phosphorylation decreased in KR158 non-target versus Lcn2 knockdown cells. Quantification of AXL phosphorylation is shown on the right.

(B) Immunoblot showing phosphorylation of AXL at Y702 showed decreased phosphorylation in CPA control versus Lcn2 knockdown cells. See Figure S7B for the entire blot.

(C) Schematic of pathway with RSL3 inhibition.

(D) Perturbation of GPX4 via RSL3 does not change AXL phosphorylation via immunoblot.

(E) Lipid peroxidation was increased via Image-It lipid peroxidation kit. Statistics measured by paired t test (*p < 0.05).

(F) Schematic of pathway with RSL3 inhibition. All data points are shown, with the median indicated.

(G) Perturbation of AXL phosphorylation via R428 decreases GPX4 levels via immunoblot.

(H) Lipid peroxidation was increased upon measurement via Image-It Lipid Peroxidation kit. Statistics measured by paired t test (**p < 0.01). All data points are shown, with the median indicated.

(I) Quantifications of immunoblot of phosphorylated AXL relative to total AXL, normalized to actin. Full blot in Figure S7K. Statistics were measured by paired t test (*p < 0.05). All data points are shown, with the bars indicating the range.

(J) Schematic showing the four-pronged experiment. Mice were either implanted with non-target control cells or Lcn2 knockdown cells. Mice were then treated with vehicle control (corn oil) or 125 mg/kg of R428.

(K) Survival analysis of CPA non-target control cells with vehicle treatment or R428 treatment, and Lcn2 knockdown constructs with vehicle treatment or R428 treatment. Significance measured by Gehan-Breslow-Wilcoxon test.

To eliminate the possibility that the reduction of AXL phosphorylation was due to decreased secretion of its canonical ligand, GAS6, we measured GAS6 levels via ELISA and observed no detectable GAS6 secretion with the corresponding Lcn2 knockdowns (Figures S7F and S7G). While there were no detectable levels of GAS6 via ELISA, Gas6 was downregulated by quantitative reverse-transcription PCR in Lcn2 knockdown cells compared to non-target control cells (Figure S7H). We also investigated several potential AXL regulators, including OPCML, which dephosphorylates AXL and TENC1, a protein that has been shown to interact with AXL.^68,69 The mRNA expression of both potential regulators was decreased (Figure S7H), suggesting that Lcn2 knockdown is affecting other, non-canonical AXL pathways as well.

To better understand the signaling relationship involving LCN2 and AXL, we perturbed each component using genetic and pharmacological approaches. We first established the IC₅₀ dose of both RSL3 and the small molecule AXL inhibitor bemcentinib (R428) in the CPA cell line (Figures S7I and S7J). R428 specifically binds to the catalytic domain of AXL and inhibits its kinase activity.⁷⁰ Additionally, we chose R428 to inhibit AXL because it is currently being evaluated in multiple phase I and II clinical trials (NCT03824080, NCT02922777, NCT02424617, and NCT02488408) and is brain penetrant.⁷¹ To identify whether GPX4 inhibition altered AXL phosphorylation, we treated parental cells with vehicle or RSL3 (Figure 5C). If GPX4 were upstream of AXL, treatment with RSL3 would result in decreased AXL phosphorylation. There were no changes in AXL phosphorylation via immunoblot (Figure 5D), even though treatment with RSL3 increased lipid peroxidation (Figure 5E), indicating that GPX4 was not upstream of AXL phosphorylation. To identify whether AXL phosphorylation was upstream of GPX4, we inhibited AXL phosphorylation with R428 (Figure 5F). We discovered that R428 treatment decreased GPX4 expression via immunoblot (Figure 5G); lipid peroxidation also increased with R428 treatment (Figure 5H). We then compared benign patient tissue to GBM patient tissue from Cleveland Clinic. Strikingly, GBM patient tissue showed little total AXL with a clear pattern of cleaved phosphorylated AXL, which was in line with our findings in mouse glioma cells (Figure S7K). When we compared the ratio of all phosphorylated AXL forms to total AXL, GBM tissue had a significantly higher ratio of phosphorylated AXL to total AXL (Figure 5I). We also observed a significant positive correlation between Lcn2 and Axl gene expression (data obtained from GlioVis; Figure S6L).

