Metabolic inhibition induces pyroptosis in uveal melanoma

Abstract

Few treatment options are available for metastatic uveal melanoma (UM) patients. Although the bispecific tebentafusp is FDA-approved, immunotherapy has largely failed, likely given the poorly immunogenic nature of UM. Treatment options that improve the recognition of UM by the immune system may be key to reducing disease burden. We investigated whether UM has the ability to undergo pyroptosis, a form of immunogenic cell death. Publicly available patient data and cell line analysis showed that UM expressed the machinery needed for pyroptosis, including gasdermins D and E (GSDMD and E), caspases 1, 3, 4, and 8 (CASP1, 3, 4, and 8), and ninjurin1 (NINJ1). We induced cleavage of gasdermins in UM cell lines treated with metabolic inhibitors. In particular, the CPT1 inhibitor, etomoxir, induced propidium iodide uptake, caspase 3 cleavage and the release of HMGB1 and IL-1β, indicating that the observed cleavage of gasdermins led to pyroptosis. Importantly, a gene-signature reflecting CPT1A activity correlated with poor prognosis in UM patients and knockdown of CPT1A also induced pyroptosis. Etomoxir-induced pyroptosis was GSDME-dependent, but GSDMD-independent and a pyroptosis gene-signature correlated with immune infiltration and improved response to immune checkpoint blockade in a set of UM patients. Together, these data show that metabolic inhibitors can induce pyroptosis in UM cell lines, potentially offering an approach to enhance inflammation-mediated immune targeting in metastatic UM patients.

Introduction

Uveal melanoma (UM) is the most common eye cancer in adults. While treatment of the primary tumor with radiation plaque therapy or enucleation has a high success rate, ~50% of patients develop metastases (1, 2). A major risk factor for UM metastatic progression is the functional loss of BRCA1-associated protein 1 (BAP1), and up to 84% of metastatic UM tumors are BAP1-deficient (3). Metastatic UM patients respond poorly to immune checkpoint blockade (ICB) and targeted therapies such as MEK inhibitors (4). The FDA-approved treatment, tebentafusp, a bispecific towards gp100 brings T cells into UM tumors. While tebentafusp improves patient survival, the therapy is only used in HLA-A*02:01 patients (5). Additionally, while darovasertib and crizotinib have recently displayed encouraging results in phase II clinical trials in UM patient, these drugs are not yet FDA approved. Thus, there is an urgent need for additional therapies for metastatic UM patients.

Designing new therapeutic approaches to make UM more immunogenic and infiltrated with tumor-suppressive immune cells is crucial. One possible avenue is the induction of immunogenic cell death, which leads to antitumor responses (6, 7). Pyroptosis is a form of immunogenic cell death that involves caspase-dependent cleavage of gasdermin proteins. Once cleaved, gasdermins form pores in the cell membrane, leading to osmotic swelling and release of inflammatory cytokines and soluble damage-associated molecular patterns (DAMPs) such as high-mobility group 1 (HMGB1) and interleukin 1β (IL-1β). These events are followed by full membrane rupture mediated by Ninjurin-1 (8–10). Consequently, pyroptosis initiation has the potential to propagate robust antitumor immune responses (11). In cutaneous melanoma, targeted BRAF inhibitors and MEK inhibitors promote the cleavage of gasdermin E (6), leading to immune cell infiltration of tumors and antitumor immune responses, and pyroptosis gene-signatures positively correlate with immunotherapy outcomes (12, 13). Thus, pyroptosis-inducing agents may represent a unique treatment strategy for UM leading to improved anti-tumor immunity or response to immune-based therapeutics like tebentafusp.

Here, we investigated the ability of UM cells to undergo pyroptosis. We found that UM patient samples and cell lines expressed all the machinery needed for pyroptosis, regardless of BAP1 status. Building on our previous findings regarding UM sensitivity to metabolic inhibitors (14–16), we found that the CPT1 inhibitor, etomoxir, or knockdown of CPT1A induced GSDME-dependent pyroptosis in BAP1-deficient UM cell lines. Importantly, gene signatures corresponding to CPT1A activity were higher in BAP1-deficient cells and UM patients with poor prognosis. Additionally, pyroptosis gene signature scores (PScore) increased during immune checkpoint blockade treatment (ICB) and correlated with immune infiltration signatures in UM patients with stable disease but not progressive disease. Thus, metabolic inhibition can induce pyroptosis in BAP1-deficient UM cells, presenting a potential therapeutic approach for metastatic UM patients.

