Pore formation in regulated cell death

Abstract The discovery of alternative signaling pathways that regulate cell death has revealed multiple strategies for promoting cell death with diverse consequences at the tissue and organism level. Despite the divergence in the molecular components involved, membrane permeabilization is a common theme in the execution of regulated cell death. In apoptosis, the permeabilization of the outer mitochondrial membrane by BAX and BAK releases apoptotic factors that initiate the caspase cascade and is considered the point of no return in cell death commitment. Pyroptosis and necroptosis also require the perforation of the plasma membrane at the execution step, which involves Gasdermins in pyroptosis, and MLKL in the case of necroptosis. Although BAX/BAK, Gasdermins and MLKL share certain molecular features like oligomerization, they form pores in different cellular membranes via distinct mechanisms. Here, we compare and contrast how BAX/BAK, Gasdermins, and MLKL alter membrane permeability from a structural and biophysical perspective and discuss the general principles of membrane permeabilization in the execution of regulated cell death.


Pore formation in membranes is a conserved strategy to kill cells
Biological membranes define biochemical environments and are fundamental for the existence of life (Deamer, 2016). Selective transport between these enclosed spaces is highly regulated and sustained disruption in membrane integrity is, as a consequence, a point of no return in cell death (Youle & Strasser, 2008;Kunzelmann, 2016;Broz et al, 2020). Yet, alterations in membrane permeability are relevant for a large number of biological processes including immunity, metabolism, and infection (Kagan, 2012;McCormack et al, 2013).
A membrane pore can be defined as any local membrane perturbation that allows the passive flow of molecules (Schwarz & Robert, 1992). Pore-forming proteins (PFPs) represent a large and structurally diverse family of proteins that have the common function of altering membrane permeability by creating pores. They can act exogenously as secreted soluble proteins that permeabilize the plasma membrane of their target cells. This includes most poreforming toxins (PFTs), which are some of the most potent virulence factors found in nature (Ros & Garcia-Saez, 2015;Dal Peraro & van der Goot, 2016) or perforin, which is released by cytotoxic T cells and Natural Killer cells (Voskoboinik et al, 2015;Prinz et al, 2020). PFPs can also be intracellular executioners as components of cell death signaling pathways (Espiritu et al, 2019;. For instance, Bcl-2-associated X protein (BAX) and BCL2antagonist/killer 1 (BAK) form pores that lead to mitochondrial outer membrane permeabilization (MOMP) during intrinsic apoptosis. Gasdermins (GSDMs) execute pyroptosis by a mechanism that culminates with pore opening at the plasma membrane. In necroptosis, mixed lineage kinase domain-like (MLKL) induces plasma membrane permeabilization through a yet unclear mechanism that could also be linked to pore formation Cosentino & Garcia-Saez, 2017;Flores-Romero & Garcia-Saez, 2019a).
PFPs are usually classified in a or b, depending on the secondary structure of the protein segments forming the pores (Ros & Garcia-Saez, 2015;Cosentino et al, 2016). In addition, PFPs can build different types of pores, as defined by the presence or absence of lipids in the pore structure (Ros & Garcia-Saez, 2015;Gilbert, 2016). They are classified as protein-lined if they are constituted only of proteins, as pure lipid pores, or as protein/lipid pores, when they contain both types of molecules. In protein-lined pores, the lumen is solely covered by transmembrane segments of proteins organized into aor b-barrel "walls" (Dal Peraro & van der Goot, 2016;Gilbert, 2016). Importantly, in this type of pores, the membrane or certain membrane lipids can play a functional role in protein recruitment, assembly, and folding (Rojko & Anderluh, 2015;Gilbert, 2016). In contrast, membrane lipids together with the amphipathic regions of proteins or peptides form the edge of protein/lipid pores (Ludtke et al, 1996;Matsuzaki et al, 1996). Here, lipids rearrange from their bilayer distribution to a non-lamellar assembly that creates a continuous surface in which the membrane bends at the pore boundary like a torus. Protein/lipid or toroidal pore opening is promoted by proteins or protein fragments that generate membrane tension. PFP accumulation at or next to the pore rim then reduces the line tension and stabilizes the open pore state Puech et al, 2003) (Fig 1). Pure lipid pores are also toroidal, but in the absence of proteins, their opening probability and lifetime are very low. They mainly occur upon strong membrane perturbations such as mechanical and electrical tension or osmotic swelling (Tieleman et al, 2003;Tieleman & Marrink, 2006).
In regulated cell death, a diverse repertoire of endogenous prodeath effectors reveals a plethora of strategies evolved to permeabilize cellular membranes. The architecture of the resulting pores is defined by specific protein and lipid compositions, as well as by intramolecular interactions. This determines pore heterogeneity, size, and stability, and thereby the type and extent of molecules that can be released through them. As a consequence, the properties of membrane pores impact the signaling cascades that are activated downstream of membrane permeabilization. (i) Pore-forming peptides bind avidly to the accessible interface of the lipid bilayer and occupy a volume only in the interfacial region, which causes asymmetric stretching and membrane thinning (Δh). As a consequence, the membrane is stressed and destabilized, so that defects in the lipid bilayer become more likely and eventually a pore is formed (ii). Once the pore is open, a line tension appears at the pore edge due to the extra energy cost associated with the reorientation of the lipids into a highly curved boundary. This tension increases with the pore perimeter and is therefore a line tension. The initial pore grows quickly as long as the membrane tension dominates. But as the pore size grows, so does the counterbalancing line tension too. Furthermore, with the open pore, the peptides redistribute in the membrane by diffusion through the pore to the other monolayer, which reduces the membrane tension due to asymmetric distribution of the peptides. At a certain moment, the line tension becomes predominant and the pore size starts to decrease. However, if the pore-forming peptides bound near the pore rim are able to reduce the line tension, an equilibrium can be reached with a smaller but stable pore (iii). R = radius, Δh = change in the thickness of the membrane (absence vs. presence of the protein/peptide mass). Adapted from .