Finally, we tested R428 treatment in vivo to determine whether the combination of Lcn2 knockdown and R428 treatment could prolong survival. We first generated a dose escalation curve with doses found in the literature;⁷⁰ from this, we selected an optimal dose of 125 mg/kg, once per day (Figures S5M and S5N). We then orthotopically implanted either non-target control cells or Lcn2 knockdown cells in mice and observed that when R428 was used as a single agent in non-target control cells, there was no difference in median survival (Figures 5J and 5K). However, when R428 was combined with Lcn2 knockdown, median survival was significantly increased. These results were repeated to the same effect in different knockdown models, where survival was almost three times the original survival of non-target control cells treated with vehicle (Figure S8). These data suggest that inhibition of AXL phosphorylation through R428 in combination with Lcn2 knockdown can potentially be translated into the clinic. Taken together, these data reveal a LCN2-AXL-GPX4 signaling network that links LCN2 to ferroptosis through AXL phosphorylation.

DISCUSSION

Here, we provide evidence that LCN2 may function to inhibit ferroptosis via GPX4 regulation in GBM. GBM standard of care, which consists of maximal safe surgical resection along with radiation and temozolomide, canonically targets apoptosis.³ However, radiation and temozolomide often fail due to the complex heterogeneity in the GBM tumor itself.⁶ To address these limitations, there is increasing interest in investigating other forms of cell death. For example, we have previously shown that GBM cells are sensitive to lysosome-mediated cell death,⁷² and it is likely that many other forms of cell death, including ferroptosis, are also suppressed in GBM.

In this work, we propose LCN2 as a modulator of iron levels, which in turn affects ferroptosis induction. Our system shows a dependence on GPX4 and not on system Xc- or erastin sensitivity. Others have shown that transferrin and transferrin receptors are required for erastin sensitivity⁷³; we find that knockdown of Lcn2 decreases transferrin receptor 1 and ferroportin, which may explain the dependence on RSL3 and GPX4 for ferroptosis induction over system Xc, after LCN2 knockdown.

AXL overexpression has been well documented as a predictor of cancer prognosis, and several AXL inhibitors have advanced to phase I–III clinical trials (NCT02488408, NCT02922777, NCT02424617, and NCT03824080).^71,74 Despite the ability of AXL to serve as a cancer biomarker and its advancement in clinical trials, AXL inhibitors have shown limited clinical efficacy as monotherapies (NCT02488408).^75,76 Most recently, the AXL-specific inhibitor R428 was evaluated in a phase 1b/2a study in relapsed or refractory acute myeloid leukemia. Although R428 monotherapy exhibited weak efficacy, its combination with low-dose cytarabine demonstrated significantly improved response rates. These findings highlight the potential of AXL inhibition when used in combination with other therapies.^75,77

Consistent with these clinical observations, our in vivo data show that AXL inhibition alone is insufficient to extend survival. However, survival was extended when AXL was inhibited in combination with Lcn2 depletion. The enhanced efficacy of R428 in the context of Lcn2 knockdown underscores a mechanistic interaction that supports further investigation into combination therapies that use LCN2 as a biomarker for AXL inhibition. Traditionally, AXL inhibition has been discussed in terms of apoptotic induction^60,78; here, we demonstrate that AXL inhibition also intersects with ferroptosis.

Here, we demonstrate that AXL may play a role in ferroptosis in GBM and that AXL phosphorylation and LCN2 are functionally linked. Lcn2 knockdown redefines iron contents within the cell, modulates antioxidant and lipid processing through the GPX4 master regulator, and can be further leveraged with AXL inhibition to increase median survival in GBM animal models. The observations made here provide evidence for a signaling network that links LCN2, AXL, and ferroptosis, representing a potential next-generation therapeutic target.

