Conformational switching of the pseudokinase domain promotes human MLKL tetramerization and cell death by necroptosis

Necroptotic cell death is mediated by the most terminal known effector of the pathway, MLKL. Precisely how phosphorylation of the MLKL pseudokinase domain activation loop by the upstream kinase, RIPK3, induces unmasking of the N-terminal executioner four-helix bundle (4HB) domain of MLKL, higher-order assemblies, and permeabilization of plasma membranes remains poorly understood. Here, we reveal the existence of a basal monomeric MLKL conformer present in human cells prior to exposure to a necroptotic stimulus. Following activation, toggling within the MLKL pseudokinase domain promotes 4HB domain disengagement from the pseudokinase domain αC helix and pseudocatalytic loop, to enable formation of a necroptosis-inducing tetramer. In contrast to mouse MLKL, substitution of RIPK3 substrate sites in the human MLKL pseudokinase domain completely abrogated necroptotic signaling. Therefore, while the pseudokinase domains of mouse and human MLKL function as molecular switches to control MLKL activation, the underlying mechanism differs between species.

N ecroptosis is a regulated cell death mechanism in which the plasma membrane is compromised, allowing escape of the cell's contents and instigation of an inflammatory response. In contrast to apoptosis, necroptosis is a caspaseindependent cell death pathway, which is executed by the terminal protein, mixed lineage kinase domain-like (MLKL), following activation by the conventional protein kinase, receptor interacting protein kinase (RIPK)-3 [1][2][3] . RIPK3 hetero-oligomerizes with another protein kinase, RIPK1, to form the "necrosome": a high molecular weight signaling platform that has been variously reported to activate MLKL via stable recruitment 2,3 or by a transient enzyme-substrate interaction 1 with RIPK3. RIPK3mediated phosphorylation of the activation loop residues, mouse S345 1,4,5 or human T357/S358 2 , in the MLKL pseudokinase domain (PsKD) is widely thought to be the trigger for MLKL activation. MLKL phosphorylation promotes oligomerization, translocation to the inner plasma membrane, and ensuing necroptotic cell death by membrane permeabilization [6][7][8][9][10][11][12] , although the precise molecular details of this event are the subject of ongoing debate.
MLKL comprises an N-terminal four-helix bundle (4HB) domain connected to a C-terminal PsKD via a two-helix linker, which we termed the brace helices 1 . The N-terminal 4HB domain was shown by us and others using cellular and biochemical assays to be responsible for lipid engagement and membrane permeabilization 7,10,12,13 . The PsKD is so-named because it has a conserved kinase-like fold 1,5,14 ; however, it lacks the residues that are necessary for catalytic activity in canonical protein kinases. As a result, we hypothesized that the PsKD functions as a molecular switch, where phosphorylation of the MLKL PsKD activation loop by RIPK3 leads to a conformation change and relief of an inhibitory protein-protein interaction with the executioner 4HB domain 1,11,15 . Despite lacking catalytic activity, MLKL has retained the ability to bind ATP 1,14,16 , although the role of nucleotide binding in modulating the molecular switch and regulating MLKL's necroptotic function are currently unclear.
How the PsKD might suppress the executioner function of the 4HB domain, and the nature of the conformational change, if any, that occurs upon MLKL activation, remain unknown. Much of our understanding of mechanism extends from the structure of full-length mouse MLKL 1 , where the 4HB domain is solvent exposed and would thus be available to participate in necroptotic killing. In contrast, no structure of full-length human MLKL has been reported to date, and while recombinant mouse MLKL forms trimers in solution 11 , the stoichiometry of human MLKL oligomers has been a matter of debate. Intriguingly, in contrast to mouse MLKL, human MLKL 4HB domain expression does not induce cell death in mouse or human cells in the absence of forced oligomerization 7,17 . Additionally, in contrast to recombinant mouse MLKL 4HB+brace and full-length human MLKL, human MLKL 4HB+brace exhibited modest activity in liposome permeabilization assays in vitro 7 , implying a crucial role for the PsKD in a necroptotic human MLKL assembly.
Using biophysics, mass spectrometry (MS), and cellular assays, we reveal a role for the PsKD in directing the transition of human MLKL from a basal monomeric state to a pro-necroptotic tetramer. Wild-type hMLKL assembled into tetramers in vitro, robustly permeabilized liposomes, and reconstituted necroptotic signaling in MLKL −/− U937 cells. In contrast, hMLKL PsKD mutants, including some identified in cancers, stabilized a monomeric state, leading to deficiencies in membrane permeabilization in liposome dye release assays and in cell death induction. These data support the idea that mutations or binding of ligands, such as ATP, within the PsKD favor a monomeric form of hMLKL that exists basally in the cytoplasm prior to the cell receiving a necroptotic stimulus. Using crosslinking MS (XL-MS) to direct molecular modeling, we found the 4HB domain was bound in cis by the suppressor PsKD, which thereby restricted its executioner activity. The transition to the hMLKL tetramer was modeled from small-angle X-ray scattering (SAXS) data and molecular dynamics (MD), with intermolecular crosslinks as constraints, yielding a model in which the second brace helix serves an important role in higher-order assembly: an assertion further supported by hydrogen-deuterium exchange MS (HDX-MS).
Previous studies have reported that hMLKL oligomerization, and thus activation, can be promoted by introduction of negatively charged residues to mimic hRIPK3-mediated phosphorylation within the PsKD activation loop, T357 and S358 2,12,18 . However, reconstitution of MLKL −/− U937 and HT29 cells with hMLKL constructs encoding phosphomimetic (Glu or Asp) mutations of T357 and S358 neither induced constitutive cell death nor reconstituted cell death upon treatment with necroptotic stimuli. These data indicate that, contrary to its mouse counterpart, activation of human MLKL relies on additional steps beyond modification by RIPK3 for necroptotic signaling to ensue. Overall, this work provides insights into how residues within the MLKL PsKD promote MLKL structural change and the self-association essential for necroptosis.

Results
hMLKL exists in three conformational states. In cells, cytosolic MLKL is monomeric prior to phosphorylation within the pseudoactive site by RIPK3 7,11 . This led us to hypothesize that rearrangement within the PsKD, induced by phosphorylation of the activation loop, drives oligomer formation in vivo. Recombinant full-length hMLKL has been reported to exist in an oligomer:monomer equilibrium 12 , suggesting a dynamic interconversion in solution. E351 is part of the GFE motif present in MLKL orthologs, which is the equivalent of the Mg 2+ -binding DFG motif in active kinases. E351 resides adjacent to activation loop RIPK3 substrate residues, T357 and S358, proximal to other features common to kinases: the VAIK 230 motif and catalytic loop HRD motif counterpart (HG 330 K 331 ) (Fig. 1a). We previously observed that introduction of an E351K mutation, which was first identified in a lung carcinoma 19 , into the recombinant PsKD led to enhanced ATP binding 14 . To determine whether enhanced ATP binding within the PsKD might perturb the oligomer: monomer equilibrium, we compared full-length E351K and wildtype hMLKL. While wild-type hMLKL eluted as two distinct peaks from size exclusion chromatography (SEC), as expected 12 , E351K eluted as a single species at a retention time equivalent to the 55 kDa wild-type hMLKL monomer ( Supplementary Fig. 1a, b). We analyzed the self-association of full-length wild-type and E351K hMLKL at 1, 0.5, and 0.25 mg/mL using sedimentation velocity (SV) analytical ultracentrifugation (AUC). At each concentration, two species were observed for wild-type hMLKL with sedimentation coefficients of 3.5 S and 8.2 S, while only one 3.5 S species was observed for E351K ( Supplementary Fig. 1c, d). While the addition of 300 μM ATP did not alter the positions of either protein at 1 mg/mL, ATP skewed the equilibrium of wild-type hMLKL toward the smaller 3.5 S species (Fig. 1b). Under these conditions, E351K remained monomeric (Fig. 1c).
