Direct visualization of native GSDMD pores reveals lipid-driven stabilization during pyroptosis

Abstract

Gasdermin D (GSDMD) executes pyroptosis by forming membrane pores, yet how these structures assemble and are regulated in cells has remained technically inaccessible. We introduce polymer-supported plasma membranes (PSPMs), which preserve native PM properties while providing cytosolic access for nanoscopic imaging. Combining PSPMs with DNA-PAINT super-resolution microscopy, we visualize human and mouse GSDMD nanostructures directly at the PM of pyroptotic cells and uncover species-specific differences in pore size. Quantitative analyses reveal that GSDMD assembles into heterogeneous macromolecular architectures, including ring-shaped structures, which correlate with PM permeabilization. The palmitoylation-deficient C191A mutant retains minor membrane association but fails to form complete rings, indicating that ring assembly, more than membrane binding, determines pore activity. Last, we identify PI(3,4,5)P₃ as a key regulator of pore stabilization. Its early increase during pyroptosis promotes growth of large rings, and mutations in PI(3,4,5)P₃-interacting residues undermine assembly. These findings define the native architecture of GSDMD pores and reveal lipid-dependent stabilization as a central mechanism regulating pyroptotic membrane permeabilization.

PI(3,4,5)P₃ lipid remodeling during pyroptotic cell death structurally stabilizes GSDMD pores at the plasma membrane.

INTRODUCTION

Gasdermins (GSDMs) are key executors of pyroptosis, a cell death program that plays a critical role in the innate immune response to pathogen attacks or endogenous insults (1–4). GSDMs have the ability to perforate cellular membranes through cleavage-mediated activation by specific proteases (1, 3, 4). In the canonical and noncanonical inflammasome pathways of pyroptosis, the GSDMD family member is cleaved by pro-inflammatory caspases, such as caspase-1 and caspase-4/5 (5, 6). Cleavage by caspases releases the pore-forming N-terminal domain from the autoinhibitory C-terminal domain, allowing its subsequent translocation to the plasma membrane (PM) to form pores (7–13). GSDMD pores mediate the release of inflammatory cytokines [e.g., interleukin-1β (IL-1β) and IL-18] to recruit immune cells and may induce cell lysis (6, 14–16). In addition to triggering an inflammatory response and clearing infected or injured cells, excessive pyroptosis has been associated with various inflammatory and autoimmune diseases (4, 17).

The cryo–electron microscopy (cryo-EM) structure of GSDMD pores in detergent-reconstituted systems suggests the assembly of a prepore on the membrane, which eventually inserts into the membrane to form a functional homogeneous annular pore composed of ~33 subunits with an inner diameter of 21.5 nm (18). In contrast, atomic force microscopy (AFM) in supported lipid bilayers revealed arc-like, slit-like, and ring GSDMD shapes, all capable of perforating the membrane (12, 19). This evidence suggests considerable plasticity of GSDMD pore assembly and opens the possibility of pore size regulation during pyroptosis to control the release of cell contents. The modulation of the density of GSDMD pores at the PM by activating membrane repair machineries to eliminate GSDMD pores (20, 21) and the regulation of GSDMD functionality by posttranslational modifications (22–26) could be ways to control the release of mature IL-1β/18 through GSDMD pores while preventing cell lysis and detrimental pyroptosis (27, 28). Furthermore, lipid interactions may contribute substantially to GSDMD function and membrane targeting specificity (8, 19, 29, 30). The presence of phosphatidylinositol 4,5-bisphosphate [PI(4,5)P₂] at the PM enhances GSDMD activity and may facilitate the assembly of the final ring structures (8, 19). Phosphoinositides (PIPs) may even regulate pyroptosis through a highly dynamic process of opening and closing individual GSDMD pores controlled by calcium influx and local PIP metabolism (29).

To elucidate the physiological implications of these regulatory principles, it is essential to resolve GSDMD structures directly at the PM of pyroptotic cells and to assess the influence of modulatory factors on pore formation. Super-resolution fluorescence imaging techniques such as DNA point accumulation for imaging in nanoscale topography (DNA-PAINT) provide the resolution needed to visualize the small GSDMD pores. However, the notable morphological changes of the PM during pyroptosis and the highly fluorescent cytosolic background of cells expressing labeled GSDMD have made this task challenging. To tackle this challenge, we have devised polymer-supported plasma membrane (PSPM) for quantifying the structure and stoichiometry of GSDMD pores in their native PM environment with molecular resolution. PSPMs are intact, flat PM sheets generated by cells bound to polymer-coated surfaces, with the cell body subsequently removed. By combining PSPMs with DNA-PAINT and quantitative PAINT (qPAINT) super-resolution microscopy (31, 32), we uncovered a heterogeneous array of GSDMD assemblies, ranging from undefined clusters (indicative of small oligomeric assemblies) to incomplete and complete rings that vary in size, subunit number, and ability to permeabilize the PM. Notably, our analysis reveals structural differences between human and mouse GSDMD architectures, with mouse assemblies forming smaller ring-like shapes. Low GSDMD expression reduces the density of GSDMD structures at the PM without altering their size or shape, highlighting that pore formation is intrinsically robust once initiated and that GSDMD abundance controls the extent of membrane permeabilization and cytokine release rather than the functionality of individual pores. Building on previous evidence that PIPs regulate GSDMD pore dynamics (29, 30), we observed a notable early increase in phosphatidylinositol 3,4,5-triphosphate [PI(3,4,5)P₃] during pyroptosis. Guided by atomistic molecular dynamics (MD) simulations and validated experimentally using GSDMD mutants, we show that PI(3,4,5)P₃ stabilizes functional GSDMD ring structures by bridging adjacent subunits, revealing a dynamic lipid-dependent mechanism that couples PIP signaling to pore growth and stabilization. Overall, by considering the PM and its dynamic lipid environment, our quantitative ultrastructural analysis of GSDMD pores in pyroptotic cells supports a model in which PI(3,4,5)P₃ actively contributes to the regulatory complexity of pore formation. These findings highlight a previously underappreciated role of this lipid in directly participating in and modulating the pyroptotic pore, with important implications for understanding the molecular regulation of inflammatory cell death.

RESULTS

Development of PSPMs to overcome imaging limitations in GSDMD pore visualization

To visualize GSDMD pores using DNA-PAINT, we used human embryonic kidney (HEK) 293T cells, which lack endogenous GSDMD and allow efficient transfection, stably equipped with a DmrB-Casp-1 construct for inducible and controlled activation of caspase-1 by dimerization [hereafter HEK293T Casp-1 cells; (21)]. These cells were transfected with human or mouse GSDMD (hGSDMD or mGSDMD, respectively) containing a monomeric enhanced green fluorescent protein (mEGFP) insertion between the N- and the C-terminal domains, immediately upstream of the caspase cleavage site [GSDMD-mEGFP, as previously reported; (33)]. Pyroptosis was induced by dimerization of DmrB-mCas1 by adding a dimerizer to the medium (21). After 90 min of treatment, most cells displayed typical pyroptotic morphology and ToPro3-Iodide (ToPro) uptake, indicating PM permeabilization (fig. S1, A to C). Pyroptotic cells exhibited bright fluorescent puncta at the PM, indicative of GSDMD pore formation; however, a diffuse background signal of GSDMD-mEGFP was consistently detected (fig. S1D). Cells were then fixed, permeabilized, and incubated with an anti-GFP nanobody carrying a DNA docking strand (GFPnb-DS) for DNA-PAINT imaging. Super-resolution imaging revealed a strong spatial correlation between DNA-PAINT localizations and GSDMD-mEGFP fluorescence confirming the specificity of GFPnb-DS labeling (fig. S1E). To improve resolution and minimize nonspecific localizations due to localization errors in DNA-PAINT recordings, images were postprocessed as described in Materials and Methods (“DNA-PAINT”) (figs. S1E and S2). Despite this, GSDMD assemblies appeared unresolved, and expected pore-like ring shapes could not be visualized (fig. S1F). This likely resulted from the cell swelling and partial detachment from the substrate that characterize pyroptotic cells, together with the high cytosolic background of labeled GSDMD, both of which compromised DNA-PAINT imaging quality.

To overcome these issues, we developed PSPMs. PSPMs have the main advantage to allow the imaging of events on an immobilized and ultraflat PM, thus, allowing the visualization of individual pore structures not biased by axial variations due to membrane invaginations and partial detachment from the cell substrate. To prepare PSPMs, HEK293T Casp-1 cells expressing GSDMD-mEGFP were additionally transfected with a HaloTag-mTagBFP (blue fluorescent protein)-TMD (transmembrane domain) construct serving as a tethering protein (Fig. 1A). Cells were cultured on a surface coated with poly-l-lysine-graft–polyethylene glycol functionalized with the HaloTag ligand (PLL-PEG-HTL). This ensured irreversible tethering of cell surface HaloTag-mTagBFP-TMD through the covalent HaloTag-HTL interaction while providing an ultrathin biocompatible polymer support (34). Following the induction of pyroptosis and GSDMD pore formation in the treated group, both pyroptotic and healthy control cells were exposed to the actin polymerization inhibitor latrunculin B for 5 min (35) and then subjected to shear forces by vigorous pipetting to remove the cell body (Fig. 1A). For imaging by scanning electron microscopy (SEM), AFM and DNA-PAINT, the resulting PSPMs were fixed directly afterwards. We first ensured the formation of intact PSPMs in healthy control cells by fluorescence microscopy, SEM, and AFM (Fig. 1, B to D). In PSPMs, the cell body and cytosolic inactive GSDMD-mEGFP were removed, while the PM was retained on the polymer support, as indicated by the presence of a farnesylated mCherry construct serving as a membrane marker (Fig. 1B). SEM and AFM confirmed a very flat topography of these PSPMs, with an average height of ~13 nm above the polymer cushion (Fig. 1, C to E). Elevated regions at the periphery of the cell can probably be attributed to collapsed PM from the top of the cell. However, toward the center of the cell, PSPMs were homogeneous in their topography (Fig. 1, C and D). We further tested the integrity of the PSPMs by measuring membrane fluidity before fixation (fig. S1, G and H) using fluorescence recovery after photobleaching (FRAP) in HEK293T Casp-1 cells additionally expressing farnesylated mCherry (fig. S1G). Farnesyl-mCherry showed maximal recovery after 80 s with an average recovery half-time of 24.46 ± 0.55 s, resulting in an estimated diffusion coefficient of 0.23 ± 0.005 μm²/s (fig. S1H). To visualize mobility across the PM sheet and to obtain a more accurate diffusion rate for membrane-associated proteins in PSPMs, we performed single-molecule tracking analysis. PSPMs allow efficient labeling of proteins facing the intracellular side. We then labeled a small portion of farnesyl-mCherry with an anti-mCherry nanobody conjugated to Dy647 to enable long-term single-molecule tracking (fig. S1I). The labeled farnesyl-mCherry moved across the entire PM sheet with an average diffusion constant of 0.37 ± 0.05 μm²/s (fig. S1J). Considering intrinsic differences between cell types, this result is consistent with previous data of farnesylated EGFP in the PM of HeLa cells with a diffusion coefficient of ~0.47 ± 0.46 μm²/s (36). Next, we evaluated whether changes in membrane curvature due to membrane tethering could prevent the formation of functional GSDMD structures. To this end, we compared the timing of pyroptosis in tethered versus untethered cells. Cells in both conditions showed comparable morphological alterations and similar timing of cell death detected by ToPro uptake (Fig. 1, F and G).

Fig. 1. Development of PSPMs for GSDMD pore visualization.

