Ferroptosis induces heterogeneous death profiles that are controlled by lysosome rupture

Summary

Ferroptosis is a lipid peroxide-dependent form of cell death that occurs in degenerative conditions and may be leveraged for cancer therapy. While numerous regulators are known to control its cell-autonomous execution, ferroptosis also has a collective property that involves propagation between cells, and this regulation has remained more obscure. Different modes of ferroptosis induction involving inhibition of the anti-ferroptotic enzyme GPX4, or depletion of glutathione can impact the collective death response differently, but mechanisms underlying “single cell” versus “propagative” ferroptosis are not well understood. Here we discover significant lysosome rupture occurring during propagative ferroptosis, and identify glutathione depletion as sufficient to convert GPX4 inhibition from an individual cell to collective response. We find that induction of single cell ferroptosis involves heterogenous death profiles, with necrosis and apoptosis occurring in parallel within cell populations. These findings identify factors that control propagation and underscore lysosomes as critical to the execution of ferroptosis.

Graphical Abstract

eTOC

Das et al. show that the death mechanism ferroptosis is regulated in cell populations by the parallel induction of necrotic and apoptotic modes of execution. Death propagation, a unique feature of necrosis, involves stress on lysosomal membranes that leads to rupture and is sensitized by depletion of glutathione.

Introduction

Ferroptosis is an iron and lipid peroxide-dependent form of cell death that is damaging to normal tissues but is a therapeutic target for cancer due to its ability to eliminate cells that are otherwise resistant to death^1,2. We^3,4 and others^5–7 have shown that ferroptosis also has a unique, non-cell-autonomous activity to propagate between cells and eliminate large populations in a wave-like manner. While cell-autonomous regulators of ferroptosis are now well described⁸, how this collective death phenotype is controlled is not well understood.

Numerous mechanisms have been shown to modulate ferroptosis sensitivity by affecting cellular iron content, redox potential, or the lipid composition of membranes⁸. Ferroptosis can also be induced by depleting cells of glutathione, or acutely inhibiting the activity of the antioxidant enzyme GPX4, which leads to the accumulation of lipid peroxides and necrotic cell death². We³ and others⁹ have shown that the delivery of iron into lysosomes, for example through endocytosis of specialized nanoparticles called C’ dots, can also induce ferroptosis, suggesting that this cellular compartment could play an important role in regulating cell death³. Recent studies have also shown that lysosomal membranes are a target of lipid peroxidation and their destabilization can contribute significantly to ferroptosis induction^10–12.

While death propagation is still poorly understood, it has been shown that only particular forms of ferroptosis can lead to a collective response that eliminates large cell populations, for example treatments that deplete cells of glutathione, or load excess iron into cells through lysosomes. Direct inhibition of GPX4, on the other hand, leads to what has been described as “single cell” ferroptosis, which is less propagative and resembles, with respect to population dynamics, other forms of death that are regulated cell-autonomously like apoptosis⁴. Here we investigate cell population responses to ferroptosis induction and uncover a role for lysosome stress in regulating collective ferroptosis. We further show that heterogeneous forms of death occur within cell populations in response to GPX4 inhibition, and identify glutathione as a key factor whose depletion can convert a single cell death induced by GPX4 inhibition to a collective form of death.

Results

Ferroptotic cell populations exhibit heterogenous death profiles

To investigate population dynamics in ferroptosis, time-lapse imaging was first performed with GPX4-inhibited cells to carefully examine cell morphologies as individual cells underwent death through “single cell” ferroptosis⁴. During propagative ferroptosis, death has been shown to be uniformly necrotic in morphology and kills large cell populations^3,4. However, we found that cell death induced by treatment with the GPX4 inhibitor ML162 often failed to kill a majority of cells and was also morphologically complex, with necrosis occurring alongside deaths with classic apoptotic features, including cell blebbing and fragmentation, delayed membrane rupture, and the presence of condensed or fragmented nuclei (Fig. 1A–C). Cell deaths with mixed features of necrosis and apoptosis, for example cell blebbing and contraction in the absence of nuclear changes, were also observed (Fig. 1B). Heterogenous death profiles occurred in every cell type we examined, but to different extents. HT1080 fibrosarcoma cells exhibited mostly necrotic morphologies with some apoptotic features, whereas HeLa cervical carcinoma cells, A549 lung adenocarcinoma cells, U2OS osteosarcoma cells, HaCaT keratinocytes, and MCF10A mammary epithelial cells showed significant heterogeneity, in some cases with necrosis representing a minor fraction of cell deaths (Fig. 1B). Time-lapse imaging of MCF10A cells expressing a fluorescent reporter of the nucleus (H2B-mCherry) revealed minimal evidence of cell or nuclear condensation or fragmentation prior to cell rupture for necrotic deaths (Fig. 1C and Supplementary Video S1). Deaths with apoptotic morphologies, on the other hand, exhibited clear nuclear condensation or fragmentation occurring with cell blebbing (Fig. 1C and Supplementary Video S1). Heterogeneous, cell population-scale death profiles were evident even within individual colonies grown from cells plated at low density, suggesting that different cell death fates are unlikely to be controlled genetically (Fig. 1D, Supplementary Video S2). To further explore how GPX4 inhibition leads to heterogeneous death fates, we focused mechanistic studies on MCF10A cells, which exhibit clear necrotic or apoptotic deaths in response to GPX4 inhibition with minimal involvement of mixed morphologies (Fig. 1B, C), and which we have shown previously can undergo wave-like ferroptosis propagation^3,4.

Figure 1. GPX4 inhibition results in complex morphological patterns of death.

(A) Percent cell death from 24-hour time-lapse imaging is shown for the indicated cell lines treated with ML162. Bars show mean percent death from three independent experiments; error bars show SD (MCF10A: 100 cells per replicate; HaCat: minimum of 86 cells per replicate; A549: minimum of 100 cells per replicate; HT1080: minimum of 79 cells per replicate; Hela: 100 cells per replicate; U2OS: 150 cells per replicate). (B) Cell deaths recorded in part A are categorized by morphological type. Images show representative examples of ML162-treated HaCat cells undergoing necrotic (top), apoptotic (middle), or mixed morphological (bottom) types of death. Sytox orange labeling of the nucleus indicates plasma membrane rupture and nuclear morphology. (C) Images show examples of ML162-treated MCF10A cells expressing H2B-mCherry and undergoing necrosis (left) or apoptosis (right). Insets shown on top show time-lapse images of nuclear and cell morphology; times are indicated in minutes. Note the apoptotic cell nucleus labels with Sytox green after more than 8 hours. See Supplementary Video S1. (D) Representative cell colony exhibiting heterogeneous morphological death patterns in response to treatment with ML162, from time-lapse analysis. Necrotic cells are highlighted by Sytox green labeling. H2B-mCherry-labeled nuclei are shown in red. Insets show example of apoptotic and necrotic cell from same colony; bottom images show nuclear morphology. See Supplementary Video S2. All scale bars = 10μm.

Death subtypes are shared by sibling cells and differ in propagative activity

To begin to define cell properties that might underlie heterogeneous deaths, the relative cell cycle position of individual cells was first determined by time-lapse imaging and examined for effects on ML162-induced cell death. While a small effect of relative cell cycle position was revealed (Fig. 2A), no significant correlations with death morphology were observed. We also examined cell relationships by quantifying the fates of sibling cells that are directly related from a last mitotic event, a parameter that has been shown to correlate with sensitivity to death in response to treatment with the apoptosis inducer TRAIL¹³. Sibling cells showed a correlation for death type, with nearly 75% concordance between these cell pairs for necrotic or apoptotic deaths, suggesting that different death fates can be transmitted between cells in transient lineages that are carried through mitosis (Fig. 2B–E and Supplementary Videos S3 and S4).

Figure 2. Morphological death types are specified by transient lineages.