Materials and Methods

Cell lines and cell culture:

MP cell lines were routinely tested for mycoplasma and authenticated by STR analysis. STR was most recently performed in March 2024. MP38 (RRID:CVCL_4D11), MP46 (RRID:CVCL_4D13), and MP65 (RRID:CVCL_4D14) cell lines were obtained from Dr. Sergio Roman-Roman (Paris, France) and confirmed to harbor GNAQ, GNA11, or BAP1 mutation by Sanger DNA sequencing (42). MP cell lines were maintained in RPMI 1640 supplemented with 20% FBS and 1% penicillin/streptomycin. Cell lines were passaged for a maximum of 20 passages.

siRNA knockdown:

MP cell lines were reverse-transfected for 72 hours with siRNAs at a final concentration of 10 nM using Lipofectamine RNAiMAX^™ (Invitrogen^™). Nontargeting control (Cat # D-001810-01-05, seq: 5’-UGGUUUACAUGUCGACUAAUU-3’), GSDME-targeting (Cat #s: D-011844-04, D-011844-18, seqs: 5’-CUACGGUGUCAUUGAGUUA-3’, 5’-GCAAGGUGGCUUCGAGAAC-3’), GSDMD-targeting (Cat #s: D-016207-01, D-016207-19, seqs: 5’-GCACCUCAAUGAAUGUGUA-3’, 5’-CAAUAAAGGUGGCAUACGA-3’) and CPT1A-targeting (Cat #s: D-009749-01, D-009749-01, seqs: 5’- GAGAGAACCUCAUCAAUUU-3’, 5’- GAAGAAGGAUACAGAAGUG-3’) siRNAs were purchased from Dharmacon, Lafayette, CO.

Drug treatments:

Treatments were performed in Opti-MEM^™I Reduced Serum Medium (Gibco^™, Cat # 31985070) or serum free RPMI 1640 supplemented with 1 μM Propidium Iodide (PI, Sigma-Aldrich, St. Louis, MO) for Incucyte^®-analysis experiments. WZB117 (Tocris, Minneapolis, MN), 6-aminonicotinamide (Cayman Chemical Company, Ann Arbor, MI), and IACS-010759 (Chemie-Tek, Indianapolis, IN) were dissolved in DMSO. Etomoxir (Sigma-Aldrich, St. Louis, MO), was dissolved in sterile water (sterile water was used as the vehicle control group).

Incucyte^®:

Incucyte^® imaging was performed on Incucyte^® model S3 (Sartorius, Newark, DE). Propidium Iodide (PI)-uptake measurements were derived by dividing the PI-positive area (‘surface fit’ segmentation, RCU threshold = 0.75) by cellular confluence area (‘AI Confluence’ segmentation).

Western blotting:

Protein lysates were prepared in Laemmli sample buffer with 2-mercaptoethanol (BME, 2.5%), separated by SDS-PAGE, and proteins were transferred to polyvinylidene difluoride membranes. For western blots using anti-CPT1A, we strictly followed the instructions provided by the manufacturer for western blotting (information on the antibody below). Cell supernatants were collected in serum free Opti-MEM^™ medium. After centrifugation to remove cell debris, supernatants were concentrated 15× using Amicon Ultra 10K (Sigma-Aldrich). Concentrated supernatants were mixed with Laemmli sample buffer (Bio-Rad) with BME and analyzed via Western blotting. Supernatant membranes were stained with Ponceau-S (Sigma-Aldrich, Cat# P7170) prior to blocking and antibody incubation to ensure even loading. Alternatively, supernatant loading was assessed by Coomassie blue (Fig. 5C). For Fig. 1, lysates were analyzed by SDS-PAGE to ensure even protein loading. Primary antibodies: anti-BAP1 (Cell Signaling Technology Cat# 13271, RRID:AB_2798168), anti-CASP1 (Cell Signaling Technology Cat# 3866, RRID:AB_2069051), anti-CASP3 (Cell Signaling Technology Cat# 14220, RRID:AB_2798429), anti-CASP4 (Cell Signaling Technology Cat# 4450, RRID:AB_1950386), anti-CASP8 (Cell Signaling Technology Cat# 9746, RRID:AB_2275120), anti-GSDMD (Cell Signaling Technology Cat# 97558, RRID:AB_2864253), anti-HMGB1 (Cell Signaling Technology Cat# 6893, RRID:AB_10827882), anti-HSP90 (Cell Signaling Technology Cat# 4877, RRID:AB_2233307), anti-IL-1β (Cell Signaling Technology Cat# 12703, RRID:AB_2737350), anti-GSDME (Abcam Cat# ab215191, RRID:AB_2737000), anti-PARP (Cell Signaling Technology Cat# 9542, RRID:AB_2160739), anti-CPT1A (Cell Signaling Technology Cat# 12252, RRID:AB_2797857), anti-Vinculin (Santa Cruz Biotechnology Cat# sc-73614, RRID:AB_1131294), and anti-β-actin (Sigma-Aldrich Cat# A2066, RRID:AB_476693). Immunoreactivity was detected using HRP-conjugated secondary antibodies: Goat Anti-Rabbit HRP (Millipore Cat# 401315-2ML, RRID:AB_437787) or Goat Anti-Mouse HRP (Millipore Cat# 401215, RRID:AB_10682749) and chemiluminescence substrate (Thermo Fisher Scientific, Cat# 34578, 34096) on a BioRad ChemiDoc Imager (Bio-Rad).