(A) Scheme of the assay to produce GSDMD pores-containing PSPMs from pyroptotic cells for interrogation by advanced imaging techniques. (B) TIRF images of a representative healthy control HEK293T cell expressing the HaloTag-mTagBFP-TMD anchor (cyan), inactive, cytosolic mGSDMD-mEGFP (green), and the membrane marker farnesyl-mCherry (yellow) before PSPM synthesis (top row) and after PSPM generation (bottom row). Scale bars, 20 μm. (C) Representative SEM images of a PSPM from a control HEK293T cell. Scale bars, 20 μm, 5 μm, and 500 nm. (D) Representative DIC (differential interference contrast), TIRF (top row), and AFM (bottom row) images of a PSPM generated from a control HEK293T cell expressing the HaloTag-mTagBFP-TMD anchor protein (cyan). Scale bars, 20 μm. The full-color height range of the AFM topograph is from low (brown-orange) to high (yellow-white). (E) AFM height profile of the polymer-coated surface and PSPM along the indicated line (cyan) from the AFM image in (D) and (inset) average analysis of the height of PSPMs above the polymer-coated surface (n = 4 samples, 7 cells, 35 measured regions). Average PSPM heights were calculated from the flat, central regions of the membrane, explicitly excluding elevated peripheral areas. (F) Representative confocal images of pyroptosis induction in HEK293T cells expressing mGSDMD-mEGFP WT (green) without (left) and with (right) tethering by expression of the HaloTag-mTagBFP-TMD (cyan) construct and cell seeding on a PLL-PEG-HTL–coated surface. Pyroptosis is monitored by morphological changes (BF) and ToPro3-Iodide staining (red). Scale bars, 30 μm. (G) Quantification of PM permeabilization of HEK293T cells as in (F) by normalized fluorescence intensity of ToPro3-Iodide over 90 min (n = 2 experiments with >20 cells analyzed per experiment). Lines in the graph correspond to the average values from all measured cells and colored areas to data variability (means ± SD).

Together, these results show that PSPMs form intact, flat, and functional PM sheets that give us direct access to the cytosolic side of the PM for imaging the native PM environment, without the requirement of harsh treatments of the cells and without compromising the kinetics of pyroptosis at the time resolution of our experiments.

GSDMD assembles into heterogeneous nanostructures at the PM of pyroptotic cells

To identify pyroptotic cells after cell body removal for DNA-PAINT imaging, HEK293T Casp-1 cells expressing hGSDMD-mEGFP and HaloTag-mTagBFP-TMD were cultured on gridded, PLL-PEG-HTL–coated microscopy slides. Pyroptosis was induced for 90 min, after which PSPMs were prepared and fixed in less than 15 min (Fig. 2, A and B). Pyroptotic PSPMs displayed hGSDMD-mEGFP puncta at the PM, indicative of oligomer formation; however, the background signal observed in intact cells was markedly reduced in PSPMs (Fig. 2, B and C, and fig. S1D). The super-resolution imaging of hGSDMD-mEGFP in PSPMs revealed a strong correlation between DNA-PAINT localizations and GSDMD-mEGFP fluorescence signal. These localizations were retained in the postprocessing (Fig. 2C). In contrast, untreated cells showed no visible bright hGSDMD-mEGFP oligomers at the PM and exhibited minimal GSDMD-specific DNA-PAINT signal (fig. S3, A to C).

Fig. 2. GSDMD assembles into heterogeneous nanostructures at the PM of pyroptotic cells.

(A) Scheme of the correlative assay to monitor GSDMD structures (green) in the PM of pyroptotic cells with DNA-PAINT. HEK293T cells stably expressing DmrB-mCas1 and transiently expressing hGSDMD-mEGFP are cultured on a functionalized surface with a grid. Pyroptosis is induced by artificial dimerization of DmrB-mCas1 allowing activation of GSDMD indicated by cell permeabilization (red) and GSDMD oligomer (green dots) formation. Afterwards, PSPMs are generated allowing the removal of cytosolic fluorescent background and efficient labeling for DNA-PAINT (yellow). The grid allows identifying pyroptotic cells after cell body removal and PSPM formation for correlative DNA-PAINT imaging. (B) Representative TIRF images of HEK293T cells (dashed lines) showing the process from pyroptosis induction indicated by morphological changes (DIC), ToPro3-Iodide staining (red), and hGSDMD-mEGFP oligomers formation (green) to PSPM production (HaloTag-mTagBFP-TMD anchor, cyan) and fixation for DNA-PAINT. Scale bars, 20 μm. (C) Left: Representative images of hGSDMD-mEGFP oligomers on PSPMs (top) and corresponding DNA-PAINT images before (middle) and after (bottom) postprocessing. Scale bars, 20 μm. Right: Zoom-in images of the area indicated on the left with merge of hGSDMD-mEGFP dots (green) with diffraction-limited picked structures (magenta) (top), DNA-PAINT localizations (middle), and super-resolved picked structures (bottom, with zoom-in on a representative structure). Scale bars, 1 μm. (D) Gallery of hGSDMD structures in PSPMs of pyroptotic HEK293T cells resolved with DNA-PAINT. Scale bars, 20 nm. (E) Pie chart of average hGSDMD structure type distribution (n = 4 experiments, 10 cells). Variability shown in fig. S2G. (F) Quantification of the relative distribution of the radius of hGSDMD-mEGFP complete ring structures in PSPMs of pyroptotic HEK293T cells with an average radius of ~16.42 nm (n = 70 structures; n = 11 cells; five experiments).

Although DNA-PAINT imaging did not resolve individual GSDMD subunits within the pore due to their close spatial proximity, it revealed pore-like assemblies with distinct ring-shaped morphology. We observed different macromolecular architectures rather than one type only, as would be expected if GSDMD assembled exclusively into preformed rings. We classified these structures as rings, incomplete rings, and undefined clusters, which likely represent small, unstructured GSDMD oligomers, using the program ASAP (Fig. 2, D and E, and fig. S3D; see the “Structural classification of super-resolved structures” section in Materials and Methods) (37). With radial profiling analysis (37), we further characterized these assemblies and found that ring-like hGSDMD structures exhibited an unexpectedly broad size distribution, with an average radius of 16 nm (Fig. 2F). Given the localization of GFP on the outer side of the GSDMD pore, these measurements refer to the outer rather than the inner ring radius. Consistently, the average ring radius of 16 nm closely matches the outer radius reported in the hGSDMD cryo-EM structure (18).

To ensure that our observations were not dependent on the imaging technique, we resolved hGSDMD structures in PSPMs after pyroptosis using an alternative super-resolution method, MINFLUX-nanoscopy (fig. S4, A to F). This technique achieves spatial resolution in biological samples down to ~3 nm (38, 39). By combining MINFLUX nanoscopy with DNA-PAINT labeling (40), we observed hGSDMD assemblies comparable to those detected by DNA-PAINT imaging (fig. S4, D to F).

Last, consistent with recent reports (23), we analyzed the pore forming activity of full-length hGSDMD. While full-length wild-type (WT) GSDMD under basal conditions remained fully autoinhibited, the induction of pyroptosis in HEK293T Casp-1 cells expressing the cleavage-deficient mutant hGSDMD-D275A-mEGFP at similar expression levels (fig. S4G) resulted in minor PM permeabilization, indicating partial destabilization of autoinhibition, even in the absence of cleavage (fig. S4H). DNA-PAINT imaging revealed significantly fewer, but still detectable, GSDMD-mEGFP signals and a low density of resolved GSDMD structures at the PM, primarily resulting in small, undefined clusters, with impaired formation of ring-like structures and a reduced average radius of 10 nm (fig. S4, I to L). These results demonstrate that full-length GSDMD can partially translocate to the PM and oligomerize when autoinhibition is destabilized by caspase binding or stress-induced modifications. However, at comparable expression levels, full-length GSDMD fails to achieve the same robust pore formation and PM permeabilization observed for cleaved GSDMD-NT.

Together, these results demonstrate that active GSDMD forms highly heterogeneous structures at the PM that differ in both shape and size. This structural diversity supports the notion that GSDMD pores may be dynamically modulated to enable selective content release. Moreover, these data also validate PSPMs as a robust and reliable platform for visualizing native GSDMD pore architectures with super-resolution precision.

Mouse GSDMD forms smaller ring structures compared to human GSDMD

Previous studies have highlighted distinct features between human and mouse GSDMD (18, 41, 42). However, to date, all structural studies of GSDMD pores have focused on the human variant (8, 18, 19). We then used our approach to resolve the structure of mouse GSDMD (mGSDMD) pores directly within its native PM, without the need for cryo-EM reconstruction (fig. S5). Same as hGSDMD, mGSDMD oligomers assembled into various macromolecular architectures (Fig. 3A and fig. S5, D and E). Strikingly, mGSDMD structures were significantly smaller than their human counterparts, with an average outer radius of 12 nm (Fig. 3B and fig. S5F). To gain further insight into mGSDMD oligomerization, we determined the number of subunits composing mGSDMD structures by qPAINT, a method based on the frequency of DNA-PAINT localizations (31). This method requires the use of DNA origami coupled to the same docking strand as the anti-GFP nanobodies as a calibration tool. For more precise quantification, the DNA origamis were deposited on the same substrate used for preparing PSPMs. For this purpose, PLL-PEG-Biotin was included into the PLL-PEG-HTL surface coating to enable capturing of DNA origamis via streptavidin (Fig. 3, C to E). This approach allowed us to simultaneously image DNA origamis and mGSDMD structures in PSPMs by DNA-PAINT (Fig. 3E). Quantitative qPAINT analysis revealed that ring-shaped mGSDMD assemblies varied widely in stoichiometry, ranging from 12 to 46 subunits with an average number of 23 subunits (Fig. 3F). Notably, the ring stoichiometry quantified here closely aligns with that of other GSDMs with similar pore size (43, 44).

Fig. 3. Mouse GSDMD ring structures and stoichiometries.

(A) Relative distribution of mGSDMD structure types over the total number of structures in all measured cells (n = 8 cells, three experiments). (B) Quantification of the relative distribution of the radius of mGSDMD-mEGFP complete ring structures in PSPMs of pyroptotic HEK293T cells with an average radius of 12.06 nm (n = 74 structures, eight cells and three experiments). (C) Scheme of orthogonal surface functionalization for simultaneous DNA-origami immobilization (red) and tethering of PSPMs of pyroptotic cells for stoichiometry analysis of super-resolved GSDMD structures (green) by qPAINT. (D) Representative TIRF images of a HEK293T cell grown on an orthogonal functionalized surface before (left) and after (right) PSPM generation (HaloTag-mTagBFP-TMD anchor, cyan) with mGSDMD-mEGFP oligomers (green). Scale bars, 20 μm. (E) Top row: DNA-PAINT localizations of the PSPM shown in (D) together with immobilized DNA-origamis serving as a calibration for qPAINT. Overview of all DNA-PAINT localizations of the patterned surface (left; scale bar, 20 μm) and zoomed-in images of DNA-origami (1, middle) and GSDMD structures (2, right) after postprocessing (scale bars, 500 nm). Bottom row: Representative super-resolution images of post-processed DNA-origami (left) and GSDMD ring structures. Scale bars, 20 nm. (F) Quantification of the relative distribution of the number of mGSDMD-mEGFP subunits in complete ring structures in PSPMs of pyroptotic HEK293T cells analyzed by qPAINT. Rings have, on average, ~23 subunits (n = 194 rings, eights cells and three experiments).