BAX and BAK pores in apoptosis: old friends with new habits
The toroidal pore of BAX and BAK Currently, it is well established that the BCL2 family members BAX, BAK, and perhaps BOK are the executioners of MOMP and thereby fundamental effectors of the intrinsic apoptotic pathway (Moldoveanu & Czabotar, 2019). In healthy cells, BAX and BAK exist as inactive, monomeric proteins that constitutively shuttle between cytosol and mitochondria, with BAX being mostly cytosolic and BAK mitochondria-associated (Edlich et al, 2011;Schellenberg et al, 2013;Todt et al, 2015;Lauterwasser et al, 2016). Upon apoptosis induction, both proteins accumulate at and insert into the mitochondrial membrane, undergo conformational rearrangements, oligomerize and form pores that release pro-apoptotic factors such as cytochrome c and SMAC/DIABLO (Nechushtan et al, 2001;Rehm et al, 2003;Zhou et al, 2005;Edlich et al, 2011). However, the relative order of events, the structural intermediates involved, as well as the molecular properties of the membrane openings mediated by BAX-type proteins remains controversial (Moldoveanu & Czabotar, 2019;Flores-Romero & Garcia-Saez, 2019a).
Despite their functional heterogeneity, the 3D structures of all multidomain BCL2 proteins (including not only BAX, BAK, and BOK, but also the pro-survival family members) present the same globular a-helical fold, with a predominantly hydrophobic central hairpin that is flanked on both sides by pairs of amphipathic ahelices. This peculiar folding is strikingly similar to the pore-forming domain of bacterial a-PFTs such as the colicins and diphtheria toxin (Muchmore et al, 1996;Suzuki et al, 2000;Petros et al, 2004;Moldoveanu et al, 2013;Ke et al, 2018). Considering these structural similarities and their ability to allow currents through artificial membrane systems known as black lipid membranes, it was initially proposed that BAX-type proteins induce MOMP by generating membrane pores or channels Sattler et al, 1997;Schlesinger et al, 1997;Basanez et al, 1999). Different observations have since then supported the hypothesis that active BAX and BAK form long-lived toroidal pores. In vitro, BAX decreases the lifetime of planar membranes and forms pores with variable conductance (Basanez et al, 1999) that are affected by the physical properties of the membrane and the presence of lipids with intrinsic monolayer curvature (Basanez et al, 2002;Terrones et al, 2004). A peptide derived from helix a5 of BAX forms membrane pores with lipid molecules in the lumen as demonstrated by X-ray diffraction (Qian et al, 2008) and conductance experiments in black lipid membranes (Garcia-Saez et al, 2005). This is in agreement with the transbilayer lipid movement coupled with membrane permeabilization induced by BAX (Epand et al, 2003;Garcia-Saez et al, 2006) (Fig 2A and B). These lines of evidence converge into the current view of a protein/lipid pore of tunable size constituted by BAX/BAK homodimers, where the size of BAX and BAK pores is not constant, but it evolves with time and depends on protein concentration. In this model, the central hairpin of helices a5 and a6 lies on the membrane surface at the edge of the pore (Qian et al, 2008;Basanez et al, 2012;Bleicken et al, 2013;Mandal et al, 2016;Cosentino & Garcia-Saez, 2017;Bleicken et al, 2018) (Fig 2A and B). Not only BAX/BAK molecules, but also the mechanical properties of the membrane play a role in the size and stability of the pores .
According to the toroidal pore model, the initial asymmetric insertion of BAX a-helices into the cytosolic leaflet of the MOM stresses the membrane and generates membrane tension. Once a threshold tension is reached (which may be locally enhanced by protein concentration or oligomerization), the energy required to reorganize lipids out of the bilayer structure becomes thermally accessible and a pore opens, which dissipates membrane tension . Lipids reassemble into a torus around the membrane pore to avoid exposure of the hydrophobic acyl chains to the water environment. The two membrane monolayers form a continuous surface at the pore edge with negative curvature in the plane of the membrane and positive curvature in the plane perpendicular to the membrane. The bending of the lipids at the pore rim has an energetic cost that is directly proportional to the length of the pore, giving rise to line tension (Ludtke et al, 1996;Matsuzaki et al, 1996). Line tension acts as the driving force for pore closure and opposes membrane tension. As a result, toroidal pores are metastable structures whose lifetime is governed by the balance between membrane tension and line tension (Valcarcel et al, 2001;Yang et al, 2001;Karatekin et al, 2003;Fuertes et al, 2010b). In support of this model, the rim of BAX pores is formed by both lipid headgroups and protein molecules (Kuwana et al, 2016;Salvador-Gallego et al, 2016), with BAX molecules decreasing the line tension to maintain the pore stably open (Basanez et al, 1999;Garcia-Saez et al, 2007;Fuertes et al, 2010a;Fuertes, 2011;Unsay et al, 2017;Bleicken et al, 2018) (Figs 1 and 2B).