(A) Relative cell cycle position determined by time of last mitosis for cells undergoing the indicated death responses to ML162, shown by timing from last division to treatment with ML162. Data points show individual cells (n=114) from two biological replicates; bars indicate SD. No significant differences are observed between death types (One-Way ANOVA). (B) Sibling cells show concordant morphological death fates significantly more frequently than random cell pairs. Bars show the proportion of matching fates for the indicated cell groups. A chi-squared test of independence was used to compare agreement rates between the two groups, revealing a statistically significant association between sibling status and fate matching (χ² = 40.86, ****p<0.0001). (C) Pairwise analysis of death fates; sibling A and sibling B have matched fates in center blue boxes (n=57 sibling pairs). (D, E) Representative images of sibling cell analyses. Cells were time-lapse imaged by DIC microscopy (top images) to determine sibling relationships and then treated with ML162 and imaged for nuclear morphology (H2B-mCherry) and membrane permeabilization (Sytox green) to indicate death type (bottom images). Examples of sibling cells undergoing necrosis (D) versus apoptosis (E) are shown. Arrows indicate sibling cells; times of death for each sibling relative to the starting image are shown in minutes. See Supplementary Videos S3 and S4. Scale bars = 10μm. (F) Schematic illustration of random (left) versus spatially segregated (right) cell death fate organization. In random distributions (left), cells are surrounded by a proportional number of apoptotic and necrotic neighbors. In spatially segregated regions, cells are surrounded by other cells with the same death fate (right). Shaded grey areas illustrate examples of a segregated (right) versus random (left) spatial death fates. (G) Four biological replicates of colonies with heterogeneous death fates were spatially segregated with SSI values greater than 1. All segregation indices were statistically significant with p-values < 0.001 (***). Scale bars correspond to 100μm. (H) SPI was calculated for each cell fate for the same four biological replicates. In each replicate, necrotic death exhibits a higher propagation (SPI) than the corresponding apoptotic death. All biological replicates exhibited statistically significant SPI values with p-values p < 0.001 (***) or p-values < 0.03 (*).

Given that sibling cells share common death fates, and they also tend to grow in close proximity, we anticipated that sibling relationships might induce segregated regions of homogeneous cell death within colonies, as can be observed in Figure 1D. To quantitatively verify the observation of spatially segregated death types, inspired by Sadahiro (2019)¹⁴, we devised the Spatial Segregation Index (SSI), a measurement for the non-random spatial organization of death fates in colonies (Supplementary Fig. S1). When death fates are randomly distributed in space, each cell would be expected to have a number of neighboring cells undergoing apoptosis and necrosis in proportions that roughly align with the fraction of these death subtypes across the total population (Fig. 2F, left). However, if death fates are segregated, then each cell would be expected to have a larger fraction of neighboring cells with the same fate (Fig. 2F, right). For each cell we calculated the fraction of neighboring cells with the same death fate and defined the SSI as the ratio of the mean fraction of concordant neighboring deaths to the fraction of death fates within the total population. When death fates are randomly distributed, this ratio is approximately 1, indicating no spatial segregation. SSI values greater than 1 reflect the extent (in fold change) to which cells with the same death fate are more likely to be located near each other than expected by chance. For example, an SSI value of 1.5 indicates a segregation fold change of 1.5 with respect to the random fate distribution. The SSI therefore provides a quantitative readout of spatial organization of heterogeneous death patterning. To quantify the spatial organization of heterogeneous deaths induced by ML162, we applied SSI to different death fates (necrosis versus deaths with apoptotic features) in five independent colonies, including two closely localized colonies with nearly pure but opposing fates, and three additional colonies with intermixed, heterogeneous fates. In all cases, both death fates yielded values consistently greater than 1, and were deemed statistically significant using a permutation test (see Methods), thus confirming spatial segregation for each death fate (Fig. 2G).

We next assessed the propagation potential of these two different types of death, necrosis versus deaths with apoptotic features, by applying a metric we developed previously to study ferroptosis propagation called the Spatial Propagation Index, or SPI⁴. Briefly, for each (spatially segregated) death fate, the SPI quantifies the ratio between the mean difference of death times between neighboring cells, and the expected difference between the death times of spatially randomized cells. In this case, due to the spatial segregation of death fates, and the expectation that propagation occurs over short distances, the SPI was calculated by restricting the analysis to neighboring cells with the same fate (see Methods). The SPI values, calculated separately for both fates in each of the four samples, indicate that necrotic deaths are more propagative than deaths with apoptotic features (Fig. 2H). This result is consistent with our previous results for samples with homogeneous fates⁴, and shows that regions of necrosis induced by GPX4 inhibition are locally more propagative than regions with apoptotic features within complex population-scale death responses. These results also suggest that spatial segregation of cell fates occurs due to at least two parameters: transient lineage effects resulting from sibling relationships, and localized spatial propagation occurring between adjacent necrotic cells.

Amino acid starvation sensitizes to necrotic fates

To further explore heterogeneous death responses, we considered altering the culture conditions to try to identify factors that could sensitize to propagation. We have shown previously that propagative necrosis can be induced by treatment with C’ dot nanoparticles or Ferric Ammonium Citrate (FAC), which deliver iron into cells through endocytosis, particularly when cells are also starved for amino acids³. As shown in Figure 3A, while amino acid starvation of MCF10A cells did not cause much cell death on its own, it had a significant and accumulating effect on cell death induced by GPX4 inhibition, as 3, 6 and 12 hours of starvation increasingly sensitized to ML162-induced death, first with heterogeneous death profiles, and then with increasingly more homogeneous necrosis. Starvation for 24 hours sensitized to pure ML162-induced necrosis that eliminated nearly all cells cultured either in pools or colonies, consistent with a collective death response (Fig. 3A, B and Supplementary Video S5). Like MCF10A cells, amino acid starvation also sensitized to ML162-induced necrosis in A549 lung carcinoma cells (Supplementary Fig. S2).

Figure 3. Amino acid starvation converts GPX4 inhibition-induced death into collective necrosis.

(A) Cell death percentages broken down by death type from 24-hour time-lapse analysis are shown for ML162-treated MCF10A cells. Cells were starved for amino acids for the indicated times prior to treating with ML162. Death percentage for control 24h amino acid-starved cells in the absence of ML162 is shown in left bar. Bars show means from three independent biological replicates (n=100 cells per replicate); error bars show SD. p-values for cell death induced by ML162 compared to 24h amino acid-starved untreated control, from left to right for necrosis: **p=0.0013, ****p<0.0001, **** p<0.0001, ****p<0.0001; p-values for apoptosis compared to 24h amino acid-starved untreated control, from left to right: ***p=0.0003, ***p=0.0003; p-value for 24h necrosis compared to 3h: ****p<0.0001 (Two-Way ANOVA). Right images show representative fields of view from time-lapse imaging of cell populations starved for 6 or 24 hours prior to treatment. Boxed regions shown as insets are representative apoptotic and necrotic deaths; note all cells are necrotic in 24-hour starved population. Scale bars = 10μm. (B) Image shows MCF10A colony starved for amino acids for 24 hours and then treated with ML162 and imaged by time-lapse microscopy. Necrotic cells are shown by Sytox green labeling of the H2B-mCherry nuclei (red). See Supplementary Video S5. Scale bar = 10μm. (C) Analysis of death percentages and death types in MCF10A cells cultured in full media or starved conditions as in part A, but for cells treated with nano-RSL3. Bars shows means from three independent biological replicates; error bars show SD (n=100 cells per replicate). Note Liproxstatin-1 (Lip) inhibits cell death. p-values for necrosis compared to 24h amino acid-starved untreated control, from left to right: ****p<0.0001, ****p<0.0001; p-value for 3h apoptosis compared to 24h amino acid-starved untreated control: *p=0.03; p-value for 24h + Lip compared to 24h: ****p<0.0001 (Two-Way ANOVA). (D) Representative images of fields of view from time-lapse analysis of nano-RSL3 treated cells, similar to images shown in part A. Scale bars = 10μm. (E) Representative image from time-lapse analysis of nano-RSL3 and Liproxstatin-1 treated cells, similar to images in part D. Scale bar = 10μm. (F) Left graph: cell death percentages broken down by death type from 24-hour time-lapse analysis are shown for FINO₂-treated MCF10A cells (key for death types shown in parts A and C). Cells were starved for amino acids for the indicated times prior to treatment. Death percentage for control 24h amino acid-starved cells in the absence of FINO₂ is shown in left bar. Bars show means from three independent biological replicates (n=100 cells per replicate); error bars show SD. p-value for necrosis induced by FINO₂ in full media compared to 24h amino acid-starved untreated control, ***p=0.0009; p-values for necrosis induced by FINO₂ in starved conditions plus or minus zVAD-fmk, compared to full media, from left to right, **p<0.0014, **p<0.0074; p-value for apoptosis in full media compared to 24h amino acid-starved untreated control, *p=0.0122; p-value for apoptosis induced by FINO₂ in starved conditions plus zVAD compared to full media, ****p<0.0001. Right graph: cell death percentages are shown for FINO₂-induced deaths in 24 hour starved cells in the presence or absence of Liproxstatin-1 (Lip), DFO, and ConA, broken down by death type from 15-hour time-lapse analysis. P-values for necrosis of FINO₂-treated compared to control, ****p<0.0001; Lip compared to FINO₂-treated, ****p<0.0001; DFO compared to FINO₂-treated, ****p<0.0001; and ConA versus FINO₂-treated ****p<0.0001 (Two-Way ANOVA). (G) Graph shows amounts of glutathione (μM) from MCF10A cells in full media or starved for amino acids for the indicated times. Data points show individual biological replicates; bars show means, error bars show SEM. **p=0.0042 (One-Way ANOVA with Dunnett’s multiple comparisons test).