To accurately determine the molecular weight of the higherorder species of MLKL, we performed native MS analysis on the wild-type hMLKL oligomer SEC fraction. The mass spectrum revealed a charge envelope consistent with a tetrameric protein complex (216,190 Da) (Fig. 1d). In contrast, the most abundant species in the E351K mass spectrum corresponds to a monomer (55,350 Da) in addition to a very small population of dimer   (Fig. 1e). These results confirm that hMLKL forms a tetrameric complex, consistent with the larger species observed by SV-AUC, which is disrupted by the introduction of the E351K mutation in the PsKD. Bilayer permeabilization driven by wild-type and E351K hMLKL was measured using a liposome dye release assay 7 . E351K and the N-terminal domain (hNTD; comprising residues 2-154) induced modest dye release (~15% of maximum) (Fig. 1f), while wild-type hMLKL caused rapid 5(6)-carboxyfluorescein release to~80% of maximum dye release. Addition of ATP to the system enhanced E351K and lowered wild-type hMLKL activity to converge at~40% maximum dye release (Fig. 1f). Taken together with the AUC and native MS results, the activity of these proteins in the liposome dye release assays indicates that recombinant hMLKL forms a tetrameric structure in solution that is destabilized by conformational changes induced by ATP binding or mutations in the vicinity of the PsKD activation loop. Moreover, the liposome permeabilization activity of E351K hMLKL suggests that MLKL can transition between three distinct conformational states: a basal monomer, a nucleotide-loaded transition-state monomer with intermediate liposome permeabilization activity, and the necroptotic tetramer.  The 4HB domain packs against the PsKD αC helix in monomeric hMLKL. We hypothesized E351K hMLKL reflects the basal monomeric conformation of cytosolic hMLKL prior to necroptotic stimuli and sought to model the structure using a chemical XL-MS approach. Using 4-(4,6-dimethoxy-1,3,5-triazin-2-yl)-4-methylmorpholinium chloride (DMTMM) coupling, "zerolength" crosslinks between proximal aspartic/glutamic acid and lysine side chains 20 were identified by MS (Supplementary Data 1, Supplementary Fig. 2a).
From an initial open model of hMLKL 21 (Fig. 1a), generated using the hMLKL N-terminal domain nuclear magnetic resonance structure 13 and PsKD crystal structure 14 by homology with the crystal structure of full-length mouse MLKL 1 , the lowest-energy large-amplitude normal mode that brought the 4HB and PsKDs into contact with one another was traced. This unbiased domain reorientation yielded a structure in good agreement with the chemical crosslinking results. High-scoring links corresponded to Cα-Cα separations less than the upperbound limit of 30 Å (Supplementary Data 1). Thus the resulting model reveals a compact structure in which the 4HB domain is packed against the αC helix and pseudocatalytic loop of the PsKD (Fig. 2a).
Addition of ATP to E351K hMLKL resulted in a transition to a conformer with increased ability to disrupt liposomes (Fig. 1f). To analyze changes in secondary structure (via hydrogen bonding) and dynamics of this transition from a basal to a pro-necroptotic state, we used HDX-MS in the presence or absence of ATP. ATP binding to E351K hMLKL resulted in a reduction of deuterium exchange in the regions surrounding the ATP-binding pocket in the PsKD (Fig. 2b-d; Supplementary Fig. 3) and increased in deuterium exchange in the PsKD αC helix (Arg-244 to Leu-263) and in the α3 and α4 helices from the 4HB domain that abut the αC helix in our model ( Fig. 2b-d). This suggests that, upon ATP binding, the 4HB and αC helix become disengaged leading to an increase in secondary structure dynamics and a reduction in stability of the α-helices in each domain as the protein transitions to a pro-necroptotic conformation.
hMLKL tetramer assembles into a daisy chain configuration. In previous mutational studies of the human MLKL PsKD, we observed that introduction of a K331N substitution in the pseudocatalytic loop decreased ATP affinity 14 . Here, in the context of full-length hMLKL, K331N substitution markedly favored tetramer formation over monomer by SV-AUC sedimentation experiments ( Supplementary Fig. 4a), thus identifying K331N hMLKL(2-471) as an ideal candidate for biophysical characterization of the hMLKL tetramer. Recombinant K331N permeabilized liposomes equivalently to wild-type hMLKL in the absence of ATP, with only modest perturbation by the addition of ATP ( Fig. 3a), indicating that this mutation promotes a hMLKL conformer that is predisposed to forming higher-order assemblies. To define the in-solution structure of the hMLKL tetramer, we performed SAXS analysis on recombinant K331N hMLKL eluted by inline SEC into the path of the X-ray beam (statistics shown in Supplementary Table 1). Scatter profiles corresponding to the apex of the tetramer SEC peak were averaged and background subtracted, with Guinier analysis indicating that the data were monodisperse (Fig. 3b) with a maximum particle dimension (D max ) of 170 Å (Fig. 3c). We fitted a tetrameric model of hMLKL to these scattering data by rigid-body modeling of the reported hMLKL model 21 using maximum 35 Å distance restraints between K157-K157, K173-K173 and K305-K305 pairs in neighboring protomers (Fig. 3d). These contacts were experimentally derived from crosslinks identified in hMLKL tetramers by MS following treatment with the primary amine crosslinkers, disuccinimidyl suberate (DSS) and Bis(sulfosuccinmidyl) suberate (BS 3 ) ( Supplementary Fig. 2b, c; Supplementary Data 1). These restraints are few, because only the like-to-like crosslinks mediated by DSS and BS 3 could be unambiguously assigned as intersubunit linkages within the hMLKL tetramer. Molecular dynamics refinement and simulated annealing minimization were used to further refine the tetramer model. Theoretical scatter calculated for our model was consistent with the experimental data (χ = 0.542 using CRYSOL; Fig. 3b). The tetramer model ( Fig. 3d) revealed an arrangement of protomers in which residues previously reported to be involved in lipid engagement 17 were surface exposed (Fig. 3e, f) and all 4HB domains presented in a plane that would allow simultaneous lipid engagement (Fig. 3e, f) The daisy chain-like tetramer arrangement of hMLKL formed through heterologous interfaces, rather than a dimer of dimers with isologous interfaces, is consistent with the AUC and native MS analyses, which suggest that the protein exists in monomer and tetramer states but not dimers and trimers.