Figure 5. CPT1A knockdown induces pyroptosis in UM cell lines.

MP38 and MP65 UM cell lines were transfected with non-targeting control (siControl) or CPT1A-targeting siRNAs (siCPT1A #1, siCPT1A #2) for 72 hours prior to culture in serum-free media for duration of the experiment. (A) Representative phase images merged with propidium iodide staining (PI, red) of UM cell lines MP38 (left panel) or MP65 (right panel) after 72 hours. (B) AUC boxplots representing the sum of PI positive area (PI⁺ area) relative to total confluence (PI+/Phase) over 72 hours in UM cell lines MP38 (left) or MP65 (right) as determined by Incucyte-S3 analysis (mean centered across experiments, p-values indicated, one-way ANOVA and Tukey’s HSD, n = 3). (C-D) Cleavage of GSDME and GSDMD, and CPT1A expression were assessed by western blots from whole cell lysates collected from MP38 or MP65 cells after 48 hours. HSP90 serves as loading control. Secreted HMGB1 in was assessed by western blot and quantified by densitometry normalized to Coomassie Blue staining loading control (mean centered across experiments, p-values indicated, one-way ANOVA and Tukey’s HSD, n = 3).

Figure 1. Pyroptotic gene expression in uveal melanoma.

(A-C) mRNA expression of GSDMD, GSDME, and NINJ1 (A-C, respectively) across TCGA tumor types, ordered by median (Log2 RSEM, Batch normalized from Illumina HiSeq_RNA-SeqV2). (D) Western blots of BAP1, GSDMD, GSDME and caspases CASP1, CASP3, CASP4, CASP8, and loading controls (HSP90, β-actin, or vinculin) in UM cell line lysates.

Publicly available data sources:

UM patient tumor scRNA-seq data (23) shown in Fig. 3A were obtained from the GEO database (accession: GSE139829). TCGA Pan-Cancer Atlas batch-corrected RNA-seq data (Figs. 1,3) were obtained from Hoadley et al (43). Batch-corrected RNA-seq and associated clinical data from matched UM patient samples undergoing new course of ICB therapy shown in Fig. 7 was obtained from Campbell et al (29). ‘CPT1A-inhib down’ (CPT1A -activity) gene signature shown in Fig. 3 were gathered from Ahn et al (22) (Table S1). PSscore gene signature shown in Fig. 7 was obtained from Chen et al (12) (Table S1). Immune population gene signatures shown in Fig. 7 were obtained from Jerby-Arnon et (28). UM SCNA molecular classifications (Figs. 7, S2) were obtained from Robertson et al (24). Gene set scores were calculated from bulk RNA-seq data using Gene Set Variation Analysis with the gsva (v1.40.1) package (44).

Figure 3. High CPT1A-activity scores correlate with high-risk indicators and poor prognosis in UM.

Gene set variation analysis was performed on the indicated UM RNA-seq data set with a gene signature representing ‘CPT1A-activity’ (CPT1A-inhib. down signature score, see Table S1) (27). (A) Violin plots of CPT1A-activity scores generated from UM single cell RNA-seq data (GSE139829) grouped by BAP1 status (BAP1-WT = red/pink, BAP1-deficient = blue/green). A Welch’s two sample t-test was performed using the median GSVA score value per patient to determine statistical difference between BAP1-WT and BAP1-Def groups (p-value indicated). (B) Boxplots of GSVA CPT1A-activity scores from (TCGA-UVM) grouped by somatic copy number alteration (SCNA) molecular subtypes (24) 1 and 2 (BAP1-WT = red/pink, n = 38) vs 3 and 4 (BAP1-deficient = blue/green n = 42, two sample t-test with Bonferroni adjustment, p-value indicated). (C) Kaplan-Meier plot and log-rank p value for clinical event of UM metastasis among CPT1A activity clusters with increasing CPT1A ‘activity’ scores (clusters 1 to 3, determined by unsupervised k-means clustering, see Fig. S1 and Table S1). The number of cases and UM metastasis events for each cluster are indicated and tick marks correspond to censoring events (date of last follow-up).

Figure 7. Pyroptosis gene-signature scores correlate with improved response to immune checkpoint blockade in UM.