GSDMD ring formation correlates with PM permeabilization and is impaired in palmitoylation-deficient, membrane-associated mutants

The finding that GSDMD forms oligomers of different sizes and shapes at the PM prompted us to investigate which assemblies are relevant for membrane permeabilization during pyroptosis. To address this, we compared hGSDMD-mEGFP WT with the inactive mutant hGSDMD-C191A-mEGFP, which has been reported to impair pyroptosis due to loss of the palmitoylation site required for membrane targeting (23, 25). Despite higher expression levels of hGSDMD-C191A compared to hGSDMD WT (Fig. 4A), confocal microscopy revealed marked reduced cell death following pyroptosis induction (Fig. 4, B and C). Unexpectedly, PSPMs prepared from cells expressing hGSDMD-C191A-mEGFP still displayed GSDMD puncta at the PM (Fig. 4, D and E), indicating that the mutant can still associate with the membrane to some extent. Super-resolution analysis showed that the overall density of hGSDMD-C191A assemblies was comparable to the WT (Fig. 4F); however, their nanoscopic architecture was profoundly altered, with a marked reduction in ring-like structures and a predominance of small, undefined clusters (Fig. 4, G and H). A similar phenotype was observed for the mouse ortholog, mGSDMD-C192A (Fig. 4, I and J). Overall, these data indicate that ring-like assemblies represent the functional form of GSDMD pores during pyroptosis. In addition, they provide direct structural evidence that defective pore formation in palmitoylation-deficient GSDMD mutants is not simply due to impaired membrane binding but rather results from a failure to assemble into complete ring architectures, thereby establishing a clear structural-functional link in GSDMD-mediated membrane permeabilization. The residual, slower ToPro uptake observed in C191A-expressing cells after 75 min of pyroptosis (Fig. 4C) likely reflects reduced dye flux through fewer and/or smaller, low-conductance pores. Whether these small oligomers can mediate limited membrane permeabilization, as previously suggested (29, 30, 45, 46), remains to be determined.

Fig. 4. GSDMD ring structure formation correlates with PM permeabilization.

(A) Expression levels of hGSDMD-WT-mEGFP and hGSDMD-C191A-mEGFP in HEK293T Casp-1 cells quantified by mean mEGFP intensity (WT: n = 196 cells; C191A: n = 185 cells; P = 5 × 10⁻⁵). (B and C) Representative confocal images (B) and quantification (C) of PM permeabilization (ToPro3-Iodide intensity, red, a.u., arbitrary units) in dimerizer-treated HEK293T Casp-1 cells expressing hGSDMD-C191A-mEGFP (green). Scale bars, 20 μm (WT: four experiments, C191A: two experiments; >20 cells per experiment). Lines represent mean values across cells and shaded areas indicate variability (means ± SD). (D) Representative TIRF microscopy images of dimerizer treatment with ToPro3-Iodide staining (red), subsequent PSPM formation (HaloTag-mTagBFP-TMD anchor, cyan), and fixation for DNA-PAINT. Scale bars, 20 μm. (E) Representative DNA-PAINT images of hGSDMD-C191A-mEGFP in PSPMs. Top: Overview images showing hGSDMD-C191A-mEGFP oligomers before DNA-PAINT (left), DNA-PAINT localizations (middle), and picked GSDMD structures (right). Bottom: Zoom-in of the indicated region showing merged hGSDMD-C191A-mEGFP puncta (green) with diffraction-limited picked structures (magenta), DNA-PAINT localizations (middle), and super-resolved picked structures (right; inset shows representative structure). Scale bars, 20 μm (top) and 1 μm (bottom). (F to H) Density of super-resolved hGSDMD-WT and hGSDMD-C191A structures across the PM after postprocessing of DNA-PAINT localizations (F), representative structure images (G, Scale bar 20 nm), and relative distribution of structural classes (H) in dimerizer-treated HEK293T Casp-1 cells expressing hGSDMD-WT-mEGFP or hGSDMD-C191A-mEGFP (density: WT, n = 8 cells, four experiments; C191A, n = 8 cells, three experiments; P = 0.199. structures: WT, n = 10 cells, four experiments; C191A, n = 6 cells, three experiments; P = 1 × 10⁻⁶). (I and J) Density of super-resolved mGSDMD-WT and mGSDMD-C192A structures across the PM (I) and relative distribution of structural classes (J) in dimerizer-treated HEK293T Casp-1 cells expressing mGSDMD-WT-mEGFP (n = 8 cells, 3 experiments, 493 structures) or mGSDMD-C192A-mEGFP (n = 7 cells, 3 experiments, 1294 structures; P = 1 × 10⁻⁵). Density: WT, n = 8 cells, three experiments; C192A, n = 7 cells, three experiments; P = 0.617. n.s., not significant and ***P < 0.001 (Student’s t test).

GSDMD expression levels modulate membrane permeabilization without altering pore architecture

Given the highly heterogeneous shape and stoichiometry of GSDMD pores at the PM, we asked whether this variability depends on the availability of GSDMD during pyroptosis. Although reduced GSDMD transcription has been shown to prevent GSDMD-mediated pyroptosis (47, 48), the impact of reduced GSDMD expression on the kinetics and architecture of GSDMD pore formation remains unclear. To address this, we compared pore formation under reduced mGSDMD expression by generating HEK293T Casp-1 cells stably expressing mGSDMD-mEGFP under a weaker phosphoglycerate kinase (PGK) promoter (HEK293T DS PGK) (49), as opposed to the higher expression achieved with the cytomegalovirus (CMV) promoter in HEK293T DS cells. Fluorescence quantification confirmed reduced GSDMD-mEGFP expression in PGK-driven cells (fig. S6A). As expected, HEK293T DS PGK cells exhibited attenuated PM permeabilization and reduced LDH release following pyroptosis induction (fig. S6, B to D). DNA-PAINT analysis on the PSPMs generated from these cells (fig. S6, E and F) revealed a markedly lower density of these structures at the PM (fig. S6G). Strikingly, however, neither the relative proportions of ring structures nor their size distribution was affected by reduced expression levels (fig. S6, H and I). These findings indicate that while GSDMD abundance dictates the extent of pore formation at the membrane, the architectural heterogeneity of pores—whether they form and what size they adopt—appears to be intrinsically robust and largely independent of expression level.

Temporal analysis reveals increasing ring size during pyroptosis

To investigate the kinetics of pore assembly, we analyzed the temporal evolution of GSDMD structures at the PM examining PSPMs obtained from cells at defined intervals after pyroptosis induction (0, 15, 30, 60, 90, and 180 min) (Fig. 5). The overall density of GSDMD assemblies increased steadily, peaking at 90 min before declining at 180 min, possibly reflecting late-stage membrane remodeling and structural reorganization or partial disassembly of GSDMD pores (Fig. 5, A and B). The relative distribution of structural classes remained largely stable throughout, although a slight increase in incomplete rings was observed at the expense of small clusters (Fig. 5C), suggesting a gradual transition between oligomeric intermediates. The average radius of complete rings increased significantly at later stages (Fig. 5D), indicating that early during pyroptosis smaller ring assemblies predominate, whereas later conditions favor formation of larger rings. This temporal shift in ring size likely reflects the stochastic timing of pore nucleation and the progressive oligomerization of pores over time. Newly nucleated pores are smaller, whereas pores that nucleated earlier have had more time to complete oligomerization, reaching the architecture observed at steady state.

Fig. 5. Temporal analysis reveals increasing ring size during pyroptosis.

(A) Representative TIRF microscopy images of pyroptotic HEK293T Casp-1 cells [DIC and PM permeabilization monitored by ToPro3-Iodide uptake (red)] and PSPMs [hGSDMD-WT-mEGFP oligomers (green) and DNA-PAINT localization before and after postprocessing] at different time points (Ctrl, 15 min, 30 min, 60 min 90 min, and 180 min) after pyroptosis induction. Scale bars, 20 μm. (B) Density of super-resolved hGSDMD structures across the PM in HEK293T Casp-1 cells at different time points after pyroptosis induction (Ctrl: n = 9 cells, three experiments; 15 min: n = 9 cells, three experiments; 30 min: n = 9 cells, three experiments; 60 min: n = 8 cells, three experiments; 90 min: n = 8 cells, four experiments; 180 min: n = 7 cells, three experiments) [Ctrl: P(15 min) = 0.316, P(30 min) = 0.04, P(60 min) = 0.0012, P(90 min) = 3 × 10⁻⁵, P(180 min) = 0.0004; 15 min: P(30 min) = 0.147, P(60 min) = 0.0024, P(90 min) = 5 × 10⁻⁵, P(180 min) = 0.0012; 30 min: P(60 min) = 0.0097, P(90 min) = 0.0001, P(180 min) = 0.027; 60 min: P(90 min) = 0.066, P(180 min) = 0.263; 90 min: P(180 min) = 0.0059]. (C) Comparison between the relative distributions of structural classes in HEK293T Casp-1 cells expressing hGSDMD-WT-mEGFP at the different time points (Ctrl: n = 7 cells, two experiments; 15 min: n = 9 cells, three experiments; 30 min: n = 7 cells, three experiments; 60 min: n = 8 cells, three experiments; 90 min: n = 10 cells, four experiments; 180 min: n = 7 cells, three experiments). Only statistical significant differences are indicated [untreated (untr.): P(15 min) = 0.0049, P(30 min) = 1.3 × 10⁻⁴, P(60 min) = 2 × 10⁻⁴, P(90 min) = 8 × 10⁻⁸, P(180 min) = 5 × 10⁻⁷; 15 min: P(30 min) = 0.26, P(60 min) = 0.069, P(90 min) = 9 × 10⁻⁴, P(180 min) = 0.008; 30 min: P(60 min) = 0.397, P(90 min) = 0.016, P(180 min) = 0.077; 60 min: P(90 min) = 0.2, P(180 min) = 0.512; 90 min: P(180 min) = 0.476]. (D) Quantification of the relative distribution of the radius of hGSDMD-WT-mEGFP complete ring structures in PSPMs of pyroptotic HEK293T Casp-1 cells at the different time points (rings: 15 min: ~11.62 nm, n = 29; 30 min: ~11.14 nm, n = 24; 60 min: ~14.32 nm, n = 49; 90 min: ~16.42 nm, n = 70; 180 min: ~11.17 nm, n = 52) [P(15-30 min) = 0.843, P(30-60 min) = 0.003, P(60-90 min) = 0.024, P(90-180 min) = 1.3 × 10⁻⁴]. *P < 0.05, **P < 0.01, and ***P < 0.001 (Student’s t test).

PI(3,4,5)P₃ metabolism regulates pyroptosis

Building on the ability of PSPMs to resolve GSDMD nanostructures directly at the PM of pyroptotic cells, we hypothesized that the local membrane nanoenvironment may contribute to regulating GSDMD assembly. Prior studies have implicated PIPs, especially PI(4,5)P₂, in supporting GSDMD-mediated permeabilization; however, the role for other PIPs, namely, PI(3,4,5)P₃, has remained more controversial (8, 19, 29, 30). To investigate whether PI(3,4,5)P₃ dynamics contribute to GSDMD activity, we monitored its metabolism during pyroptosis and evaluated its impact on GSDMD pore density and architecture at the PM. To this end, we coexpressed the PI(4,5)P₂ sensor pleckstrin homology domain of phospholipase C delta 1 fused to an infrared fluorescent protein (PH-PLCδ1-iRFP) (50, 51) and the PI(3,4,5)P₃ sensor PH-Akt-EGFP (52, 53) together with mGSDMD-mCherry in HEK293T Casp-1 cells and monitored the permeabilization of pyroptotic cells by SYTOX Blue staining (fig. S7A). During the early stage of pyroptosis (~10 to 20 min), we observed a progressive redistribution of the PH-PLCδ1-iRFP sensor from the PM to the cytosol, indicating a decrease of PI(4,5)P₂ at the PM (fig. S7, A and B). In contrast, at the same time, PI(3,4,5)P₃ levels increased rapidly after pyroptosis induction, as reflected by the corresponding rise in membrane-associated PH-Akt-EGFP signal (fig. S7, A and B), while the intensity of both lipid sensors remained unaltered in untreated cells (fig. S7C). These observations were also recapitulated for human GSDMD. Here, PI(3,4,5)P₃ was visualized by staining HEK293T cells expressing hGSDMD-mEGFP with purified PH-Akt-mScarlet after fixation at defined time points of pyroptosis (Ctrl, 10, 20, and 30 min) (Fig. 6A). This approach minimized cytosolic background from unbound sensor while enhancing the detection of PM-associated PI(3,4,5)P₃. Consistent with our live-cell measurements, we detected a significant increase in PI(3,4,5)P₃ enrichment at the PM within the first 30 min of pyroptosis induction compared to control cells (Fig. 6B).