To understand the dynamic changes that occur to MLKL upon tetramer formation, we performed HDX-MS on the E351K (monomer) and wild-type (tetramer) hMLKL proteins (Fig. 3g, h; Supplementary Fig. 2d-f). We observed markedly reduced deuterium exchange in wild-type relative to E351K hMLKL in the brace region (residues 151 and 194) (Fig. 3h) and increased deuterium exchange in the PsKD αC helix and in the 4HB domain α4 helix within the tetramer (Fig. 3g), consistent with 4HB domain reorientation away from the PsKD within the tetramer.
PsKD switch mutations compromise liposome permeabilization. To understand how changes in the PsKD of MLKL control the transition between monomer and tetramer, we introduced a b Mirror plot demonstrating deuteration (percentage of theoretical maximum) of E351K hMLKL peptides±ATP over the indicated time points. Error bars represent SEM of triplicate experiments. Each tick on the x axis represents an individual peptide. Overlapping or adjacent peptides were identified for all regions of the protein except for a small section of the N-terminal 4HB. A line break in this region has been introduced to illustrate the gap in sequence coverage (see Supplementary Fig. 3a for peptide sequence coverage). Overlapping regions of peptide coverage are illustrated using tiled lines below the x axis. All SEM values did not exceed ±5.96 from n = 3 independent time course experiments; error bars omitted for clarity. c Differential deuterium uptake of E351K hMLKL peptides ±ATP over the indicated time points (calculated + ATP deuteration minus -ATP deuteration). As in b, each tick represents an individual peptide with a line break introduced to illustrate a gap in sequence coverage. Peptides corresponding to overlapping sequences are shown as tiles below the plot. The corresponding regions of hMLKL the peptides map to are shown schematically between c and d, with sequence elements that undergo differential exchange±ATP highlighted in blue. d Differential hydrogen-deuterium exchange (E351K hMLKL±ATP, 60 min time point) for each analyzed peptide was averaged across overlapping sequences per amino acid residue (excluding the first 2 residues of each peptide due to high rates of back exchange) and mapped onto the E351K hMLKL model (see Supplementary Fig. 3b). Dark gray coloring represents regions of the protein excluded from analysis owing to insufficient sequence coverage NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-04714-7 ARTICLE NATURE COMMUNICATIONS | (2018) 9:2422 | DOI: 10.1038/s41467-018-04714-7 | www.nature.com/naturecommunications series of mutations into full-length hMLKL guided by XL-MS, HDX-MS, and residues mutated in human cancer specimens (Fig. 4a, Table 1). Mutations in hMLKL PsKD annotated in human tumors are of interest because defective necroptosis may underlie persistence of tumors. To benchmark activities in liposome permeabilization assays, we compared recombinant wild-type or mutant full-length hMLKL to a previously reported control mutant, K16A/R17A hMLKL, which exhibits compromised liposome permeabilization 17 (Fig. 4b).
The activation loop of the hMLKL PsKD contains E351 of the GFE ("DFG") motif, K354/T355, and the RIPK3 substrates, T357 and S358 (Fig. 4a). In our model, E351 is surrounded by other residues of interest: K230 of the VAIK motif, which positions ATP for catalysis in active protein kinases and was previously   presence (red) of 500 μM ATP. Dye release was monitored by spectrophotometry at 485 nm over 30 min. Data represent mean ± SEM of three independent assays. b Small angle scattering profile of averaged and background subtracted data from the apex of inline size exclusion peak (black circles). Agreement between the experimental and theoretical scatter (red circles) calculated from a rigid body tetramer model in SASREF is reflected in the χ value of 0.54. Inset: Guinier plot of experimental SAXS data for q.Rg < 1.3, illustrating that aggregated species do not contribute substantively to the scatter. c Interatomic distance distribution plot, P(r), profile calculated from scattering data with GNOM. Maximum particle dimension, D max , was estimated as 170 Å. d Arrangement of subunits within the hMLKL tetramer. e Electrostatic surface of one subunit in the tetramer. Charge distribution is consistent with available presentation of residues with positive charge in the 4HB domain to engage lipid (lipid-binding residues defined in ref. 17 14 ; G330 of the HGK motif (equivalent to the catalytic "HRD" motif in active kinases); and the αC helix residue, E258. Mutation of each of these residues led to deficits in liposome permeabilization (Fig. 4), which could not be overcome by composite mutations, such as E258K/E351K. E258 formed DMTMM crosslinks with K351 within the E351K hMLKL monomeric mutant ( Supplementary Fig. 2a, Supplementary Data 1), indicative of a salt bridge present in these mutants that locks the protein in a suppressed conformation. Phosphorylation of the activation loop residues, T357 and S358, by RIPK3 or replacement with phosphomimetic Asp or Glu residues was reported to promote oligomer formation of hMLKL 2,12 . We introduced phosphomimetic Glu residues into both wild-type and E351K hMLKL. T357E/S358E hMLKL behaved similarly to wild-type hMLKL except that its ability to permeabilize liposomes was not diminished in the presence of ATP. Alanine substitution of the adjacent K354/T355 led to an hMLKL with similar behavior in liposome assays (Fig. 4c), with similar lack of attenuation upon ATP. The effects of activation loop substitution are consistent with the idea that phosphomimetic mutations, and perhaps phosphorylation itself, stabilize MLKL tetramers and ATP does not induce dissociation of this stabilized form (Fig. 4h). Somewhat surprisingly, T357E/S358E/ E351K hMLKL was unable to disrupt liposomes and was unresponsive to ATP, indicating a block in the molecular switch-driven transition between the three conformational states of hMLKL (Fig. 4i).
In both the wild-type and E351K hMLKL monomers, a pair of lysines adjacent to the aforementioned E258 in the αC helix, K255 and K256 (Fig. 4a), were observed to form zero-length (DMTMM) crosslinks with E111 in the 4HB domain (Supplementary Data 1). These crosslinks were not detected in the wildtype hMLKL tetramer, raising the possibility that K255/K256 contribute to maintaining the monomeric conformation. The K255A/K256A and D107A/E111A hMLKL mutants exhibited liposome permeabilization comparable to wild-type hMLKL (Fig. 4j, k). Liposome permeabilization by these mutants was only modestly impacted by ATP and, to comparable extents, consistent with reciprocal interaction of these sites within the 4HB domain-restrained monomeric conformation (Fig. 4j, k). Deficits in liposome permeabilization observed among PsKD mutants were not attributable to compromised lipid binding, because E258K and G330E hMLKL bound membrane lipid arrays with a preference for phosphatidylserine, phosphatidylinositol-4phosphate, and cardiolipin, like wild-type and D107A/E111A hMLKL ( Supplementary Fig. 5), and consistent with similar reports for wild-type hMLKL 10,12 .
Our recent studies have implicated the brace helices in communicating activating signals from the PsKD to the executioner 4HB domain and as an important contributor to MLKL oligomerization 22 . Mutation of brace residue, K157, which formed DSS/BS 3 crosslinks to K157 of another protomer within the wild-type hMLKL tetramer ( Supplementary Fig. 2a, b), induced deficits in liposome permeabilization (~40% dye egress vs 80% for wild-type hMLKL; Fig. 4l).