In addition to ML162, administration of the structurally distinct GPX4 inhibitor RSL3 also led to deaths that were heterogeneous in full media and sensitized to homogeneous necrosis by amino acid starvation over time (Fig. 3C,D). Treatment with the lipophilic antioxidant and ferroptosis inhibitor Liproxstatin-1 nearly completely inhibited cell death, consistent with the key role of lipid peroxidation in death induced by GPX4 inhibition even in starved cells (Fig. 3C,E). We finally considered if treatment with another distinct ferroptosis inducing compound, the endoperoxide FINO₂, which induces ferroptosis by oxidizing iron and indirectly inhibiting GPX4¹⁵, might also lead to heterogeneous deaths that would be influenced by amino acid availability. As shown in Figure 3F, treatment with FINO₂ induced markedly heterogeneous death responses with apoptosis and necrosis occurring in parallel, in a manner where necrosis was again sensitized by amino acid starvation. Death responses were inhibited by treatment with Liproxstatin-1 and the iron chelator deferoxamine (DFO), as well as the lysosomal v-ATPase inhibitor Concanamycin A (ConA) (Fig. 3F), demonstrating a role for iron and lysosome function in regulating cell death. Apoptosis was also specifically inhibited by treatment with the pan-caspase inhibitor zVAD-fmk (Fig. 3F).

Death subtypes are controlled by levels of glutathione

To investigate how amino acid starvation can promote necrotic death, we considered if it might sensitize cells by depleting glutathione, the tri-peptide co-factor of GPX4, which we found was depleted over time in amino acid-starved cells (Fig. 3G). As shown in Figure 4A, cysteine/cystine supplementation was able to rescue the effect of amino acid starvation, reducing the amount of ML162-induced death while also enhancing death heterogeneity. Treatment of cysteine/cysteine-rescued cells with the glutathione synthesis inhibitor buthionine sulfoximine (BSO) re-sensitized to complete death induction with homogeneous necrotic morphologies, consistent with a model that depletion of glutathione over time sensitizes to collective ferroptosis in response to acute GPX4 inhibition.

Figure 4. Death responses are controlled by glutathione.

(A) Cysteine/cystine supplementation rescues death of MCF10A cells starved for 3 hours and treated with ML162, in a BSO-inhibitable manner. Bars show mean percent cell death broken down by death type from three independent biological replicates; error bars show SD (n=100 cells per replicate). p-values from left to right: ***p=0.0005, ***p=0.0003 (One-Way ANOVA). Images show representative fields of view from time-lapse analysis. Scale bars = 10μm. (B) GPX4^KO HT1080 cells show heterogeneous deaths in response to Trolox washout. Images show representative field of view from timelapse microscopy with necrotic (black box) and apoptotic (blue box) fates highlighted in right images. Times are shown as minutes (min). Scale bar = 10μm. See Supplementary Video S6. (C) BSO sensitizes GPX4^KO HT1080 cells to necrotic fates after Trolox washout. Left graph: bars show mean percent cell death broken down by death type from three independent biological replicates; error bars show SD (n=100 cells per replicate). p-values from left to right for necrosis: ****p<0.0001, **p=0.0075, ****p<0.0001; p-values for apoptosis: ****p<0.0001 (Two-Way ANOVA). Right graphs: data show relative glutathione levels (left) and GSH/GSSG ratios (right) after Trolox washout in the presence or absence of BSO. Data are from three independent replicates; error bars show SEM. Left: *p=0.0237 (Paired t test); Right: *p=0.0223 (Paired t test).

To further explore how glutathione controls heterogeneous death responses, we turned to GPX4 knockout HT1080 cells, published previously¹⁶, which are normally grown in the presence of the antioxidant Trolox to inhibit the accumulation of lethal ROS, but can be induced to undergo ferroptosis by Trolox washout. Death responses to Trolox washout were imaged by time-lapse microscopy, which revealed heterogeneous fates with apoptotic modes of execution occurring alongside necrosis (Fig. 4B and Supplementary Video S6). Heterogeneous death responses were inhibited by treatment with Liproxstatin-1, while apoptotic deaths were specifically blocked by treatment with the pan-caspase inhibitor zVAD-fmk (Fig. 4C). Necrotic fates were sensitized by the addition of BSO to the culture medium following Trolox washout, indicating that glutathione inhibition can sensitize toward necrotic death fates even in GPX4 knockout cells (Fig. 4C). Like amino acid-starved MCF10A cells, GPX4 knockout HT1080 cells exhibited decreased levels of glutathione, as well as increased glutathione oxidation, in response to treatment with BSO after four hours of Trolox washout and prior to the initiation of cell death (Fig. 4C).

Lysosome stress occurs in GPX4-inhibited cells and lysosome rupture correlates with necrosis

To examine how mixed death responses might be regulated intracellularly, we further sought to investigate markers of lysosomal stress, as lysosomes have been implicated in ferroptosis regulation but with complex roles^12,17,18. We first expressed a reporter of lysosome rupture, a GFP-tagged Galectin 3 protein (GFP-Gal3), which is cytosolic when lysosomes are intact, but binds to the luminal face of lysosomes that undergo rupture, forming bright puncta that are observable by light microscopy¹⁹. As shown in Figures 5A and B and Supplementary Video S7, necrotic deaths were observed to have large numbers of lysosomes that ruptured prior to Sytox labeling. Deaths with apoptotic-like morphologies, on the other hand, had virtually no evidence of lysosome rupture, while mixed morphological deaths, with features of both apoptosis (e.g. cell blebbing) and necrosis (absence of nuclear changes), displayed intermediate amounts of rupture, demonstrating that lysosome rupture is a feature that correlates closely with different morphological types of death in GPX4-inhibited cells (Fig. 5A, B and Supplementary Video S7). Cells cultured as colonies and starved for amino acids for 24 hours prior to treatment with ML162 exhibited collective necrosis that was characterized by large numbers of GFP-Gal3 puncta appearing in every cell, indicating that necrotic responses sensitized by a loss of glutathione also exhibit significant lysosome rupture (Fig. 5C and Supplementary Video S8). We next expressed a GFP-tagged version of the TFEB transcription factor (Transcription Factor EB), a key regulator of lysosome biogenesis whose translocation to the nucleus is regulated by lysosomal signaling and can be induced by lysosome stress, as well as lysosome damage^20–22. TFEB was observed to translocate into the nucleus in response to GPX4 inhibition in all cells, irrespective of death type, but with different kinetics. TFEB translocated most rapidly in necrotic cells compared to mixed and apoptotic morphologies, which showed an increasing delay in TFEB activation (Fig. 5D,E and Supplementary Video S9). Taken together, these data demonstrate that the inhibition of GPX4 leads broadly to lysosomal stress and damage, but in a heterogenous manner, where the relative extent of lysosome stress correlates with different death responses.

Figure 5. Ferroptosis is regulated by lysosome damage.