Fig. 6. PI(3,4,5)P₃ regulates pyroptosis and enlarges GSDMD pores at the PM.

(A and B) Representative confocal images (A) and quantification (B) of PI(3,4,5)P₃ levels at the PM monitored with Akt-PH-mScarlet (yellow) in fixed, pyroptotic HEK293T Casp-1 cells expressing hGSDMD-mEGFP (green) [n = 3 experiments, >15 cells; P(Ctrl: 10 min) = 0.881, P(Ctrl: 20 min) = 0.003, and P(Ctrl: 30 min) = 0.0007]. (C) Quantification of PM permeabilization in HEK293T Casp-1 cells expressing hGSDMD-mEGFP and mScarlet-TMD-Pten by ToPro3-Iodide intensity over 90 min (n = 4 experiments with >10 cells analyzed per experiment) [Pten: P(30 min) = 0.356, P(60 min) = 0.052, P(75 min) = 0.025]. Lines indicate means; shaded areas represent variability (means ± SD). (D to H) Pyroptosis induction (ToPro3-Iodide, red), PSPM formation (HaloTag-mtagBFP-TMD, cyan) (D; scale bar, 20 μm), and DNA-PAINT (E-H) in HEK293T Casp-1 cells expressing hGSDMD-mEGFP (green) with PIP₃ depletion by overexpression of mScarlet-TMD-Pten (yellow). (E) Top row: Overview of hGSDMD-mEGFP (left) and DNA-PAINT localizations before (middle) and after (right) postprocessing. Scale bars, 20 μm. Bottom row: Areas indicated above. Correlation of hGSDMD-mEGFP (green) and diffraction-limited structures (magenta), DNA-PAINT localizations before (middle) and after (right) postprocessing. Scale bars, 1 μm. Density of hGSDMD structures across the PM (F, hGSDMD WT: n = 8 cells, four experiments; hGSDM WT + Pten: n = 7 cells, three experiments; P = 0.262). Comparison of the relative distribution of GSDMD structural classes (G, hGSDMD WT pyroptotic: n = 10 cells, four experiments; hGSDMD WT pyroptotic + Pten: n = 7 cells, three experiments; P = 8 × 10⁻⁴) and radius distribution of complete ring structures [H, rings: average radius of ~16.42 nm without Pten (n = 70) and ~ 10.13 nm with Pten (n = 40); P = 5 × 10⁻¹¹]. (I to K) GUV permeabilization by recombinant mGSDMD of POPC:PI(4,5)P₂ or POPC:PI(3,4,5)P₃ 95:05 mol % GUVs labeled with recombinant PLCΔ1-2x-PH-HaloTag + HTL-549 or Akt-PH-mScarlet, respectively, permeabilized (red) monitored by dextran uptake [10-kDa dextran–Alexa Fluor 647 (magenta) and 150-kDa dextran-fluorescein (green) with representative confocal images (I) and fraction of permeabilized GUVs for the 10-kDa (J) and 150-kDa dextran (K) (n = 3)]. White arrowheads indicate permeabilized GUVs for both dextrans and cyan arrowheads for only the 10-kDa dextran. Scale bars, 50 μm. *P < 0.05, **P < 0.01, and ***P < 0.001 (Student’s t test).

To assess the functional relevance of PI(3,4,5)P₃ at the PM during pyroptosis, we inhibited its production in mGSDMD pyroptotic cells using the PI3-kinase (PI3K) inhibitor Wortmannin (Wtm) (54). Wtm treatment effectively abolished PI(3,4,5)P₃ enrichment at the PM (fig. S7D). Cells lacking PI(3,4,5)P₃ showed a pronounced reduction and delay in membrane permeabilization after pyroptosis induction (fig. S7E). Notably, Wtm treatment without pyroptosis induction was well tolerated by the cells and did not affect cell morphology or basal membrane integrity, indicating that the observed defects were not caused by nonspecific toxicity (fig. S7F). Because Wtm treatment can have pleiotropic effects, we alternatively inhibited PI(3,4,5)P₃ production by overexpressing a membrane-bound version of the PI(3,4,5)P₃ phosphatase Pten (fig. S7G) (55). Consistent with Wtm treatment, Pten overexpression significantly impaired pyroptosis-induced membrane permeabilization in cells expressing either mouse or human GSDMD (Fig. 6C and fig. S7, E and F).

We also attempted to increase, rather than deplete, PI(3,4,5)P₃ levels at the PM by overexpressing PI3K. However, increased PI(3,4,5)P₃ production did not further enhance PM permeabilization by hGSDMD upon pyroptosis induction (fig. S7I). This lack of effect is likely due to the fact that PI(3,4,5)P₃ levels are already strongly up-regulated early during pyroptosis, potentially reaching concentrations sufficient to support maximal GSDMD activity. Neither Pten nor PI3K overexpression affected cell viability in untreated cells (fig. S7J).

Together, these findings establish PI(3,4,5)P₃ as a critical regulator of efficient membrane permeabilization during GSDMD-driven pyroptosis.

PI(3,4,5)P₃ stabilizes and enlarges GSDMD ring pores at the PM

To further elucidate the role of PI(3,4,5)P₃ in GSDMD pore formation, we performed DNA-PAINT on PSPMs from pyroptotic cells in which PI(3,4,5)P₃ levels were reduced. The depletion of PI(3,4,5)P₃ by either Pten overexpression or Wtm treatment did not impair the appearance of GSDMD puncta after pyroptosis induction (Fig. 6, D and E) nor did it alter the overall density of GSDMD structures at the PM (Fig. 6F and fig. S7K). However, PI(3,4,5)P₃ depletion resulted in a significant reduction in the proportion of ring-shaped GSDMD relative to pyroptotic cells with normal PI(3,4,5)P₃ levels (Fig. 6G and fig. S7L). In addition to reducing ring frequency, PI(3,4,5)P₃ depletion also decreased ring size. In Wtm-treated cells, mGSDMD ring structures were slightly smaller (~11-nm radius) than those formed in the presence of PI(3,4,5)P₃ (~12 nm; fig. S7M). This effect was more pronounced in Pten-overexpressing cells, where mGSDMD rings displayed an average radius of ~9 nm (fig. S7M). A similar but stronger effect was observed for hGSDMD where Pten overexpression reduced the radius of hGSDMD rings from ~16 to ~10 nm (Fig. 6H).

Increasing PI(3,4,5)P₃ levels by PI3K overexpression did not enhance GSDMD ring formation compared to pyroptotic cells with unaltered PI(3,4,5)P₃ levels. Instead, unexpectedly, PI3K overexpression reduced both the density and the size of GSDMD rings at the PM (fig. S7, N to P). This reduction could be indirectly related to changes in PI(4,5)P₂ availability, which has been proposed to facilitate initial GSDMD membrane recruitment by increasing its local concentration (25, 29).

To validate the specific role of PI(3,4,5)P₃ in stabilizing GSDMD pores by direct comparison with PI(4,5)P₂ and without interference of other cellular factors, we reconstituted GSDMD-mediated permeabilization in minimalist membrane systems. Giant unilamellar vesicles (GUVs), containing either PI(3,4,5)P₃ or PI(4,5)P₂, were exposed to active GSDMD, and the uptake of dextrans of different sizes (10 and 150 kDa) was monitored over time (Fig. 6, I to K). While both lipid compositions supported similar uptake of the smaller 10-kDa dextran (Fig. 6J), GUVs containing PI(3,4,5)P₃ displayed significantly enhanced uptake of 150-kDa dextran compared to PI(4,5)P₂-containing GUVs, indicating the formation of larger pores (Fig. 6K). Together, these results demonstrate that although GSDMD pores can form in the absence of PI(3,4,5)P₃, the latter plays a crucial role in stabilizing and enlarging GSDMD pore structures, thereby promoting efficient membrane permeabilization during pyroptosis.

Mutations disrupting PI(3,4,5)P₃ interactions with GSDMD destabilize ring pores

To rationalize the severely hampered GSDMD pore formation in PI(3,4,5)P₃-depleted cells, we developed an MD simulation system. In a previous study, we identified highly positively charged clusters of arginine and lysine residues on both sides of human GSDMD that interact with PI(4,5)P₂ and hypothesized that the high negative charge of PIPs could stabilize intersubunit interface and hence promote GSDMD assembly (30). To quantify the effects of the different PIPs on GSDM ring stability, we simulated 16-meric half-rings of the structurally characterized human GSDMD embedded in a dioleoylphosphatidylcholine (DOPC) membrane. In three different simulation setups, the interfaces between the adjacent subunits were occupied by PI(4,5)P₂, PI(3,4,5)P₃, and no PIPs for reference, respectively. Already during equilibration, the lipid bilayer detached from the inner surface of the GSDMD arc, as seen before (30). The resulting open membrane edge is associated with a line tension that exerts a force of about 44 pN on the tips of the GSDMD arc (56). Because of this large mechanical tension, the interfaces in the rings cracked within the first few hundred nanoseconds of our MD simulations (movies S1 to S3). By analyzing the extent and time course of interface cracking, we then quantify the stabilizing effects of PI(4,5)P₂ and PI(3,4,5)P₃ at the subunit interfaces. These MD simulations of subunit cracking are akin to a force spectroscopy experiment but with line tension as force generator. In the absence of PIPs, half-rings typically cracked at four to five sites and the assembly equilibrated as a round pore. The incorporation of PIPs at the interfaces reduced the number of cracks to one to three with PI(4,5)P₂ and even further to one to two when PI(3,4,5)P₃ was present. These results indicate a clear lipid charge–dependent stabilizing effect, with PI(3,4,5)P₃ providing the strongest reinforcement of the GSDMD assembly (Fig. 7, A and B). In addition, pores containing PIPs displayed more asymmetric shapes with a still visible membrane edge. The increased stability of the assemblies containing bridging PIPs was evident from the approximated free energy of cracking (Fig. 7C). Compared to the assemblies without PIPs, in PI(4,5)P₂ systems, the distances between the centers of mass of adjacent subunits were more stable even below 28 Å, which broadened the free energy minimum. With bridging PI(3,4,5)P₃, the distance between two adjacent subunits was more stably maintained around 27.5 Å, resulting in a narrower free energy minimum and a greatly increased barrier to cracking. Hence, with PI(3,4,5)P₃ bound to the interface, “near cracking“ distances (~30 Å) between two neighbors were reached less frequently. We further characterized the electrostatic environment around GSDMD. At the interface, the charge of the basic clusters mentioned above is not sufficiently balanced by acidic protein residues. Therefore, same as the entire bottom of GSDMD, the interface region between two subunits exhibits a strong positive electrostatic potential (Fig. 7D). We also analyzed the specific binding modes exhibited by different PIPs in the noncracked binding site (Fig. 7E). In a previous study, we identified lateral binding sites of hGSDMD comprising five charged residues (K43, R53, K55, R153, and K235) interacting with PI(4,5)P₂. Driven by electrostatics, both the PI(4,5)P₂ and PI(3,4,5)P₃ lipids, with their four and six negative charges, respectively, interact with the highly positively charged interface. However, unlike PI(4,5)P₂, PI(3,4,5)P₃ can accommodate the charges of all five residues simultaneously. These extensive interactions lead to an increase in contacts between PI(3,4,5)P₃ and K43, R153, and K235 compared to PI(4,5)P₂ (Fig. 7, E and F). We propose that this increase in charge density of PI(3,4,5)P₃ compared to PI(4,5)P₂ is capable to electrostatically stabilizing the interface sufficiently to counterbalance the cracking force caused by the line tension. In turn, this may stabilize the opening of GSDMD pores, thus favoring the growth of ring GSDMD structures (Fig. 6). To validate the MD simulation predictions, we mutated GSDMD residues identified as interacting with PI(3,4,5)P₃ and assessed their impact on GSDMD pore formation using DNA-PAINT analysis in PSPMs. Specifically, we mutated residues R153 and K235 (R153A and K235A), which the simulations predicted to interact more strongly with PI(3,4,5)P₃ than PI(4,5)P₂, and R53 (R53A), which showed no preferential interaction between the two lipids (Fig. 7F). Intriguingly, all mutants exhibited reduced PM permeabilization compared to hGSDMD WT upon pyroptosis induction, with R153A and K235A mutations resulting in a more notable reduction in pyroptosis than R53A (Fig. 7G). Consistent with this functional impairment, all three mutants showed a decrease in density (Fig. 7H) and impaired formation of ring structures (Fig. 7I). Mutant rings were also significantly smaller, reflecting the destabilization of the pore architecture (Fig. 7J). The most significant reduction in GSDMD ring size was observed for the K235A mutant, consistent with K235 being the only lipid-interacting residue located on the other neighboring GSDMD subunit within the interface (Fig. 7, E and J). Together, these results mirror the effects observed upon PI(3,4,5)P₃ depletion and reinforce the MD-based prediction that PI(3,4,5)P₃ directly stabilizes GSDMD pore assembly. These findings support a model in which PI(3,4,5)P₃ strengthens the intersubunit interface by engaging multiple positively charged residues across adjacent GSDMD subunits, thereby promoting the formation of stable, larger pore structures.