PsKD mutations restrict molecular switch toggling by ATP. The observation that ATP binding destabilized full-length wildtype hMLKL tetramer formation and attenuated liposome permeabilization activity led us to examine whether ATP binding might restrict MLKL activation. Using a thermal stability shift assay 16,23 , we measured the affinity of full-length wild-type and mutant hMLKL for ATP (Table 1; Supplementary Fig. 4). While wild-type hMLKL bound ATP with a K d of 36 μM, curiously, the liposome permeabilizing activities of T357E/S358E, E258K, and G330E hMLKL were unaffected by ATP (Fig. 4) despite comparable affinities, suggesting that these mutations perturb (or override) any conformational change that may impact MLKL activation upon ATP binding. In contrast, E351K hMLKL bound ATP with a K d of 26 μM, and ATP binding induced elevated liposome permeabilization activity (Fig. 1f), consistent with a molecular switch that can be toggled by ATP in E351K hMLKL to relieve the repression of hMLKL activation. Unsurprisingly, K230M, K331N, K354A/T355A, and T357E/S358E/E351K hMLKL did not markedly respond to ATP in the liposome dye release assays, consistent with their ATP-binding deficits ( Supplementary Fig. 4). Overall, residues within the PsKD play key roles in governing hMLKL activation, and their mutation can severely impact (or override) the sensitivity of hMLKL to ATP-binding-induced conformational changes. We expected mutations outside the PsKD to have little impact on ATP binding. Monomer promoting PsKD mutations delay necroptosis.
Having identified residues in the hMLKL PsKD as important determinants of activation in in vitro assays, we sought to establish their contribution to cellular necroptosis signaling. To this end, we generated MLKL −/− U937 cells by CRISPR editing using a non-integrating lentivirus and reconstituted the necroptosis pathway by expressing wild-type or mutant hMLKL for 16 h via a doxycycline-inducible lentiviral vector before sensitivity to death stimuli was assessed. MLKL knockout was verified by western blot (Fig. 5a) and next-generation sequencing, and manifested in negligible cell death (Fig. 5b), as quantified by propidium iodide (PI) uptake using flow cytometry, following stimulation with the necroptotic stimulus, TSI (comprising tumor necrosis factor (TNF) (T), the Smac mimetic Compound A (S), and the pan-caspase inhibitor, emricasan/IDN-6556 (I); as reported previously 22,24 ). Induced expression of wild-type hMLKL in MLKL −/− U937 cells restored sensitivity to the necroptotic stimulus, TSI, resulting in~55% cell death 24 h posttreatment (Fig. 5b, c). We further validated this cellular system as a platform for examining the impact of mutating MLKL on necroptotic signaling by introducing the mutant, K16A/R17A hMLKL. Consistent with an earlier study 17 , K16A/R17A hMLKL exhibited retarded cell death relative to wild-type hMLKL, with reduced cell death observed at 6 h but comparable death to wildtype hMLKL at 12 and 24 h time points (Fig. 5c). Importantly, wild-type and all mutant hMLKL constructs examined herein were expressed at equivalent levels, as assessed by western blot analysis ( Supplementary Fig. 6), indicating that any deficits in necroptotic signaling are not a consequence of reduced protein expression. All cell lines showed comparable sensitivity to the apoptotic stimulus, TS, in the presence and absence of doxycyline-induced protein expression. As expected, doxycycline was essential for responsiveness to the necroptotic stimulus, TSI, in reconstituted MLKL −/− U937 lines ( Supplementary Fig. 7). Pseudoactive site integrity was important to hMLKL activity in liposome assays (Fig. 4), and mutations to E351, E258, G330, and K230 have been observed in various tumors (Table 1), raising the possibility that their mutation may compromise necroptosis. Consistent with this idea, relative to wild-type hMLKL, expression of E351K or E258K hMLKL in MLKL −/− U937 cells (Fig. 5c), but not parental U937 cells (Supplementary Fig. 7j, l), led to delayed kinetics of necroptotic death comparable to K16A/ R17A hMLKL. This was borne out in Blue Native polyacrylamide gel electrophoresis (BN-PAGE) analyses, where hallmarks of hMLKL activation-translocation of hMLKL from cytoplasm (C) to membrane (M) fraction, the membrane fraction assembling into higher-order species and undergoing phosphorylation on the activation loop residue S358-was delayed for the E258K hMLKL mutant relative to wild type ( Fig. 5d; Supplementary Fig. 8). While wild-type hMLKL was phosphorylated, oligomerized, and membrane translocated robustly from 4 h post-stimulation, E258K hMLKL was only modestly oligomerized and membrane translocated, with a correspondingly weak phospho-S358 signal detected ( Supplementary Fig. 8). We concluded that, because E258K hMLKL was phosphorylated, albeit weakly over this timeframe, it has retained the capacity to engage RIPK3. Comparatively, E258K/E351K hMLKL was even more defective, with maximal~50% death observed at 24 h post-TSI. E258K/ E351K hMLKL exhibited attenuated membrane translocation, higher-order hMLKL complex assembly and no S358 phosphorylation by BN-PAGE post-TSI up to 10 h poststimulation, while robust signals were observed for the wildtype hMLKL counterpart at 4 h post-stimulation (Fig. 5c, d;  Supplementary Fig. 8). Defective necroptosis was similarly profound in the K230M and G330E mutants where levels of death were only~50% of those of wild-type hMLKL at 24 h poststimulation (Fig. 5c), while the neighboring mutant, K331N, exhibited no deficits in necroptotic signaling (Fig. 5c). We observed that the αC helix undergoes enhanced solvent exchange in monomeric E351K hMLKL upon addition of ATP and accordingly examined the role of the proximal αC helix residue, K249 (Fig. 4a), as a probe for conformational changes induced by the molecular switch mechanism. Because of the importance of K230 and E351 to MLKL activation, it was unsurprising that mutation of the proximal αC helix residue, K249, also led to delayed necroptotic signaling (Fig. 5c).
In our MLKL monomer model arising from XL-MS, the αC helix residues, K255 and K256, are adjacent to D107 and E111. We examined the contributions of these contacts to hMLKL activation by reconstituting MLKL −/− U937 cells with K255A/ K256A hMLKL or D107A/E111A hMLKL. While mutation of K255/K256 were dispensable, the integrity of D107/E111 proved essential to hMLKL-mediated cell death. Membrane-associated high molecular weight assemblies are a hallmark of hMLKL activation, but interestingly, despite failing to signal for cell death, D107A/E111A hMLKL assembled into the higher molecular complexes in both cytoplasmic and membrane fractions (Fig. 5d), and curiously S358 was phosphorylated in both fractions ( Supplementary Fig. 8). D107 and E111 are the hMLKL counterparts of D106 and E110 previously implicated as essential for mouse MLKL-mediated necroptosis 11 , indicating that these residues play conserved roles in recruiting necroptotic coeffectors required for MLKL translocation to, and/or permeabilization of, membranes. The identities of these additional factors remain of outstanding interest.
