MLKL trafficking and accumulation at the plasma membrane control the kinetics and threshold for necroptosis

Mixed lineage kinase domain-like (MLKL) is the terminal protein in the pro-inflammatory necroptotic cell death program. RIPK3-mediated phosphorylation is thought to initiate MLKL oligomerization, membrane translocation and membrane disruption, although the precise choreography of events is incompletely understood. Here, we use single-cell imaging approaches to map the chronology of endogenous human MLKL activation during necroptosis. During the effector phase of necroptosis, we observe that phosphorylated MLKL assembles into higher order species on presumed cytoplasmic necrosomes. Subsequently, MLKL co-traffics with tight junction proteins to the cell periphery via Golgi-microtubule-actin-dependent mechanisms. MLKL and tight junction proteins then steadily co-accumulate at the plasma membrane as heterogeneous micron-sized hotspots. Our studies identify MLKL trafficking and plasma membrane accumulation as crucial necroptosis checkpoints. Furthermore, the accumulation of phosphorylated MLKL at intercellular junctions accelerates necroptosis between neighbouring cells, which may be relevant to inflammatory bowel disease and other necroptosis-mediated enteropathies. Mixed lineage kinase domain-like (MLKL) is the terminal protein in the pro-inflammatory necroptotic cell death program. Here the authors show that MLKL trafficking and plasma membrane accumulation are crucial necroptosis checkpoints, and that accumulation of phosphorylated MLKL at intercellular junctions promotes necroptosis.

Although our understanding of the necroptotic pathway is still in its infancy, much attention has been paid to the upstream events that govern MLKL activation and the terminal event whereby MLKL disrupts membrane integrity. By comparison, the molecular mechanisms of intervening steps, such as MLKL's translocation to the plasma membrane to trigger necroptosis, are poorly understood. To date, proteins of the ESCRT-III complex, Flotillins and Alix:Syntenin-1 have been reported to regulate MLKL localization [29][30][31] , mostly in the context of endosomal compartmentalization or exosome shedding of activated MLKL to negate cell death. Which, if any, of these pathways regulate MLKL trafficking to the plasma membrane, where membrane integrity is disrupted to induce cell death, remains unclear. To address this gap in knowledge, we devised an immunofluorescence-based technique to temporally track and quantify the activation and translocation of endogenous human MLKL during necroptosis at the single-cell level. Using this approach, we have defined the rate-limiting mechanisms that control the trafficking of MLKL from the necrosome to the plasma membrane. We identified a Golgi-microtubule-actin-dependent pathway as a crucial regulatory step during the effector phase of necroptotic signaling, which controls MLKL translocation to the plasma membrane in epithelial and hematopoietic cell lines. Interestingly, MLKL shares the same trafficking pathway as the tight junction-associated protein, Zonula Occludens-1 (ZO-1), and co-accumulates with tight junction proteins as large hotspots at the plasma membrane during epithelial cell necroptosis. Furthermore, once at the plasma membrane, we find that stabilizing the peri-junctional organization of ZO-1 potently counteracts the ability of MLKL to cause epithelial cell necroptosis. Conversely, destabilizing the peri-junctional organization of ZO-1 sensitizes epithelial cells to necroptosis. Hence, the trafficking and membranolytic threshold for MLKL is mechanistically linked to tight junction proteins during epithelial cell necroptosis. Prompted by this interplay, we show that peri-junctional damage to the plasma membrane occurs during epithelial cell necroptosis, which in turn propagates membrane damage and accelerates the death of neighboring cells.
Our findings uncover two new checkpoints in necroptosis: one is a trafficking mechanism that regulates the kinetics of necroptosis; and the other controls the amount of MLKL needed to disrupt plasma membranes, which in turn facilitates neighboring cell death in a cell-extrinsic manner. These post-necrosomal checkpoints represent additional targets for therapeutic intervention in the necroptosis pathway.

Results
Single-cell methodology to study MLKL activation. To gauge the contribution of regulatory mechanisms downstream of necrosome assembly and RIPK3 activation, we compared the kinetics of RIPK1, RIPK3 and MLKL phosphorylation and plasma membrane lysis in multiple human epithelial cell lines treated with the necroptosis-inducing stimulus TNF, Smacmimetic and IDN-6556 (hereafter referred to as TSI; Fig. 1a, b and Supplementary Fig. 1). Immunoblot analyses showed that RIPK1 was rapidly phosphorylated at S166 (pRIPK1) within 1.5 h of TSI-treatment (Fig. 1a). Subsequently, RIPK3 phosphorylation at S227 (pRIPK3) and MLKL phosphorylation at S358 (pMLKL) was evident after 3 h of TSI-treatment and increased until 6 h post-TSI treatment (Fig. 1a, b). Notably, all cell lines exhibited a temporal gap between the emergence of pMLKL and the onset of plasma membrane lysis (as measured by LDH release), which was most pronounced in the HT29 colorectal cancer cell line (Fig. 1a, b and Supplementary Fig. 1). Emergence of pMLKL was mirrored by the onset of pMLKL oligomerization and membrane association (Fig. 1c), which all preceded cell death by more than 3 h (Fig. 1d) in HT29 cells following TSI-treatment. This temporal gap between the hallmarks of MLKL activation (namely MLKL phosphorylation, oligomerization and membrane association) and subsequent cell death has been noted before 32 ; suggesting that rate-limiting mechanisms exist for necroptosis downstream of MLKL recruitment to, and activation within, the necrosome.