Fig. 7. Mutations disrupting PI(3,4,5)P₃ interactions with GSDMD destabilize pores.

(A to C) Stability of GSDMD assemblies in 1-μs unbiased MD simulations without, with PI(4,5)P₂, or with PI(3,4,5)P₃. (A) Representative snapshots showing the protein (blue cartoon) and PI(4,5)P₂ or PI(3,4,5)P₃ densities (orange and green, respectively); black triangles indicate cracks. (B) Number of cracks across three simulations with probability distributions during the final 700 ns (right); the shaded region indicates the equilibration phase excluded from analysis. (C) Free energy of cracking as a function of the center-of-mass distance between neighboring subunits (n = 3, solid line; shading: SEM). Rare cracking events in the PI(3,4,5)P₃ system prevent reliable error estimates in the cracked region (r > 31.5 Å; shaded area). (D) hGSDMD dimer electrostatic potential surface [calculated with APBS independent on PI(3,4,5)P₃ with a PI(3,4,5)P₃ molecule bound to the interface]. (E) Snapshots of hGSDMD dimers with PI(4,5)P₂ (left) or PI(3,4,5)P₃ (right) [binding sites shown and colored (blue) according to the average number of contacts]. Protein residues with at least one atom that has more than 0.2 average contacts are shown in licorice representation. (F) Average number of contacts between PI(4,5)P₂ and PI(3,4,5)P₃ heavy atoms with each GSDMD residue (n = 3). (G to J) Quantification of PM permeabilization by normalized fluorescence intensity of ToPro3-Iodide (G, WT and K235A, n = 4; R53A and R153A, n = 2 experiments with >20 cells analyzed per experiment) [K235A: P(30 min) = 0.168, P(60 min) = 0.003, P(75 min) = 0.0001] and DNA-PAINT (H to J) in HEK293T Casp-1 cells expressing hGSDMD WT or hGSDMD-R53A, hGSDMD-R153A, and hGSDMD-K235A. (H) Density of super-resolved structures (hGSDMD WT: n = 8 cells, four experiments; hGSDMD-R53A: n = 8 cells, three experiments; P = 0.028; hGSDMD-R153A: n = 10 cells, three experiments, P = 0.00124; hGSDMD-K235A: n = 9 cells, three experiments, P = 0.00193). (I) Comparison of the relative distribution of different structure classes (hGSDMD WT: n = 10 cells, four experiments; hGSDMD-R53A: n = 8 cells, three experiments, P = 0.0019; hGSDMD-R153A: n = 10 cells, three experiments, P = 0.00015; hGSDMD-K235A: n = 9 cells, three experiments, P = 0.0089). (J) Quantification of the relative size distribution of complete ring structures (WT: ~16.42 nm, n = 70; R53A: ~12.37 nm, n = 30, P = 2 × 10⁻⁴; R153A: ~14.24 nm, n = 37, P = 0.0047; K235A: ~10.68 nm, n = 31, 5 × 10⁻⁸). *P < 0.05, **P < 0.01, and ***P < 0.001 (Student’s t test).

DISCUSSION

Here, we establish an experimental framework for visualizing native GSDMD pores in the PM of pyroptotic cells and reveal a previously unrecognized role of the lipid PI(3,4,5)P₃ to pore stabilization. By combining PSPMs with DNA-PAINT microscopy, we overcome long-standing challenges in resolving GSDMD oligomers in cells, including pyroptotic swelling, membrane detachment, and high cytosolic background from labeled GSDMD. PSPMs generate immobilized, ultraflat and integral PM sheets that can be obtained in less than 5 min directly at the microscope, preserve physiological protein mobility, and allow direct access to the cytosolic leaflet for nanoscale labeling and multimodal imaging. This enables quantitative structural analysis of pore architectures under native lipid conditions, without the harsh treatments required for traditional unroofing approaches (57–60). While we cannot fully exclude minor alterations in membrane properties due to PM tethering and PSPM production, GSDMD pores form in intact, adherent cells prior to cell body removal, which is performed only afterward during PSPM preparation and followed by immediate fixation, minimizing the influence of subsequent manipulations on previously formed structures. Our data indicate that tethering neither delay nor impair cell death, and GSDMD assemblies exhibit normal temporal evolution throughout pore formation; in addition, PSPM preparation is unlikely to distort pore architecture, as actin disruption occurs naturally during pyroptosis (61).

Using this approach, we resolve GSDMD assemblies with a localization precision of ~7 nm and show that pores at the PM adopt highly heterogeneous shapes, sizes, and stoichiometries, consistent with previous observations in artificial membranes (8, 12, 19) and with the structural diversity reported for other pore-forming proteins (62, 63). While we did not directly measure permeabilization of incomplete rings, their abundance correlates with pyroptotic activity and temporal redistribution during pore formation, and partial rings have been shown to form functional intermediates in GSDMD and other pore-forming proteins on supported lipid bilayers (19, 62).

We also uncover substantial species-specific differences with human pores being significantly larger than mouse pores, despite conservation of the oligomerization interface (41). This indicates that determinants of pore size extend beyond the core residue interaction and likely involve species-specific structural features that influence oligomer curvature and packing. Subtle residue differences, domain orientation, membrane insertion angle, or NT-domain flexibility can profoundly alter pore diameter without requiring changes in the interface residues themselves (42). On the basis of reported inner and outer diameters of different GSDMs (18, 43, 44, 64), the gap between the outer and inner radii is generally ~5 nm. Applying this relationship to mouse GSDMD, which we estimated having an outer radius of ~12 nm, its inner radius is expected to be around 7 nm, about 4 nm smaller than the human one. This size is similar to GSDMB pores, which exhibit a radius of 7 to 8 nm and 24- to 27-fold symmetry (43, 44). Our qPAINT-based stoichiometry analysis support this model as mGSDMD rings contain ~23 subunits, on average, consistent with their smaller radius and the expected stoichiometry of GSDM pores of comparable size. These results also validate our integrated calibration strategy for quantifying protein copy number within individual pores.

By correlating pore architecture with functional outcome, we show that ring-like structures are the assemblies most strongly associated with membrane permeabilization. The palmitoylation-deficient mutant C191A, which displays impaired pyroptosis, still targets the PM, although to a lesser extent than the GSDMD WT, yet predominantly forms small, nonring oligomers. Given that S-palmitoylation at Cys¹⁹¹ enhances GSDMD membrane engagement (22, 23, 25, 26), our data suggest that palmitoylation does not act as an absolute targeting signal but instead stabilizes the membrane residence of otherwise weakly bound monomers. Loss of this modification lowers the effective membrane concentration of GSDMD, thereby affecting the oligomerization process.

Temporal and concentration-dependent analyses further illuminate the principles governing pore formation. While reduced GSDMD expression decreases the density of GSDMD structures that form at the membrane, it does not alter their size or shape, indicating that pore nucleation, not oligomer growth, is the concentration-sensitive step. Our temporal analysis shows that once nucleation occurs, pores continue to mature, with ring structures enlarging over time. These findings support a model in which nucleation is rate-limiting, whereas oligomer growth is intrinsically robust and proceeds efficiently once initiated even at low GSDMD abundance. This regulatory logic allows cells to form fully assembled pores capable of releasing IL-1β/18 even at low GSDMD levels while limiting the overall number of pores and thus preserving cell viability under sublytic conditions. These results have physiological implications for the regulation of pyroptosis based on the modulation of GSDMD expression (47, 48). While a correlation between cell viability and GSDMD pore density was shown in the context of ESCRT-mediated removal of GSDMD pores (21), our data suggest that, since pore size is not affected, the modulation of GSDMD levels may be a viable strategy to sustain IL-1β/18–mediated inflammatory response of pyroptosis and cell viability, even in settings where membrane repair machineries cannot be activated [e.g., due to pathogen-mediated inhibition of repair mechanisms (65)]. Last, the direct correlation between pore number and cell fate reported here helps establish a quantitative threshold between sublytic and lytic states.

Last, we identify PI(3,4,5)P₃ as a key lipid regulator of GSDMD pore stabilization. PI(3,4,5)P₃ levels rise early during pyroptosis and its depletion, either by PI3K inhibition or by Pten overexpression, strongly impairs membrane permeabilization. DNA-PAINT reveals that PI(3,4,5)P₃ is not required for pore nucleation but is essential for stabilizing rings and enabling the formation of larger pores. Minimal membrane systems corroborate this, showing that PI(3,4,5)P₃, more than PI(4,5)P₂, promotes the uptake of large dextrans, consistent with larger functional pores. MD simulations demonstrate that PI(3,4,5)P₃ forms more extensive electrostatic interactions than PI(4,5)P₂ with positively charged residues at the intersubunit interface, increasing the energetic penalty for cracking and stabilizing the pore interface. Mutations of these residues replicate the effects of PI(3,4,5)P₃ depletion, directly validating the lipid-bridging mechanism. Together, these findings establish that PIPs play distinct and synergistic roles during GSDMD pore formation: Because GSDMD density at the PM is not affected by the lack of PI(3,4,5)P₃, we hypothesize that PI(4,5)P₂ is sufficient to target GSDMD to the PM but not to stabilize GSDMD assemblies, whereas PI(3,4,5)P₃ stabilizes intersubunit interfaces to promote pore opening and growth. Moreover, they further highlight that the active modulation of PI(3,4,5)P₃ levels at the PM is a key mechanism by which pyroptotic cells ensure efficient GSDMD pore formation, allowing cells to tune pore stability through local lipid metabolism. Notably, recent studies show that PI3Kγ (a class I PI3K) directly modulates nucleotide-binding oligomerization domain (NOD), leucine-rich repeat (LRR)-containing proteins (NLR) family pyrin domain containing 3 (NLRP3)/GSDMD-mediated pyroptosis in septic myocardial injury and noncanonical pyroptosis in abdominal aortic aneurysm, with PI3Kγ inhibition reducing pyroptotic signaling and IL-1β release (66, 67). These findings support a physiological role for PI3K/PIP₃ in tuning pyroptotic responses under inflammatory conditions and are consistent with our conclusion that PI(3,4,5)P₃ regulates pore size and stability.

