ESCRT-dependent membrane repair negatively regulates pyroptosis downstream of GSDMD activation

See allHide authors and affiliations

Science  23 Nov 2018:
Vol. 362, Issue 6417, pp. 956-960
DOI: 10.1126/science.aar7607

Fine-tuning pyroptosis with ESCRT-III

Pyroptosis is an inflammatory form of cell death induced by select caspases downstream of inflammasome complexes. These caspases cleave gasdermin D (GSDMD), whose N-terminal fragments quickly form large permeability pores that induce cell death. However, a large percentage of cells with active inflammasomes are resistant to pyroptosis. Rühl et al. found that the membrane-remodeling ESCRT-III machinery was recruited to the plasma membrane upon GSDMD activation. ESCRT-III–dependent membrane repair limited proinflammatory cytokine secretion and pyroptosis after activation of inflammasomes.

Science, this issue p. 956


Pyroptosis is a lytic form of cell death that is induced by inflammatory caspases upon activation of the canonical or noncanonical inflammasome pathways. These caspases cleave gasdermin D (GSDMD) to generate an N-terminal GSDMD fragment, which executes pyroptosis by forming membrane pores. We found that calcium influx through GSDMD pores serves as a signal for cells to initiate membrane repair by recruiting the endosomal sorting complexes required for transport (ESCRT) machinery to damaged membrane areas, such as the plasma membrane. Inhibition of the ESCRT-III machinery strongly enhances pyroptosis and interleukin-1β release in both human and murine cells after canonical or noncanonical inflammasome activation. These results not only attribute an anti-inflammatory role to membrane repair by the ESCRT-III system but also provide insight into general cellular survival mechanisms during pyroptosis.

Gasdermin D (GSDMD) is a pore-forming protein that induces pyroptosis, a necrotic form of cell death that is initiated after inflammasome activation (13). The detection of pathogen- or host-derived danger signals by inflammasomes triggers the activation of inflammatory caspases (caspase-1 and -11 in mice and caspase-1 and -4 in humans), which cleave GSDMD to release autoinhibition on its N-terminal domain (GSDMDNT) (1, 2). The GSDMDNT targets the plasma membrane and organelles, where it forms large pores to initiate pyroptosis (47). Damage to the plasma membrane does not necessarily result in cell death, as it has been observed that the influx of Ca2+ ions from the extracellular milieu triggers repair programs involving either the endocytosis of a damaged membrane or its shedding in the form of ectosomes (811). This latter mechanism relies on components of endosomal sorting complexes required for transport (ESCRT-0 and -III and associated factors that control ESCRT recruitment and disassembly) and has been specifically implicated in the restoration of plasma membrane integrity during necroptosis (12) or upon chemical- or laser-induced damage (10, 11).

To investigate if cells repair membranes damaged by GSDMD pores, we imaged the disruption of the electrochemical gradient (by using the Ca2+ dye Fluo-8) and loss of membrane integrity [by using propidium iodide (PI)] in mouse bone marrow–derived macrophages (BMDMs). These cells were either untreated or transfected with lipopolysaccharide (LPS) to activate caspase-11. Within 2 hours, up to 50% of LPS-transfected wild-type (WT) BMDMs underwent pyroptosis, as demonstrated by a strong PI signal (PIhi) (fig. S1A). A marked spike of Fluo-8 signal preceded the PIhi signal, indicating a loss of membrane integrity and Ca2+ influx (Fig. 1, A and B; fig. S1, B to D; and movies S1 and S2). In contrast, Fluo-8 or PI signals did not change in Casp11/ or Gsdmd/ BMDMs (Fig. 1B; fig. S1, C and D; and movies S1 and S2). Unexpectedly, we found that Ca2+ influx–negative WT BMDMs that maintained the electrochemical gradient gradually acquired low levels of PI (PIlo) (Fig. 1C). This PIlo signal reached only 10% of the PIhi signal observed in the WT BMDMs that had lost membrane integrity upon LPS transfection. Furthermore, neither unstimulated WT BMDMs (Fig. 1A) nor LPS-transfected Casp11/ or Gsdmd/ cells showed this PIlo signal (Fig. 1C). These findings confirmed that the low-level PI influx was caused by GSDMD pores and suggested that the PIlo cells had repaired their plasma membrane to prevent lysis, given the absence of a Fluo-8 signal peak.