In addition to K157-K157 BS 3 /DSS crosslinks between subunits within the hMLKL tetramer, K157 formed zero-length crosslinks with D300 in the C-lobe of the PsKD, suggesting a possible role in both communicating conformational changes from the PsK to the 4HB domain. Consistent with the deficits observed in liposome permeabilization assays, K157A hMLKL exhibited delayed necroptosis kinetics in reconstituted MLKL −/− U937 cells, with only~50% of the death observed for wild-type hMLKL at 6 h (Fig. 5c).
Phosphomimetic PsKD mutants do not induce necroptosis. In contrast to mouse MLKL 1,11 , introduction of point mutations in the hMLKL PsKD did not induce stimulus-independent death but rather led to defective necroptotic activity (Fig. 5c, d). To further probe the differences in activation mechanisms between mouse and human MLKL, we introduced phosphomimetic (acidic) or phosphoablating (alanine) residues at T357 and S358 in the hMLKL PsK activation loop to, respectively, emulate or block the key activation step, phosphorylation by RIPK3. Introduction of the phosphomimetic substitution, S345D, into mouse MLKL to emulate RIPK3-mediated phosphorylation leads to robust stimulus-independent death in mouse cells 1,8,25 . By contrast, not only did phosphomimetic T357E/S358E and T357E/S358D or phosphoablating T357A/S358A hMLKL mutants fail to induce constitutive cell death, each could not reconstitute necroptosis signaling in MLKL −/− U937 (and HT29) cells and no cell death was observed 24 (and 48) hours post-stimulation with TSI (Fig. 5e). This was further reflected in an absence of detectable T357E/S358E hMLKL oligomerization and membrane translocation by BN-PAGE at 12 h post-TSI stimulation (Fig. 5d).
Notably, similar results were obtained using MLKL −/− HT29 cells, where inducibly expressed wild-type, but not T357E/S358E or T357E/S358D, hMLKL constructs could reconstitute the necroptosis signaling pathway (Fig. 5f). Furthermore, recombinant wild-type, but not T357E/S358E, hMLKL robustly bound immobilized Streptavidin-binding peptide (SBP)-tagged hRIPK3 kinase domain in surface plasmon resonance (SPR) experiments (Fig. 5g). These data suggest deficits in necrosome  (Fig. 6). Defective T357E/S358E hMLKL function is not attributable to the mutation compromising MLKL folding or structure, because the crystal structure of T357E/S358E hMLKL PsKD (residues 190-471) solved to 2.8 Å resolution was comparable to that of wild-type hMLKL PsK (RMSD 0.78 Å) ( Supplementary Fig. 4b, c; Supplementary Table 2). Additionally, introduction of T357E/S358E substitutions failed to overcome deficits in necroptotic signaling in E351K hMLKL, with negligible cell death observed 24 h post-stimulation with TSI. Introduction of single phosphomimetic and phosphoablating mutations at T357 or S358, on the other hand, did not preclude reconstitution of necroptotic signaling in MLKL −/− U937 cells. All single residue mutants were capable of restoring signaling to different extents, with T357D most resembling wild-type hMLKL and mutations to S358 the most deleterious (Fig. 5e). Collectively, these data indicate that, while both phosphorylation sites are required for the complete activity of MLKL, S358 is most crucial.

Discussion
MLKL is ubiquitously expressed 1,26 and, as such, mechanisms must exist to prevent MLKL activation and necroptotic cell death in the absence of an upstream trigger. Here we reveal the existence of a basal, monomeric conformer in which the executioner 4HB domain interacts with the PsKD via an interface centered around the αC helix to attenuate its killing activity. Using XL-MS and HDX, coupled with in vitro liposome permeabilization assays and biophysical and cellular studies, we implicated a core of pseudoactive site residues centered around the activation loop and αC helix of the human MLKL PsKD in controlling assembly of hMLKL into the tetramers that mediate necroptotic cell death.
The stoichiometry of the human MLKL oligomer has been a matter of debate in the literature, with trimers, tetramers, hexamers, and octamers variously proposed 6,[9][10][11][12]27,28 , with few studies using high-resolution approaches. Here we used native MS to define the hMLKL oligomer as a tetramer. Further evidence for hMLKL-forming tetramers was obtained from SAXS data, which in combination with XL-MS and MD, enabled lowresolution modeling of the hMLKL tetramer. These data are consistent with a daisy chain assembly, in which the 4HB domain and brace helices abut the foot of the PsKD C-lobe in the neighboring hMLKL protomer. However, detailed knowledge of the precise intersubunit interactions that govern tetramer assembly await a high-resolution tetramer structure. Nonetheless,  Fig. 6 The necrosome recruits hMLKL via RIPK3 before RIPK3-mediated phosphorylation induces hMLKL reorganization and/or disengagement, membrane translocation, and necroptotic death. a A revised model for MLKL activation. Human MLKL relies on recruitment to the necrosome via RIPK3, which scaffolds assembly of necroptotic MLKL tetramers and their reorganization and/or disengagement following RIPK3-mediated phosphorylation. b In contrast to mouse MLKL, which can be activated by mutations that mimic RIPK3-mediated phosphorylation, hMLKL activity is compromised by pseudoactive site and activation loop mutations within our tetramer model, the four 4HB domains (and component residues implicated in lipid binding) are largely surface exposed and are orientated at the outer torus surface, which would allow simultaneous membrane docking. Although not observed in our studies, we cannot exclude the possibility that higher-order oligomers, such as the octamers reported to arise from tetramers via an interface surrounding the 4HB domain α3-α4 loop residue, C86 27 , may arise from hMLKL tetramers in cells.

MLKL
Our findings indicate that in solution hMLKL populates either tetramer or monomer forms and suggest an intrinsic difference from mouse MLKL, whose 4HB+brace was found to form homotrimers in our earlier AUC 11 and SAXS 22 studies. Interspecies differences in oligomeric status are not surprising considering the evolutionary divergence in oligomeric status of other effectors in other death pathways, including the apoptosome scaffold APAF1 29 . In the case of human MLKL, our observations that ATP binding destabilized the hMLKL tetramer and modulated liposome permeabilization indicate that hMLKL can occupy three distinct states: a monomeric form with attenuated activity, which we propose corresponds to the preactivation form; a tetramer, which corresponds to the form that mediates necroptosis; and an intermediate form, in which the PsKD molecular switch mechanism is locked, preventing communication of changes from the PsKD to the executioner 4HB domain.