We chose the HT29 line, in the first instance, to investigate the post-necrosomal regulation of MLKL. Current methods for assessing cellular MLKL activation provide critical insight into the induction phase of necroptosis (i.e., 1.5-4.5 h TSI-treatment), but are less informative during the effector phase of necroptosis (e.g., ≥4.5 h TSI-treatment; Fig. 1). Thus, we developed a quantitative single-cell approach to measure the activation and translocation of endogenous human MLKL. To this end, we generated two monoclonal antibodiesthe 7G2 and 10C2 clonesthat recognize distinct epitopes within human MLKL ( Fig. 2a and Supplementary Fig. 2a, b). We used these antibodies, together with an established anti-MLKL pS358 antibody 23 , to optimize the immunofluorescent detection of MLKL. Immunofluorescence using either the 7G2 or 10C2 antibodies showed that MLKL resides in small cytoplasmic puncta in the basal state, which then coalesce into larger cytoplasmic clusters during the effector phase of necroptosis ( Fig. 2b-f). These clusters of MLKL grow around congregations of RIPK1 ( Fig. 2g and Supplementary Movie 1), indicating that they are necrosome-related. Consistent with this finding, small puncta of MLKL pS358 originated from these cytoplasmic necrosomal clusters ( Fig. 2e-g) and progressively accumulated as hotspots of MLKL pS358 at the plasma membrane ( Fig. 2b-g). The formation of MLKL clusters and hotspots appears to be a hallmark of necroptosis as these structures were observed in two epithelial cell lines treated with different necroptotic stimuli, but not during extrinsic or intrinsic apoptosis ( Supplementary Fig. 2c-f). As expected, loss of RIPK1, RIPK3 or MLKL completely abrogated formation of hotspots and clusters ( Supplementary Fig. 2d).
Some unexpected features of the clusters and hotspots were noted. Firstly, rather than associating uniformly with the plasma membrane, MLKL pS358 typically accumulates as 1-2 hotspots per cell in HT29 and HCC2998 cultures undergoing necroptosis ( Fig. 2 and Supplementary Fig. 2c-f). These hotspots were~0.5-3 μm in diameter, preferentially located to intercellular junctions, and had an irregular ring-like architecture containing heterogeneous substructure (Fig. 2e, f and Supplementary Movie 2). This architecture is distinct from other membrane-associated executioner complexes, such as apoptotic pore complexes, the Gasdermin family and membrane attack complex/perforin family [33][34][35][36] , and does not support the idea that MLKL forms regularly-structured membrane pores. Secondly, only a modest degree of overlap was observed between the MLKL 10C2 and MLKL pS358 immunosignals ( Supplementary Fig. 2g), which suggests that these immunosignals reflect different chronological states of MLKL activation (Fig. 2b). To investigate this possibility, we employed two inhibitors that act downstream of MLKL phosphorylation: the synthetic protein ligand, Monobody 37 (Mb37) 37 , and the compound, necrosulfonamide (NSA) 6 . Mb37 binds the killer four-helix bundle domain (Fig. 2a) of human MLKL to block membrane translocation and cell death, but does not inhibit MLKL phosphorylation or oligomerization 37 . Like Mb37, NSA does not block MLKL phosphorylation, but is instead thought to prevent MLKL from associating with membranes 23,38 . Indeed, phosphorylated MLKL could be readily detected by immunoblot in HT29 cells treated with necroptotic stimuli in the presence of Mb37 (as shown in ref. 37 HT29 cells were treated with media alone or the  necroptotic stimuli, TNF and Smac-mimetic and pan-caspase inhibitor, IDN-6556, (TSI) for the indicated time period and chronological RIPK1, RIPK3, and MLKL phosphorylation (a), plasma membrane lysis (b), MLKL oligomerization and membrane-association (c), and plasma membrane damage with cell death (d) were measured. a Whole-cell lysates were fractioned by SDS-PAGE and immunoblotted to determine the extent of RIPK1, RIPK3 and MLKL phosphorylation over time. The displayed immunoblots are representative of n = 2 independent experiments for RIPK1/3 and n = 4 independent experiments for MLKL. b Left-hand y-axis plots the ratio of the pMLKL and MLKL immunoblot signals (relative to the t = 6 h time point value) across n = 4 independent experiments. Mean ± SD. Right-hand y-axis plots the extracellular release of lactate dehydrogenase (LDH) as an index of plasma membrane lysis (relative to detergent-treated cells) across n = 9 independent experiments. Mean ± SEM. All data are included in the Source Data file. c Subcellular fractions of the cytosol (C) and membrane (M) were subjected to Blue Native-PAGE and immunoblotted to assess MLKL phosphorylation, oligomerization and membrane-association. Fractionation was verified by probing for the membrane protein, voltage-dependent anion channel 1 (VDAC1), and the cytosolic protein, glyceraldehyde-3-phosphosphate dehydrogenase (GAPDH). Data are representative of n = 2 independent experiments. d Sequential time-lapse micrographs of Annexin V binding (Annexin V; green) and propidium iodide-uptake (PI; red) as indices of plasma membrane damage and cell death, respectively. Micrographs are representative of n = 4 independent experiments.  Fig. 1). Despite this, neither clusters nor hotspots of MLKL were observed in the presence of Mb37 (Fig. 3a, b) or NSA ( Fig. 3c-g) confirming that these MLKL-containing structures are formed downstream of MLKL phosphorylation, oligomerisation and membrane-association (Fig. 2b).
We propose that membrane-associated MLKL 10C2 clusters accumulate around RIPK1-containing necrosomes in the cytosol during the effector phase of necroptosis (Fig. 2b, g). These clusters resemble the previously described RIPK3-containing puncta that form during necroptosis 6     vesicle-associated MLKL pS358 then emerges from these clusters and traffics to the cell periphery where it accumulates as hotspots at the plasma membrane (Fig. 2b). The increases in the occurrence/size of the MLKL 10C2 clusters and the occurrence/ size/intensity of the MLKL pS358 hotspots mirrors plasma membrane lysis (compare Fig. 2d and Supplementary Fig. 2h with Fig. 1b). In line with this conclusion, adding NSA during the effector phase of necroptosis (i.e., ≥5 h of TSI-treatment in HT29 cells when oligomeric pMLKL is already membrane-associated on vesicles) not only profoundly reduced cell death (Fig. 3d), but also markedly reduced the size, intensity and occurrence of MLKL pS358 hotspots on the plasma membrane ( Fig. 3f, g). Hence, our immunofluorescence protocol provides quantitative insights into intracellular trafficking events that occur downstream of the phosphorylation, oligomerization and membranetranslocation of MLKL.