Fig. 1 GSDMD-induced calcium flux negatively regulates pyroptosis.

(A and B) Fluo-8 (Ca2+ indicator) and PI signals in unstimulated or LPS-transfected WT and Gsdmd/ BMDMs. Graphs show average data from n = 22, 22, 28, and 30 cells. Ft, Fluorescence at time t; F0, Fluorescence at time 0. (C) Acquisition of PIlo signal in Ca2+ influx–negative WT, Casp11/, and Gsdmd/ BMDMs treated as described for panel (A). Graphs show average data from n = 29, 29, and 26 cells. (D) LDH release from BAPTA-AM–treated BMDMs mock transfected or transfected with LPS for 2 hours. (E) GSDMD processing in hemagglutinin (HA)–GSDMD–transgenic iBMDMs 2 hours post-LPS transfection. GAPDH, glyceraldehyde-3-phosphate dehydrogenase. (F and G) LDH release and GSDMD processing in untreated or BAPTA-AM–treated HeLa cells infected with ∆sifA S. typhimurium for 5 hours. MOI, multiplicity of infection. The graphs in panels (D) and (F) show means ± SD. **P < 0.01; ***P < 0.001; **** P < 0.0001; ns, not significant (Student’s t test). Results are representative of at least three independent experiments.

We then used the Ca2+ chelators BAPTA-AM and EDTA to block membrane repair, as previously reported (10, 11). Ca2+ chelators increased the levels of cell lysis and lactate dehydrogenase (LDH) release from WT BMDMs after LPS transfection in a dose-dependent manner (Fig. 1D and figs. S1A and S2). Notably, Ca2+ chelation was not cytotoxic (Fig. 1D and fig. S2A) and did not enhance LPS transfection or caspase-11 activation, as assessed by immunoblotting for GSDMDNT (Fig. 1E and fig. S3A). BAPTA-AM also increased caspase-4–dependent pyroptosis without significantly altering GSDMD processing in HeLa cells infected with a ΔsifA deletion strain of Salmonella enterica serovar Typhimurium (hereafter S. typhimurium) (Fig. 1, F and G, and fig. S3B) (13). Thus, Ca2+ chelation enhanced pyroptosis in human and mouse cells, potentially by preventing the initiation of plasma membrane repair.

ESCRT proteins target membranes during membrane repair to form a punctate pattern (10, 12). Consistently, CHMP4 puncta were observed in HeLa cells expressing CHMP4 fused to green fluorescent protein (CHMP4-GFP) from an endogenous promoter (14) after S. typhimurium infection, GSDMDNT expression, or perforin treatment (control), in contrast with untreated cells (fig. S4 and movie S3). Moreover, Ca2+ chelation significantly reduced the number of cells with CHMP4-GFP–positive puncta (Fig. 2A), indicating that ESCRT assembly requires Ca2+ influx via GSDMD pores. Next, we transfected CHMP4-mCherry or CHMP4-GFP fusion proteins into human embryonic kidney (HEK) 293T cells stably expressing caspase-1 fused with a modified FKBP domain (DmrB–Casp-1). After dimerization and activation of DmrB–Casp-1 with B/B homodimerizer, cells underwent pyroptosis as demonstrated by morphological changes and plasma membrane annexin-V staining (Fig. 2, B and C, and movies S4 to S6). CHMP4 relocalized to a punctate pattern in DmrB–Casp-1-transgenic HEK293T cells expressing either high [Fig. 2, B and C, and fig. S5, A to C (arrows)] or low (fig. S5B) levels of CHMP4. Similar CHMP4 puncta were detected after activation of DmrB–Casp-11 (fig. S6A). Although many CHMP4 puncta clearly localized to the plasma membrane (Fig. 2, B and C; figs. S4 to S6; and movie S6), puncta also formed in the cytoplasm, presumably on intracellular organelles. During necroptosis, the ESCRT machinery promotes the formation of annexin-V+ plasma membrane vesicles, which have been proposed to remove damaged areas of the plasma membrane (12). Similar annexin-V+ vesicles were detected budding from HEK293T cells after activation of DmrB–Casp-1 (Fig. 2, C and D; fig. S5D; and movie S5) or DmrB–Casp-11 (fig. S6B). Notably, these vesicles showed clear CHMP4 puncta at the neck (Fig. 2C and figs. S5D and S6B). When GSDMD-GFPinternal (fig. S7A) was coexpressed, it localized close to CHMP4-mCherry puncta on the cell periphery (fig. S7B) and on the membrane of annexin-V+ vesicles (Fig. 2D and fig. S7C), suggesting that ESCRT-induced ectosomes remove GSDMD pores from the plasma membrane.