Gene set variation analysis (GSVA) was performed on RNA-seq data from matched UM patient samples before (pretreatment) and during (on-treatment) a new course of immune checkpoint blockade (ICB, described in detail in Table S2) (29) using immune cell gene sets (28) or pyroptosis gene-signature (‘PScore’) (12). (A) Heatmap of GSVA scores corresponding to PScore and significantly correlated immune signatures are shown in matched UM patient samples (patient number 1–7) grouped by treatment and disease response status (progressive disease = red/pink, stable disease = blue/green). Correlation between PScore and immune signatures are shown (left side, Linear mixed effect model correlation, FDR-adjusted p-values indicated). (B) Boxplots of PScore overlayed with line-slope plots indicating PScore response to ICB treatment in matched patient samples with progressive (red/pink, n = 3) or stable (blue/green, n = 4) disease (p-values indicated), LME modeling was used for statistical analysis (see methods).

Analysis of single cell RNA-seq data:

The Seurat package (v3.1.4) (45) was used to analyze the data in Fig. 3 (GSE139829). Malignant cells were identified using the methods described previously (23). Raw count data were normalized using the SCTransform (46) with regression based on the percent mitochondrial content. Signature scores were calculated using the AddModuleScore function in Seurat (v5.1.0) (46). The ggplot2 package (v3.4.3 https://ggplot2.tidyverse.org) was used to generate violin plots. The Welch two sample t-test was performed on the median ‘CPT1A-inhib down’ score value across cells within each tumor sample to test for differences between malignant cells from BAP1 wild-type and deficient tumor groups. Data analyses were performed in R (v3.6.0, v4.3.1, v4.3.2 http://www.R-project.org/).

Statistical Analysis:

Data for area under the curve (AUC) of Incucyte-S3 data were mean-centered across experimental N. A one-way ANOVA with Tukey’s HSD was used for statistical analysis for gasdermin cleavage (Fig. 2D–E), CPT1A activity scores (Fig. S2C, see ‘Publicly available data‘ section), AUC data in (Figs. 4, 5), and HMGB1 densitometry data (Fig. 5D) or two-way ANOVA (Fig. 6D). A robust two-way ANOVA model with MM-type estimation was used for mean-centered AUC data in (Fig. 6B). The robust high breakdown point MM-type estimators for ANOVA (47–49) were used because of potential outliers and deviations from normality. Densitometry was performed on supernatant western blots for HMGB1 and normalized to densitometry of common bands in Ponceau-S blots or Coomassie blue gels and mean centered across experimental N (Figs. 5D, 6D).

Figure 2. Metabolic inhibitors induce gasdermin cleavage in UM cell lines.

(A) UM cell lines MP38, MP46, and MP65 were treated with 100 μM etomoxir (ETO), 50 nM IACS, 40 μM WZB117 (WZB), and 100 μM 6-aminonicotinamide (6AN) for 48 hours. Western blots of GSDME, GSDMD, and loading controls (vinculin and HSP90, n = 3). Bands corresponding to full-length or cleaved N-terminal gasdermin domains are denoted ‘FL’ and ‘NT’ and the GSDMD 43 kDa band is denoted ‘~43 kDa’. (B-C) Quantification of cleaved N-terminal gasdermin domains in the treated UM cell lines from panel A (significant p-values are displayed and non-significant p-values denoted ‘n.s.’ or not shown, one-way ANOVA and Tukey’s HSD, n = 3).

Figure 4. Etomoxir induces pyroptosis in UM cell lines.

(A+D) Representative phase images merged with propidium iodide staining (PI, red) of UM cell lines MP38 (A) or MP46 (D) taken 72 hours post treatment with vehicle control (DMSO) or 75 μM etomoxir (ETO). (B+E) Line plots representing the portion of PI positive area (PI⁺ area) relative to total confluence (phase area) over time in UM cell lines MP38 (B) or MP46 (E) under treatment with vehicle control (DMSO, blue lines), or increasing doses of ETO (50–100 μM, light to dark red lines) as determined by Incucyte-S3 analysis (mean centered across experiments, n = 5). (C+F) Area under the curve (AUC) of Incucyte-S3 data in B+E representing the sum PI⁺ area relative to total confluence (PI+/Phase) over 72 hours of treatment for MP38 (C) and MP46 (F, p-values indicated, one-way ANOVA and Tukey’s HSD, n = 5). (G) Cleavage of GSDME and GSDMD, PARP, or CASP-3 was assessed by western blots of whole-cell lysates prepared from MP38 and MP46 cells treated with vehicle control (DMSO) or ETO for 72 hours at the indicated concentration. (H). Levels of inflammatory proteins, HMGB1 and IL-1β, in supernatants from cell treatments in G were assessed by western blot. Ponceau-S (Pon-S) staining was used as a loading control (n = 3).