MLKL hotspots accelerate necroptosis in neighboring cells. To assess whether focal accumulation of MLKL pS358 at hotspots affects plasma membrane integrity, we treated HT29 cells with necroptotic stimuli and visualized the resulting plasma membrane damage reflected by Annexin V binding to externalized phosphatidylserine, and cell death via TOPRO3-uptake using lattice light sheet microscopy. Akin to the accumulation of MLKL pS358 at the plasma membrane ( Notably, Annexin V + sites appeared within the immediate vicinity of MLKL pS358 hotspots on the plasma membrane during necroptosis ( Fig. 4d and Supplementary Movie 4), raising the prospect that MLKL pS358 hotspots correlate with sites of membrane damage during epithelial cell necroptosis. Additionally, Annexin V + membrane protrusions (previously described as bubbles 29 ), were observed 1.5-4.5 h after the induction of necroptosis ( Fig. 4a, b). These protrusions decreased in number as cells swelled near to time-of-death; they were heterogenous in size, shape, Annexin V binding intensity, and whether they were shed or retained by the cell (Fig. 4a-c). In addition, the protrusions did not necessarily appear at the same location as the Annexin V + hotspots ( Fig. 4a-c). Interestingly, HT29 cells died 1-3 h after the appearance of Annexin V + hotspots, with a cataclysmic plasma membrane disruption event at a single site, which we term "membrane blowout", being the most common feature coinciding with death ( Fig. 4a-c). Collectively, these observations are consistent with our hypothesis that pMLKL accumulates and damages the plasma membrane at hotspots in epithelial cells, which eventually triggers local membrane blowout and necroptotic death. Given that MLKL-mediated membrane damage preferentially occurs at intercellular junctions, we next considered whether necroptosis could be communicated to neighboring cells and influence their susceptibility to necroptotic death. To address this, we used widefield time-lapse imaging to track plasma membrane damage and cell death events across a large population of HT29 cells undergoing TNF-induced necroptosis. These experiments showed that membrane damage, as indicated by phosphatidylserine exposure, could propagate from one cell to its abutting neighbor, like a chain reaction (Fig. 5a). Despite the onset of death after propagation of membrane damage to a neighboring cell being highly variable, large-scale analysis showed that adjacent HT29 cells died via necroptosis~25% faster than their non-adjacent counterparts (Fig. 5a-c). Importantly, no promotion of death occurred between neighboring HT29 cells undergoing TNF-induced apoptosis, which is an MLKLindependent form of cell death (Fig. 5b, c). To ascertain whether necroptosis of one cell was sufficient to damage neighboring cells, we treated co-cultures of wild-type and MLKL −/− HT29 cells with necroptotic stimuli and used high spatio-temporal resolution lattice light sheet microscopy to visualize membrane damage over time. Unlike the propagation of membrane damage between neighboring wild-type HT29 cells, there was no evidence for the propagation of Annexin V binding from a wild-type necroptotic cell to a neighboring MLKL −/− cell (Fig. 5d). Moreover, in these co-culture experiments, no MLKL −/− HT29 cells (N = 7 cells from n = 3 independent experiments) died in response to necroptotic stimuli. Together, these data suggest that junctional accumulation of MLKL accelerates death of adjacent epithelial cells, but only when both neighboring cells are primed to undergo necroptosis. We propose that this is a junction-related ancillary mechanism for accelerating necroptosis between neighboring cells.
MLKL is actively trafficked to the plasma membrane. Having ascertained that hotspots are a primary site of membrane damage during necroptosis, we postulated that the temporal gap between MLKL phosphorylation and cell death reflects the time needed for MLKL to traffic from the necrosome to these hotspots. To test this idea, we examined the role of various trafficking mechanisms across the induction and effector phases of necroptosis using a range of inhibitors (Fig. 6a). We were primarily interested in inhibitors that interfered with steps downstream of MLKL phosphorylation, and were thus capable of blocking cell death when added during either the induction or effector phases of necroptosis. Interestingly, we observed that inhibitors that destabilize trafficking along the Golgi, microtubule, and actin networks attenuated plasma membrane lysis when added during either the induction or effector phases of necroptosis ( Fig. 6b and Supplementary Fig. 3a). Inhibiting these trafficking pathways with a cocktail of Nocodazole, Cytochalasin B and Brefeldin A (NCB) was a particularly potent approach to suppress plasma membrane lysis at any timepoint during necroptosis (Fig. 6a, b and Supplementary Fig. 3a). Importantly, addition of either NSA or NCB five hours after TSI-treatment, when oligomeric MLKL pS358 had already begun to accumulate at the plasma membrane, inhibited subsequent necroptotic death of three human epithelial cell lines: HT29, HCC2998 and SW620 (Fig. 6c). NCB also attenuated TNF-induced necroptosis in the mouse YAMC epithelial cell line ( Supplementary Fig. 3b).
It is salient that addition of NCB during the effector phase of necroptosis immediately stalled subsequent cell death (Fig. 6c). This suggests that NCB may interfere with pMLKL trafficking and thereby prevent the threshold for plasma membrane lysis from being reached. By comparison, NCB did not reduce the onset of plasma membrane damage or cell death during the MLKL-independent processes of extrinsic-or intrinsic-apoptosis ( Supplementary Fig. 3c). Moreover, NCB did not alter MLKL phosphorylation (as measured by immunoblot analyses; Fig. 6d) or prevent MLKL oligomerization and membrane translocation (as assessed by Blue-Native PAGE; Supplementary Fig. 3d) during necroptosis in HT29 cells. Collectively, these observations are consistent with the notion that NCB interrupts the trafficking of MLKL from the necrosome to the plasma membrane, rather than influencing MLKL activation at the necrosome or by affecting mechanisms that generically influence plasma membrane integrity and cell death.