BAX/BAK membrane topology in the context of the toroidal pore Once in the membrane, activated-BAX/BAK undergoes a rearrangement of their globular fold that implies first the opening of their Nterminal region and then, in the case of BAX, the dislodgment and transmembrane insertion of the tail anchoring domain in helix a9 Griffiths et al, 2001). This is followed by exposure of their BH3 domain, which is thought to occur concomitantly with the reorganization of BAX/BAK into two functionally different regions, namely the dimerization/core (helices a2-a5) and the piercing/latch (helices a6-a8) domains (Dewson et al, 2008;Dewson et al, 2012;Czabotar et al, 2013;Moldoveanu et al, 2013;Bleicken et al, 2014;Flores-Romero et al, 2017). Although the structural organization of BAX dimers in the membrane remains controversial (Westphal et al, 2014;Mandal et al, 2016;Cosentino & Garcia-Saez, 2017;Bleicken et al, 2018), some of the models proposed provide an explanation for how BAX/BAK may generate the membrane stress required for pore opening and alleviate the line tension for pore stabilization (Bleicken et al, 2014;Mandal et al, 2016; Fig 2B). Partial opening and insertion of the hairpin of helices a5-a6 in BAX/BAK dimers within the lipid headgroup region of the cytosolic leaflet of the MOM would initially generate positive curvature and membrane stress leading to pore opening (Czabotar et al, 2013;Bleicken et al, 2014). The crescent-like shape of these two ahelices in the context of the pore rim would then act as a scaffolding chaperone that stabilizes the high lipid curvature, decreases the line tension, and maintains the open pore state (Mandal et al, 2016;Bleicken et al, 2018; Fig 2B). The formation of high-order oligomers with such membrane disposition would potentiate the stabilizing effect (Subburaj et al, 2015).
In agreement with this, atomic force microscopy (AFM) experiments on supported bilayers, electron microscopy (EM) assays in

BAX and BAK supramolecular structures and functions beyond MOMP
Historically, the segregation of BAX and BAK into discrete puncta at mitochondria, also known as apoptotic foci, has been linked with the apoptotic phenotype (Nechushtan et al, 2001;Karbowski et al, 2002). High-resolution imaging techniques have recently allowed deciphering the riddles of these supramolecular structures, revealing distinct molecular architectures such as lines, arcs, and full rings (Grosse et al, 2016;Salvador-Gallego et al, 2016). Rings and arcs of BAX similar to those found at the MOM perforated the membrane in supported lipid bilayers indicating that these higher-order oligomerization states do not necessarily require other mitochondrial proteins ( Fig 2C). BAX seems to oligomerize by subsequent addition of dimers (Subburaj et al, 2015), but the formation and evolution of these oligomers at the supramolecular level are vaguely understood. BAX molecules might first organize into linear and arc-shaped structures, with some of them evolving to complete rings. Alternatively, lines and arcs might correspond to kinetically trapped assemblies in the process of pore formation. Several inter-dimer binding surfaces have been described for BAX-type proteins, including helices a6 and a9 (Dewson et al, 2009;Zhang et al, 2010) or the interface of helices a3/a5 (Mandal et al, 2016), but none of them appears to be indispensable for BAX/BAK oligomerization. Considering that BAX/BAK self-assembly seems to be crucial for reducing line tension and pore stabilization, one could envision that the membrane itself may play a role in the organization and dynamics of BAX/BAK supramolecular arrangements by contributing to the state of minimal energy. In this scenario, the energy cost of the membrane perturbations (perhaps including initial pore opening) induced by the insertion of BAX dimers into the membrane could be reduced by the coalescence of these membrane alterations and associated BAX/BAK molecules. Such a model would explain the higher-order assembly of BAX/BAK dimers via membrane-mediated interactions (Harroun et al, 1999;Reynwar et al, 2007;Shlomovitz & Gov, 2009;Cowan et al, 2020) and provide a mechanistic basis for the pore growth observed during apoptosis (Riley et al, 2018;. Connected with BAX foci, it has been proposed that BAX and BAK pores of hundreds of nanometers in diameter can also induce mitochondrial inner membrane (MIM) permeabilization and extrusion into the cytosol, leading to the release of mitochondrial DNA (mtDNA) and the activation of cGAS/STING inflammatory pathway (McArthur et al, 2018;Riley et al, 2018; Fig 2D). These discoveries have challenged our understanding of the role of BAX/BAK in apoptosis beyond that of inducing MOMP for caspase activation, which now expands to inflammation and cell-to-cell communication. The mechanism how BAX and BAK promote MIM poration and mtDNA release remains unknown. While some studies suggest that monomers or dimers are sufficient to form functional pores in model membranes (Kushnareva et al, 2012;Xu et al, 2013), it seems unlikely that BAX/BAK monomers or dimers would be able to induce the large membrane disruptions required for mtDNA release. One could envision a scenario in which low order oligomers of BAX/ BAK induce openings at the MOM that release cytochrome c and other proteins, while large supramolecular structures may enable other cell functions including mtDNA release and inflammatory responses. The continuous growth of BAX pores upon MOMP would thereby allow a controlled release of the mitochondrial components to regulate in time these alternative functions and with them, the inflammatory outcome of apoptosis (Cosentino & Garcia-Saez, 2018;McArthur et al, 2018;Riley et al, 2018).