Limitations of the study

Our findings show that LCN2 can be manipulated to reduce tumor cell growth. There are, however, limitations to our current work. LCN2 is a secreted protein, which can be inherently difficult to therapeutically target. LCN2 also has multiple potential receptors, including SLC22A17, MC4R, and LRP2.⁷⁹ Due to the complex relationship between LCN2 and its receptors, we initially focused on the cell-intrinsic role of LCN2, but receptor targeting, or an anti-LCN2 antibody, would be a natural future step that would also incorporate the tumor microenvironment. Taken together, the limitations of the current study discussed above represent priorities for future studies.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Sex as a biological variable

Our study investigated both male and female animals; similar findings were observed in both sexes.

Cell culture

Syngeneic mouse GBM cell lines (OL61, KR158, CPA) were grown in adherent conditions with RPMI 1640 with 10% fetal bovine serum (ThermoFisher) and 1% penicillin-streptomycin (Cleveland Clinic Media Preparation Core). Media was changed every other day until cells reached 50% confluency, at which point media was changed daily. Cells were passaged with trypsin-EDTA (Cleveland Clinic Media Preparation Core) when 85–90% confluent. Cells were grown in humidified incubators at 37°C and 5% CO2. OL61 and CPA were generously provided by the Castro-Lowenstein laboratory (University of Michigan). KR158 was generously provided from the Deleyrolle Laboratory (Mayo Clinic). Primary mouse astrocytes were derived from three-day old C57BL/6 mice as previously described.⁸¹ All cells were regularly tested for Mycoplasma spp (Lonza).

For desired cell counts, cells were dissociated into single cell suspension with trypsin and resuspended, and 30 μL of cell suspension was taken for counting. An equivalent volume of Trypan Blue (ThermoFisher) was added to the cell suspension. A total of 10 μL of the mixture was taken and applied to a cell counting slide (Bio-Rad). Live cell counts were obtained with the Bio-Rad TC-20 Automated Cell Counter.

Intracranial tumor cell implantation

All animal studies were approved by the Institutional Animal Care and Use Committee of Cleveland Clinic. Male and female mice (C57BL/6 (RRID:IMSR_JAX:000664)) were purchased from the Jackson Laboratories. NSG (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ) mice were bred in-house (Biological Resources Unit, Lerner Research Institute; Cleveland Clinic). LCN2-deficient (B6.129P2-Lcn2^tm1Aade/AkiJ) mice were purchased from Jackson Laboratories and then bred in-house. All mice were implanted between 6 and 9 weeks of age. For tumor cell implantation, all mice were anesthetized with aerosolized isoflurane (2.5%). A total of 15,000 tumor cells was injected 3.5 mm ventral, 0.5 mm rostral, and 1.8 mm lateral to bregma using a stereotaxic apparatus (Kopf Instruments). Animals were monitored for endpoint (neurological and behavioral symptoms) daily in compliance with institutional guidelines. Overall survival time was based on detection of neurological symptoms, and log rank test was utilized to statistically compare survival differences.

Primers used to validate Lcn2 genotype were according to Jackson Laboratories (Table S1).

Patient mRNA expression and survival

Data for the IDH-wildtype subset of GBM patients of TCGA was downloaded from the GlioVis portal (http://gliovis.bioinfo.cnio.es). GlioVis⁸² was used to extract LCN2 mRNA expression for the “TCGA_GBMLGG” dataset.⁸³ R version 4.0.5, ggplot2 version 3.4.2, and dplyr version 1.1.2 were used to create a boxplot of LCN2 mRNA expression across astrocytoma, oligodendroglioma, and GBM histologies. Tumors with the histopathological diagnosis of oligoastrocytoma, owing to the recent reclassification of brain tumors by the WHO in 2021, were excluded from this analysis. Single sample gene set enrichment analysis (ssGSEA) scores as implemented through the fgsea package (version 1.32.0) in R was used to compute signature scores from a manually selected set of signatures representative of modes of cell death taken from the molecular signatures database (MSigDB).^84,85 Gene expression data was taken from the Cancer Genome Atlas (TCGA) GBM cohort and the Chinese Glioma Genome Atlas (CGGA), batch 2 (325 samples). Clinical information for each sample was obtained, and only those with a histopathological diagnosis of WHO Grade 4 GBM, IDH wild type, were used. ssGSEA score as a continuous univariate predictor of overall survival was examined using a Cox proportional hazards model (survival package in R, version 3.7), with hazard ratio (>1 indicative of worse prognosis) subsequently visualized in a forest plot (forestplot package 3.1.6).