We next examined the physical proximity of MLKL to components of NCB-sensitive trafficking pathways in HT29 cells undergoing necroptosis. We observed no substantive overlap between MLKL and the Golgi-marker Vti1a (Fig. 7a). In contrast, a substantial proportion of MLKL 10C2 clusters and MLKL pS358 vesicles were positioned along tubulin and actin filaments running from the nucleus to MLKL p358 hotspots at the plasma membrane (Fig. 7a). Moreover, these MLKL p358 hotspots co-stained for the presence of cortical actin (Fig. 7a). These data are consistent with MLKL translocating in a polarized manner from cytoplasmic necrosomes along actin and tubulin filaments to the plasma membrane.
Notably, the described trafficking pathway for MLKL is not epithelial cell-specific, as NCB also slowed the rate of TNFinduced necroptosis in the Molt-4 (human acute T-lymphoblastic leukemia) and U937 (human monocytic lymphoma) cell lines ( Supplementary Fig. 3e). NCB, however, did not markedly alter necroptotic cell death in the Colo-205 (human colorectal adenocarcinoma) or RPMI-8226 (human multiple myeloma line) cell lines (Supplementary Fig. 3e). Thus, a NCB-sensitive trafficking mechanism for MLKL is employed by many, but not all, cell types during necroptosis.
Unlike NSA, NCB does not prevent MLKL from associating with membranes, as MLKL clusters and hotspots were still readily detectable by immunofluorescence in HT29 cells that had been co-treated with TSI and NCB (Fig. 7b). Quantitation showed that the addition of NCB during necroptosis decreased the size of MLKL clusters and hotspots ( Fig. 7c and Supplementary Fig. 3f). As the formation of clusters and hotspots occurs downstream of MLKL phosphorylation, these findings suggest that NCB slows both the emergence of membrane-associated MLKL 10C2 and MLKL pS358 from the necrosome and its subsequent movement towards plasma membrane hotspots. NCB also depolarized the trafficking of MLKL and randomized the distribution of MLKL pS358 vesicles within the cell and at the plasma membrane (Fig. 7b). Together, our data illustrate that MLKL traffics from the necrosome to plasma membrane hotspots via Golgi-, actin-and microtubule-dependent mechanisms. This post-necrosomal trafficking event is a formerly-unrecognized rate-limiting checkpoint in necroptosis.  Tight junction proteins and MLKL are co-trafficked. To deduce the nature of the membrane-associated MLKL pS358 puncta and hotspots that form during necroptosis, we co-stained HT29 cells undergoing necroptosis for MLKL pS358 and various junction, organelle or microdomain markers. Negligible colocalization was observed between MLKL pS358 and markers of desmosomes (Desmoglein-2), adherens junctions (E-cadherin), endosomes and caveolae (Flotillin-1), mitochondria (TOMM20), lysosomes (LAMP2), proteasome (20S) or the chaperone HSP90 39 (Supplementary Fig. 4a). However, colocalization was evident between MLKL pS358 and three tight junction markers: JAM-A, occludin and ZO-1 ( Fig. 8a and Supplementary Fig. 4b). Colocalization between MLKL pS358 and tight junction markers was also observed in HCC2998 cells undergoing necroptosis ( Supplementary Fig. 4c). Strikingly, ZO-1 colocalized with MLKL pS358 both at hotspots and intracellular puncta, suggesting that MLKL is co-trafficked with tight junction proteins to the plasma membrane during necroptosis in HT29 cells (Fig. 8a). Furthermore, this colocalization was quantitative in nature, with pixel-level proportionality observed between the MLKL pS358 and ZO-1 immunosignals (Fig. 8b, c).   Fig. 6 Intracellular trafficking mechanisms govern necroptotic death kinetics. a, b, d The indicated necroptosis-, trafficking-or organellar-inhibitors were added to HT29 cells with TNF, Smac mimetic and IDN-6556 (TSI) or 5-6 h after TSI-treatment. Plasma membrane lysis (b) and MLKL phosphorylation (d) were examined 7.5 h post-TSI treatment (as depicted schematically in a). b Chart summarizes the impact of inhibitors on lactate dehydrogenase (LDH) release (relative to TSI-treated cells). Color coding indicates mean values across n = 3-4 independent experiments; '0% inhibition' indicates no LDH release above untreated cells and '100% exacerbation' indicates a doubling of LDH release above TSI-treated cells. Chlorpromazine was toxic upon prolonged exposure (gray). *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001 and p-values between 0.05 and 0.1 are shown as determined by one-way ANOVA with Dunnett's correction for multiple comparisons. Independent datapoints underlying (b) are shown in Supplementary Fig. 3a. c HT29, HCC2998 and SW620 cells were left untreated (control) or TSI-treated. Necrosulfonamide (NSA) or Nocodazole, Cytochalasin B, Brefeldin A (NCB) was added after 5 h of TSI-treatment. Propidium iodide (PI) + (dead) cells per mm 2 over time were measured using an IncuCyte S3 System (mean of N = 3 fields with >600 cells per group from one experiment with the underlying biological replicates shown as dots; data representative of n = 2 independent experiments). d Immunoblots of whole-cell lysates indicate that NCB does not affect TSI-induced MLKL phosphorylation (representative of n = 2 independent experiments). NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16887-1 ARTICLE NATURE COMMUNICATIONS | (2020) 11:3151 | https://doi.org/10.1038/s41467-020-16887-1 | www.nature.com/naturecommunications impermeable peptides that indirectly modulate tight junctions. AT-1001 preserves the peri-junctional organization of ZO-1 and protects barrier integrity during epithelial challenge 41 . Conversely, AT-1002 is thought to disrupt the peri-junctional