However, it is important to note that additional components are present at BAX/BAK foci at mitochondria, such as Dynamin-related protein 1 (Drp1) and Mitofusins (Karbowski et al, 2002;Ugarte-Uribe & Garcia-Saez, 2017;Ugarte-Uribe et al, 2018;Hertlein et al, 2020), Optic atrophy-1(OPA1)/ Metalloendopeptidase OMA1 (OMA1) (Yamaguchi et al, 2008;Jiang et al, 2014), or Voltage-Dependent Anion Channel (VDAC) (Lauterwasser et al, 2016;Kim et al, 2019), which might contribute to this phenomenon. As MIM permeabilization appears to occur after MOMP, the driving force that makes the mitochondrial interior squeeze out through BAX/ BAK macro-pores might involve mechanical (osmotic) forces. This could possibly be due to or happen in combination with additional mitochondrial alterations like the dismantling of mitochondrial cristae. The mitochondrion-specific lipid cardiolipin (CL), which seems to regulate the action of several BCL2 proteins (Terrones ◀ Figure 2. BAX/BAK toroidal pore. (A) Schematic representations of the protein/lipid model shown as a 3D view cut through the membrane pore. Gray layers represent lipid headgroups of the bilayer, the acyl chains are shown in red and protein helices by dark cylinders (top). The corresponding normalized electron density distributions of acyl chains in lipid bilayers containing BAX a5 (bottom). Note that, unlike in a protein channel, in a toroidal pore: (i) the surface of the pore is lined by lipid headgroups, (ii) membrane monolayers are bent at the pore edge, and (iii) the two leaflets of the bilayer become continuous. Taken from (Qian et al, 2008). (B) Structural representation of membraneembedded BAX in the context of a toroidal pore based on (Bleicken et al, 2014;Bleicken et al, 2018). BAX is represented with nine cylinders corresponding to its nine helices. BH3 domain and C-terminal/tail anchoring domain are depicted in orange and green, respectively. One monomer is shown in gray (1-9) and the other is depicted in white (1ʹ-9ʹ). The relative orientation of the helices a9 remains unresolved. (C) BAX oligomers are organized into line, arc, and rings. Each panel shows the schematic representation (left) and the AFM images (right) of BAX assemblies. Both arcs and rings but not lines, reveal a circular depression (black) that spans the lipid membrane (dark orange). BAX molecules around the pore rim (yellow) protrude above the membrane plane, as confirmed by the height cross-sections shown below each image (corresponding to the white line in the AFM images). The topography of the arc structure reveals a pore only partially surrounded by BAX molecules, while lipids alone form the rest of the pore rim. Based on (Salvador-Gallego et al, 2016). (D) Model for the temporal control of content release during MOMP. Upon apoptotic stimuli, BAX and BAK permeabilize the MOM and induce the release of apoptotic factors, for example, cyt C. The consequent MIM permeabilization and the widening of BAX/BAK pores induce the release of mtDNA in the cytosol. In absence of caspase activity, this leads to the activation of the cGAS/STING signaling pathway. Based on (Cosentino & Garcia-Saez, 2018).
ª 2020 The Authors The EMBO Journal 39: e105753 | 2020 ) and is also implicated in mitochondrial functions including organelle ultrastructure (Schlame & Ren, 2009), could also play a role in MOM/MIM permeabilization. Because of its unique structural properties (e.g., two negative charges, a relatively small head group and four acyl chains), CL can form highly curved inverted hexagonal structures (Grijalba et al, 1999;Ortiz et al, 1999;Unsay et al, 2013) and laterally segregate into defined nanodomains in vitro (Kawai et al, 2004;Sorice et al, 2009). These additional elements might modulate the formation, size, shape, and kinetics of BAX/BAK assemblies in apoptotic foci. At the functional level, it is reasonable to argue that the regulation of the extent and kinetics of BAX/BAK-induced MOMP may elicit different scenarios, which could be: (i) genomic instability and cancer, associated with partial release of cytochrome c and minority MOMP , (ii) immunologically silent apoptosis, when MOMP, caspase activation and cell removal due to "eat me signals" are fast and complete (Depraetere Gasdermins are potent pore-formers at the core of pyroptosis The GSDM family is a new class of PFPs GSDMs represent a family of proteins that comprises six members in humans: GSDM A, B, C, D, and E (also known as DFNA5), and PJVK (also known as DFNB59), and ten in mice (GSDM A1-3, C1-4, D, E, and PJVK). Some of these proteins have proved to be essential for the highly inflammatory pathway of pyroptosis (Aglietti & Dueber, 2017;Galluzzi et al, 2018;Broz et al, 2020). Pyroptotic cell death is characterized by extensive cell swelling and membrane blebbing in absence of cell detachment de Vasconcelos et al, 2019), which resembles the phenotype induced by PFTs during their attack to the plasma membrane of target cells (Garcia-Saez et al, 2011;Ros et al, 2017). This, together with the essential role of GSDMs in pyroptosis, suggested that GSDMs have an intrinsic and potent pore-forming activity that mediates osmotic lysis in pyroptosis. Accordingly, the N-terminal domain (GSDM NT ) of GSDMs alone displays pore-forming activity in liposomes, which has been the basis to define GSDMs as the minimal machinery for pyroptosis execution (Aglietti et al, 2016;Ding et al, 2016;Liu et al, 2016).
Intense research during the recent years has provided insight into the structure of GSDMs (Fig 3A-D). All GSDMs (except DFNB59) display a two-domain architecture formed by an N-terminal (GSDM NT ) and a C-terminal (GSDM CT ) domain, separated by a linker region (Fig 3A; Broz et al, 2020). The crystal structure of fulllength GSDMA3 (Ding et al, 2016) and GSDMD (Kuang et al, 2017) revealed that the GSDM NT is inhibited by inter-domain interactions with juxtaposed regions of the GSDM CT . The a-helical fold of GSDM CT interacts with the helix a1 and a short b-hairpin located on the concave side of the b-sheet of GSDM NT . Additionally, the short a-helix at the end of the b-sheet of GSDM NT protrudes from the globular structure to interact with GSDM CT (Ding et al, 2016). For many GSDMs, caspase-mediated proteolytic processing induces the dissociation of the GSDM NT from its auto-inhibitory C-domain (Fig 3A; Kuang et al, 2017;Liu et al, 2018;Liu et al, 2019). Free monomers of GSDM NT translocate then to the inner leaflet of the plasma membrane and induce pyroptotic pores. Remarkably, the sequence and 3D-structure of all GSDMs differ significantly from any other known PFP Broz et al, 2020). Therefore, GSDMs have emerged as a new group of PFPs with a common function in pyroptosis.