Fig. 2 The ESCRT machinery translocates to the plasma membrane during pyroptosis.

(A) Fixed-cell microscopy images and CHMP4-GFP puncta quantification of HeLa cells expressing CHMP4-GFP and Dox-inducible GSDMDNT treated with 1 μg/ml of Dox for 4.5 hours. DIC, differential interference contrast. (B and C) Time-lapse confocal images of either CHMP4-mCherry–expressing (B) or CHMP4-GFP–expressing (C) DmrB–Casp-1–transgenic HEK293T cells stained with annexin-V as membrane marker. Insets show CHMP4 localizing to the plasma membrane or necks of budding vesicles. (D) Close-up of annexin-V+ vesicle released from CHMP4-mCherry–expressing and GSDMD-GFPinternal–expressing DmrB–Casp-1–transgenic HEK293T cells after a 1-hour homodimerizer treatment. Graphs show means ± SD. *P < 0.05; **P < 0.01 (Student’s t test). Results are representative of at least three independent experiments. Arrowheads indicate CHMP4 puncta. Scale bars, 5 μm (A), 10 μm [(B) and (C)], or 1 μm (D).

We then examined the impact of ESCRT-III inactivation on inflammasome effector functions. We transiently expressed either WT or dominant-negative CHMP3 or VPS4A in immortalized mouse BMDMs (iBMDMs) or HeLa cells (fig. S8, A to C), because prolonged ESCRT depletion is cytotoxic (10, 12). The expression of dominant-negative VPS4AE228Q mutant protein (Glu228→Gln) or the CHMP31-179 fragment inhibited ESCRT disassembly, as verified by a punctate pattern (fig. S8D), and reduced cell viability after 20 to 24 hours of expression (fig. S8E). This prompted us to restrict our experiments to a maximum of 15 hours postinduction with doxycycline (Dox). The expression of VPS4AE228Q in WT iBMDMs resulted in two- to fourfold-higher levels of cell death after LPS transfection compared with VPS4AWT-expressing controls (Fig. 3, A and B; fig. S9; and movies S7 and S8) and also decreased recovery after LPS transfection (fig. S10). ESCRT inactivation also strongly enhanced interleukin-1β (IL-1β) release after LPS transfection (Fig. 3C), but notably, it enhanced neither LDH nor cytokine release from untreated, Casp11/, or Gsdmd/ cells (fig. S11). IL-1β release after noncanonical inflammasome activation depends on GSDMD-induced K+ efflux and subsequent activation of the NLRP3–caspase-1 pathway (15, 16). Because LPS-transfected WT iBMDMs expressing VPS4AE228Q had higher levels of processed, activated caspase-1 and released more mature IL-1β than VPS4AWT-expressing control cells (Fig. 3, C and D), we hypothesized that this was due to enhanced K+ efflux via increased GSDMD pore formation. Consistently, and in agreement with previous reports (15), cytokine release but not cell death was attenuated by treatment with the NLRP3 inhibitor MCC950 or high extracellular K+ (Fig. 3E and fig. S12A).

Fig. 3 ESCRT inactivation enhances noncanonical inflammasome-induced pyroptosis.