organization of ZO-1 and induce epithelial barrier dysfunction 41 . To address whether modulating tight junctions influences necroptotic signaling, we induced necroptosis in HT29 cells in the presence of AT-1001 or AT-1002. AT-1001 attenuated plasma membrane lysis by~75%, presumably owing to membrane rigidification at hotspots thereby raising the threshold of pMLKL required for membrane perturbation. Conversely, AT-1002 increased plasma membrane lysis by~90% in HT29 cells treated with a necroptotic stimulus (Fig. 8d), presumably due to AT-1002 compromising membrane stability at hotspots and thereby lowering the threshold for pMLKL-mediated disruption. AT-1001 also reduced the extent of necroptotic death in three other epithelial cell lines: human HCC2998 cells, human Colo-205 cells and mouse YAMC cells ( Supplementary Fig. 4d-f). Importantly, immunoblot analyses show that AT-1001 and AT-1002 did not alter TSI-induced MLKL phosphorylation, suggesting that these peptides act upon a post-necrosomal checkpoint (Fig. 8e). Furthermore, no effect was observed on accumulation of MLKL pS358 into plasma membrane hotspots by microscopy in HT29 cells (Fig. 8f). In line with their reported mechanisms of action, AT-1001 increased peri-junctional ZO-1 whereas AT-1002 decreases ZO-1 organization (Supplementary Fig. 4g). Unexpectedly, AT-1001 also protected the human leukemia line, Molt-4, from necroptosis ( Supplementary Fig. 4f); this was surprising because Molt-4 cells do not express detectable levels of ZO-1 and thus are unlikely to form intercellular tight junctions (Supplementary Fig. 5a)  direct actions upon tight junctions 42 . These observations support the idea that co-regulated trafficking of MLKL and tight junction components underlies the coalescence of MLKL pS358 and ZO-1 in epithelial cell plasma membranes. Reciprocally, our data also indicate that MLKL regulates tight junction formation during necroptosis in epithelial cells. Here, necroptotic stimulation of HT29 cells increased the accumulation of ZO-1 in plasma membrane hotspots in a manner that relied on the expression of MLKL and could be blocked with the MLKL inhibitor, NSA (Fig. 8g, h). Thus, bona fide co-regulation exists between MLKL and tight junction proteins during epithelial cell necroptosis. The capacity of tight junction protein trafficking to influence the membranolytic threshold for MLKL represents a new layer of regulation for the necroptotic pathway. Monocytic U937 cells serve as a negative control for the junctional checkpoints because, compared to epithelial cells, they lack ZO-1 expression ( Supplementary Fig. 5a), are refractory to the anti-necroptotic effect of AT-1001 (Supplementary Figs. 4f and 5b, c), and thus exhibit a relatively short temporal gap between the onset of MLKL phosphorylation and death (compare Fig. 1a, b with Supplementary Fig. 5b, c). Furthermore, membrane damage in U937 cells does not manifest as hotspots with ensuing membrane blowout. In stark contrast, necroptotic membrane damage in U937 cells increased in a relatively uniform manner across the plasma membrane, with ensuing death coinciding with the formation of small Annexin V + apertures ( Supplementary Fig. 5d, e). Additionally, no facilitation of TNFinduced necroptosis was observed between neighboring U937 cells ( Supplementary Fig. 5f). These vastly different features of TNF-induced necroptosis in U937 cells suggest that this cell line is not subjected to the same trafficking and junction-related checkpoints that influence epithelial cell necroptosis.

Discussion
Here, at the sub-cellular level, we mapped the choreography of activation, trafficking and plasma membrane damage mediated by endogenous human MLKL to execute necroptotic cell death. Based on these data, we propose the following model of epithelial necroptosis ( Fig. 9): (1) Under basal conditions, MLKL resides in small (<100 nm) puncta that are distributed evenly throughout the cytoplasm. (2) Between 1.5 and 3 h post-TSI, MLKL phosphorylation, oligomerization and membrane translocation increased as detected by immunoblot analysis. Notably, neither clusters nor hotspots of MLKL are detectable by immunofluorescence at this stage. (3) Between 3 and 4.5 h post-TSI, gelbased analyses show that maximum steady-state levels of MLKL phosphorylation, oligomerization and membrane-association are achieved. MLKL coalesces into large cytoplasmic clusters around congregations of RIPK1, which we propose to be necrosomes. The peri-nuclear location of these clusters, and the observation that Golgi destabilisers attenuated necroptosis, leads us to speculate that clusters form transiently on the surface of the Golgi apparatus. Following this, Golgi-derived vesicles of MLKL pS358 emerge from these necrosomal clusters, traffic to the plasma membrane along actin and microtubule filaments, and accumulate with tight junction proteins at nascent hotspots. Concomitantly, damage to the plasma membrane emerges at hotspots during this time. (4) Between 4.5 and 7.5 h post-TSI, clustering at necrosomes and vesicular trafficking to the plasma membrane continues, resulting in a marked increase in the number, size and density of MLKL pS358 hotspots. Coordinate plasma membrane damage accumulates at MLKL pS358 hotspots until the membranolytic threshold is surpassed, triggering membrane blowout and cell death. As necroptotic hotspots are sites of intercellular junctions in epithelial cells, the accumulation of membrane damage at hotspots has deleterious consequences for adjacent epithelial cells whereby necroptosis of one cell accelerates the death of its abutting neighbors.