In summary, our work provides mechanistic insights into the structural organization and lipid regulation of GSDMD pores. By visualizing pore architectures directly in their native PM environment, we demonstrate that the formation and stabilization of functional pores are actively regulated by changes in the local PM lipid composition. The molecular basis for this effect results from the ability of the highly negatively charged PI(3,4,5)P₃ to form strong electrostatic interactions with GSDMD, thereby reinforcing intersubunit interfaces and promoting the maturation of large, conductive pores. Beyond establishing a role for PI(3,4,5)P₃ in pyroptosis, our findings raise the intriguing possibility that contexts with altered PI3K signaling, such as cancer, may influence the efficiency of GSDMD pore formation. In these settings, elevated PI(3,4,5)P₃ levels could, in principle, potentiate pore stabilization if GSDMD becomes activated, providing a mechanistic link between oncogenic signaling and inflammatory cell death pathways. More broadly, by demonstrating that PI(3,4,5)P₃ directly participates in pore stabilization, we identify this lipid as an active structural component of the pyroptotic pore and highlight PIP metabolism as a promising axis for modulating inflammatory cell death in disease.

MATERIALS AND METHODS

Cell culture

HEK293T stable DmrB-mCas1 (Tet-On vector) and double-stable DmrB-mCas1 (Tet-On) + mGSDMD-N-mEGFP-C [with mEGFP just before the caspase cleavage site, as in (33)] were cultured at 37°C and 5% CO₂ in minimum essential medium (MEM) Eagle (PAN Biotech, P04-09500) supplemented with 10% Tet-On system approved fetal bovine serum (Gibco, A4736401), 1% nonessential amino acids (MEM NEAA, PanBiotech, P08-32100) and 1% Hepes buffer 1 M (PanBiotech, P05-01100). Cells were transfected at 80% confluency by calcium phosphate precipitation overnight. The day before microscopy, cells were detached by treatment with trypsin/EDTA (Capricorn Scientific, TRY-1B10) at room temperature (RT) and seeded on the respective microscopy supports.

Pyroptosis induction and permeabilization kinetics

Sixteen to 24 hours before the experiments, cells were treated with doxycycline (500 ng/ml; Sigma-Aldrich, D3447) to induce expression of DmrB-mCas1. Pyroptosis was induced by addition of 500 nM B/B-homodimerizer (Takara Bio, AP20187) at 37°C and 5% CO₂. To observe pyroptotic cell permeabilization, the sample was treated with 0.5 μM ToPro3 Iodide (Thermo Fisher Scientific, T3605) or 3 μM SYTOX Blue Cell Death stain (Thermo Fisher Scientific, S34857). Pyroptosis kinetics measurements were performed by acquiring images before pyroptosis induction and afterward with intervals of 15 min at a Spinning disk confocal microscope (CellVoyager CQ1 Benchtop High, Yokogawa) at 37°C with a dry 40× objective.

Polymer-supported PMs

To generate PSPMs, gridded coverslips (ibidi, 10817) were intensively cleaned with isopropanol before and after plasma cleaning (Plasma Cleaner femto 1A, Diener electronics) at 100% output power for 15 min. Subsequently, the slides were coated with a 30/70% (w/w) mixture of poly-l-lysine coupled to a polyethylene glycol functionalized with either a HaloTag-ligand (PLL-PEG-HTL) or an RGD-peptide (PLL-PEG-RGD), respectively (68). Cells for PSPM synthesis were transfected with pDisplay-HaloTag-mTagBFP-TMD-GSlinker (TMD sequence: ASALAALAALAALAALAALAALAKSSRL) and seeded on the functionalized slides the day before microscopy. While the RGD-peptide functionalization allows attachment of the cells by integrin interactions, the covalent HTL-HaloTag interaction allows stable tethering of the PM to the surface. To minimize side effects due to membrane anchoring, the percentage of HaloTag-ligand binding PM on the polymer-coated surface was kept at 30%. For final PSPM generation, cells were treated with 10 μM latrunculin B (Abcam, ab144291) for 5 min at 37°C, and subsequently, the cell body was removed by shear forces through heavy pipetting directly at the microscope. Cell debris was removed by washing the sample 3× with 1 ml of phosphate-buffered saline (PBS) (Dulbecco’s PBS, PanBiotech, P04-35500). Afterward, PSPMs were fixed with 4% paraformaldehyde.

Scanning electron microscopy

For SEM, PSPMs of HEK293T DS cells were prepared as described above. After the removal of the cell body, PSPMs were fixed with 2.5% glutaraldehyde (Science Services, Germany) in 0.2 M Hepes buffer (pH 7.2; Roth, Germany) for 30 min at room temperature and then washed twice with 0.2 M Hepes buffer (pH 7.2) and once with Milli-Q water (Merck Millipore). Samples were stained with aqueous 0.1% uranyl acetate (Science Services) for 20 min, washed once with Milli-Q water (Merck Millipore), and dehydrated in a graded ethanol series of 50, 70, 80, and 90% and twice in 100% ethanol for 4 min each. Last, PSPMs were critical point dried in 100% ethanol in a critical-point dryer (Leica CPD300, Leica, Austria) with liquid carbon dioxide as transition fluid, then glued onto Leit-tabs (Plano, Germany) mounted on aluminum stubs (Plano, Germany), and sputter-coated with a 3-nm-thin gold layer (ACD600, Leica, Austria). SEM images were acquired with a Zeiss Auriga FEG-SEM (Zeiss, Germany) operating at an accelerating voltage of 4 kV, with an InLens detector at 4.2-mm working distance.

TIRF microscopy

All microscopy imaging experiments were performed at an inverted Olympus IX-81 microscope equipped with a motorized quad-line TIR-illumination condenser (cellTIRF-4-Line, Olympus), a motorized xy-stage (IM 120x80, Märzhäuser), a 100x oil immersion objective [UAPON 100× total internal reflection fluorescence (TIRF), numerical aperture (NA) of 1.49, Olympus], a large incubator with temperature control (TempController 2000-2, CellVivo) and a CO₂ controller (CO₂-controller 2000, CellVivo). To ensure PM tethering by HaloTag anchor expression, mTagBFP excitation was achieved by a 405-nm diode laser (Olympus), GSDMD-mEGFP oligomer formation was monitored by excitation with a 488-nm diode-pumped solid-state laser (Olympus), and Pten expression was ensured by excitation of mScarlet with a 561-nm diode-pumped solid-state laser (Olympus), both set to TIR conditions. Cell permeabilization was monitored by nuclear staining by the cell-impermeable dye ToPro3 Iodide via epimode excitation with a 640-nm diode laser (Olympus). Laser lines were filtered by clean-up filters (405 nm: BrightLine HC 390/40, Semrock; 488 nm: BrightLine HC 482/18, Semrock; 561 nm: BrightLine HC 561/14, Semrock; and 640 nm: BrightLine HC 640/1, Semrock). Fluorescence emission was filtered by a bandpass filter (BrightLine HC 446/523/500/677) and, in addition, single bandpass filters (BrightLine HC 390/40, 482/18, 561/14 and 640/14, Semrock) for each channel, respectively, before detection with a scientific complementary metal-oxide semiconductor camera (ORCAFlash 4.0 V3, Hamamatsu). Images were acquired with the software CellSens 3.2 (Olympus) with an exposure time of 32 ms and 2 × 2 pixel binning, resulting in a pixel size of 130 nm.

FRAP and single-molecule tracking

PSPM lipid mobility was investigated by FRAP and single-molecule tracking analysis at RT. First, PSPM generation and TIRF microscopy were performed as described above using HEK293T cells stably expressing DmrB-mCas1 and mGSDMD-mEGFP and transiently expressing the HaloTag-mTagBFP-TMD anchor and farnesyl-mCherry as a membrane marker. After PSPM generation directly at the microscope, a circular region with a radius of 5 μm was bleached using the 405-nm laser set to 100% for 10 s. Immediately afterward, the fluorescence recovery of farnesyl-mCherry was detected over 200 s with 1-s intervals. The apparent diffusion coefficient was derived by the photobleaching corrected fluorescence recovery over time

$$\text{Norm}.\text{intensity} \left(t\right)=\frac{I_{\text{Ctrl}}\left(0\right)}{I_{\text{FRAP}}\left(0\right)}·\frac{I_{\text{FRAP}}\left(t\right)}{I_{\text{Ctrl}}\left(t\right)}$$ (1)

A monoexponential fit provided the half-life τ_1/2, which, together with the radius r, leads to the diffusion coefficient by the Soumpasis equation (69)

$$D=0.224\ast \frac{r^2}{{\mathrm{\tau }}_{1/2}}$$ (2)

To analyze the dynamics at single-molecule level by tracking, we substoichiometrically labeled farnesyl-mCherry by 10 pM of a Dy647-conjugated LaM-anti-mCherry nanobody. After incubation for 10 min, the diffusion of single farnesyl-mCherry molecules was monitored for 500 frames with a 30-Hz frame rate. The diffusion coefficient was obtained by mean square displacement (MSD) analysis using the software SLIMFast 4C (34), and trajectories were extracted using the software TrackIt (70).

Atomic force microscopy

AFM experiments were performed on a JPK nanowizard atomic force system (JPK Instruments) mounted on an inverted Olympus IX-71 microscope (Olympus). PSPMs were imaged using silicon nitride cantilevers (SNL-10, Bruker) with a typical spring constant of 0.09 N/m in intermittent contact mode. Here, the cantilever oscillation was fine-tuned between 2 and 5 kHz, and the amplitude was set to 1 V. Imaging was performed at RT with a scan rate between 0.6 and 1 Hz. Height profiles were obtained using a smoothing function with the JPKSPM data processing (JPK instruments).