Consistent with the importance of the "pseudoactive" site in hMLKL to activation, mutations to K230, E258, G330, and E351 have been identified in tumor genomes. Not only did K230M, E258K, G330E, and E351K mutations compromise hMLKL activity in in vitro liposome permeabilization assays but also necroptotic signaling in reconstituted MLKL −/− U937 cells. Importantly, it is the network of residues centered on the activation loop and αC helix (Fig. 4a) that underpins hMLKL activation, rather than simply ATP-binding propensity. This parallels the situation in mouse MLKL where mutation of the key ATPbinding residue, K219 (the mouse counterpart of human K230), and a partner residue in the activation loop not involved in ATP binding, Q343, can toggle the molecular switch to induce constitutive activation of mMLKL 1 . In hMLKL, mutation of pseudoactive site residues induce the opposite outcome, giving rise to deficits in hMLKL activation. Together, these studies indicate that ATP-binding propensity is a hallmark of a PsKD that can behave as a conformational switch but that ATP binding itself is not necessarily a physiological modulator of MLKL activation. Rather, mutations that favor maintenance of monomeric forms of human MLKL lead to deficits in cell death as a consequence of delayed engagement and phosphorylation by necrosomal hRIPK3. Because E258K hMLKL can still undergo RIPK3-mediated phosphorylation, this supports the idea that monomeric hMLKL mutants have retained the capacity to engage RIPK3. However, the delayed kinetics of oligomerization and RIPK3mediated phosphorylation of these mutants are consistent with the notion that they are locked in a pre-necroptotic monomeric conformation in the cytosol, which slows their recruitment to RIPK3 and subsequent activation in cells.
In contrast to our earlier studies of mouse MLKL 1,11 , none of the mutations we introduced into the human MLKL PsKD in the present work led to stimulus-independent cell death. Fundamental differences between the mouse and human necroptosis pathways are not altogether surprising considering 71.5%, 60.5%, and 61% amino acid identity between mouse and human RIPK1, RIPK3, and MLKL, respectively, and the inability of human MLKL to reconstitute the necroptosis pathway in mouse Mlkl −/− fibroblasts 7 . However, the inability of hMLKL PsKD mutants to induce cell death in our study was surprising in light of previous reports 2,12,18,30-32 . The basis for these differences are unclear, although it is possible that the choice of cell line, expression vector, and mode, and whether endogenous MLKL is present, are contributing factors. Moreover, not only did the T357E/S358E or T357E/S358D hMLKL double mutants fail to induce constitutive cell death when expressed in MLKL −/− U937 and HT29 cells in our hands but it also completely abrogated sensitivity to necroptotic stimuli. Our earlier studies hinted at this possibility: expression of T357E/S358E hMLKL in parental U937s did not induce constitutive necroptosis but instead led to a dominantnegative effect when cells were treated with necroptotic stimuli 7 , presumably owing to hetero-oligomerization with endogenous hMLKL poisoning assembly of active oligomers. While reconstitution of MLKL −/− U937 and HT29 cells with T358E/S358E hMLKL led to abrogation of necroptotic killing, individual substitution of S358 with Glu or Ala and T357 with Ala led to defective, but not abrogated, sensitivity to necroptotic stimuli in U937 cells. In contrast to mouse MLKL, where a transient "hitand-run" interaction with RIPK3 to phosphorylate MLKL is sufficient for its activation and can be emulated by the constitutive killer S345D mutation 1,4,11 , our findings suggest a model in which RIPK3 recruits hMLKL to the necrosome via a stable interaction. This assembly precedes RIPK3-mediated phosphorylation to facilitate hMLKL tetramer assembly, reorganization, or disengagement from the necrosome, to allow translocation to the plasma membrane where permeabilization ensues (Fig. 6a). Defective necroptosis arose upon mutation of the hMLKL PsKD pseudoactive site or RIPK3 substrate sites (Fig. 6b), which could be attributed to deficits in RIPK3-mediated recruitment of hMLKL to the necrosome. In this study, we have unveiled crucial, here-to-fore unrealized, mechanistic differences in activation between mouse and human MLKL. An appreciation of speciesdependent differences underlying the PsKD molecular switch mechanism is essential to understanding how MLKL drives necroptosis and will prove invaluable to structure-guided design of modulatory small molecules.

Methods
Protein expression and purification. Full-length human MLKL and mutants were expressed in Sf21 insect cells as previously described 7 . Briefly, DNA encoding wildtype (synthesized by DNA2.0) or mutant hMLKL (residues 2-471) were cloned into a pFastBac-derived vector as in-frame fusions with a TEV protease-cleavable N-terminal GST tag. Bacmids generated using the Bac-to-Bac system (Invitrogen) following transformation of DH10MultiBac Escherichia coli cells were used to transfect Sf21 insect cells to generate P1 baculovirus according to the Bac-to-Bac manual (Invitrogen). Following a second viral amplification step, P2 virus (10% v/ v) was used to infect 0.5 L Sf21 cultures proteins in 2.8 L Fernbach flasks shaking at 90 rpm, 27°C for 48 h. Human MLKL N-terminal domain (2-154) was expressed via a pGEX-derived vector encoding an in-frame fusion with an N-terminal, TEV protease-cleavable GST tag in E. coli BL21 Codon Plus, as previously described 7 . Cell pellets were resuspended in lysis buffer (20 mM Tris, pH 7.5, 200 mM NaCl, 10% glycerol, 0.5 mM TCEP) and cells lysed via sonication. Cell debris was separated from soluble protein via centrifugation at 45,000 × g and lysate filtered with a 0.45 μm filter prior to incubation with pre-equilibrated glutathione agarose (UBPBio) at 4°C for 1 h with gentle agitation. Beads were then collected via centrifugation and washed exhaustively with lysis buffer and mixed with 200 μg of TEV for 2 h at 20°C or overnight at 4°C. The supernatant was filtered through a 0.45-μm filter, concentrated via centrifugal ultrafiltration (30 kDa molecular weight cutoff; Millipore) and loaded onto a Superdex S200 gel filtration column (GE Healthcare) equilibrated in gel filtration buffer (20 mM HEPES pH 7.5, 200 mM NaCl, 5% glycerol). For AUC experiments, glycerol was omitted from the buffer. Proteins eluted from SEC column were assessed for purity by reducing sodium dodecyl sulfate (SDS)-PAGE with Sim-plyBlue SafeStain visualization (ThermoFisher). Except where indicated, when the protein eluted as oligomer and monomer peaks, the two peaks were pooled. Proteins were concentrated to approximately 5 mg/mL and snap frozen for storage at −80°C. N-terminally SBP-tagged hRIPK3 (2-356; DNA synthesized by ATUM) was expressed in Sf21 cells via a bacmid generated from pFastBac Htb (Invitrogen), which encodes a TEV protease-cleavable N-terminal His 6 fusion, using methods analogous to those described for mouse RIPK3 33 .
Liposome dye release assays. Lipids were purchased from Avanti Polar Lipids (Alabaster, AL, USA) and resuspended in chloroform at 20 mg/mL to emulate a plasma membrane composition (20% POPE, 40% POPC, 10% phosphoinositol, 20% DOPS, 10% POPG), as previously reported 7 . Dried lipids were resuspended in 1 mL 50 mM 5(6)-carboxyfluorescein solutions, vortexed, and sonicated and then liposomes extruded through a 100 nm pore membrane. Prior to use, liposomes were purified from excess 5(6)-carboxyfluorescein using PD-10 column (GE Healthcare) equilibrated with 10 mM HEPES pH 7.5, 135 mM KCl. Samples in the presence of ATP were preincubated with 500 μM ATP on ice for 30 min. For dye release assays, 8 μM of liposomes were mixed with 1 μM protein in Falcon 96-well plates and immediately measured every 60 s (excitation 485 nm, emission 535 nm; HiDex Plate reader) at 20°C for 30 min. 100% dye release was determined by the incubation of liposomes with 1% Triton X-100 and data were represented as a percentage of maximum release. All assays were performed in triplicate; data are presented as a mean ± SD of three independent assays.