(A) The presence of ruptured lysosomes indicated by GFP-Gal3 puncta correlates with death type. Graph shows number of GFP-Gal3 puncta per cell prior to cell death determined by time-lapse imaging; data points show individual cells, bars show means from four independent biological replicates, error bars show SD (n = 314 cells). Images below graph show representative cell regions either without (apoptosis) or with (necrosis) GFP-Gal3 puncta. ****p<0.0001 (One-Way ANOVA). (B) Representative images of GFP-Gal3-expressing MCF10A cells undergoing ML162-induced necrosis (left) or apoptosis (right). Times are indicated as minutes. Box regions highlight insets show in part A. See Supplementary Video S7. (C) Amino acid starvation sensitizes to lysosome rupture in response to ML162. Graph shows percent cells from colonies treated with ML162 in full media (left) or 24-hour amino acid-starved conditions (right) exhibiting lysosome rupture (with at least three GFP-Gal3 puncta), determined by 24-hour time-lapse imaging. Data points show means from three independent replicates; bars show SD. At least 100 cells were analyzed for each replicate. *p=0.0199 (Welch’s t test). Images show representative colony of GFP-Gal3-expressing MCF10A cells starved for 24 hours and then treated with ML162. Time-lapse from colony imaging is shown in large images from top left (0 min) to right (90 min) and bottom left (305 min), indicated by black arrowheads. Insets show boxed individual cell over time. See Supplementary Video S8. (D) TFEB is translocated to the nucleus in response to treatment with ML162. Images show MCF10A cells expressing TFEB-GFP and undergoing necrosis (left) or apoptosis (right) in response to treatment with ML162. Times are shown as minutes. White hatched regions highlight cell boundaries; blue circles highlight nuclei. See Supplementary Video S9. (E) Top graph shows percent cells with nuclear translocation of TFEB-GFP in control or ML162-treated conditions, broken down by death type determined by 24-hour time-lapse imaging. Bars show means from three independent biological replicates. At least 60 cells were analyzed for each replicate. Bottom graph shows the timing of nuclear TFEB-GFP translocation following treatment with ML162, broken down by death type from the data shown in top graph. Data points represent individual cells, bars show means from three independent biological replicates and error bars show SD. ****p<0.0001. At least 60 cells were analyzed for each replicate (One-Way ANOVA). (F) Treatment with zVAD-fmk inhibits apoptosis of ML162-treated MCF10A cells. Cells were treated with zVAD-fmk one hour prior to ML162 addition. Graph shows cell death percentages broken down by death type from 24-hour time-lapse analysis. Bars show means from three independent biological replicates; error bars show SD (n=100 cells per replicate). Control untreated cells are shown in right bar. p-value for apoptosis ***p=0.0067; note, necrosis in zVAD-fmk--treated cells shows no significant change (ns) (Two-Way ANOVA). (G) Cells rescued from apoptosis by treatment with zVAD-fmk show GFP-Gal3 puncta indicating the presence of ruptured lysosomes. Top graph shows percent viable cells from 24-hour time-lapse analysis with three or more GFP-Gal3 puncta appearing during imaging after treatment with ML162, in the presence or absence of zVAD-fmk. Bottom graph shows number of GFP-Gal3 puncta per cell in ML162 and zVAD-fmk-treated viable or necrotic cells. For both graphs, bars show means and error bars represent SD, *p=0.036; ****p<0.0001 (Welch’s t-test). Right images show representative GFP-Gal3 puncta appearing in viable cells treated with ML162 and zVAD-fmk. Inset shown in top image. See Supplementary Video S10. All scale bars = 10μm.

Caspase inhibition rescues apoptosis and does not increase necrosis

To further explore heterogeneous population-scale death responses, we considered the possibility that lysosome stress may eventually lead to more necrotic deaths over time in cell populations if apoptosis, which is not linked to lysosome rupture, was blocked. To investigate this, we treated cells with an inhibitor of caspases, zVAD-fmk, and quantified death rates and morphological features by long-term time-lapse imaging. As shown in Figure 5F, treatment with zVAD-fmk significantly reduced the percentage of cells that underwent apoptosis; however, this did not lead to increased rates of necrosis over 24 hours and had an overall decreasing effect on death rates. Interestingly, imaging of GFP-Gal3 in cells treated with ML162 and zVAD-fmk revealed the presence of ruptured lysosomes appearing in viable cells over time (Fig. 5G and Supplementary Video S10). The number of ruptures per cell, indicated by quantifying GFP-Gal3 puncta, was significantly lower in viable, apoptosis-inhibited cells compared to those that underwent necrosis (Fig. 5G), suggesting again that heterogeneous fates in GPX4-inhibited cells correlate tightly with differential levels of lysosome stress, and that apoptotic cells do not reach the levels of lysosome rupture that are linked to necrosis.

Lysosomal enzymes regulate necrotic death

To further explore the significance of the link between lysosome rupture and ferroptotic death, we considered if the release of cathepsin proteases from lysosomes could regulate death execution, as cathepsins have previously been implicated in ferroptosis¹². Cells expressing GFP-Gal3 were treated with ML162 and inhibitors of cathepsin proteases (zFA-fmk) or cathepsin B (CA-074), and the fates of cells with ruptured lysosomes were quantified. As shown in Figure 6A and Supplementary Video S11, while all cells exhibiting lysosome rupture underwent necrosis in control conditions, a proportion of cells treated with cathepsin inhibitors died instead with apoptotic or mixed morphologies, and a small number were rescued from death. Cathepsin-inhibited cells that died with necrotic morphologies also showed a delay in membrane rupture, as measured by the relative timing of release of GFP from the cytosol compared to Sytox labeling of the nucleus (Fig. 6B and Supplementary Video S12). We further examined death responses to FINO₂ utilizing MCF10A cells expressing H2B-mCherry, and found, again, that cathepsin inhibition significantly increased the time that it takes for necrotic cells to rupture, visualized in this case by retention of the H2B-mCherry signal in the cytoplasm (Fig. 6C). Taken together, these findings support a model where lysosome damage occurring in response to ferroptosis inducers leads to the release of lysosomal enzymes, including cathepsin B, whose activity contributes to rupture of the plasma membrane during the execution of necrosis.

Figure 6. Ferroptotic death responses are regulated by lysosomal enzymes.