Protein analysis of banked human samples

Flash frozen ex vivo patient samples were collected by the Cleveland Clinic Rose Ella Burkhardt Brain Tumor and Neuro-Oncology Center neurosurgical team after obtaining patient written consent. The studies were conducted in accordance with recognized ethical guidelines and approved by the Cleveland Clinic Institutional Review Board (IRB 2559). A total of 7 biological male and 7 biological female GBM samples and 7 biological male and 7 biological female non-tumor control tissue (epilepsy resection), approximately age-matched, were collected.

Study approval

All animal procedures were performed in accordance with the guidelines and protocols approved by the Institutional Animal Care and Use Committee at the Cleveland Clinic. Human samples were acquired in accordance with recognized ethical guidelines and approved by the Cleveland Clinic Institutional Review Board (IRB 2559).

METHOD DETAILS

Cell viability and functional assays

For genetically modified cells, relative cell viability was measured using the CellTiter Glo (Promega) viability assay following the protocol established by the manufacturer. Cells were plated at the desired density (500 cells/well for CPA, 1000 cells/well for KR158) in a 96-well white-walled plate. All conditions were plated in 10 technical replicates. Readings for luminescence were taken on Day 0 and Day 3 and normalized to Day 0 to account for any plating irregularity. Luminescence was measured on a VICTOR Nivo multimode plate reader (PerkinElmer). Each experiment was conducted in at least biological triplicate.

To determine viability with drug treatment, cells were plated at the desired density in 50 μL of media. Drugs (at twice desired concentration) or vehicle (DMSO) were then added in equivalent volumes (50 μL) the next morning. Cells were assayed for viability at time 0 and 3 days later with CellTiter Glo (Promega). Luminescence was measured on a VICTOR Nivo multimode plate reader (PerkinElmer). Relative luminescence was calculated for day 3 relative to day 0.

To measure reactive oxygen species (ROS), cells were incubated and assayed with CellRox (ThermoFisher) according to the manufacturer’s instructions, and samples were run on an LSR Fortessa flow cytometer (BD Biosciences) with a minimum of 10,000 events collected. Single cells were gated, and ROS levels were measured as gMFI (geometric mean fluorescence intensity) using FlowJo software (v10, BD Biosciences).

To measure early and late cell death, cells were labeled with FITC-annexin V (BioLegend) and DRAQ7 (Invitrogen) according to manufacturer’s protocols. Single cells were gated, and the percentages of Annexin V- and/or DRAQ7-positive cells were determined. Levels were measured as gMFI (geometric mean fluorescence intensity) using FlowJo software (v10, BD Biosciences).

Lipid peroxidation

To measure lipid peroxidation of LCN2 knockdown cells, cells were plated at the desired density (500 cells/well) in media in a 96-well white-walled plate. Three days later, cells were washed once with PBS, stained with 10 μM Lipid Peroxidation Sensor (Image-It Lipid Peroxidation Kit, Invitrogen) for 30 min in serum-free HBSS in 37°C, then washed three times with PBS. Cells were then assayed for lipid peroxidation by measuring fluorescence emission at 590 nm and 510 nm on a VICTOR Nivo multimode plate reader (PerkinElmer). The ratio of the signals from 590 nm to 510 nm channels was used to quantify lipid peroxidation. Each experiment was conducted in biological triplicate.

To measure lipid peroxidation of parental cells treated with R428 or RSL3, parental cells were treated with increasing serial dilutions of drug to create a 24-h IC₅₀ assessment as explained below in section: IC₅₀ assessment of drugs. Once the concentrations were established (0.55 μM for R428, and 0.80 μM for Rsl3), cells were plated (3000 cells/well) in 10 replicates in a white-walled 96-well plate in 50 μL per well media and left to adhere overnight. The following day, 50 μL media containing twice the desired drug concentration, or vehicle (DMSO) was plated. Cells were then stained and prepped for lipid peroxidation as explained above.