Somewhat surprisingly, inhibitors of autophagolysosomal function, nuclear export, clathrin-mediated endocytosis or Cdc42 all altered the extent of cell death when added simultaneously with TSI. Because these inhibitors had negligible effects on the  epifluorescence (b, c) microscopy. a Micrograph is a maximum intensity projection and representative of n = 5 independent experiments. Diagonal closed arrowhead indicates an intracellular MLKL pS358 vesicle that co-stains for ZO-1. Diagonal arrow indicates a MLKL 10C2 cluster that does not co-stain for ZO-1 or MLKL pS358 . b 2D histogram depicts the frequency of MLKL pS358 and ZO-1 immunosignal pixel co-intensities across N = 514 cells from n = 2 independent experiments. c Pearson's correlation of MLKL pS358 and ZO-1 immunosignals. Two negative controls are shown (MLKL pS358 versus MLKL 10C2 or randomly redistributed MLKL pS358 immunosignals). Graph shows the mean (with underlying datapoints as dots) from n = 6 independent fields and N = 514 cells.   Fig. 9 Model of TNF-induced epithelial cell necroptosis chronology. Shown is a graphic summarizing the subcellular relocation and accumulation of RIPK1, MLKL 10C2 or 7G2 , MLKL pS358 and ZO-1 that occurs during necroptosis in HT29 cells. The onset and increases in phosphatidylserine (PS) exposure and the eventual onset of plasma membrane blowout are also indicated. Based on the data from this study, a hypothetical explanation for these chronological events during necroptosis is provided.
effector phase of necroptosis, it is likely they act upstream or directly upon the necrosome. Consistent with this notion, others have shown that inhibiting autophagolysosomal function enhances RIPK3-mediated MLKL phosphorylation 43 . It has also been shown that inhibiting nuclear export suppresses necroptosis 44 ; however, in this instance, nucleocytoplasmic shuttling downstream of MLKL phosphorylation was proposed, which seems incongruous with our finding that Leptomycin B does not influence the effector stage of necroptosis. More broadly, our observations highlight that multiple forms of trafficking regulate the induction phase of necroptosis [45][46][47][48][49][50] . By comparison, only the MLKL inhibitor, NSA, or destabilisers of the interrelated microtubule-, actin-and Golgi-trafficking pathways significantly inhibited the effector phase of necroptosis. Because vesicular trafficking of tight junction proteins relies upon these same three pathways to reach the plasma membrane 51 , as a unifying explanation, we propose that MLKL and tight junction proteins converge at the necrosome and thereafter employ the same trafficking mechanism to translocate to the plasma membrane. Several other forms of regulation exist for necroptosis downstream of RIPK3-mediated MLKL phosphorylation 50 . For example, inositol phosphate-6 was proposed to promote necroptosis by mediating pMLKL oligomerization and membrane-association 52,53 ; TAM kinases were recently proposed to mediate pMLKL oligomerization 54 ; HSP70 and HSP90 likely contribute to pMLKL oligomerization 39,55 ; and a trafficking chaperone was proposed to mediate pMLKL membrane translocation via a binding site upon the α4 helix of the human MLKL killer 4HB domain 37 . As the trafficking and junction-related checkpoints described herein influence the effector phase of necroptosis, when pMLKL is already in an oligomeric membraneassociated state, they would likely operate downstream of TAM kinases and inositol phosphate-6 in the necroptotic signaling cascade, potentially in concert with other chaperones.
Internalization and externalization events also regulate necroptosis downstream of the necrosome. For example, Flotillinmediated endocytosis was proposed to suppress necroptosis by removing MLKL from the plasma membrane and redirecting it for lysosomal degradation 31 . A key role for this degradative pathway, however, was not apparent in our study as Flotillin-1 was largely absent from the plasma membrane during the effector phase of necroptosis, MLKL pS358 did not colocalize to a significant extent with LAMP-2-containing lysosomes, and inhibiting lysosomal degradation during the effector phase of necroptosis did not significantly alter the extent of necroptotic cell death. Exocytosis was also proposed to counteract the effector phase of necroptosis via ESCRT-, ALIX:syntenin-1-or Rab27a/bmediated expulsion of MLKL-containing bubbles to diminish the MLKL residing at the plasma membrane [29][30][31]53,56 . Here, we observed that Annexin V + membrane bubbles first appeared during the induction phase of necroptosis, then decreased just prior to cell death, consistent with the notion that this phenomenon may allay necroptosis 29 . Nevertheless, our data suggest that membrane bubbling during necroptosis is an incredibly dynamic and heterogenous phenomenon with protrusions variously extending, retracting and shedding in a fashion seemingly independent of the primary sites of MLKL accumulation and membrane damage.
This study provides direct evidence of a mechanism that licenses necroptotic cell death by trafficking MLKL from the necrosome to the plasma membrane. Our data support the idea that a membranolytic threshold must be overcome for MLKL to execute necroptosis. This threshold is met by the accumulation of MLKL into irregularly-shaped hotspots, which is in contrast to highly-ordered protein assemblies, such as Bak/Bax or Gasdermins, which execute other forms of cell death 36 . Remarkably, the post-necrosomal trafficking steps that control the membranolytic threshold also influence the translocation of tight junction proteins during epithelial cell necroptosis. The MLKL hotspots described herein more closely resemble primordial junctions rather than mature epithelial tight junctions 40 . Despite this, epithelial necroptosis has been previously associated with tissue barrier loss; although the basis for this association was not explored [57][58][59][60] . Physiologically, our data raise the prospect that necroptotic signaling might be propagated along epithelial tissues, such as the colon, which is consistent with the implication of necroptosis in inflammatory bowel disease 57,58,61 . Moreover, as AT-1001 ameliorates multiple preclinical and clinical examples of gastro-inflammation 41,62,63 and is undergoing Phase III clinical trials for celiac disease, a role for necroptosis in these indications should be investigated. Because MLKL and tight junction proteins share common trafficking pathways, we propose that the two post-necrosomal checkpoints described herein, (1) trafficking of MLKL via the Golgi/microtubule/actin networks and (2) accumulation of MLKL at plasma membrane hotspots, represent candidates for therapeutic intervention in necroptotic diseases.