(A to D) PI staining, LDH release, IL-1β release, and caspase-1 processing from VPS4AWT-transgenic or VPS4AE228Q-transgenic iBMDMs 2 hours after LPS transfection. UT, untreated. (E and F) LDH release, IL-1β release, and GSDMD processing from VPS4AWT-transgenic or VPS4AE228Q-transgenic iBMDMs 2 hours post-LPS transfection in the presence or absence of 2.5 μM MCC950. (G) LDH and IL-1β release 4 hours post-LPS transfection from Ripk3/ iBMDMs treated with the corresponding small interfering RNA. NT, nontargeting siRNA. Graphs show means ± SD. **P < 0.01; ***P < 0.001, ****P < 0.0001; ns, not significant (Student’s t test). Results are representative of at least three independent experiments. VPS4AWT/E228Q expression in panels (A) to (F) was induced with 0.5 μg/ml of Dox for 6 hours.

We next assayed GSDMD processing and found that GSDMDNT levels were slightly enhanced in VPS4AE228Q-transgenic iBMDMs compared with VPS4AWT-transgenic iBMDMs at high LPS concentrations (Fig. 3F). We postulated that the enhanced GSDMD processing was caused by enhanced NLRP3 inflammasome activation (Fig. 3E and fig. S12A) and was thus caspase-1 dependent. Indeed, GSDMD processing did not differ between VPS4AE228Q- and VPS4AWT-expressing iBMDMs in the presence of MCC950 (Fig. 3F and fig. S12B), confirming that ESCRT negatively regulates pyroptosis downstream of caspase-11 activation and GSDMD processing.

To study the role of ESCRT-associated proteins, we knocked down the expression of ESCRT-I and -III proteins and of two proteins, ALG2 and ALIX, that have been implicated in the Ca2+-dependent recruitment of ESCRTs (8). As spontaneous necroptosis occurs upon ESCRT depletion (12), we performed all knockdown experiments in Ripk3/ iBMDMs (12), which remained viable despite successful protein depletion (fig. S13, A to E). The knockdown of CHMP3, VPS4A, and VPS4B significantly enhanced pyroptosis and IL-1β release in Ripk3/ iBMDMs transfected with LPS (Fig. 3G). By contrast, single or double knockdown of ALIX and ALG2 had no impact on pyroptosis (Fig. 3G). ALG2 and ALIX may act redundantly with TSG101 (17), but simultaneous depletion was not possible due to the cytotoxicity of TSG101 depletion in BMDMs (fig. S13F).

ESCRT inactivation also enhanced pyroptosis in human cells infected with ΔsifA S. typhimurium (fig. S14, A to D). In this experiment, ESCRTs appeared to have an additional function upstream of inflammasomes, as ESCRT inactivation resulted in higher levels of ruptured, galectin-3+ Salmonella-containing vacuoles (fig. S15, A to C), despite equal levels of bacterial invasion (fig. S14, E and F). This implied that ESCRTs repair damaged vacuoles and endosomes, as suggested recently (17), and thereby control bacterial escape into the cytosol. However, the elevated levels of cytosolic bacteria in ESCRT-depleted cells showed only a negligible impact on GSDMD processing at the time points examined (fig. S15, D to F).

We next asked if ESCRTs also regulate cell death after canonical inflammasome activation. Ca2+ chelation significantly enhanced pyroptosis after DNA transfection of BMDMs (fig. S16, A and B), without changing caspase-1 processing (fig. S16, C and D). Furthermore, ESCRT depletion significantly enhanced cell death and IL-1β release in iBMDMs after NLRC4 activation with log-phase S. typhimurium bacteria but did not enhance ASC speck formation or processing of caspase-1 or GSDMD (Fig. 4, A and B, and fig. S17). This indicated that ESCRTs restrict pyroptosis downstream of GSDMD. To further exclude any effects of ESCRT depletion on the upstream signals that control caspase-1 activation, we used DmrB–Casp-1–transgenic HEK293T cells. Activation of DmrB–Casp-1 induced pyroptosis with faster kinetics in cells expressing dominant-negative VPS4AE228Q or CHMP31-179 than in cells expressing the WT proteins, particularly if DmrB–Casp-1 activity was limited by adding an inhibitory washout compound (Fig. 4C and fig. S18A). B/B homodimerizer enhanced pyroptosis in ESCRT-deficient cells compared with ESCRT-proficient cells across a variety of concentrations (Fig. 4D and fig. S18B). Notably, ESCRT inactivation did not affect GSDMD processing (Fig. 4E and fig. S18C) and enhanced pyroptosis only in DmrB–Casp-1–transgenic HEK293T cells when GSDMD was coexpressed (fig. S18, D and E). Finally, ESCRT inactivation also enhanced IL-1β release, as shown in pro–IL-1β–transgenic HEK293T cells (Fig. 4F and fig. S18F).