Cell number and proliferation

Cell number was assayed in real-time with the IncuCyte S3 Live-Cell Analysis System (Sartorius). Cells were plated at the desired density (500 cells/well for CPA, 1000 cells/well for KR158) in a 96-well clear plate. All conditions were plated in 10 technical replicates. Eight hours post plating, the first images were taken. Data were captured every 6–12 h over a period of 3 days and analyzed with IncuCyte 2021C software.

To measure proliferation over time, cells were stained with 1:1000 CellTrace Violet (Invitrogen) prior to culturing. Cells were collected on Day 0 and Day 1. CellTrace Violet was measured using the violet 450 nm laser on the LSR Fortessa flow cytometer (BD Biosciences) with a minimum of 10,000 events collected. Single cells were gated, and proliferation was measured as gMFI (geometric mean fluorescence intensity) using FlowJo software (v10, BD Biosciences).

IC₅₀ assessment of drugs and drug matrix

The half maximal inhibitory concentrations (IC₅₀) for RSL3 (SelleckChem), erastin (SelleckChem), and R428 (SelleckChem) were established across multiple cell models. Cells were plated at the desired concentration in 50 μL of media. The next morning, twice the desired concentration of drug (in 50 μL media) was added in a serial dilution for the final desired concentration in each well. Readings for luminescence were taken at time 0 using CellTiter Glo (Promega) as described above. Three days later, cell viability was again determined using CellTiter Glo (Promega) per the manufacturer’s instructions. Luminescence was measured on a VICTOR Nivo multimode plate reader (PerkinElmer). Relative cell number was calculated from the ratio of day 3 luminescence relative to day 0. IC₅₀ was calculated using [Inhibitor] vs. normalized response curve in Graphpad Prism 9.

For RSL3 rescue experiments, the established RSL3 IC₅₀ concentration was used. Ferrostatin-1 (5 μM; SelleckChem) and liproxstatin-1 (1 μM; SelleckChem) concentrations were as determined in the literature.^86,87 ZVAD-FMK (25 μM; InvivoGen), necrostatin-1 (0.5 μM), and PAC-1 (0.25 μM) working concentrations were determined according to the manufacturer’s instructions.

Quantitative real-time PCR

Cells were extracted for RNA using an RNeasy kit (Qiagen; 74134). RNA concentrations were measured using a NanoDrop spectrophotometer. cDNA was synthesized using qScript cDNA SuperMix (Quanta Biosciences; 101414-102). qPCR was performed in Fast SYBR Green Mastermix (Applied Biosystems; 01120793) and an Applied Biosystems QuantStudio 3. Primer sequences are shown in Table S2. During qPCR analysis, threshold cycle values were normalized to Gapdh.

Protein analysis

Cells were washed with PBS, then dissociated with trypsin, collected, and centrifuged at 400xg for 4 min. Cells were then washed in PBS and snap frozen in liquid nitrogen. Cells were lysed with 1x NP-40 lysis buffer (0.5% IGEPAL (Sigma Aldrich); 10 mM Tris-Cl, pH 7.5; 1 mM EDTA, pH 8.0; 150 mM NaCl) with added protease inhibitor (Sigma-Aldrich; P8340), phosphatase inhibitor (Sigma-Aldrich; P5725), and PMSF (Sigma; PMSF). Cells were denatured, and protein (20 μg) was loaded on 12% SDS-PAGE gels, electrophoresed, and transferred onto PVDF membranes. Blots were blocked in 5% BSA in TBS-T and probed with primary antibodies diluted at 1:1000 in 5% BSA in TBS-T and Starbright B-700-conjugated secondary antibodies (Bio-Rad) diluted 1:2500 in 5% BSA in TBS-T. Rhodamine-actin used was used at 1:2500 in 5% BSA in TBS-T. Membranes were imaged on a Bio-Rad ChemiDoc Imaging System.

For banked human patient samples, tissue was analyzed by a board-certified neuropathologist to ensure adequate tumor content in the GBM samples and lack of tumor content in the non-tumor control tissue. Frozen samples were homogenized with 1x NP-40 lysis buffer, with added protease inhibitor (Sigma-Aldrich; P8340), phosphatase inhibitor (Sigma-Aldrich; P5725), and PMSF (Sigma; PMSF). For immunoblot, protein (30 μg) was loaded on 12% SDS-PAGE gels, electrophoresed, and transferred onto PVDF membranes. Blots were blocked in 5% BSA in TBS-T, probed with primary antibodies, and Starbright B-700-conjugated secondary antibodies. Membranes were imaged on the Bio-Rad ChemiDoc MP Imaging System.