(A) Top graph shows death fates for MCF10A cells expressing GFP-Gal3 and treated with ML162 and the cathepsin inhibitors zFA-fmk or CA074. Cells were treated with cathepsin inhibitors one hour prior to ML162 addition. Data points represent the timing of the indicated cell fates for individual cells exhibiting lysosome rupture (with at least three GFP-Gal3 puncta), determined by 24-hour time-lapse imaging. The timing of necrosis was determined by loss of GFP from the cytoplasm (see Supplementary Video S11). Apopt refers to cell deaths with the presence of apoptotic features (cell blebbing or nuclear condensation or fragmentation). The bottom graph shows mean percent death types for the same conditions as in the top graph for cells with lysosome rupture. Bars indicate means from three independent biological replicates; error bars show SD (n = at least 170 cells in each replicate). p-values from left to right: *p<0.016, ***p=0.0002, ***p=0.0008, ****p<0.0001, **p=0.0068 (Two-Way ANOVA). (B) Representative images from part A showing necrotic and apoptotic deaths in cells with lysosome rupture. Left image shows example of necrosis with rapid membrane rupture, determined by the timing of loss of cytoplasmic GFP with respect to Sytox labeling of the nucleus. Middle image shows necrotic death with slow membrane rupture, a more common outcome in cathepsin-inhibited cells (see graph in part A). White regions show cell borders, blue boxed regions highlight insets show at bottom for left and middle images. See Supplementary Videos S11 and S12. Scale bars = 10μm. (C) Graph shows timing of necrotic rupture for MCF10A cells expressing H2B-mCherry and treated with FINO₂ in the presence or absence of zFA-fmk. Data points show individual necrotic ruptures determine by the release of mCherry from the cytoplasm in timelapse imaging. Data are from three independent replicates; total n=156 cells for control and 170 cells for zFA-fmk. Bars show means and error bars show SD. ****p<0.0001 (Welch’s t test). Images show representative necrotic cell with delayed rupture, indicated by mCherry fluorescence contained in the cytoplasm. Top images show individual DIC and red channels from inset indicated in bottom merged image. Scale bar = 10μm. (D) Graph shows Lysotracker Deep Red intensity in cells stained with C11-BODIPY and imaged for 510nm fluorescence indicating lipid peroxidation. Cells with C11-BODIPY-510 on lysosomes do not exhibit peroxidation at the plasma membrane, whereas cells with C11-BODIPY-510 at the plasma membrane have low levels of Lysotracker staining indicating lysosome rupture. Data points show individual measurements for n=34 cells; bars show means and error bars SD. ***p<0.0002 (Welch’s t test). (E) Images show C11-BODIPY staining of FINO₂-treated MCF10A cells stained with Lysotracker Deep Red. Left top image shows C11–510 at lysosomes marked with Lysotracker; Lysotracker and C11–590 fluorescence are shown in insets. Top right images show C11–510 at the plasma membrane. Note Lysotracker staining (top right panel) is lost from this cell; C11–590 is shown underneath. Bottom left images show representative cell with lysosome rupture marked by GFP-Gal3 puncta that has loss of Lysotracker staining. Bottom right images show representative area of cytoplasm stained with C11-BODIPY in untreated cell; note no C11–510 staining is found at lysosomes under control conditions. All scale bars = 10μm. (F) Model, top row: ferroptosis induction by acute GPX4 inhibition involves morphologically complex death fates that manifest as necrosis occurring alongside apoptosis and deaths with mixed morphological features. Lysosome rupture (shown as puncta) correlates with necrotic morphology and decreases in deaths with apoptotic features. The transcription factor TFEB is activated in response to GPX4 inhibition in all cells. Bottom row: Collective ferroptosis occurs in GPX4-inhibited cells if they are first starved for amino acids, a condition that reduces levels of glutathione. Starved cells treated with GPX4 inhibitors exhibit uniform necrotic morphologies with a high degree of lysosome rupture.

We finally sought to examine if lipid peroxidation occurs at lysosomal membranes by using an imaging-based approach with the lipid peroxidation reporter C11-BODIPY. In FINO₂-treated cell populations stained with Lysotracker to mark acidified compartments, we indeed observed cells with lipid peroxidation at lysosomes (Fig. 6D, E). Interestingly, cells with lysosomal lipid peroxidation did not show evidence of peroxidation at the plasma membrane, while conversely, cells exhibiting a loss of Lysotracker staining, which is indicative of lysosome rupture as confirmed in GFP-Gal3-expressing cells, did show lipid peroxidation at the plasma membrane (Fig. 6D, E). As lysosome rupture precedes necrotic death, these data are consistent with the model that lipid peroxidation occurring at lysosomal membranes leads to lysosome rupture and, ultimately, necrotic death that is associated with lipid peroxidation at the plasma membrane. Altogether, our study shows that ferroptosis induction involves complex, population-scale patterns of cell death, where the sensitivity of individual cells to necrotic or apoptotic fates is a property that is shared between siblings. Collective necrosis, an activity that is promoted by the depletion of glutathione over time, is defined by significant lysosome rupture, and occurs through a mechanism that is regulated by lysosomal enzymes (see model, Fig. 6F).

Discussion

Here we investigated the dynamics of ferroptosis and uncovered complex regulation underlying both the elimination of individual cells through autonomous mechanisms and large groups of cells through collective necrosis. By carefully examining cell death morphologies over time, we found that the treatment of cells with GPX4 inhibitors, which we⁴ and others⁵ have shown has a limited ability to induce propagation, leads to complex and heterogenous death profiles, in which individual cells undergo morphologically distinct types of death in parallel to each other; from pure necrosis, to classic apoptosis, and deaths with mixed morphological features. Heterogeneous death profiles occur in cell populations and also colonies, in a manner where sibling cells often share the same fate, suggesting that transient lineage relationships can dictate how cells die in response to acute GPX4 inhibition. We show that apoptotic deaths in GPX4-inhibited cells fail to propagate like necrotic deaths, suggesting that the failure of GPX4 inhibitors to induce propagation in previous studies may be linked, at least in part, to the induction of heterogeneous deaths. We further find that starvation for amino acids can sensitize to pure necrosis by depleting glutathione, demonstrating that reduced antioxidant capacity is sufficient to tip the balance toward collective necrosis.

Our study also reveals that lysosomes become damaged during ferroptosis, and damage that is severe enough to cause organelle rupture leads to the execution of necrotic death that is regulated by lysosomal cathepsins. This conclusion is supported by previous reports that have also found lysosome membrane damage and permeabilization occurring in ferroptosis^10–12,23. While cathepsin B has been shown to be required for ferroptosis induction¹², we find here that cathepsins regulate the morphological types of death that occur in response to GPX4 inhibition, rather than the amount of death per se. Our imaging studies also reveal that inhibition of cathepsin activity slows the rate of necrotic cell rupture. Curiously, we have observed lysosomes rupturing close to the plasma membrane during ferroptosis propagation (see Supplementary Video S13), suggesting that the release of lysosomal enzymes, as well as iron¹⁰, in close proximity to the membrane could promote cell rupture and potentially affect nearby cells.

Why some cells and not others in a population are sensitive to lysosome rupture in full media conditions is, at present, unknown. We show that glutathione depletion can tip this balance, suggesting that heterogeneity between cells for antioxidant capacity could account for the observed differences. This, along with differences between cells in lysosomal activity or lysosome content, could contribute to set relative thresholds of lysosome “rupturability”, the latter suggested by a recent report that treatment with lysosomotropic agents can also sensitize to rupture¹⁰. Rupturability may also relate to particular cell states that are defined by expression signatures that mark cells with increased iron uptake^11,24. In cells treated with zVAD-fmk, a small number of ruptured lysosomes were observed, supporting a model that all cells in the population experience lysosome stress, but those that undergo apoptosis do not reach a threshold of rupture needed to cause necrosis, and instead die through alternative mechanisms.

Our data also reveal that sibling relationships correlate with death type, suggesting that lysosome rupturability is a property that persists long enough to be transmitted through mitosis, but may be otherwise short-lived, as individual cells within colonies that are not related by last division show more discordant death fates. Previously, it has been found that sibling cells share setpoints of sensitivity to apoptosis induced by treatment with TNF-Related Apoptosis Inducing Ligand (TRAIL), due to similar levels of expression of proteins that regulate apoptosis signaling¹³. Our data support a model that sensitivity to lysosome rupture relates siblings to a shared likelihood of dying by necrosis when challenged by acute GPX4 inhibition. We also show that different death fates are spatially clustered, an effect potentially linked to sibling relationships that would contribute to establishing such patterns. Regions of necrosis are also influenced by a collective property that can spread death locally; however, we note that apoptotic death fates are also shared by siblings despite the more limited potential of this form of death to propagate. Finally, localized positional effects unrelated to cell lineage or propagation may also play a role in spatial patterning. Future studies may reveal how sibling relationships, localized propagative activity, and other positional effects might interact to control the dynamics of heterogeneous death responses in cell populations.

Death propagation is an emerging feature of ferroptosis that may underlie, at least in part, its effects on tissue degeneration in pathological conditions such as acute kidney or cardiac ischemic injury, stroke, or neurodegeneration^25–28. Cardiac injury and stroke have also been shown to involve heterogeneous death responses, involving mixtures of ferroptosis and apoptosis, suggesting a possible clinical link to heterogeneous population-scale fates^8,29–31. The propagative activity of ferroptosis could further be beneficial for cancer therapy, as otherwise treatment resistant cells might become eliminated if neighboring cells die by this mechanism^27,32. While cell-cell contacts were recently shown to facilitate the spreading of lipid radicals between small clusters of cells, cell adhesions have also been shown to be dispensable for cell population-scale propagations that have a wave-like appearance⁴, and are capable of transiting large gaps between cells of more than 100 microns⁵. Our study reveals that lysosome rupture is a key feature of collective ferroptosis that can eliminate either small or large groups of cells, depending on levels of glutathione, a finding potentially linked to a recent report that ferroptosis propagation can involve a loss of the membrane-localized SLC7A11 cystine transporter, which would further suppress glutathione levels and could facilitate death spreading⁷. We show that glutathione suppression can sensitize even GPX4 knockout cells to collective necrosis, which demonstrates that this effect occurs independently of GPX4 and may relate to other glutathione-dependent enzymatic activities³³, or potentially the direct buffering of labile iron by glutathione itself³⁴. Increased lysosomal iron has been shown to sensitize lysosomal membranes to damage by enhancing the generation of ROS through fenton chemistry, a model favored here as well, as we show that lipid peroxidation occurs at lysosomal membranes prior to rupture²³. For lysosomes, we have shown previously that they become multivesicular in response to stress on the limiting membrane, the result of microautophagy activity that infolds portions of the limiting membrane, forming intraluminal vesicles²⁰. In this case, the rupture of lysosomes could contribute to vesicle pools that may spread lipid peroxides or other cellular material inside, and potentially also between cells^35–37, a model that could be pursued in further studies.