Fig. 4 ESCRT negatively regulates pyroptosis and cytokine release downstream of caspase-1.

(A and B) LDH release, IL-1β release, and processing of caspase-1 and GSDMD from VPS4AWT/E228Q-transgenic iBMDMs infected with log-phase S. typhimurium cells for 1 hour. (C) Time course of PI staining in DmrB–Casp-1–transgenic HEK293T cells expressing FLAG-hGSDMD-V5 and Dox-inducible VPS4A or CHMP3 constructs. Cells were treated with 25 nM B/B homodimerizer (B/B HD) at t = 0 and with a 50-fold excess of washout compound at t = 30 min. (D) LDH release at t = 120 min by DmrB–Casp-1–transgenic HEK293T cells treated as described for panel (C) except with different B/B HD concentrations. Immunoblots show equal expression levels of VPS4A proteins. (E) GSDMD processing at t = 60 min of DmrB–Casp-1–transgenic HEK293T cells treated as described for panel (C) except 50 nM B/B HD was used. (F) LDH and IL-1β release at t = 90 min from DmrB–Casp-1–transgenic and pro–IL-1β double-transgenic HEK293T cells expressing FLAG-hGSDMD-V5 and Dox-inducible VPS4A constructs. Cells were treated with B/B HD at t = 0 and with a 50-fold excess of washout compound at t = 30 min. Graphs show means ± SD. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001 (Student’s t test) [(A), (D), and (F)] or a two-way analysis of variance (C). Results are representative of at least three independent experiments. Protein expression was induced for 6 hours [0.5 μg/ml of Dox for (A) and (B)] or 16 hours [1 μg/ml of Dox for (C) to (F)].

Thus, in a variety of cell lines and systems, the ESCRT machinery dampens GSDMD pore–mediated cell death and negatively regulates IL-1β secretion after canonical or noncanonical inflammasome activation. We propose that ESCRTs play a central role in removing GSDMD pores from the plasma membrane in the form of ectosomes and that membrane repair could allow cells to restrict pyroptosis while permitting limited GSDMD-dependent cytokine release (18, 19). Our results place ESCRTs downstream of caspase activation and GSDMD processing, but they do not exclude the possibility that ESCRTs also control the activation of inflammasomes (e.g., by repairing damaged bacteria-containing vacuoles and thus restricting cytosolic access of bacteria-derived ligands). Further experiments are needed to assess this aspect of ESCRTs and to determine if ESCRT-induced vesicles transport cytokines and danger signals.

Supplementary Materials

Materials and Methods

Figs. S1 to S18

Movies S1 to S8

References and Notes

Acknowledgments: We thank A. Hyman (MPI-CBG, Dresden, Germany) for CHMP4-GFP HeLa cells, A. Oberst (University of Washington, Seattle, WA, USA) for the DmrB–caspase-1 construct, J. Pieters (Biozentrum, Basel, Switzerland) for support and discussions, and C. Ramon-Barros for technical assistance. Microscopy analysis was performed at the Biozentrum and UNIL Imaging Core Facilities. Funding: This work was supported by grants from the Swiss National Science Foundation (310030_175576) and the ERC (ERC-2017-CoG –770988, InflamCellDeath) to P.B. and a Werner Siemens Fellowship (Fellowships for Excellence, University of Basel) to S.R. Author contributions: S.R. and P.B. designed the study. S.R., K.S., B.D., R.H., J.C.S., and P.B. performed experiments, analyzed data, and wrote the manuscript. Competing interests: We declare no competing interests. Data and materials availability: All data are available in the article or in the supplementary materials. Cell lines and plasmids are available from P.B. upon request.

Stay Connected to Science

Navigate This Article