Enzyme-linked immunosorbent assay

Cells were plated and grown to 95% confluency in T75 flasks. At time 0, media was changed, and 10 mL warm, fresh media was added to each T75 flask. Twenty-four hours later, media was collected, and cells were dissociated into a single-cell suspension. Live cell count was taken using Trypan blue. Collected media was diluted to fit in the range of the standard curve (4.69–300 pg/mL; ab199083; Abcam). Concentration (pg/mL) was normalized to live cell count.

For human samples, tissue was collected and processed as above (ab215541; Abcam). Concentration (pg/mL) was normalized to total protein in the sample.

Lcn2 genetic manipulation via short hairpin RNA knockdown

MISSION pLKO.1-puro Non-Mammalian shRNA Control Plasmid (SHC002) and Lcn2 shRNA plasmids TRCN0000055328 (referenced as KD1), TRCN0000055331 (referenced as KD2), TRCN0000055332 (referenced as KD3) were purchased (Sigma) as glycerol stocks. Plasmids were purified using a plasmid maxiprep method per manufacturer’s instructions (ThermoFisher). Plasmids for psPAX2 (Addgene plasmid #12260; http://n2t.net/addgene:12260; RRID:Addgene_12260) and pMD2.G (Addgene plasmid #12259; http://n2t.net/addgene:12259; RRID:Addgene_12259) were also prepared accordingly. Lentivirus was created using calcium phosphate transfection. The best two knockdowns as assessed by ELISA were selected for each experiment.

Iron measurements

Protocol was adapted from previously described.⁸⁰ Standards were prepared with Iron Standard (High Purity Standards): 1000 μg/dL, 800 μg/dL, 600 μg/dL, 400 μg/dL, 300 μg/dL, 200 μg/dL, 100 μg/dL, 0 μg/dL. Equal numbers of cells were mixed with protein precipitation solution (1:1 1N HCl; Sigma 320331 and 10% trichloroacetic acid; Sigma T8657) and heated at 95°C for 1 h to release iron. Precipitated proteins were removed by centrifugation at 16,000 x g for 10 min at 5°C. Total iron was measured by mixing supernatant with equal amounts of chromogenic solution (0.5 mM Ferrozine; MedChemExpress HY-137805, 1.5 M sodium acetate; ThermoFisher A13184-30, 0.1% [v/v] thioglycolic acid; Fisher Scientific M005225G). Absorbance was measured at 562 nm. To measure Fe(II), equal volume of chromogenic solution without thioglycolic acid was mixed with supernatant. Absorbance was measured at 562 nm.

Histology

For immunofluorescence staining, the protocol was used as previously described.⁸⁸ Brains were fixed in 4% paraformaldehyde (PFA), dehydrated in 30% sucrose solution, and embedded in optimal cutting temperature compound, snap frozen, then sectioned (30 μm) thick via a Leica CM3050 S precooled to −20°C. Tissue sections were incubated overnight at 4°C with the following primary antibodies: anti-F-actin, chicken anti-GFP, rabbit anti-phospho-histone H3 and rabbit anti-cleaved caspase-3. The following secondary antibodies were used: donkey anti-chicken Alexa Fluor 488 overnight at 4°C and WGA Alexa Fluor 680 (4 μg ml–1 in HBSS-T with magnesium and calcium) or goat anti-rabbit 647 for 1 h at room temperature.

For hematoxylin and eosin staining, brains fixed in 4% paraformaldehyde (PFA) and dehydrated in 70% sucrose solution were submitted to the Cleveland Clinic Imaging Core for sectioning and staining.