The structure of a GSDMA3 NT pore was recently determined by cryo-EM ( Fig 3C)  . Together with the crystal structure of full-length GSDMA3 (Ding et al, 2016), these findings were the basis for a model of the transformation that this protein undergoes from the auto-inhibited form to the active membrane-embedded state. Cleaved GSDM NT associates with negatively charged lipids at the plasma membrane via a cluster of positively charged amino acids in helix a1 . Interaction with the lipid environment facilitates a conformational change that involves refolding into a new four-stranded amphiphilic b-sheet and generation of three oligomerization interfaces. The establishment of welldefined protein-protein interactions drives the stabilization of a ringshaped b-barrel pore embedded in the membrane. The amphipathic b-sheets are bundled together in a transmembrane configuration where the hydrophobic surfaces are in contact with the membrane core, while the hydrophilic ones are facing the pore lumen (Ding et al, 2016). However, the structural arrangement of the lipids in the vicinity of GSDM NT pores remains unknown.
Time-resolved AFM studies of the pores formed by human GSDMD NT in model membranes provided new information about the dynamics of GSDMD assembly and pore opening (Fig 3B; Sborgi et al, 2016;Mulvihill et al, 2018). These data support a model where, instead of oligomer pre-assembly in solution, GSDMD NT ◀ Figure 3. GSDMs pores evolve from toroidal to barrel structures.
(A) Crystal structure of GSDMA3 in its auto-inhibited form (PDB: 5B5R). The GSDMA3 NT and GSDMA3 CT domains are colored pink and blue, respectively. Inter-domain interactions between the GSDMA3 NT and the GSDMA3 CT keep the protein in an auto-inhibited state (Ding et al, 2016). (B) GSMDs involves the form arc-, slit, and ringshaped GSDMD NT oligomers as imaged using time-resolved AFM (Mulvihill et al, 2018). (C) Cryo-EM structure of the GSDMA3 membrane pore (PDB: 6CB8). Atomic model of the 27-fold symmetric GSDMA3 pore at 3.8 Å resolution . (D) Model of pore formation by GSDMs. After cleavage, monomers of GSDM NT translocate to the inner leaflet of the plasma membrane and then self-associate into arcs or slit structures that resemble toroidal pores and later evolve into ring-shape protein-lined pores with a b-barrel configuration.
ª 2020 The Authors The EMBO Journal 39: e105753 | 2020 monomers first bind to the membrane and then self-associate into arcs or slit structures that later evolve into ring-shaped pores. The formation of stable pores with a defined ring structure seems to be a robust and intrinsic property of GSDMD NT , as changes in protein concentration (Ding et al, 2016;Liu et al, 2016) and membrane lipid composition (Mulvihill et al, 2018) affect protein binding to the membrane but not pore size. Arcs and slits are also able to perforate the membrane (Mulvihill et al, 2018), which indicates the direct involvement of lipids in these intermediate pore structures. As discussed for BAX, protein-lipid pores can be induced when amphipathic regions of proteins bind to or insert into the membrane, lowering the energy to form a pore (Cosentino & Garcia-Saez, 2017). In this scenario, it is reasonable to assume that the perimeter of the pore wall would be partially covered by GSDMs molecules and partially with a semi-toroidal disposition of the membrane in the opposite side (Fig 3B and D). This stark similitude supports a conserved strategy of pore formation by large b-PFPs and makes it reasonable to extrapolate some features of the MACPF/CDC pores to pyroptotic pores. Evidence obtained with different members of the MACPF/CDC family suggests that these proteins follow a mechanism of pore formation in which the rolling insertion of oligomers results in the flow of lipids from the pore rim back to the bilayer (Sonnen et al, 2014;Gilbert, 2016;Podobnik et al, 2016). In this process, lipids return from the semi-toroidal edges in the pore to the bilayer structure, rather than being punched out into solution during oligomer insertion (Gilbert et al, 2013;Gilbert, 2016). Considering the dynamics of GSDMs pores, it is tempting to speculate that GSDMs could follow a similar mechanism of lipid clearance that involves the evolution of intermediate protein-lipid semi-toroidal structures to a fully protein-lined pore during pyroptosis (Fig 3D; Sonnen Gilbert, 2016).
The inflammatory pore of GSDMs GSDMs pores are non-selective membrane channels that mediate the secretion of intracellular components that act as activators of the immune system (Zanoni et al, 2016;Evavold et al, 2018). Chief among them, the inflammatory cytokines IL-1b and IL-18 can be released via these pores through non-conventional mechanisms in a cell death-independent manner . The passive transport of molecules through GSDMs pores also leads to osmotic imbalance, cell bursting, and final death, which allows the unspecific release of larger damage-associated molecular patterns (DAMPS) from cells (Broz et al, 2020;Ros et al, 2020). One interesting aspect that remains a matter of debate is whether GSDMs pores could impact the events that anticipate cell death by regulating the size of the molecules that are released from pyroptotic cells. To play a significant role in controlling contents efflux, GSDMs pores should precede cell bursting with a sufficiently long lag time, which may be regulated by opposing cellular processes. Indeed, the endosomal sorting complex required for transport (ESCRT) machinery has been shown to counterbalance GSDMD NT pores during pyroptosis (Ruhl et al, 2018). This and other membrane repair mechanisms may, as a result, be relevant not only for protecting against cell death but also for regulating pore number and cell death dynamics. In this context, pore formation by GSDMs would emerge as a mechanism to tightly control the kinetics and extent of release of inflammatory molecules from pyroptotic cells.

Staying open-minded: How does MLKL execute necroptosis?
MLKL is not similar to any other PFP MLKL is the only effector exclusively implicated in necroptosis and the most downstream component of the pathway known so far (Galluzzi et al, 2018;Espiritu et al, 2019;. It is a cytosolic, monomeric pseudo-kinase with no day-job in healthy cells known to date, although evidence suggests that it may be implicated in membrane trafficking (Yoon et al, 2017). During necroptosis, it becomes activated by phosphorylation by receptorinteracting protein 3 (RIP3). MLKL activation triggers its oligomerization and association with the inner leaflet of the plasma membrane, which eventually leads to membrane permeabilization and cell death (Wallach et al, 2016;Petrie et al, 2017;Petrie et al, 2018). Despite its central role in necroptosis, MLKL is a very intriguing protein and the mechanism how it induces plasma membrane permeabilization remains a topic of debate.