DNA-PAINT

PSPMs were fixed after PSPM generation with 4% paraformaldehyde (PFA) solution for 15 min at RT and subsequently washed 5× with 1 ml of PBS (PanBiotech, P04-35500) before storage at 4°C. For DNA-PAINT preparation, the samples were reduced with sodium borohydride (1 mg/ml) for 7 min at RT and washed 4× for 5 min with 1 ml of washing buffer (Massive Photonics) before incubation with 50 nM anti-GFP nanobody + docking strand 3 or docking strand F3 (MASSIVE-TAG-Q anti-GFP DNA-PAINT Kit, Massive Photonics) dissolved in antibody-incubation buffer (Massive Photonics) for 2 hours at RT or at 4°C overnight. Afterward, the sample was washed 3× with 5 ml of washing buffer and incubated with a 1:1000 dilution of 90-nm gold nanorods (Cytodiagnostics, G-90-20) serving as fiducial markers for 5 min at RT and washed again with washing buffer. To ensure the removal of unspecifically bound nanobodies, the sample was washed with imaging buffer (Massive Photonics) before supplementation with 1 nM Cy3b-conjugated imager strand 3 of imager strand F3 (MASSIVE-TAG-Q anti-GFP DNA-PAINT Kit, Massive Photonics) dissolved in imaging buffer. All DNA-PAINT experiments were performed at the setup described in the “TIRF microscopy” section. The temperature was kept stable at 27°C, and buffer evaporation was prevented by humidification. The Imager was excited with a 561-nm laser adjusted to a power density of ~100 W/cm² in the focal plane. For each recording, typically 40,000 frames were acquired with 200 ms (Imager strand 3) or 50 ms (Imager strand-F3) exposure time and 2 × 2 pixel binning. During acquisition, the focus plane was stabilized with a hardware autofocus system (IX2-ZDC2, Olympus). DNA-PAINT raw datasets were processed with the Picasso software suite (71). For analysis, the last 30,000 frames of each dataset were considered to minimize drift and unspecific signals. First, DNA-PAINT signals were identified and localized in “Picasso Localize.” The box size was set to 7 pixels, and the minimal netto gradient (Min. Net. Gradient) was typically adjusted to 2000 to 5000 to filter unwanted background signals and signals with a weak signal-to-noise ratio. Photon conversion parameters were set as follows: EM gain of 1, baseline of 400, sensitivity of 0.46, quantum efficiency of 0.72, and pixel size of 130 nm. Single emitter signals were fitted with integrated Gaussian fit. To ensure sufficient resolution and eliminate nonspecific localization artifacts, we post-processed the imaging data (Fig. 2C and fig. S2). Following 30,000 frames of stable imaging, indicated by a consistent number of localizations and precision below 9.75 nm (Fig. 2, A to C), we corrected for drift using 90-nm gold fiducials. Afterward, localizations were linked within 6.5 nm and with dark times below the average localization lifetime (two to four frames) (Fig. 2, D and E). To identify GSDMD specific localizations, we automatically identified localization clusters smaller than 0.5 pixels that exhibited reappearing localizations throughout the experiment. Specifically, bright super-resolved structures were picked with a pick diameter of 0.5 pixels and filtered by trace for frequent sampling over all frames to select for specific GSDMD interactions. With the “pick similar” function, specific structures with a frequent sampling pattern were picked automatically with a standard deviation (STDEV) of localization number of 1.0. For further removal of accumulations of unspecific localizations, we filtered all picked structures by the number of localizations. Experiments with DNA-origami monomeric binding sites (Fig. 3) showed each strand yields 10 to 20 localizations more than 30,000 frames under the same imaging conditions. Therefore, we further filtered identified structures to include only those with 30 to 500 total localizations, ensuring the exclusion of nonspecific localizations while accounting for low-order oligomers (Fig. 2, E to G). We confirmed the specificity of DNA-PAINT localizations by showing that most correlate with GSDMD-GFP signals (Fig. 2C). GSDMD-GFP signals without matching DNA-PAINT localizations likely reflected monomeric or low-order oligomers that were filtered out due to localization values falling below the localization threshold, potentially as a result of limited nanobody labeling efficiency. Conversely, unmatched DNA-PAINT signals likely represented true GSDMD assemblies for which the GFP signal was lost due to photobleaching during the 90-min imaging period following pyroptosis induction. From the remaining structures, 50 random overview images per dataset were saved. In these images, each signal was rendered at the median localization precision within the field of view and displayed at a zoom factor of 100 (pixel size of 1.3 nm). The images were saved as 8-bit tagged image file format (TIFF) files for structural analysis with Automated Structures Analysis Program (ASAP) (see below) (Fig. 2, H and I).

Structural classification of super-resolved structures

Overview images of DNA-PAINT super-resolved GSDMD structures were analyzed for structural classification. First, we applied the software ASAP (37). Structures were identified by “connectivity” with the following parameters: thresholding method: fixed; threshold multiplier: 0.1; no cleaner; and identification range size: 200 to 10,000. The exported identification files with a pixel size of 1.3 nm (Zoom 100, see the “DNA-PAINT” section) were analyzed by radial profiling (plotting the average intensity along concentric circles and determining the radius corresponding to the highest intensity; Fig. 2I) with a maximum ring size of 50 pixels, and no operations were selected. On the basis of observations on the appearance of GSDMD structures, the software was trained to classify all identified structures into the following shape classifications: complete rings, incomplete rings, as well as linear and globular undefined clusters. Training for automated structural classification was performed with the learner “discriminant” and by the descriptor “Raw Radial Profile”. Afterward, all identified super-resolved GSDMD structures were automatically classified into the shapes according to the training process and subsequently verified for proper classification.

Quantitative PAINT

Stoichiometric analysis by qPAINT was performed by comparing the binding frequency of each GSDMD structure with the average binding frequency of a monomeric reference site. This reference, which contains a defined number of imager-binding sites, was used to convert the measured binding frequency of each GSDMD structure into the corresponding number of subunits based on the relative frequency of DNA-PAINT localizations (31). As calibration standards, we used DNA-origami structures containing single docking sites coupled to the same docking strand as the anti-GFP nanobodies. The DNA-origami calibration structures consisted of a 3 × 4 array of docking sites spaced 20-nm apart. DNA-origami structures were assembled according to the protocol given in using DNA-oligonucleotides designed with “Picasso design” (71). To optimize the stoichiometric analysis, DNA-origami structures were immobilized next to tethered PSPMs on an orthogonal functionalized surface. Orthogonal functionalization was achieved by micro-contact printing of PLL-PEG-HTL on plasma cleaned cover slides with PDMS stamps (68) creating square patterns of 100 μm by 100 μm in size, allowing tethering of PSPMs, and subsequent backfilling with PLL-PEG-Biotin for DNA-origami immobilization. Cell seeding and incubation, pyroptosis induction, and PSPM generation were performed as described above. For simultaneous DNA-PAINT imaging, the sample was first prepared as described above, and DNA-origami immobilization was performed as described in (71). After incubation with the anti-GFP nanobody + docking strand 3, the sample was washed 2× with 500 μl of buffer A⁺ followed by incubation with streptavidin (1 mg/ml; Streptavidin UltraPure, PanREAC AppliChem, A1495005) dissolved in buffer A⁺ for 5 min. Unbound streptavidin was washed out 2× with 500 μl of buffer A⁺, and the buffer was exchanged by washing 2× with 500 μl of buffer B⁺. A 50 μl of a 1:10 dilution of purified DNA-origamis in buffer B⁺ was supplemented for 30 min at RT, and unbound molecules were removed by washing with 500 μl of buffer B⁺. Afterward, the Imager strand-Cy3b was used at 1 nM in buffer B⁺ for DNA-PAINT imaging. Imaging and DNA-PAINT postprocessing were performed as described above, but here the DNA-origamis themselves served as fiducial markers. For qPAINT calibration, the average dark mean time of a single binding site for an individual experiment was gained by taking the localization frequencies of >10,000 DNA-origami single binding sites per experiment and fitting all dark mean times. For each structure classified as a ring, the individual dark times were used to calculate the number of binding sites by comparing them with the average dark time of a single binding site in the same experiment. To determine the number of GSDMD subunits per structure, the calculated number of anti-GFP nanobody binding sites was corrected for the labeling efficiency of ~47% (nonapplicable for DNA-origami binding sites as, here, the imager strand is part of the structure and not connected via nanobody binding) (72).

MINFLUX—DNA-PAINT

MINFLUX imaging data were recorded on a commercial Abberior MINFLUX microscope (Abberior Instruments). Details of the microscope and components are described in (73). Briefly, the microscope is built around an inverted microscope body (IX83, Olympus) and equipped with the following laser lines that were required for these experiments: 405 nm (confocal), 488 nm (confocal), and 642 nm (confocal and MINFLUX). The system uses two stabilization systems—sample stabilization, which uses a 980-nm infrared laser and an xyz sample piezo (Piezoconcept), and beamline monitoring. The system uses a 100× 1.45 NA UPLXAPO oil objective (Olympus). Detection was performed in two spectral channels using two avalanche photodiodes (APDs), one at 650 to 680 nm and one at 685 to 720 nm, from which the signal was added up. The microscope was controlled using the Imspector microscope control software [v16.3.15635, Abberior Instruments, (74)]. The microscope was aligned (MINFLUX beam PSF, beam overlapping, and pinhole position) before each measurement session using a sample containing gold beads and fluorescent beads immobilized on a cover slip (Abberior Instruments). For the MINFLUX measurements, the default MINFLUX sequence for two-dimensional imaging with a hexagonal targeted coordinate pattern supplied with the system was used (provided as a supplementary file, while the most important parameters are listed in Table 1).

Table 1. Parameter MINFLUX—DNA-PAINT imaging.

Iteration	Target Control Point length (nm)	Photon limit	Dwell time (ms)	Pattern repeats	Count Flux Ratio limit	Background threshold (kHz)	Laser power factor
0 (preloc.)	–	160	1	1	–	15	1
1	288	150	1	5	–	10	1
2	151	100	1	5	0.8	10	2
3	76	100	1	5	0.8	10	4
4	40	150	1	5	2	10	6

PSPMs from pyroptotic HEK293T Casp-1 cells were scanned in confocal mode using the 488-nm excitation laser, imaging hGSDMD-mEGFP oligomers. A region of interest was chosen in the resulting confocal image, of a size ranging between 4 to 12 × 4 to 12 μm². The binding and unbinding of the DNA-PAINT imager strands (see the “DNA-PAINT” section) were confirmed before the start of each measurement with rapid confocal scanning using the 642-nm excitation laser. The two stabilization systems were started before each image acquisition, and the sample stabilization precision was typically ~1 nm. MINFLUX excitation laser power and pinhole size were additionally adjusted to the optimal values for the sample—642-nm laser power of 20 to 60 μW in the first iteration and pinhole diameter 0.67 airy units (AU) to minimize background from unbound imager strands. Each image was acquired over a time period of 5 to 17 hours, depending on the size of the region of interest. Single-molecule localization events displayed in MINFLUX images have been combined into aggregates of 1000 photons.