Figure 6. Etomoxir-induced pyroptotic DAMP release is dependent on GSDME.

UM cell lines were transfected with non-targeting control (siControl), GSDME-targeting or GSDMD targeting siRNA for 72 hours prior to treatment with vehicle control (DMSO) or 75 μM etomoxir (ETO). (A) Representative phase images merged with propidium iodide staining (PI, red) of UM cell lines MP38 (upper panels) or MP46 (lower panels) 48 hours after treatment. (B) Area under the curve (AUC) of Incucyte-S3 data after treatment with etomoxir for MP38 (left plot, first 48 hours, p-values indicated, n = 4) or MP46 (right plot, first 72 hours, p-values indicated, n = 3). Data in ‘B’ were analyzed with a robust two-way ANOVA model with MM-type estimation (see methods). (C-D) Cleavage or knockdown efficiency of GSDME was assessed by western blot of whole-cell lysates (upper blots) in MP38 (left panels) and MP65 (right panels) cell lines. Levels of HMGB1 in supernatants were assessed by western blot using Ponceau-S (Pon-S) staining as a loading control (lower blots) and quantified by densitometry in panel ‘D’ for MP38 and MP65 (p-values indicated, Two-way ANOVA and Tukey’s HSD, n = 3).

Clinical outcome data from the UM cases in the TCGA data used to determine metastatic rates by Kaplan-Meier analysis were obtained from supplementary data files from Robertson et al (24). Clinical outcome was defined as the time in days between primary UM diagnosis and date of documented metastatic disease or last follow-up. Of the 80 patients within the TCGA-UVM dataset, 10 patients were removed from the analysis because the time between primary and metastatic diagnoses was unknown. The optimal number of CPT1A ‘activity’ clusters was determined from the GSVA scores using the elbow method and plotted with the factoextra (v4.4.1) package (Fig. S2A). Based on the outcome of the elbow plot, K-Means clustering was performed, with the number of clusters (centers) set to 3 with the cluster (v4.4.1) package. The Kaplan-Meier plot and Log-Rank test in Fig 3C were generated using ‘survival’ (v3.7–0) and ‘ggsurvfit’ (v1.1.0) R packages and censor tick marks represent date of last follow-up.

ICB-treated Patient Data: Batch-corrected transcriptome and associated clinical data from matched UM patient samples undergoing new course of ICB therapy shown in Figure 7 was obtained from Campbell et al (35). Previous ICB treatment (where applicable) and ICB treatment details for each patient are shown in Table S2. Gene set scores were calculated using the gsva (v1.40.1) package (50).

To estimate correlation between PScores and immune signatures adjusting for correlation between paired observations from the same patients (Fig. 7A), we modelled values of each immune signature (GSVA score) in a separate linear mixed effects (LME) model with the fixed effect of PScore and random effect of patient. The slope estimated from the LME was used to compute the correlation coefficient. The reported p-values for testing the null hypothesis of zero slope (and equivalently zero correlation coefficient) for all immune signatures were adjusted for multiple testing controlling for the False Discovery Rate (50).

Pre- and post-ICB PScores were modeled in a linear mixed effects (LME) model with the fixed effects of treatment (pre vs. post) and disease response status (Progressive n = 3, Stable n = 4) and random effect of patient. The model was used to evaluate Pre- and post-ICB mean difference in progressive and stable disease groups. The boxplot with lines connecting subjects was created with the ggplot2 (v3.5.1) package (Fig. 7B). Statistics were performed in R (The R Foundation for Statistical Computing http://www.R-project.org).

Data and material accessibility:

All data generated in this study are available within the article and the supplementary data files. All code used in this study is available at: https://github.com/AplinLabBioinformatics/varney_eto

Results

High expression of pyroptosis-associated genes in uveal melanoma

To investigate the potential of UM tumor cells to undergo pyroptosis, we examined expression of key pyroptosis-related genes across 32 cancer types within the TCGA PanCancer dataset. Comparative mRNA analysis revealed that UM tumors rank third and seventh in mRNA expression (Log2 RSEM) of gasdermin family members D and E (GSDMD, GSDME), respectively (Fig. 1A–B). Furthermore, UM exhibited the highest expression of Ninjurin-1 (NINJ1) (Fig. 1C). To confirm expression of pyroptosis elements in UM, we analyzed a panel of UM cell lysates for protein expression and observed near ubiquitous expression of both GSDMD and GSDME in both BAP1 WT and BAP1-deficient lines (Fig. 1D). Despite testing multiple antibodies, we were unable to determine the protein expression of NINJ1.