The 3D structure of inactive MLKL is comprised by two domains: an N-terminal four helix bundle domain (4HB) and a Cterminal pseudo-kinase domain (psK) bridged by a flexible brace region (Murphy et al, 2013;Su et al, 2014;Fig 4A). Intensive research over the last years has demonstrated that the 4HB of MLKL acts as the killer domain, whereas the psK domain and the brace region are critical to keep in check the cell death-inducing capacity of MLKL (Hildebrand et al, 2014;Su et al, 2014;Arnez et al, 2015;Tanzer et al, 2016). The 4HB domain of MLKL is arranged into a coiled-coil of amphipathic a-helices in which the hydrophobic faces are hidden in the core of the structure (Murphy et al, 2013;Su et al, 2014). This configuration has 3D homology with the HeLo-like domain found in fungal Heterokaryon incompatibility protein S (HET-S), which form pores by a mechanism involving the insertion of their N-terminal a-helix in the membrane (Daskalov et al, 2016).

of 18
The EMBO Journal 39: e105753 | 2020 ª 2020 The Authors Upon activation, MLKL is thought to undergo a conformational change that facilitates its weak interaction with PIPs at the plasma membrane through a cluster of positively charged residues located in the 4HB (Dondelinger et al, 2014). High-affinity sites are then exposed, which stabilizes stronger interactions with PIPs via a rolling-over mechanism (Quarato et al, 2016). Besides this, very little is known about the assembly pathway and oligomeric state of MLKL at plasma membrane. Different studies have proposed that MLKL can form trimers, tetramers, hexamers, and even higher-order oligomers (Cai et al, 2014;Chen et al, 2014;Huang et al, 2017;Liu et al, 2017;Petrie et al, 2018). It is reasonable to speculate that the different reported oligomers could be intermediate states of higherorder complexes.
The supramolecular structure of membrane MLKL oligomers promoting membrane permeabilization remains undefined too. Early studies proposed that MLKL activates endogenous ion channels, which would be the actual responsible for cell death (Cai et al, 2014;Chen et al, 2014). However, this model does not seem to hold true in light of more recent data showing that specific ion channels are not essential for necroptosis (Chen et al, 2014;Xia et al, 2016;Ousingsawat et al, 2017). A more general hypothesis is that MLKL directly mediates the permeabilization of the plasma membrane either as a result of its partial insertion into the lipid bilayer (Su et al, 2014) or by forming defined channels or pores (Fig 4B;Xia et al, 2016). The evidence that membrane nanopores represent a core mechanism in necroptosis supports a possible pore-forming function for MLKL (Ros et al, 2017). However, whether MLKL is directly or indirectly forming necroptotic pores, alone or together with other cellular components, is something that still remains to be elucidated.
Alternative models for pore formation by MLKL One distinctive feature of pore-forming domains is the presence of a highly hydrophobic stretch that helps partitioning into the hydrophobic core of the membrane. Amphipathic a-helices can also function as pore-forming domains since they are ideal for binding to membrane interfaces (Ros & Garcia-Saez, 2015;Grage et al, 2016). To predict potential pore-forming segments in MLKL, we built the Trimers? Tetramers? Octamers? …? P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P P psK © EMBO 10 of 18 The EMBO Journal 39: e105753 | 2020 ª 2020 The Authors hydropathy profile of the 4HB (Fig 5A). We found that none of the a-helices in the 4HB of MLKL is sufficiently hydrophobic to be predicted to insert in membranes. This is in contrast to other a-PFPs, including BAX or the HeLo-like proteins, which questions the pore-forming activity of MLKL. We also calculated the mean hydrophobic moment (µ) as a quantitative indicator of amphipathic nature of the individual helices in the 4HB (Fig 5B). The values obtained were comparable to those of single-helix pore-forming domains of other PFPs analyzed, although in MLKL, the hydrophobic patches of the four a-helices cooperate in the assembly of the helical bundle fold and may thus not necessarily behave as poreforming domains. Unfortunately, we still lack the structure and topology of MLKL in membranes. It remains to be disclosed whether MLKL activation facilitates the unfolding of the 4HB and the exposure of normally hidden hydrophobic surfaces. This could lead to formation a protein-lined transmembrane pore in which the polar surfaces of the amphipathic a-helices face the aqueous lumen and the hydrophobic sides are in contact with the membrane core. Alternatively, the amphipathic a-helices of the 4HB might remain lying at the membrane interface region without spanning the bilayer, as it has been proposed ( Fig 5C) (Su et al, 2014;Petrie et al, 2017). In such scenario, MLKL could mediate pore formation via alternative mechanisms: 1 One possibility could be that the 4HB helices of MLKL behave as surfactants and create a carpet on the membrane surface (Fig 5C), thereby destabilizing the bilayer structure and permeabilizing the membrane once a concentration threshold at the membrane is reached, like in the case of detergents or in form of amyloid fibers (Liu et al, 2017). In this regard, MLKL could follow a mechanism similar to the antimicrobial peptide LL37 or the bacterial cytotoxic PSMa3 peptide, where amphipathic a-helices are densely packed forming amyloid fibers that induce membrane destabilization (Tayeb-Fligelman et al, 2017;preprint: Engelberg & Landau, 2020). 2 MLKL may act instead as an a-PFP by inducing membrane curvature without fully spanning the hydrophobic core of the membrane (Fig 5C). A similar mechanism has been proposed for different lytic peptides and proteins such as melittin and magainin (Sengupta et al, 2008;Lee et al, 2011) and actinoporins (Alvarez et al, 2003;Mesa-Galloso et al, 2019;Fig 5C).