Native mass spectrometry. Recombinant protein was buffer exchanged into 250 mM ammonium acetate using a Micro Bio-Spin 6 column (Bio-Rad, USA). MS spectra were acquired on a Bruker Maxis II ETD using an off-line nano electrospray ionization source (Bruker, Bremen, Germany). Samples were introduced into the ion source via a nanoflow capillary emitter tip prepared in-house with spray current introduced via a titanium wire in the rear of the capillary. The optimized instrument parameters used were as follows: Spectra rate 0.5 Hz, Capillary voltage 1000 V, Dry Temp 60°C, Funnel 1 RF 400Vpp, Multipole RF 800Vpp, isCID 0 eV, Quadrupole (Ion Energy 1 eV, Low Mass 1000 m/z), Collision Cell (Energy 1 eV, RF 3000Vpp, Transfer Time 120 μs, Pre Pulse Storage 20 μs, Gas Flow Rate 20%). Spectra were analyzed using the Unidec software 34 .
Crosslinking mass spectrometry. Wild-type or E351K hMLKL (2.5 μg) were crosslinked in the presence and absence of 500 μM ATP. DSS and BS 3 were used at 0.1, 0.25, 0.5, 1.0, and 2.0 mM while DMTMM was used at 100, 30, 15, and 7.5 mM. All reactions were incubated at 20°C for 30 min and stopped by addition of reducing SDS loading dye, before boiling and resolution by 4-20% Tris-Glycine SDS-PAGE (BioRad, CA). Protein bands were visualized with SimplyBlue SafeStain (ThermoFisher) and manually excised for in-gel reduction with 10 mM dithiothreitol (DTT; Sigma) for 30 min, alkylated for 30 min with 50 mM iodo-acetamide (Sigma), and digested with 375 ng trypsin gold (Promega) for 16 h at 37°C. The extracted peptide solutions were then acidified (0.1% formic acid) and concentrated to 10 μL by centrifugal lyophilization using a SpeedVac AES 1010 (Savant). Extracted peptides were injected and separated by reversed-phase liquid chromatography (LC) on a M-class UHPLC system (Waters, USA) using a 250 mm × 75 μm column (1.7 μm C18, packed emitter tip; Ion Opticks, Australia) with a linear 90-min gradient at a flow rate of 400 nL/min from 98% solvent A (0.1% Formic acid in Milli-Q water) to 35% solvent B (0.1% Formic acid, 99.9% acetonitrile). The nano-UPLC was coupled on-line to a Q-Exactive Orbitrap mass spectrometer equipped with a nano-electron spray ionization source (Thermo Fisher Scientific, Bremen, Germany) or an Impact II UHR-QqTOF (Bruker, Bremen, Germany). High-mass accuracy MS data were obtained in a data-dependent acquisition mode.
Raw files were analyzed using MaxQuant (version 1.5.5.1). The database search was performed using the Uniprot Homo sapiens database plus common contaminants with strict trypsin specificity allowing up to two missed cleavages. MaxQuant APL files were converted to MGF files using the APL to MGF convertor software (https://www.wehi.edu.au/people/andrew-webb/1298/apl-mgf-converter). Crosslinked peptides were identified from the MGF files using the StavroX software (version 3.6.0.1). Lysines, protein N-termini, serines, threonines, and tyrosines were set as reaction sites of the crosslinker NHS esters (DSS and BS 3 ) or Lysines or protein N-termini linked to aspartic and glutamic acids or protein C-terminus (DMTMM). Trypsin was set as the enzyme allowing for three missed cleavages at lysines and two at arginines. Precursor precision was set at 10 ppm with fragment ion precision set at 20 ppm.
Hydrogen-deuterium exchange mass spectrometry. Experiments were initiated by 20-fold dilution of protein (1 mg/mL final) in deuterated buffer containing 10 mM Tris-HCl pD 7.5. Aliquots were taken at multiple time points (30 s, 60 s, 5 min, 10 min, 1 h and 16 h) with the hydrogen/deuterium exchange reaction suppressed by acidification of the sample to pH 2.5 using formic acid before snapfreezing in liquid nitrogen. Digestion of the protein was carried out by thawing the sample in a ten-fold dilution of H 2 O before addition of an equimolar concentration of pepsin (Sigma, USA) for 5 min on ice. Peptides were subjected to LC-MS analysis using an 1100 series HPLC (Agilent, USA) coupled to an LTQ-Orbitrap XL (Thermo, USA). Peptides were loaded onto an in-house packed, reverse-phase trap column (ReproSil-Pur C18 (Dr. Maisch GmbH, Germany), 2 × 2 mm 2 , 5 μm) before separation on an in-house packed, reverse-phase analytical column (ReproSil-Pur C18 (Dr. Maisch GmbH), 200 μm×150 mm, 3 μm) housed at 1°C. Peptides were loaded onto the trap column at 5% acetonitrile and 0.2% formic acid, and elution performed using a gradient rising from 5% to 40% acetonitrile over 12 min, then 85% acetonitrile for 5 min before reconditioning the column at 5% acetonitrile for 15 min. Spectra were acquired in positive ion mode with m/z range from 350 to 1850. Each sample was analyzed in triplicate. Experiments were performed as experimental replicates (three independent exchange time courses). Deuteration of peptides was determined by analysis of samples using HDX workbench, as previously described 35 . Additional analysis of HDX workbench output was performed using in-house scripts written in R (version 3.3.1). Deuterium exchange per residue was calculated by summing and averaging deuterium exchange across overlapping peptide regions for each residue. The first two residues of each peptide were not included in this calculation due to known issues with rapid back exchange 36 .
Modeling of the monomeric and tetrameric conformers. An initial monomeric structural model was generated using MODELLER using the PDB entries, 4BTF, 2MSV, and 4MWI, as templates against the hMLKL sequence. Conformational change in this model was predicted using normal mode analysis (NMA) employing the ProDy software 37 . Since large amplitude motions are not harmonic, an eigenvector tracking approach was used to model domain motion and reorientation. Using the X-ray crystal structure, the eigenvector corresponding to the lowestfrequency normal mode was determined using the ProDy NMA application. Atomic coordinates were displaced along this eigenvector, and the geometry minimized using the YASARA software. NMA was performed upon the new geometry, and atomic coordinates again displaced in the direction of the eigenvector with the largest overlap with the original (X-ray structure) lowest-energy eigenvector. This procedure was repeated until the two domains were in contact. To the final structure obtained following NMA tracking, 30 ns of MD simulation was applied using the YASARA software.