QUANTIFICATION AND STATISTICAL ANALYSIS

For two-group comparisons, p values were calculated using an unpaired t test, with the significance level set to p < 0.05. For multiple comparisons within one condition, a one-way ANOVA with Dunnett’s multiple comparisons test was used. For multiple comparisons in multiple conditions, a two-way ANOVA was performed. Kaplan-Meier analysis and log rank tests were performed for in vivo survival analyses. All statistical analyses were performed using GraphPad Prism 9. All in vitro experiments were carried out in at least biological triplicate with three technical replicates per experiment for each experimental group. Additional statistical details, including p values and sample size, can be found in the figure legends.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.116965.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-GPX4	Cell Signaling Technology	Cat #52455S; RRID:AB_2924984
Rabbit anti-pAXL (Y702)	ThermoFisher	Cat #PA5-64862; RRID:AB_2662770
Rabbit anti-AXL (C terminus)	LS Bio	Cat # LS-B12012-100
hFAB™ Rhodamine Anti-Actin	Bio-Rad	Cat #120041163
Goat anti-rabbit StarBright Blue 700	Bio-Rad	Cat #12004161; RRID:AB_2721073
Rabbit anti-BCL-XL	Cell Signaling Technology	Cat #2764T; RRID:AB_2228008
Rabbit anti-BIM	Cell Signaling Technology	Cat #2819S; RRID:AB_10692515
Rabbit anti-BAK	Cell Signaling Technology	Cat #3814S; RRID:AB_2290287
Rabbit anti-lipocalin-2 antibody	Abcam	Cat #ab63929; RRID:AB_1140965
Chicken anti-GFP	Aves Labs	Cat #GFP-1020; RRID:AB_10000240
Mouse anti-Beta III Tubulin	Promega	Cat #G7121; RRID:AB_430874
Rabbit anti-Phosphorylated Histone H3	Cell Signaling Technology	Cat #4499; RRID:AB_10544537
Rabbit anti-Cleaved Caspase 3	Cell Signaling Technology	Cat #9661; RRID:AB_2341188
Bacterial and virus strains
psPAX2 (lentiviral packaging plasmid)	Addgene (Didier Trono)	Cat #12260; RRID:Addgene_12260
pMD2.G (lentiviral packaging plasmid)	Addgene (Didier Trono)	Cat #12259; RRID:Addgene_12259
MISSION pLKO.1-puro Non-Mammalian shRNA Control	Sigma	TRC #SHC002
Lcn2 shRNA KD1	Sigma	TRC #TRCN0000055328
Lcn2 shRNA KD2	Sigma	TRC #TRCN0000055331
Lcn2 shRNA KD3	Sigma	TRC #TRCN0000055332
Biological samples
Human glioblastoma patient tissue	This paper, Cleveland Clinic	See STAR Methods
Human non-tumor patient tissue	This paper, Cleveland Clinic	See STAR Methods
Primary mouse astrocytes	Silver et al.80	See STAR Methods
Chemicals, peptides, and recombinant proteins
RPMI 1640 medium	Cleveland Clinic Media Preparation Core	N/A
Fetal bovine serum	Cleveland Clinic Media Preparation Core	N/A
Penicillin-Streptomycin	Cleveland Clinic Media Preparation Core	N/A
Trypsin-EDTA	Cleveland Clinic Media Preparation Core	N/A
Trypan Blue	ThermoFisher	Cat #15250061
Image-It Lipid Peroxidation Kit	Invitrogen	Cat #C10445
CellRox Deep Red	ThermoFisher	Cat #C10422
Deferoxamine (DFO)	Sigma	Cat #D9533-1G
RSL3 (GPX4 inhibitor)	SelleckChem	Cat #S8155
Erastin	SelleckChem	Cat #S7242
Ferrostatin-1 (Fer-1)	SelleckChem	Cat #S7243
Liproxstatin-1 (Lip-1)	SelleckChem	Cat #S7699
Necrostatin-1	SelleckChem	Cat #ab141053
Z-VAD-FMK	InvivoGen	Cat #tlrl-vad