In this case, the asymmetric attack of the amphipathic ahelices of the unfolded 4HB would induce membrane thinning and tension, inducing membrane curvature and toroidal pore formation (Chen et al, 2003;Grage et al, 2016;Woo & Lee, 2017). Curvature induction could be relevant not only for the membrane permeabilizing activity of MLKL but also for potential alternative functions in endosome/exosome generation (Yoon et al, 2017;Fan et al, 2019). 3 Another option would be that MLKL induces membrane curvature and permeabilization due to local high concentration together with PIPs. In this case, MLKL would behave similarly to the poorly understood fibroblast growth factor (FGF2), a mediator of the unconventional secretory pathway (Steringer et al, 2012;Dimou et al, 2019) and the HIV-1 transactivator of transcription (HIV-Tat) (Zeitler et al, 2015). In all these pore models, the alteration of the bilayer structure by MLKL could facilitate the flip-flop movement of lipids via a scramblase activity, thereby providing an explanation for the exposure of phosphatidylserine (PS) to the outer leaflet of the plasma membrane in necroptotic cells (Gong et al, 2017a,b).

Is MLKL an intrinsic killer?
At a single cell level, necroptosis is a slow process in which cells round and detach with increasing osmotic pressure Ros et al, 2017). Recent studies have shown that during the early phase of necroptosis, phosphorylated MLKL remains at the plasma membrane for a long time before cell death (Tanzer et al, 2015;Fan et al, 2019). These features contrast GSDMs and other PFPs, suggesting that MLKL may not have a potent membrane permeabilizing activity.
A question that remains open is whether this weak activity is an intrinsic property of MLKL or the result of tight regulation by the cell. Different evidence indicates that maintaining MLKL levels at the plasma membrane below a threshold prevents necroptosis (Gong et al, 2017b;Hildebrand et al, 2020;Petrie et al, 2020;Samson et al, 2020). Besides RIP3, several proteins and co-factors have been identified as modulators of MLKL activity in cells. This includes the TAM kinases (Najafov et al, 2019), the heat shock protein 90 (HSP90) (Bigenzahn et al, 2016;Jacobsen et al, 2016), ATP (Petrie et al, 2018), and the highly phosphorylated versions of soluble inositol phosphates (i.e., IP4, IP5, and IP6) (Dovey et al, 2018;McNamara et al, 2019). Moreover, membrane repair mechanisms involving exosome release, endocytosis, and exocytosis play a counteracting role in necroptosis (Yoon et al, 2017;Gong et al, 2017b;Fan et al, 2019). The fact that MLKL orthologues from different species are not exchangeable among different host cells also supports the hypothesis that host-specific factors might regulate MLKL activity in cells (Tanzer et al, 2016;Petrie et al, 2018).
During the commitment phase of necroptosis, MLKL mediates a number of cellular alterations including PS exposure, intracellular (A) Hydropathy profiles of the 4HB of MLKL, the HeLo-like N domain-containing protein from Chaetomium globosum and BAX. Profiles were built with the program Protscale (https://web.expasy.org/protscale/), using the Eisenberg scale with a window of 9 amino acid residues. Segments above 0 are predicted as hydrophobic and below 0 as hydrophilic. Predicted transmembrane segments are highlighted in red. Predictions were made using two different softwares: TMHMM (http://www.cbs.dtu.d k/services/TMHMM) and TCDB (http://www.tcdb.org/progs/?tool=hydro). (B) Amphipathic nature of the a-helices of the 4HB of MLKL. Left: Wheel projections of the ahelices of the 4HB of human (top) or mouse (bottom) MLKL. Arrows point toward the hydrophobic face of the a-helices. Projections were built using the server Heliquest (https://heliquest.ipmc.cnrs.fr/). Right: l of the amphipathic a-helices of the 4HB of MLKL, the pore-forming domain of the a-PFT sticholysin II and the lytic peptides melittin, LL37 and PSMa3. (C) Amphipathic a-helices of MLKL could alter membrane integrity by two alterative models. In a carpet model, MLKL would act as a surfactant on the membrane surface, while in the toroidal pore, MLKL would induce membrane curvature and lipid-protein pores. Cylinders indicate individual amphipathic helices of the 4HB of MLKL. Hydrophobic surfaces are depicted in red and hydrophilic in blue.

◀ ª 2020 The Authors
The EMBO Journal 39: e105753 | 2020 vesicle trafficking, exosome release and production of inflammatory cytokines, and DAMPs (Yoon et al, 2017;Gong et al, 2017a,b;Douanne et al, 2019;Espiritu et al, 2019). Given the multiple processes activated downstream of MLKL translocation to the plasma membrane, an interesting possibility would be that MLKL is not an intrinsic killer or that its weak activity is functionally relevant. In this scenario, the primary role of MLKL could be to promote these distinctive intracellular modifications in order to allow necroptotic cells to communicate in a unique manner with the environment. Whether the result of MLKL activation is cell death would then depend on the extension of MLKL-dependent membrane damage and the efficiency of the cellular mechanisms that promote or counterbalance MLKL assembly at the plasma membrane.