Limitations of Study

This study examines the death morphologies and propagative activity of ferroptosis using monolayer cultures of non-transformed and cancer cell lines, and media conditions that involve total amino acid starvation or chemical inhibition of the generation of glutathione. Further studies are needed to address whether heterogeneous death responses occur in organoid or in vivo models, and whether ferroptosis propagation in tissues is inhibited by glutathione acting independently of GPX4.

Resource Availability.

Lead Contact:

Michael Overholtzer, overhom1@mskcc.org.

Materials Availability:

This study did not generate new unique reagents.

Data and Code Availability:

• Data: Cell death data from Figure 2 have been deposited at Zenodo and are publicly available as of the date of publication at https://doi.org/10.5281/zenodo.18271042.

• Code: All original code has been deposited at Zenodo and is publicly available at https://doi.org/10.5281/zenodo.18271042 as of the date of publication.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR Methods.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell Culture

MCF10A cells (ATCC) were cultured in DMEM/F12 (Gibco, Grand Island, NY) supplemented with 5% horse serum (Atlanta Biologicals, Flowery Branch, GA), 20 ng/ml EGF (Peprotech, Rocky Hill, NJ), 10 μg/ml insulin (Sigma), 0.5 μg/ml hydrocortisone (Sigma), and 100 ng/ml cholera toxin (Sigma) with penicillin/streptomycin at 37 °C in 5% CO₂. Amino acid-free culture medium and the Cys only (Cystine.2HCl + Cysteine HCl) media was prepared by adding 20 ng/ml EGF (Peprotech), 10 μg/ml insulin (Sigma), 0.5 μg/ml hydrocortisone (Sigma), and 100 ng/ml cholera toxin (Sigma) with penicillin/streptomycin to the base media in the absence of serum (MSKCC Media Preparation Facility). HT-1080, U2OS, A549, Hela and HaCaT cells (ATCC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (MSKCC Media Preparation Facility) supplemented with 10% heat inactivated fetal bovine serum (FBS) (Sigma, St. Louis, MO) with penicillin/streptomycin (Corning, Corning, NY) at 37 °C in 5% CO₂. GPX4^KOHT1080 cells were maintained in 200μM Trolox as reported¹⁶. Cell lines were tested for mycoplasma using DAPI staining or the Mycoalert detection kit (Lonza #LT07–218) bimonthly. Note that MCF10A, U2OS, and Hela cells are female, while HT-1080, A549 and HaCaT cells are male cells.

Reagents and Constructs

L-buthionine sulfoximine (BSO) (B2515; Sigma-Aldrich) and Deferoxamine mesylate salt (DFO) (D9533; Sigma-Aldrich), was dissolved in water and stock solutions were stored at −20°C. BSO was used at 400μM and DFO was used at 100μM. All other compounds were prepared as stock solutions in DMSO and stored at −20°C except for ML162 (SelleckChem S4452, used at 4μM) which was stored at −80°C. Liproxstatin-1 (S7699; SelleckChem) was used at 200nM, zVAD-fmk (S7023; SelleckChem) was used at 100 μM, zFA-fmk (HY-P0109A; MedChemexpress) was used at 100 μM, CA-074 methyl ester (C5857; Sigma Aldrich) was used at 15 μM, Concanamycin A (HY-N1724; MedChemexpress) was used at 0.1μM, Trolox (S366; SelleckChem) was used at 200 μM, BODIPY^™ 581/591 C11 (D3861; Invitrogen) was used at a final concentration of 5μM, and FINO₂ (25096; Cayman Chemicals) was used at 20 μM. Nano-RSL3 (formulation described below) was used at a concentration of 2 μM. Lysotracker Deep Red (L12492; Invitrogen) was used at a concentration of 0.1 μM. Sytox Green (S7020; Molecular Probes) and Sytox Orange (S11368; Molecular Probes) was used at a concentration of 15nM for all experiments. GSH/GSSG-Glo^™ Assay (V6611; Promega) was used as per manufacturer’s guidelines. The GFP-Galectin-3, TFEB-GFP, and H2B-mCherry constructs and MCF10A expressing cell lines have been previously described^20,37–40.

METHOD DETAILS

Nano-RSL3 Formulation

RSL3 nanoparticles have been described⁴¹ and were prepared by a nanoprecipitation method adopted from previous work by the Heller lab^42,43. Briefly, in a microcentrifuge tube, an aliquot of 0.05 mL of RSL3 dissolved in DMSO (20 mg ml⁻¹) was added dropwise to an aqueous solution containing 550 μL water, 100 μL indocyanine green (ICG) (2 mg ml⁻¹) and 100 μL sodium bicarbonate buffer (0.1 M, pH 8). The resultant suspension was then centrifuged (20 min 30,000 rpm), the supernatant was removed, and the nanoparticle pellet was resuspended in 100 μL of sterile saline solution. Encapsulation efficiency was quantified by HPLC and calculated by (amount of total loaded drug / total amount of drug) * 100.

Live Cell Imaging

Cells were seeded on glass-bottom plates (P06G-1.5–20-F; Mattek) and treated in fresh culture media the next day. For amino acid-free conditions, cells were pre-treated with starved media or cysteine/cystine-rich media where indicated. Differential Interference Contrast (DIC) and fluorescence imaging was performed in live-cell incubation chambers maintained at 37°C and 5% CO₂. Images were acquired every 5 to 10 min for 24–48 hours using a Nikon Ti-E or Ti2-E inverted microscope attached to a CoolSNAP charge-coupled device camera (Photometrics) or sCMOS Monochrome Camera (Hamamatsu) camera and NIS Elements AR software (Nikon). Images were quantified manually in NIS Elements AR (Nikon) and processed using ImageJ⁴⁴ and Adobe Photoshop. For quantifications of sibling fates, cells were imaged by DIC only for 24 hours or longer and then treated with ML162 and imaged with DIC and fluorescence for another 24h to record cell death fates. Death fates were recorded for sibling cells that resulted from divisions prior to treatment with ML162; divisions occurring after ML162 treatment were excluded. For quantifications of mitotic timing with respect to cell death fates, times were recorded from last mitosis to treatment with ML162. For colony growth assays, cells were plated at low density in 6-well glass-bottom plates (e.g. 50 cells per well) and grown for 5–8 days prior to imaging in the presence of cell death-induced agents and Sytox dyes. For treatment with FINO₂, amino acid-starved cells (24 hours) were quantified for cell fates by time-lapse imaging for a total of 15h after treatment.

Confocal Imaging

Cells were plated on glass-bottom dishes (P35G-1.5–20-C; Mattek). For BODIPY^™ 581/591 C11 staining imaging, cells were washed twice with Hank’s Balanced Salt Solution (HBSS) (14025–092; Thermo Fisher) and stained in 5 μM C11-BODIPY581/591 in HBSS for 10 min at 37 °C and 5% CO2, and again washed twice in HBSS. Cells were imaged at 37 °C and 5% CO2 using the Leica SP8. Each cell was only imaged once to avoid photo-oxidation of the dye.