Spinning disk confocal microscopy

For monitoring PI(4,5)P₂ and PI(3,4,5)P₃ dynamics during pyroptosis, HEK296T cells stably expressing DmrB-mCas1 were transfected with pSems-GSDMD-mCherry, pcDNA3-AKT-PH-GFP or pSems-Akt-PH-mScarlet, and pEGFP-iRFP-PH-PLCδ1 via calcium phosphate transfection overnight. A day before the experiment, the cells were seeded into eight-well plates (80821, ibidi) and treated with doxycycline (500 ng/ml), and directly before the experiment, 3 μM SYTOX Blue cell death stain (Thermo Fisher Scientific, S34857) is added. Pyroptosis was induced as described above. Alternatively, HEK293T Casp-1 cells expressing hGSDMD-mEGFP were seeded into eight-well plates (80821, ibidi) and treated with doxycycline (500 ng/ml). On the next day, pyroptosis was induced as described above. The cells were fixed at different time points of pyroptosis with 3% PFA + 0.1% glutaraldehyde (GA) for 15 min at RT and afterward washed 3× with PBS. To stain PM PI(3,4,5)P₃ levels, the cells were first permeabilized with 0.2% Triton X-100 and 3% bovine serum albumin in PBS for 30 min at 4°C, washed with PBS, and lastly treated with 0.6 μM recombinant Akt-PH-mScarlet (see the “Akt-PH-mScarlet purification” section) for 1 hour at 4°C and washed 3× with PBS. The cells were imaged either before pyroptosis induction and at every 10 min afterward or once for each time point, respectively, at a fully motorized inverted spinning disc microscope based on a Zeiss Cell Observer Z1 and a CSU-X1 spinning disc unit (Yokogawa). The microscope was equipped with live-cell imaging periphery based on a home-built incubation chamber enclosing the full microscope. Heating at 37°C is managed by a proportional–integral–derivative (PID)-controlled heater (The cube, Live Imaging Services). Samples were imaged at 37°C in a humidified 5% CO₂ atmosphere (Zeiss CO₂ module S1, Zeiss humidifier module S1). To monitor cell permeabilization, SYTOX Blue was excited with a 405-nm diode laser (maximum of 50 mW), and GSDMD-mCherry expression was checked with a 561-nm diode laser (maximum of 40 mW), while the lipid markers PH-PLCδ1-iRFP and PH-AKt-mEGFP were excited using a 488-nm optically pumped semiconductor laser (maximum of 100 mW) and a 635-nm diode laser (maximum of 30 mW), respectively, to monitor lipid dynamics. Emission light passed through a 63× oil immersion objective (Alpha Plan-Apochromat, NA 1.46) and a polychroic mirror and was filtered by a set of bandpass filters (BFP: 450/25, GFP: 525/25, mCherry: dual bandpass of 500 to 554 nm and 615 to 675 nm, and iRFP: 690/50) before being detected by a Hamamatsu ORCA Flash V3, resulting in a pixel size of 86 nm. The sample was kept stable during the experiment using a motorized xyz-stage (PZ-2000 XYZ, Applied Scientific Instrumentation) and a Definite Focus System (Zeiss). Every step was performed with the software Zeiss Zen 2.6. Because no additional spectral channel was available for an independent PM marker, the PIP sensor distribution was quantified by measuring the relative enrichment of each sensor at the cell periphery. For each cell, fluorescence intensity was measured at the PM and in the cytosolic interior using Fiji. Background signal (camera offset) was subtracted from each measurement, and the PM-to-cytosol intensity ratio (PM/cytosol) was calculated for both PIP sensors. Ratios were normalized to the corresponding value at time point t = 0 for each cell. This approach provides a relative measure of PM enrichment for each lipid sensor (Figs. 6, A and B, and 7, A to C).

Akt-PH-mScarlet purification

The construct pet21a-AKT1-PH-mScarlet-H6 was expressed in Escherichia coli BL21 [CodonPlus-RIL (arginine, isoleucine, leucine)] after electroporation with 1 μl at 2.5 kV for 6 ms in a 2-mm chamber. From overnight grown precultures (100 ml), 4× 1-liter LB medium was inoculated to an optical density at 600 nm (OD_600nm) of 0.2 and incubated at 37°C with 160 rpm until OD_600nm of 0.6. The expression was induced with 1 mM isopropyl-β-d-thiogalactopyranoside, and the cells were incubated at 16°C and 160 rpm overnight and afterward harvested by centrifugation on 4500 rpm at 4°C for 15 min and stored at −20°C. Cells were thawed and lysed by sonication on ice for 3 × 5 min + 3-min break, 2 × 2 min with 2-min break, with 0.5-s active and inactive interval in 20 ml of lysis buffer per pellet [50 mM tris-HCl (pH 8.0), 150 mM NaCl, deoxyribonuclease, 1 mM phenylmethylsulfonyl fluoride, lysozyme, and protease inhibitor). The lysed cells were ultracentrifuged at 20,000 relative centrifugal force (RCF) for 45 min at 4°C. The proteins were purified from the lysed cells by nickel affinity (Ni-NTA agarose beads) using immobilized metal affinity chromatography (IMAC) buffer A [50 mM tris-HCl (pH 8.0) and 150 mM NaCl] and increasing concentrations of IMAC buffer B [50 mM tris-HCl (pH 8.0), 150 mM NaCl, and 500 mM imidazole] and size exclusion (Superdex 200 increase 300/10 GE Healthcare) chromatography using size exclusion chromatography buffer [50 mM tris-HCl (pH 8.0), 150 mM NaCl, and 1 mM dithiothreitol (DTT)] and stored in 50 mM tris, 150 mM NaCl, and 1 mM DTT (pH 8) at −80°C.

LDH assay

Lactate dehydrogenase (LDH) release cytotoxicity assay was performed in 96-well plates (Thermo Fisher Scientific, M33089) using the CyQUANT LDH Cytotoxicity assay (Invitrogen, C20301) according to the manufacturer’s protocol. Each sample and control was performed in triplicates with 20,000 HEK293T cells stably expressing DmrB-mCas1 and mGSDMD-mEGFP (with or without PGK promotor) seeded per well on the day before. The cells were treated with doxycycline and pyroptosis induction was performed as described above. The fluorescence detection of LDH release was performed using the Tecon Plate reader Infinite 200 Pro M-Plex.

GUV permeabilization assay

POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholin; Avanti Polar Lipids, 850457), PI(4,5)P₂ (Avanti polar lipids, 850165P), and PI(3,4,5)P₃ (Avanti polar lipids, 850166P) were purchased from Avanti Polar Lipids (Alabaster, AL). The lipid mixtures [95:05 mol % of POPC:PI(4,5)P₂/PI(3,4,5)P₃] were generated by solving the required amounts of each component in chloroform and thorough vortexing. These mixtures were dried with N₂ and put under vacuum for >1 hour before storage at −20°C. For GUV preparation, lipid stocks were solved in chloroform to a final concentration of 5 mg/ml. On each platinum wire of the electroformation chambers, a 2.5 μl of the lipid mixture was spread, and after solvent evaporation, the wires were immersed in 300 mM sucrose solution. Electric pulses of 1.5 V at 10 Hz were provided for 2 hours, followed by a frequency of 2 Hz for 1 hour. For GUV visualization and to ensure proper distribution of PI(4,5)P₂ and PI(3,4,5)P₃ in the liposomes, the GUVs were treated with 200 nM recombinant PLCΔ1-2x-PH-Halo + 50 nM HTL-Dy549 or 2 μM recombinant Akt-PH-mScarlet (see the “Akt-PH-mScarlet purification” section), respectively, for 1 hour. To assess GUV permeabilization by activated mGSDMD, the protein mixtures were prepared in eight-well chambers (ibidi) with 125 nM mGSDMD-8xHis and 20 nM Caspase-11delCARD with a final volume of 200 μl in PBS (PanBiotech) supplemented with 2.5 μM 10-kDa dextran–Alexa Fluor 647 (Thermo Fisher Scientific, D22914) and 150-kDa dextran-fluorescein (Sigma-Aldrich, 69658). To each well, 50 μl of GUVs was added, and the GUV permeabilization was recorded in three distinct positions per well every 15 min for 60 min in total as soon as GUVs were settled down. Imaging was performed at room temperature with a Zeiss LSM 880 confocal microscope. GUV permeabilization was analyzed with Automatic GUV analysis software (75) with the following settings: Radius_min: 10, Radius_max: 50, Radius_corr: 0.5, sensitivity: 1.24, threshold: 0.2, and inside: 50. GUVs were considered as permeabilized when the inside intensity was at least 20% of the background intensity.

MD simulations

We constructed a computational assay to quantify the stabilizing effect of PI(4,5)P₂ and PI(3,4,5)P₃ on the stability of GSDMD assemblies by adapting the MD simulation protocol of Schaefer and Hummer (30). Specifically, we probed the resistance of GSDMD arcs with or without PIPs to mechanical force created by the line tension of the open membrane edge between the two ends of the arc. Starting from the experimental structure of a complete hGSDMD ring [Protein Data Bank (PDB) ID: 6VFE (18)], we set up structures of 16-meric half-rings of hGSDMD in pore conformation and placed them in a 36 nm × 36 nm large membrane comprising 3750 DOPC lipids. We then placed a single PI(4,5)P₂ or PI(3,4,5)P₃ lipid with palmitoyl (16:0) and oleoyl (18:1) tails in each of the interfaces between neighboring subunits. We obtained PIP lipid topologies from CHARMM-GUI (76) and placed their headgroups into a previously described lateral binding site that involves K43, R53, K55, R135, and K235 (30). For reference, we additionally performed three new replicate simulations of the previously described 16-meric hGSDMD half-ring in a pure DOPC membrane without any PIPs as setup and equilibrated previously (30). We built the simulation systems as described in (30) and followed the same steepest descent energy minimization and equilibration protocols comprising three equilibration steps (5, 50, and 80 ns) with decreasing restraints on the atomic positions of the system. During the third equilibration simulation, the lipids (including the PIPs) were not subjected to any restraints. This setup without other acidic lipid species and without a reservoir of PIP lipids in the bulk membrane was chosen specifically to quantify the stabilizing effect of different PIP species on the assembly in a controlled manner. For the bulk of the membrane, we used DOPC lipids because we found earlier that their high fluidity ensures relatively rapid dynamics of the membrane-inserted GSDMs (30).

After equilibration, we performed three production simulations with starting velocities drawn independently from the Maxwell-Boltzmann distribution for each of the systems. During these production simulations, no positional restraints were applied.

All production simulations were performed using Gromacs version 2022.4 (77). The interactions were described with the CHARMM36m (78) force field and TIP3P (transferable intermolecular potential with 3 points) water (79). The MD simulations were performed at a constant pressure of 1 bar, maintained by using the semi-isotropic (X and Y dimensions coupled together) Parrinello-Rahman barostat (80) with a coupling time constant of 5 ps and a compressibility factor of 4.5 × 10⁻⁵ bar⁻¹. To maintain a constant temperature of 37°C, we applied velocity rescale thermostats (81) separately to the protein, the lipids, and the solvent atoms every 1 ps. We computed electrostatic interactions using the particle-mesh Ewald algorithm (82) with a cutoff distance of 1.2 nm for the real space electrostatics. For van der Waals interactions, we used the same cutoff distance. Bonds to hydrogen atoms were constrained using the LINCS (linear constraint solver) algorithm (83).

To analyze the effect of the presence or absence of PIPs on the stability of the GSDMD assembly, we measured the distance between the globular domains (residues 34 to 71, 123 to 159, and 215 to 239) of neighboring subunits. From the distribution p(r) of distances r between neighboring subunits, we estimated the Gibbs free energy of GSDMD interface cracking as

$$\mathrm{\Delta G}\left(r\right)=-k_{\mathrm{B}}\mathit{\text{Tlnln}} p\left(r\right)$$ (3)

where $k_{\mathrm{B}}$ is the Boltzmann’s constant and T is the absolute temperature. For this reconstruction of ΔG, we only considered the last 700 ns of the replicate simulations. The constructed free energies were shifted vertically, so that their global minimum, which corresponds to an intact GSDMD interface, is at zero. Based on the position of the maximum of the free energy landscapes, we considered a structure as cracked in the respective position if the distance r between two neighboring subunits exceeded 31.5 Å.

In addition, for neighboring subunits that do not have a cracked interface at a given time point, we also analyzed the PIP binding modes during the last 700 ns of the simulations. To do so, we counted the number of contacts of the respective bound PIP with all heavy protein atoms using a 3.6-Å cutoff and averaged them over all intact interfaces and over all considered frames. To calculate the electrostatic potential of GSDMD, we used the APBS webserver (84). For this, we uploaded a GSDMD dimer extracted from the last frame of one of the simulations with PI(3,4,5)P₃ lipids and used the PDB2PQR tool to prepare input files for APBS. Since the input structure was the result of unbiased MD, we did not remodel any hydrogen atoms or optimized their bonding network. The APBS calculation itself was performed with default parameters.

Visual analysis and rendering were done using VMD (85), UCSF ChimeraX (86), and PyMOL (87).

Statistical analysis

The significance of differences between two conditions was determined using Student’s t test. Here, n.s. indicates a nonsignificant difference, “*”, “**,” and “***” indicate a P value below 0.05, 0.01, and 0.001, respectively.

Supplementary Materials

The PDF file includes:

Figs. S1 to S7

Legends for movies S1 to S3

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S3