The tetramer model generated from SAXS data by SASREF rigid-body fit was subjected to MD refinement using the YASARA software (www.yasara.org). Initial simulated annealing minimization was proceeded by 1 ns of MD using the YASARA2 knowledge-based force field, followed by simulated annealing minimization. The N ζ atoms of lysine residues 157, 173, and 305 were restrained to the same residue in each of the neighboring monomers with a weak harmonic restraint of 10/Nm with a distance of 30 Å. All bonds to hydrogen and angles involving hydrogen were fixed, the simulation temperature was set at 298 K, and the time step used was 2 fs.
human tonicity RPMI medium and Dulbecco's modified Eagle's medium, respectively, supplemented with 8% v/v FCS, with puromycin (5 μg/mL) added for lines stably expressing MLKL constructs. Routine testing confirmed cell lines to be mycoplasma-negative. U937 cells were plated to 5 × 10 4 cells/well in 96-well plates and treated with doxycycline (20 ng/mL) overnight to induce the expression of MLKL construct. HT29 cells were assayed analogously, but in 48-well plates with protein expression induced by doxycycline for 4 h before stimulation with necroptotic stimuli. U937 cells were then treated with TNF (100 ng/mL) and the Smac-mimetic Compound A (500 nM) (TS) to induce apoptosis or TS in the presence of the pan-caspase inhibitor IDN-6556 (10 μM) to induce necroptosis for 6, 12, or 24 h; HT29 cells were examined at 48 h post-TSI stimulation. Cells were then compared to untreated cells at the same time point via PI (1 μg/mL) uptake as quantified by flow cytometry.
Western blot and Blue Native PAGE. Expression analyses were done in parallel with cell death assays, with cells seeded into 96-well plates at 5 × 10 4 cells/well, and induced overnight with doxycycline, as indicated above, or untreated. When the cell death assay reached 6 and 12 h time points post-TS/TSI treatment, cells for expression analysis were harvested using a 2× SDS Laemmli lysis buffer, sonicated, boiled at 100°C for 5 min, and then resolved by 4-20% Tris-Glycine gel (Biorad). After transfer onto polyvinylidene difluoride (PVDF), membranes were blocked with 5% skim milk and then probed with antibodies as indicated. For BN-PAGE, wild-type U937 and U937 MLKL −/− cells expressing mutant MLKL variants were seeded into 12-well plates at 5 × 10 5 cells/well 16 h before treatment. Expression of mutant MLKL constructs was induced with 20 ng/mL doxycycline for 4 h, then treated with TSI as described above for time course indicated in Fig. 6 and Supplementary Fig. 8, or left untreated, as indicated. Wild-type U937 were treated with TSI for time course indicated in Fig. 6 or left untreated. For all cell lines, untreated cells were harvested at the longest time point assessed in that experiment. Cells were fractionated into cytoplasmic and membrane fractions as previously described 7,11 . In brief, initial cell permeabilization used a buffer containing 0.025% digitonin (BIOSYNTH, Staad, Switzerland), 2 μM N-ethyl maleimide, phosphatase, and protease inhibitors. Crude membrane and cytosolic fractions were separated via centrifugation and then adjusted to a final concentration of 1% w/v digitonin. Fractions were resolved by 4-16% Bis-Tris Native PAGE gel (ThermoFisher) and then transferred onto PVDF for western blot analyses.
Small-angle X-ray scattering. SAXS data were collected at the Australian Synchrotron SAXS/WAXS beamline using an inline co-flow size-exclusion chromatography set-up 43 . Two-dimensional scattering data were radially averaged, normalized to sample transmission, and 5 × 2 s scatter patterns from the apex of the elution peak were averaged and background scatter (an average of 20 scatter patterns from early in the SEC run) was subtracted using the ScatterBrain software (Stephen Mudie, Australian Synchrotron). The ATSAS suite of software was used for Guinier analysis (PRIMUS 44 ), to calculate the pair-wise intra-atomic distance distribution function P(r) and maximum particle dimension, D max (GNOM 45 ), and rigid body modeling of the tetramer using SASREF with P4 symmetry 46 . All structural models were illustrated using PyMOL. The data collection and processing statistics are presented in Supplementary Table 1.
Analytical ultracentrifugation. Sedimentation velocity experiments were performed using an XL-I analytical ultracentrifuge (Beckman Coulter) equipped with UV/Vis scanning optics, as before 22 . Buffer reference (20 mM HEPES pH 7.5, 200 mM NaCl) and 380 μL of hMLKL sample solutions at 0.25, 0.5, or 1 mg/mL were loaded into 12 mm double-sector cells with quartz windows and the cells were mounted in an An-60Ti 4-hole rotor or An-50Ti 8-hole rotor. Experiments in the presence or absence of 300 μM ATP were performed on 1 mg/mL hMLKL samples and buffer reference containing 300 μM ATP was used. All experiments were conducted at 50,000 rpm (201,600 × g) and 20°C, and radial absorbance data were collected at 280 nm in continuous mode. Data were fitted to a continuous sedimentation coefficient distribution [c(s)] model and converted to standardized sedimentation coefficient (s 20,w ) distributions using SEDFIT 47 . The protein partial-specific volumes, buffer density, and buffer viscosity were calculated using SEDNTERP 48 .
Thermal shift assays. Thermal shift assays were performed as described previously 1,14,16 using a Corbett Real Time PCR machine with proteins diluted in 150 mM NaCl, 20 mM Tris pH 8.0, 1 mM DTT to 0.5 μg in a total reaction volume of 25 μL. SYPRO Orange (Molecular Probes, CA) was used as a probe with fluorescence detected at 530 nm. Thermal stability was examined in the presence and absence of 0.2 mM ATP, with no divalent cations added. A positive ΔT m value indicated ATP binding that protected the protein from denaturation. The K d for ATP binding was determined by titrating ATP concentrations 0-400 mM. Two transitions were observed for some variants and the K d was calculated based on the ΔT m of the first transition state. Two independent assays were performed for wildtype and mutant hMLKL proteins; averaged data ± SD are shown for each in Supplementary Fig. 4 and K d estimates in Table 1.
Surface plasmon resonance. SBP-tagged hRIPK3 kinase domain (residues 2-356) was immobilized onto a dextran chip with preimmobilized streptavidin (Biacore SA chip). Briefly, the surface was activated with 50 mM NaOH + 1 M NaCl solution and SBP-hRIPK3 injected at a concentration of 0.04 mg/mL in 20 mM HEPES buffer (pH 8.0) containing 200 mM NaCl. Data were collected using a Biacore 4000 or Biacore T200 (Biacore International, Uppsala, Sweden). As a control, the first flow cell was used as reference surface to correct for bulk refractive index, matrix effects, and nonspecific binding. The second flow cell was immobilized with SBP-hRIPK3. Solutions of full-length wild-type hMLKL and T357E/ S358E hMLKL were flowed over the chip at 1, 0.5, 0.25, 0.125, 0.0625, and 0 μM in triplicates, at 30 μL/min, and the amount of bound material as a function of time was recorded as sensorgrams. The dissociation phase was monitored in 20 mM HEPES (pH 8.0) containing 200 mM NaCl at the same flow rate. This was followed by one injection of 50 mM NaOH + 1 M NaCl to regenerate a fully active capturing surface. All experiments were performed at 25°C.
Data availability. Atomic coordinates for the T357E/S358E hMLKL  have been deposited in the Protein Data Bank under the accession number 6BWK. The data that support the findings of this study are available from the corresponding author upon request.