Quantification of Oxidized and Total Glutathione

5000 cells/well were seeded in a 96 well cell culture plate (3596; Corning) in 200 μL of media. Note that double the number of cells were seeded for the 24hr starvation condition to account for differing growth rates. Cells were lysed and analyzed for reduced glutathione (GSH) and oxidized glutathione (GSSG) levels with the GSH/GSSG-Glo Assay Kit (Promega) per manufacturer’s instructions. Luminescence (in RLU) was measured as for the reporter assay, and concentrations of total glutathione and oxidized glutathione were calculated using a set of glutathione standards. To calculate the total glutathione to oxidized glutathione ratio, the following formula was used:

$$\frac{GSH}{\mathit{\text{GSSG}}}=\frac{\mathit{\text{Net total glutathione RLU}}-\mathit{\text{Net GSSG RLU}}}{\mathit{\text{Net GSSG RLU}}/2}$$

This equation takes into account that each molecule of oxidized glutathione is stoichiometrically equivalent to two molecules of reduced glutathione.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistics and Reproducibility

Data were analyzed using Microsoft Excel (Office 2011), GraphPad Prism 10 and Python. The specific statistical tests used for each analysis are indicated in the corresponding figure legends. All experiments were independently replicated at least three times unless otherwise stated. Sample sizes (n) are reported in the figure legends. All tests were two-sided unless otherwise indicated and p-values indicated in the figure legends were considered statistically significant if < 0.05.

Data

To investigate the impact of heterogeneous cell death responses induced by ML162 on spatiotemporal death dynamics, we used four independent biological replicates of MCF10A cell colonies maintained in full media. Two independent replicates were imaged every 5 minutes for a total duration of 48 hours. ML162 was introduced to the cells after the initial 24 hours of imaging. These samples were imaged at a spatial resolution of approximately 0.643 μm/pixel. The remaining two samples were imaged every 10 minutes for 24 hours, starting immediately after ML162 treatment, at a spatial resolution of 1.290 μm/pixel. Across all replicates, initial cell counts ranged from 200 to 641.

Annotations and Preprocessing

The time of death and death fate were manually annotated based on morphology using DIC time-lapse images with the Multi-Point Tool in FIJI. For these analyses, all deaths with apoptotic features, including those with mixed morphologies, were counted together to compare to necrosis. The Sytox signal was used as an additional marker to confirm membrane rupture and necrotic death. Cells that remained alive at the end of the experiment constituted less than 6% of the total population in each of the four biological replicates and were excluded from downstream analysis. All downstream analyses were performed at a temporal resolution of 10 minutes; datasets originally acquired at 5-minute intervals were downsampled accordingly after annotation.

Spatial Segregation Index (SSI)

The Spatial Segregation Index (SSI) was defined for a given death fate in a heterogeneous death fate colony, as the ratio between the observed probability of adjacent neighboring cells sharing the same death fate and the overall proportion of that death fate in the population. This measurement was inspired by previous work of Sadahiro (2019)¹⁴. First, we used the Voronoi tessellation⁴⁵ to define the adjacent neighbors of each cell and then excluded Voronoi-neighbors that were more than 100μm apart (Supplementary Figure S1). Second, for each cell, we calculated the probability that its neighbouring cells would die with the same morphological death fate as the cell that was at test. The mean probability across all cells sharing the same morphological death fate was then divided by the overall fraction of that death fate within the experiment, yielding the spatial segregation index. Thus, the SSI reflects the fold change to which cells with the same death fate are more likely to be located near each other than expected by chance. For example, SSI = 1 Indicates that the death fates are randomly distributed with no spatial segregation. SSI > 1 indicates the extent of segregation.

To assess the statistical significance of the segregation we performed a non-parametric permutation test (aka bootstrapping). To reject the null hypothesis that different death fates were randomly distributed in space, we randomly permuted the death fates across all cells within the same replicate. i.e., each cell retained its spatial position but was assigned a random death type drawn from the original distribution of annotated death fates for that replicate. Following each permutation, we recorded the mean probability of neighboring cells sharing this same death fate. For each biological replicate (colony) and death fate, this procedure was repeated 1000 times and the p-value was calculated based on the fraction of permutations for which the derived SSI exceeded the observed SSI (Supplementary Figure S1). A p-value less than 0.05 was deemed statistically significant.

Spatial Propagation Index

The Spatial Propagation Index (SPI) was adapted, with minor modifications from⁴, to quantify local collective death propagation within each morphologically segregated death fate. First, all the adjacent neighboring cell pairs were identified, as described for SSI. Next, for each biological replicate, the mean death time difference between neighboring cell pairs sharing the same death fate was calculated yielding the experimental propagation measure denoted as as μ_expΔt.

To assess the hypothesis that there is a dependency between the times of death of neighboring cell pairs sharing the same segregated death fate, we performed a non-parametric permutation test. Specifically, we tested the null hypothesis that the timing of cell death was randomly distributed in space among cell pairs with the same death fate. We permuted the annotated death times across all such cells within each biological replicate. I.e., each cell retained its spatial position and death fate but was assigned a random time of death drawn from the same death fate times, in each biological replicate. Following each permutation, the mean time difference between neighboring cell pairs of matching death fate was calculated. This procedure was repeated 1000 times per biological replicate. The p-value was calculated as the fraction of times in which the mean time difference between the permuted neighboring cells within each segregated death fate, was smaller than the experimentally observed value, μexpΔt. We considered a p-value less than 0.05 as statistically significant. The SPI was defined as the ratio between the deviation of the experimental mean propagation (μexpΔt) from the 5th percentile of the mean randomly permuted death times (μ_perm5Δt) and μ_perm5Δt:

$$SPI=\frac{{\mu }_{\mathit{\text{perm}}5\Delta t}-{\mu }_{exp\Delta t}}{{\mu }_{\mathit{\text{perm}}5\Delta t}}$$

Higher SPI values indicate that neighboring cells are more synchronized in their death times than expected by chance, i.e., when the experimental death times are randomly shuffled across cells, suggesting the presence of local death propagation.

Supplementary Material

2

Video S1. Heterogeneous death fates occur in response to GPX4 inhibition, related to Figure 1. Images from time-lapse imaging show death responses of MCF10A cells expressing the nuclear marker H2B-mCherry to treatment with the GPX4 inhibitor ML162. Left images show DIC and right images show H2B-mCherry (red) and Sytox green fluorescence. Blue arrows indicate cells that die with apoptotic features, white arrows indicate necrosis. The indicated cells are also shown in Figure 1C. Images were collected every 5 minutes.

3

Video S2. Heterogeneous death fates occur in response to GPX4 inhibition in colonies, related to Figure 1. Images show death responses of MCF10A cells grown as a colony to treatment with the GPX4 inhibitor ML162. Merged images show DIC, H2B-mCherry (red) and Sytox green fluorescence. Colony and boxed region are also shown in Figure 1D. Images were collected every 10 minutes.

4

Video S3. Sibling cells share necrotic death fates, related to Figure 2. Images show necrotic death responses of sibling cells to treatment with the GPX4 inhibitor ML162. Cells were imaged by DIC microscopy to trace sibling pairs (left), and then treated with ML162 to induce cell death (right). Right images show H2B-mCherry (red) and Sytox green merged with DIC. Arrows indicate sibling cells. The indicated cells are also shown in Figure 2D. Images were collected every 5 minutes.

5

Video S4. Sibling cells share apoptotic death fates, related to Figure 2. Images show apoptotic death responses of sibling cells to treatment with the GPX4 inhibitor ML162. Cells were imaged by DIC microscopy to trace sibling pairs (left), and then treated with ML162 to induce cell death (right). Right images show H2B-mCherry (red) and Sytox green merged with DIC. The indicated cells are also shown in Figure 2D. Arrows indicate sibling cells. Images were collected every 5 minutes.

6

Video S5. Homogeneous necrotic death occurs in response to GPX4 inhibition in starved colonies, related to Figure 3. Images show death responses of MCF10A cells grown as a colony and amino acid-starved for 24h to treatment with the GPX4 inhibitor ML162. Merged images show DIC, H2B-mCherry (red) and Sytox green fluorescence. Colony is also shown in Figure 3B. Images were collected every 5 minutes.