Pores by BAX/BAK, GSDMs, and MLKL: the devil is in the detail BAX/BAK, GSDMs, and MLKL have evolved to have a common role as executors of regulated cell death via membrane permeabilization. There are similarities in the general mechanisms of action of these proteins. With the exception of BAK, they are all produced as soluble proteins that undergo extensive conformational changes upon activation and interaction with the membrane. Membrane binding is driven, at least partially, by electrostatic interactions of positively charged residues located on the protein surface with anionic lipids in the target membranes. Hydrophobic interactions between the bilayer and amphipathic and/or transmembrane segments of these proteins also contribute to membrane partition and insertion. Once at the membrane, all of them form oligomers that are able to disrupt the bilayer, allowing the passive flux of molecules. However, BAX/BAK, GSDMs, and MLKL strongly differ in the way they alter the structure and permeability of cellular membranes. This is reasonable, as distinct types of effectors may be required to permeabilize target membranes with different protein and lipid composition, as well as mechanical properties. To execute apoptosis, BAX is targeted to the MOM, while GSDMs and MLKL bind to the inner leaflet of the plasma membrane to mediate pyroptosis and necroptosis. The ability of BAX/BAK to open protein-lipid pores is very sensitive to the membrane lipid composition (Basanez et al, 2002;Terrones et al, 2004), which may be a regulatory mechanism during apoptosis. On the other hand, GSDMs and MLKL target the plasma membrane by a common mechanism of interaction with anionic PIPs (Parisi et al, 2018;Ros et al, 2020). These two proteins can also be found in organelles. MLKL binds to endosomes and regulates intracellular vesicle trafficking and receptor recycling (Yoon et al, 2017). Some studies have suggested that GSDMs can disrupt mitochondria via its ability to recognize CL (Ding et al, 2016;Huang et al, 2020). However, it remains unclear how GSDMs and MLKL discriminate between different organelles and which are the conditions that trigger their specific targeting to different cellular membranes.
Based on the secondary structure of the pores they form, BAX/ BAK, and perhaps MLKL, are classified as a-PFPs, while GSDMs are b-PFPs. These structural differences strongly determine the molecular properties of the membrane pores built by these proteins. a-helical proteins tend to form more flexible pores in which oligomeric structures are weakly associated with intra-chain hydrogen bonds interactions. In contrast, b-barrel pores present high stability by inter-strand hydrogen bonds formation between juxtaposed chains Ros & Garcia-Saez, 2015). These differences would explain why GSDMs have well-defined protein-protein interfaces that stabilize the b-barrel arrangement in the membrane and are amenable for structural studies. Instead, BAX and BAK assemblies are less rigid and more heterogeneous in terms of interaction interfaces and supramolecular structures, and have escaped so far high-resolution structural determination. It will be interesting to learn how the structure of MLKL oligomers in the membrane compares to that of BAX/BAK and GSDMs.
Along these lines, while a common feature of the mechanism of BAX/BAK and GSDMs is the formation of arc-shaped structures that only partially cover the pore rim, the intramolecular interactions that stabilize these protein-lipid assemblies are different. In the case of GSDMs, it seems that arcs are intermediates transitioning into complete rings in which lipids are excluded. However, the BAX/BAK pores are flexible and dynamic, and their size depends on protein density at the membrane and on lipid composition (Landeta et al, 2011;Bleicken et al, 2013;Landeta et al, 2015;Subburaj et al, 2015;Salvador-Gallego et al, 2016;Cosentino & Garcia-Saez, 2017). This, together with the difficulty to define interaction surfaces between BAX/BAK dimers, supports a model in which protein dimers and lipids could be intermixed with each other also in the ring-like structures. Further research is needed to confirm this distinctive property of BAX pores compared to GSDMs ones.
Why so many ways of piercing membranes to kill a cell?
Altogether, the different mechanisms of pore formation are likely to define the signaling processes activated downstream of membrane permeabilization in all forms of regulated cell death. On the one hand, the MOM is a membrane naturally permeable to molecules up to 5 kDa due to the presence of several porins (e.g., VDACs) and a singular lipid composition (Fleischer et al, 1967;Vance, 1990;Kagan et al, 2004). In this context, the growing nature of BAX pores may be key to enable temporal control of the release of mitochondrial contents ranging from small proteins like cytochrome c, to intermediate molecules like SMAC and to larger polymers like mtDNA (Kuwana et al, 2002;McArthur et al, 2018;Riley et al, 2018). The heterogeneity in BAX/BAK pores may therefore underline alternative cellular responses in apoptosis.
On the other hand, although GSDMs and MLKL target the same cellular membrane, the mechanism how they induce membrane permeabilization presents remarkable differences. In pyroptosis, GSDMD NT forms big and non-selective pores that mediate the release of proteins, like interleukins and other DAMPs, even before cell death. In contrast, MLKL induces the formation of smaller pores of limited permeability in necroptosis. These differences could set the basis for the different kinetics of death and morphological features of pyroptotic and necroptotic cells. It is tempting to speculate that in the tissue, two outcomes could be possible. On the one hand, content efflux might be transient and highly controlled with prevalence of plasma membrane repair and recovery in the absence of cell death. Alternatively, dominance of 12 of 18 The EMBO Journal 39: e105753 | 2020 ª 2020 The Authors loss of ion homeostasis and osmotic swelling would lead to plasma membrane bursting and complete exposure of the cellular components upon cell death. The possibility that these two scenarios may communicate distinct signals to the microenvironment is intriguing.

Conclusions and future perspectives
Despite decades of intense research, the structure and dynamic evolution of the apoptotic pores formed by BAX and BAK and the possible roles of other mitochondrial proteins and/or lipids remain open questions. What are the functional consequences of the structural flexibility of BAX/BAK supramolecular structures? Are there mechanistic and functional links between MOMP and MIMP? Our knowledge about necroptosis and pyroptosis is continuously expanding, but the functional impact that membrane permeabilization induced by GSDMs and MLKL is only starting to be grasped. We are still questioning whether MLKL is a PFP at all. How does the membrane repair machinery counterbalance the number and size of pores in pyroptosis and necroptosis? Is there a role for sub-lytic activation of GSDMs and MLKL? Pore formation has emerged as a common execution step in the signaling pathways leading to the different types of regulated cell death. By inducing or creating membrane pores of distinct nature, BAX/BAK, GSDMs, and MLKL control the movement of molecules across membranes and determine the cell's fate. They are also directly responsible for the type and timing of contents released and their effect in the organism. Therefore, understanding how membrane pores are regulated and assembled holds a significant potential to devise new strategies to not only control cell death, but also modulate the inflammatory and immunologic effects of these processes for therapy.