Pore formation by gasdermins requires proteolytic cleavage in a linker region between the N- and C-termini. While caspase-3 (CASP3) mediates this cleavage for GSDME, caspases 1, 4/5, and 8 are implicated in linker-cleavage of GSDMD (8, 17–19). We therefore assessed UM cell lysates for protein expression of these caspases (Fig. 1D). CASP3 expression was ubiquitous, while CASP1, CASP4 and CASP8 expression was heterogenous. Overall, these findings suggest that UM tumor cells express the necessary machinery for pyroptosis indicating that they may be capable of undergoing programmed inflammatory cell death.

Metabolic inhibitors induce gasdermin cleavage in BAP1-deficient UM cell lines

Growing evidence suggests that pyroptosis is positively regulated by mitochondrial dysfunction (8, 20). Recently, we reported the sensitivity of UM cell lines to metabolic inhibitors targeting fatty acid oxidation (FAO), glycolytic, or nucleotide biosynthesis pathways (15). To assess the effect of metabolic disruption on pyroptosis, we evaluated GSDMD and GSMDE cleavage in lysates of BAP1-deficient UM cell lines (MP38, MP46, and MP65) after treatment with one of the following inhibitors: etomoxir (ETO, a CPT1 inhibitor targeting FAO), IACS-010759 (IACS, a selective inhibitor of mitochondrial ETC complex I,(21)), WZB117 (an inhibitor of GLUT1, 3, 4 in the glycolytic pathway), or 6-aminonicotinamide (6-AN, an inhibitor of glucose-6-phosphate dehydrogenase involved in nucleotide biosynthesis). For these experiments, the inhibitor concentrations used were within the range we previously showed to effectively inhibit growth in UM cell lines (15).

We observed that only ETO treatment induced significant levels of GSDME cleavage in all three UM cell lines tested (Fig. 2A, quantitated in 2B). WZB induced variable levels of GSDME cleavage and was only significant in MP46. By contrast, 6-AN and IACS treatment failed to promote significant gasdermin cleavage in any of the tested UM cell lines. Specifically, ETO treatment generated a 35 kDa GSDME fragment corresponding to the N-terminal (NT) fragment associated with pore formation (Fig. 2A). However, significant levels of the 35 kDa GSDMD NT fragment were not detected with any of the treatment groups; instead, a 43 kDa band consistent with an inactivating CASP3 cleavage event was frequently observed (17–19) (Fig. 2A–C, Supplemental Figure S1). These findings suggest that metabolic inhibitors targeting CPT1 may stimulate pyroptosis in UM.

High CPT1A-activity scores correlate with high-risk indicators and poor prognosis in UM.

As CPT1A inhibition leads to GSDME cleavage in UM cell lines, we analyzed its importance in UM patients. To this end, we utilized a signature representing ‘CPT1A-activity’ (Table S1) comprised of genes that were consistently decreased during CPT1A inhibition through ETO treatment or CPT1A knockdown (22). Using this signature, we performed gene set variation analysis (GSVA) on UM single cell RNA-seq data (scRNA-seq) and bulk RNA-seq (TCGA-UVM). In a scRNA-seq dataset (23), we found that the CPT1A-activity signature was consistently higher in malignant cells from UM patient tumors that were BAP1-deficient (Fig. 3A). Given that most metastatic UM patients are BAP1-deficient, these data suggest that UM patients with high CPT1A-activity may have elevated risk for progressive disease.

To further probe the relationship between CPT1A-activity and UM prognosis, we analyzed UM patient TCGA data (TCGA-UVM) by GSVA with the CPT1A-activity signature. We grouped tumor samples by ‘BAP1 status’ using somatic copy number alteration (SCNA) molecular subsets, previously shown to reflect chromosome 3 copy number and predict metastatic risk (24). CPT1A-activity scores were significantly higher in BAP1-deficient tumors (BAP1-Def, SCNA clusters 3 + 4) compared to BAP1-WT tumors (SCNA clusters 1 + 2, Figs. 3B and Supplemental Figure S2A). To evaluate associations between CPT1A-activity and metastatic risk, we employed unsupervised K-Means clustering on CPT1A-activity scores in UM TCGA data (Supplemental Figure S2B) producing in 3 clusters (CPT1A Act. KM-Cluster) with increasing CPT1A-activity scores (Supplemental Figure S2C–D). As expected, significant differences in the rate of metastatic progression were found between the CPT1A-activity clusters, with cluster 3 (highest CPT1A activity score) showing the poorest prognosis (Fig. 3C). Together, our data suggest that CPT1A may be a target for UM patients with high-risk molecular alterations.