10C2 and 7G2 antibody production. Antibodies were generated at the Walter and Eliza Hall Institute Monoclonal Antibody Facility by immunizing Wistar rats with recombinant human MLKL (residues 190-471; expressed and purified from Sf21 insect cells using the baculovirus expression system 24 ), before splenocytes were fused with SP2/O mouse myeloid cells and arising hybridoma lines cloned. Specificity of these antibodies for human MLKL was validated via immunoblot analyses as exemplified in Supplementary Fig. 2a, b, and immunofluorescence analyses as exemplified in Supplementary Fig. 2c 24,65 and doxycycline-inducible Mb37-expressing HT29 cells 37 have been previously reported. To generate the HT29 cells expressing the short MLKL2 splice isoform (as described in ref. 66 ), wild-type and MLKL −/− HT29 cells were stably transduced with the pF-TRE3G vector (kindly supplied by Dr Toru Okamoto) into which the MLKL2 open reading frame had been introduced to allow doxycyclineinducible expression as described in ref. 22 . To generate membrane-tagged HT29 and U937 cells, the sequence encoding the mTagRFP-Membrane-1 fusion protein was amplified from Addgene plasmid #57992 using primers: 5'-CAGAATTCATG GCCACCATGCTGTGCTGTATG and 5'-CAGCTAGCTCAATTAAGTTTGTG CCCCAGTTTG. The amplified sequence was subcloned into the pF-TRE3G vector, then transfected into HEK293T cell together with the packaging constructs pCMV∂8.2 and pVSV-G to yield lentivirus. Wild-type HT29 and U937 cells were transduced with this harvested lentivirus and stable expression determined by puromycin selection and flow cytometry.
LDH release. Colorimetric LDH release assay kit (Promega #G1780) was performed according to manufacturer's instructions.
Protein production and purification. Full-length human MLKL and mutants were expressed in Sf21 insect cells using the Bac-to-Bac system (ThermoFisher Scientific) and purified using established procedures 24 . Briefly, proteins were expressed with N-terminal GST tags and captured from lysates using glutathione resin (UBP Bio). Proteins were cleaved on-resin from the GST tag using His-tagged TEV protease, before protease was removed by Ni 2+ chromatography (HisTag resin, Roche) and protein eluted from a Superdex-200 (GE Healthcare) Size exclusion chromatography column in 200 mM NaCl, 20 mM HEPES pH 7.5, 5% v/v glycerol. Protein containing fractions were spin concentrated to 5-10 mg/mL, aliquoted, snap frozen in liquid N 2 and stored at -80°C until required.
IncuCyte cell death assay. Cells were seeded into 48-well plates and treated (as in Cell treatment) in phenol red-free media with 1% FCS, G/P/S, 1 mM Na pyruvate (ThermoFisher Scientific #11360070), 1:500 dilution of AlexaFluor488-conjugated Annexin V (ThermoFisher Scientific #A13201) and 0.5 μg/mL propidium iodide (Sigma #P4170). Cells were moved into an IncuCyte S3 System (Essen Bioscience) and imaged using the 10x objective over time using the default bright-field, green and red channel settings. The number of propidium iodide-positive cells per mm 2 over time was quantified using IncuCyte S3 v2018A software (Essen Bioscience).
Immunofluorescence. Cells in 8-well μ-Slides (Ibidi #80826) were ice-chilled for 3 min, then washed in ice-cold Dulbecco's PBS (dPBS; ThermoFisher Scientific # 14190144), then fixed for 15 min in ice-cold methanol. Cells were washed twice in ice-cold dPBS, then blocked in ice-cold Tris-balanced salt solution with 0.05% v/v Triton-X100 (TBS-T) supplemented with 10% v/v donkey serum (Sigma #D9663) for >1 h. Cells were incubated in primary antibodies (2-4μg/mL anti-MLKL 10C2 or MLKL 7G2 antibodies; 1:200 to 1:250 dilution of MLKL pS358 , ZO-1, occludin or flotillin-1 antibodies; and 1:100 dilution of all other antibodies) overnight at 4°C in TBS-T with 10% v/v donkey serum. Cells were washed twice in TBS-T then incubated in the appropriate secondary antibodies supplemented with 0.1-0.5 μg/ mL Hoechst 33342 (ThermoFisher Scientific #H3570) for 3 h at room temperature with gentle rocking. Cells were washed four times in ice-cold TBS-T then stored at 4°C until being imaged. Where indicated, to demarcate the plasma membrane, 2μL of biotinylated wheat germ agglutinin (Sigma #L5142) was added to each well 10 min before fixation, and fixed wheat germ agglutinin was then detected via the addition of 1-2 μg/mL DyLight650-conjugated streptavidin (ThermoFisher Scientific #84547) during the secondary antibody incubation step. Importantly, as~95% of propidium iodide-positive cells are lost during this immunofluorescence protocol, it provides insight into events prior to necroptotic cell death.
Structured illumination or deconvolved wide field microscopy. Fixed immunostained cells in 8-well μ-Slides (Ibidi #80827) were subjected to super-resolution 3-dimensional structured illumination microscopy on a DeltaVision OMX-SR system (GE Healthcare) equipped with a 60x/1.42 N.A. PlanApo oil immersion objective (Olympus), sCMOS cameras, and 405-, 488-, 568-and 640-nm lasers, and 1.520 refractive index immersion oil. Image stacks were acquired consisting of 15 raw images per plane (5 phases, 3 angles) per color channel and a z-step size of 125 nm. Raw images for widefield deconvolution were acquired at one image per plane per color channel and a z-step size of 375 nm. Super-resolution reconstruction or widefield deconvolution and color channel alignment were performed with softWoRx v7.0 (GE Healthcare).