7

Video S6. Heterogeneous death responses occur in GPX4^KO HT1080 cells, related to Figure 4. Images from time-lapse analysis show necrotic and apoptotic death responses of GPX4^KO HT1080 cells following Trolox washout. Merged images show DIC and Sytox orange fluorescence. White box highlights a necrotic death; blue box shows apoptosis. Images are also shown in Figure 4B. Images were collected every 5 minutes.

8

Video S7. Lysosome rupture occurs in cells that die through necrosis but not apoptosis, related to Figure 5. Images from time-lapse analysis show necrotic and apoptotic death responses of MCF10A cells expressing the lysosome rupture reporter Galectin-3-GFP (Gal3-GFP) to treatment with the GPX4 inhibitor ML162. Left images show DIC, right images show GFP-Gal3 (green) and Sytox orange (red) fluorescence. Boxed regions indicate cells shown in Figure 5A. Images were collected every 5 minutes.

9

Video S8. Lysosome rupture occurs in cells that die through necrosis under amino acid-starved conditions, related to Figure 5. Images from time-lapse analysis show necrotic cells dying in a 24-hour amino acid-starved colony of MCF10A cells expressing GFP-Gal3 and treated with ML162. Images were collected every 5 minutes. Cells are also shown in Figure 5C.

10

Video S9. TFEB is activated in all ML162-treated cells but with different kinetics depending on death fate, related to Figure 5. Images show necrotic and apoptotic death responses of MCF10A cells expressing TFEB-GFP to treatment with the GPX4 inhibitor ML162. Left images show DIC, right images show TFEB-GFP (green) and Sytox orange (red) fluorescence. Arrows indicate cells shown in Figure 5D. Images were collected every 10 minutes.

11

Video S10. Lysosome rupture is observed in cells that are rescued from apoptosis, related to Figure 5. Images show MCF10A-GFP-Gal3 cells treated with zVAD-fmk and ML162. Left images show DIC, right images show GFP-Gal3 (green) and Sytox orange (red) fluorescence. Note zVAD-fmk does not inhibit necrotic deaths, and viable cells show the appearance of GFP-Gal3 puncta over time. Images were collected every 5 minutes. See Figure 5G.

12

Video S11. Cathepsin inhibition leads to deaths with apoptotic features in cells with lysosome rupture, related to Figure 6. Images show MCF10A-GFP-Gal3 cells treated with zFA-fmk and ML162. Left images show DIC, right images show GFP-Gal3 (green) and Sytox orange (red) fluorescence. The cell that undergoes death is also shown in Figure 6B. Images were collected every 5 minutes.

13

Video S12. Necrotic cells can show different rates of rupture, related to Figure 6. Time-lapse images show MCF10A-GFP-Gal3 cells undergoing necrosis in response to ML162. Top images show cell exhibiting fast rupture during necrosis, bottom images show slow rupture indicated by the retention of GFP in a Sytox-positive cell. Left images show DIC, right images show GFP-Gal3 (green) and Sytox orange (red) fluorescence. Cells are also shown in Figure 6B. Note these movies start with the first image that showed at least 3 GFP-Gal3 puncta in each cell. Images were collected every 5 minutes.

14

Video S13. Lysosomes rupture close to the plasma membrane, related to Figure 6. Time-lapse images show MCF10A-GFP-Gal3 cells undergoing necrosis in response to ML162. Left images show DIC, right images show GFP-Gal3 (green) and Sytox orange (red) fluorescence. Note the presence of Gal3-GFP puncta in close proximity to the plasma membrane as deaths propagate through this cell region. Images were collected every 5 minutes.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
DMEM/F12	Gibco, Grand Island, NY	Cat#: 11320082
DMEM	Gibco, Grand Island, NY	Cat#: 11965118
Hank’s Balanced Salt Solution (HBSS)	Thermo Fisher	Cat#:14025-092
Horse Serum	Sigma-Aldrich	Cat#: H1138-500mL Lot#: 22B148
Fetal Bovine Serum	Sigma-Aldrich	Cat#: F2442-500mL Lot#18A648
EGF	Peprotech, Rocky Hill, NJ	Cat#: AF-100-15-500ug
Insulin	Sigma	Cat#: I1882-100mg
Hydrocortisone	Sigma	Cat#: H0888-1G
Cholera toxin	Sigma	Cat#: C8052-2MG
Penicillin/streptomycin	Gibco, Grand Island, NY	Cat#: MT30-002-CI
Amino acid-free DMEM/F12 medium	MSKCC Media Preparation Facility	https://www.mskcc.org/research/ski/core-facilities/media-preparation
Cystine.2HCl + Cysteine HCl DMEM/F12 media	MSKCC Media Preparation Facility	https://www.mskcc.org/research/ski/core-facilities/media-preparation
Amino acid-free culture DMEM medium	MSKCC Media Preparation Facility	https://www.mskcc.org/research/ski/core-facilities/media-preparation
Cystine.2HCl + Cysteine HCl DMEM media	MSKCC Media Preparation Facility	https://www.mskcc.org/research/ski/core-facilities/media-preparation
L-buthionine sulfoximine (BSO)	Sigma-Aldrich	Cat#: B2515
Deferoxamine mesylate salt (DFO)	Sigma-Aldrich	Cat#: D9533
ML162	SelleckChem	Cat#: S4452
Liproxstatin-1	SelleckChem	Cat#: S7699
zVAD-fmk	SelleckChem	Cat#: S7023
zFA-fmk	MedChemExpress	Cat#: HY-P0109A
CA-074 methyl ester	Sigma Aldrich	Cat#: C5857
Concanamycin A	MedChemExpress	Cat#: HY-N1724
Trolox	SelleckChem	Cat#: S366
C11-BODIPY™ 581/591	Invitrogen	Cat#: D3861
FINO2	Cayman Chemicals	Cat#: 25096
Nano-RSL3	Heller Lab	Ruiz et al.
Lysotracker Deep Red	Invitrogen	Cat#: L12492
Sytox Green	Molecular Probes	Cat#: S7020
Sytox Orange	Molecular Probes	Cat#: S11368
Hank’s Balanced Salt Solution	Thermo Fisher	Cat#: 14025-092
Critical commercial assays
GSH/GSSG-Glo™	Promega	Cat#: V6611
Mycoalert detection kit	Lonza	Cat#: LT07-218
Deposited data
Spatial Segregation Index, code and data	Zenodo	https://doi.org/10.5281/zenodo.18271042.
Spatial Propagation Index, code and data	Zenodo	https://doi.org/10.5281/zenodo.18271042
Experimental models: Cell lines
Human: MCF10A mammary epithelial cells	ATCC	Cat#: CRL-10317
Human: U2OS	ATCC	Cat#: HTB-96
Human: HeLa	ATCC	Cat#: CCL2
Human: HT-1080	ATCC	Cat#: CCL-121
Human: A549	ATCC	Cat#: CRM-CCL-185
Human: HaCaT	ADDEXBIO TECHNOLOGIES	Cat#: T0020001
Human: MCF10A GFP-Galectin-3 MCF10A	Overholtzer Lab	Lee et al.
Human: MCF10A TFEB-GFP	Overholtzer Lab	Klein et al.
Human: MCF10A H2B-mCherry	Overholtzer Lab	Krishna et al.
Human: HT1080 GPX4KO6	Jiang Lab	Gao et al.
Software and algorithms
Microsoft Excel	Microsoft	https://www.microsoft.com/en-us/
GraphPad Prism 10	Graph Pad	https://www.graphpad.com/scientific-software/prism/
Python	Python	https://www.python.org/
Adobe Photoshop	Adobe	https://www.adobe.com/
Spatial Segregation Index	Zaritsky Lab	This study
Spatial Propagation Index	Zaritsky Lab	This study
Image J	Schneider et al.	https://imagej.nih.gov/ij/
Other
glass-bottom plates (6-well)	Mattek	Cat#: P06G-1.5-20-F
glass-bottom dishes (35mm)	Mattek	Cat#: P35G-1.5-20-C

Highlights.

• Ferroptosis induced by direct inhibition of GPX4 involves necrosis and apoptosis

• Ferroptosis propagation is linked to necrosis and sensitized by loss of glutathione

• Necrosis results from lysosome rupture and is regulated by cathepsins

• Lipid peroxidation occurring at lysosomal membranes leads to rupture