Etomoxir induces pyroptosis and secretion of inflammatory mediators in UM

To determine whether the cleavage of gasdermins in response to ETO treatment was sufficient to induce pyroptosis, we evaluated the uptake of propidium iodide (PI) in MP38 (Fig. 4A–C) and MP46 (Fig. 4D–F) UM cell lines using IncuCyte analysis. UM cells treated with ETO exhibited morphological changes including the appearance of swollen cytoplasmic bubbles (Fig. 4A, 4D), consistent with the loss of membrane integrity and osmotic cell-swelling that is attributed to gasdermin pore formation in pyroptotic cells. Notably, cells exhibiting these morphological ‘pyroptotic figures’ were positive for PI staining. The proportion of PI-positive cells increased in a dose-dependent manner over time in response to ETO (Fig. 4B, 4E). Area under the curve (AUC) analysis revealed a significant increase in the proportion of PI-positive cells at 75 µM and 100 µM for both MP38 and MP46 cells (Fig. 4C, 4F).

Next, we confirmed the cleavage of gasdermin in response to ETO treatment by Western blot. We detected cleavage of both CASP3 and GSDME and the appearance of a 43 kDa GSDMD cleavage product in ETO-treated UM lysates (Fig. 4G and Supplemental Figure S1). We additionally observed PARP cleavage in ETO-treated UM cells (Fig. 4G). ETO-induced gasdermin cleavage and PI uptake correlated with the secretion of the inflammatory mediators, HMBG1 and IL-1β, into conditioned media (Fig. 4H). While cleavage of CASP3 and PARP is canonically associated with non-inflammatory apoptosis, the secretion of inflammatory mediators and productive cleavage of GSDME suggests ETO induces pyroptotic cell death in UM cell lines.

CPT1A knockdown induces pyroptosis in UM

At higher concentrations, ETO treatment can have off-target effects and is linked to hepatotoxicity when used systemically (25, 26); hence, we tested if knocking down CPT1A, the main target of ETO, also induced pyroptosis. Knockdown of CPT1A led to increased PI uptake, GSDME cleavage, and HMGB1 release in MP38 and MP65 cell lines (Fig. 5A–D). We additionally observed generation of the ~43 kDa GSDMD cleavage product, but not the N-terminal fragment, following CPT1A knockdown. Together, these data show that CPT1A depletion phenocopies pyroptosis induced by ETO treatment, suggesting that inhibition of CPT1A either genetically or pharmacologically leads to pyroptosis in UM cells.

Etomoxir-induced pyroptosis is dependent on GSDME

CASP3 has divergent activities towards gasdermins. It inactivates GSDMD by cleaving within the N-terminal pore forming domain (Supplemental Figure S1) but activates GSDME-dependent pyroptosis via cleavage in the linker domain (27). Given the presence of cleavage-activated CASP3 and GSDME as well as the evidence of GSDMD inactivation (Fig. 4G), we hypothesized that ETO-induced pyroptosis is GSDME-dependent. To test this hypothesis, we knocked down GSDME or GSDMD from UM cells prior to treatment with ETO. The depletion of GSDME correlated with a marked reduction in the number of pyroptotic cells in response to ETO in both MP38 and MP46 UM cell lines, whereas knockdown of GSDMD did not (Fig. 6A). Consistently, AUC analysis of the proportion of PI-positive cells in response to ETO treatment was reduced by knockdown of GSDME but not GSDMD (Fig. 6B). This reduction reached statistical significance in MP46 cells using two distinct GSDME-targeting siRNAs. The findings in MP38 cells trended lower with both GSDME-targeting siRNAs but was significant with one siRNA (GSDME#18). These data suggest that GSDME facilitates ETO-induced PI-uptake in UM cell lines.

Next, we assessed the contribution of GSDME on ETO-induced HMGB1 release in MP38 and MP65 cell lines. The levels of secreted HMGB1 were significantly reduced after knockdown of GSDME in both MP38 and MP65 cell lines (Fig. 6C–D). In contrast, the levels of secreted HMGB1 after knockdown of GSDMD were comparable to control conditions following ETO treatment (Supplemental Figure S3A–D). Together, these data suggest that ETO treatment leads to a GSDME-dependent pyroptosis and secretion of HMGB1.

Pyroptosis gene signature scores correlate with improved response to immune checkpoint blockade in UM.