PAC-1	SelleckChem	Cat #S2738
Bemcentinib (R428, BGB324)	SelleckChem	Cat #S2841
1x NP-40 lysis buffer	ThermoFisher	Cat #J60766.AP
Protease inhibitor cocktail	Sigma-Aldrich	Cat #P8340
Phosphatase inhibitor cocktail	Sigma-Aldrich	Cat #P5725
PMSF	Sigma-Aldrich	Cat #10837091001
Sodium acetate	ThermoFisher	Cat #A13184-30
Thioglycolic acid	Fisher Scientific	Cat #M0052252G
Chromogenic solution (FerroZine)	MedChemExpress	Cat #HY-137805
FITC-annexin V	BioLegend	Cat #640914
DRAQ7	ThermoFisher	Cat #D15105
Critical commercial assays
CellTiter Glo Cell Viability Assay	Promega	Cat #G7573
RNeasy Mini Kit	Qiagen	Cat #74106
qScript cDNA SuperMix	Quanta Biosciences	Cat #95048-100
Fast SYBR Green Mastermix	ThermoFisher	Cat #4385618
Proteome Profiler Mouse Phospho-RTK Array Kit	R&D Systems	Cat#ARY014
Experimental models: Cell lines
OL61 (mouse glioma)	Castro-Lowenstein Laboratory, Univ. Michigan	See STAR Methods
KR158 (mouse glioma)	Deleyrolle Laboratory, Mayo Clinic	See STAR Methods
CPA (mouse glioma)	Deleyrolle Laboratory, Mayo Clinic	See STAR Methods
Experimental models: Organisms/strains
C57BL/6	JAX	RRID:IMSR_JAX:000664
NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ	Biological Resources Unit, Lerner Research Institute; Cleveland Clinic	–
B6.129P2-Lcn2tm1Aade/AkiJ	KAX	RRID:IMSR_JAX:024630
Oligonucleotides
LCN2 Forward: AGCCGTGGTAGATGGCTAGG	This paper	See Table S1
LCN2 Reverse: TTCCGCGGGTGACAGGGCTTC	This paper	See Table S1
GAPDH Forward: AGGCTGGTGACGAGGATTTG	This paper	See Table S1
GAPDH Reverse: TGTAGACCAGTATGTTGAGGTCA	This paper	See Table S1
LCN2 Forward: AGCCGTGGTAGATGGCTAGG	This paper	See Table S1
AXL Forward: ATGGCCGACATTGCCAGTG	This paper	See Table S1
AXL Reverse: CGGTAGTAATCCCCGTTGTAGA	This paper	See Table S1
OPCML Forward: CCACCCTCAGGTGTACCATAG	This paper	See Table S1
OPCML Reverse: GTAGGCGTGTTGACCAAAATGA	This paper	See Table S1
Gas6 Forward: TGCTGGCTTCCGAGTCTTC	This paper	See Table S1
Gas6 Reverse: CGGGGTCGTTCTCGAACAC	This paper	See Table S1
TENC1 Forward: TGAACCACTCAAAGCAACGCA	This paper	See Table S1
TENC1 Reverse: CGTTACATAGGTGAGGTCCAAGT	This paper	See Table S1
Recombinant DNA
psPAX2	Addgene (Didier Trono)	RRID:Addgene_12260; Cat# 12260
pMD2.G	Addgene (Didier Trono)	RRID:Addgene_12259; Cat# 12259
MISSION pLKO.1-puro Non-Mammalian shRNA Control	Sigma	TRC #SHC002
Lcn2 shRNA KD1	Sigma	TRC #TRCN0000055328
Lcn2 shRNA KD2	Sigma	TRC #TRCN0000055331
Lcn2 shRNA KD3	Sigma	TRC #TRCN0000055332
Software and algorithms
PRISM	GraphPad	https://www.graphpad.com/
FlowJo v10	BD Biosciences	https://www.flowjo.com/
QuantStudio 3	Applied Biosystems	https://www.thermofisher.com/us/en/home/technical-resources/software-downloads/quantstudio-3-5-real-time-pcr-systems.html
ImageJ	Schneider et al.	https://imagej.net/ij/
Incucyte S3	Sartorius	https://www.sartorius.com/en/products/live-cell-imaging-analysis/live-cell-analysis-software

Highlights.

• Combination of AXL inhibition and LCN2 knockdown increases survival in mouse models

• Inhibition of GPX4 and AXL enhances ferroptosis when LCN2 is depleted

• Combination of AXL inhibition and lipocalin-2 knockdown increases survival in mouse models