Lattice Light Sheet Microscopy (LLSM). Cells were seeded onto glass coverslips in phenol red-free media with 1% FCS, G/P/S, 1 mM Na pyruvate (ThermoFisher Scientific #11360070), 1-in-500 dilution of AlexaFluor488-conjugated Annexin V (ThermoFisher Scientific #A A23204) and 0.1μΜ TO-PRO-3 iodide (ThermoFisher Scientific #T3605). Cells were moved onto a custom-built LLSM system based on prior design 71 , treated with TSI (as in Cell treatment) and incubated under humidified 10% CO 2 at 37°C. Samples were illuminated using 488-, 560-, and 641nm diode lasers (MPB Communications) through an excitation objective (Special Optics, 0.65 NA). The LLSM was illuminated at the back aperture of the excitation objective through an annular mask of 0.325 inner NA and 0.4 outer NA providing a light sheet length of 20 microns. Fluorescence was collected by a detection objective (Nikon, CFI Apo 25X, 1.1 NA) and by a sCMOS cameras (Hamamatsu Orca Flash 4.0 v2). Volumes of 100 × 72 × 20 microns were acquired at a frame speed of 20 milliseconds giving a volume time of 15 sec with a time interval of 2 min between volumes. Acquired data were de-skewed then deconvolved using an iterative Richardson-Lucy algorithm. Data was compressed to h5 format using the BigDataViewer plugin 72 in ImageJ 1.52p 70 for visualization and analysis in Imaris v8-9 (Bitplane).
Colocalization quantitation. Wild-type HT29 cells were TSI-treated for 7.5 h, fixed and immunostained with the MLKL 10C2 , ZO-1 and MLKL pS358 antibodies (as in Immunofluorescence). Six randomly selected micrographs from n = 2 independent experiments were used to gauge the degree of signal co-proportionality between MLKL 10C2 , ZO1 and MLKL pS358 immunosignals at the single-pixel level. This approach is based upon the recommendations of 73 and carries a x,y-resolution limit of~350 nm. In brief, a rolling ball filter of 10 was applied to all channels and thresholded areas from all channels were processed using the 'Process>Image calculator>add' tool to create a universal mask. This mask removes low and nonspecific signals from the analysis. This mask was then processed using the Coloc 2 plugin in ImageJ 1.52p 70 . The 2-dimensional histograms generated by the Coloc 2 plugin plotting the 'MLKL 10C2 versus MLKL pS358 ' and the 'ZO-1 versus MLKL pS358 ' immunosignals were then averaged as shown in Supplementary Fig. 2g or Fig. 8b, respectively. Pearson's correlation coefficient of determination (R 2value) generated by the Coloc 2 plugin was used to further gauge the degree of colocalization. 6 fields where the MLKL pS358 immunosignal was randomly redistributed were used as additional non-correlating controls for Pearson's analysis.
Hotspot quantitation. Micrographs (captured as in 2-dimensional epifluorescence microscopy) of the MLKL pS358 immunofluorescent signal were opened in ImageJ 1.52p 70 . A rolling ball filter of 5 was applied, MLKL pS358 immunosignals thresholded (≥9000 units) and objects segmented using the 'Analyze>Analyze Particles' tool. Segmented objects with size 0.5-100 μm 2 and feret diameter>2 (i.e., elliptical objects) were considered hotspots. The number of segmented objects per 100 cells was taken as an index of hotspot occurrence. The mean size of the segmented objects was taken as an index of hotspot size. The fluorescence intensity of each segmented objected divided by its size was taken as an index of hotspot intensity. Quantitation of ZO-1 hotspots was performed using the same method, but with a rolling ball filter of 7 and thresholded immunosignals ≥7000 units were measured.
MLKL cluster quantitation. Micrographs (captured as in 2-dimensional epifluorescence microscopy) of the MLKL 7G2 immunofluorescent signal were opened in ImageJ 1.52p 70 . A rolling ball filter of 6 was applied, MLKL 7G2 immunosignals thresholded (≥2000 units) and objects segmented using the 'Analyze>Analyze Particles' tool. Segmented objects with size 0.5-50 μm 2 and circularity 0.6-1 were considered clusters. The number of segmented objects in TSI-treated MLKL-/-HT29 cells (i.e., an index of non-specific objects) was subtracted from the number of segmented objects in wild-type HT29 cells. The remaining number of segmented objects per 100 cells was taken an index of cluster occurrence. The mean size of the segmented objects was taken as an index of cluster size. The fluorescence intensity of each segmented object divided by its size was taken as an index of cluster intensity. This method of MLKL 7G2 -positive clusters was used for Fig. 2d and Supplementary Fig. S2h. A similar method was used to quantify MLKL 10C2 -positive clusters in Fig. 7c and Supplementary Fig. 3f, except that a rolling ball filter of 7 was applied and thresholded immunosignals ≥5000 units were measured.
Quantitation of spatiotemporal gaps between death events. The 'red channel' micrographs of propidium iodide-uptake over time from the aforementioned Incucyte experiments (as in Incucyte cell death assay with frames taken every 5 min) were exported into ImageJ 1.52p 70 . A rolling ball filter of 50 was applied, registration was performed using the StackReg plugin 74 , and the spatial and temporal coordinates of propidium iodide-positive cells tracked over time using the Trackmate plugin 75 . The result.txt file of coordinates was parsed using a customdesigned visual basic macro to determine the spatial and temporal gap between cell death events for all propidium iodide-positive cells in the field. One cell width was approximated to be 40 pixels and so spatial gaps of ≤40, 40-80 and 80-120 pixels were binned into 1, 2 and 3 units of cell width, respectively.
Statistical analyses. The number of independent experiments for each dataset is stipulated in the respective figure legend. The employed statistical analyses are stipulated in the respective figure legend and performed using Prism v.8.2 (GraphPad). Statistical analyses were only performed on datasets collated from at least three independent experiments.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data and customised Image J macros used by study are available on request from the authors. The source data underlying Figs. 1a-c, 3b, d, e, 4c, 5c, 6b-d, 8b-e, g, and Supplementary Figs. 1a-f, 2a, b, g, h, 3a-f, 4d-f, 5a-c, e, f are provided as a Source Data file. Source data are provided with this paper.