While ICB has provided little clinical benefit in UM (4), we tested whether a pyroptosis gene-signature score, ‘PScore’ (Supplemental Table S1.) (12), was correlated with immune cell infiltration signatures (28) and/or response to ICB in UM patients. Thus, we performed GSVA on RNA-seq data from matched patient samples from 7 UM cases before and during a new course of ICB (29). Three of the 7 patients had previous ICB experience and were switched to an alternative ICB treatment, while the remaining 4 patients were ICB-naïve (Supplemental Table S2). The PScore positively correlated with several immune signatures including cytotoxic CD8 T, various T-helper and NK cell signatures (Fig. 7A). Moreover, PScores increased in response to ICB treatment in patients with stable disease (Fig. 7B). While the number of UM patients in this data set is limited, the findings support the hypothesis that inducing pyroptosis in UM enhances immune infiltration and response to immune-based therapy. Thus, leveraging the potential to induce pyroptosis through inhibiting CPT1A pathways may improve anti-tumor immune responses in UM patients.

Discussion

Considering the poor prognosis and lack of treatment options for metastatic UM, it is crucial to identify pathways of susceptibility. Given that UM is poorly immunogenic, we tested metabolic inhibitors for their ability to induce immunogenic cell death via pyroptosis. We found that UM cells expressed the machinery needed for pyroptosis. Furthermore, metabolic inhibition targeting CPT1A induced GSDME-dependent pyroptosis and release of DAMPs in UM cell lines. Additionally, pyroptosis correlated to immune infiltration and response to ICB in UM patients. Thus, while in vivo confirmation is needed, induction of pyroptosis in metastatic UM via metabolic inhibition could sensitize UM tumors to immunotherapy.

Pyroptosis is crucial for drug efficacy and anti-tumor immune responses across multiple tumor types (11, 30–32); thus, the induction of pyroptosis may improve the prognosis of metastatic UM patients given the poor immunogenicity of UM tumors. Pyroptotic-related gene signatures positively correlate to UM patient prognosis (33–35), yet it remains unclear if UM tumors and cell lines can undergo pyroptosis. Here, we demonstrate that UM cell lines are able to undergo pyroptosis, leading to GSDME-dependent DAMP release. Further in vivo confirmation is needed to know if inducing pyroptosis in UM tumor cells alters the tumor immune microenvironment and improve immunotherapy efficacy. A limitation is that syngeneic mouse models for metastatic UM currently do not exist. It is currently unknown whether pyroptosis initiated by different gasdermin family members would produce equivalent immune modulatory effects. We observed that a CPT1A signature is elevated in high-risk UM patient tumors and that the PScore correlated with immune infiltration and response to ICB in UM patients. Thus, exploiting metabolic vulnerabilities in UM cells clinically may lead to pyroptosis and improve response to immunotherapy.

While ETO treatment of UM cells induced pyroptosis, we were unable to rule out other forms of cell death that could occur during ETO treatment or CPT1A inhibition. Pyroptosis is known to promote apoptosis through caspase-3 cleavage, may be downstream of necroptosis, and has been reported to occur simultaneously with ferroptosis in sepsis responses (36). Given that we observed PARP cleavage during ETO treatment, apoptosis is likely co-occurring with pyroptosis. Future studies will be needed to determine the spectrum of cell death mechanisms that contribute to CPT1A inhibitory effects in UM.

Greater than 90% patients harbor activating mutations in GNAQ or GNA11, which drive glycolysis and oxidative phosphorylation (OXPHOS) metabolism in UM (37). Functional loss of BAP1 in up to 84% of UM metastases is also linked to altered metabolic dependencies. Subsets of BAP1-deficient UM cell lines show either high or low OXPHOS and are differentially sensitive to inhibitors targeting different metabolic pathways (15). While these metabolic alterations appear to confer proliferative and migratory advantages (38–40), they also represent targetable vulnerabilities. Indeed, we have previously shown that targeting fatty acid synthesis and mTOR suppresses UM cell growth (41). Given the multiple metabolic underpinnings of UM, it is not surprising that UM is susceptible to metabolic inhibition, although the links to induction of immunogenic cell death are less clear. Here, we found that inhibition of CPT1, a crucial driver of FAO, induced GSDME-driven pyroptosis in three BAP1-deficent UM cell lines. Our data also suggest that inhibition of glucose transport via WZB117, induced pyroptosis in the MP46 cell line. It remains unclear if this is driven by either GSDMD or GSDME. Together, these findings suggest that inducing pyroptosis in UM tumors may be achieved through targeted metabolic inhibition.

Given the potential to induce pyroptosis in UM cells by targeting intrinsic metabolic vulnerabilities, it is hypothesized that metabolic inhibitor-based treatment could offer multiple benefits. Specifically, metabolic inhibition might not only suppress UM tumor growth but also trigger pyroptosis, thereby modifying the inflammatory tumor microenvironment and enhancing immune cell trafficking. Investigating whether these effects can be harnessed to improve the efficacy of immune-based therapies in UM patients is warranted.

Implications statement:

Induction of pyroptosis by metabolic inhibition may alter the tumor immune microenvironment and improve the efficacy of immunotherapy in uveal melanoma.