Mouse and human RIPK3 selectively activate MLKL orthologues
To examine the compatibility of orthologous MLKL proteins with the necroptosis machinery in mouse and human cells, we introduced genes encoding human (Homo sapiens), mouse (Mus musculus), rat (Rattus norvegicus), horse (Equus caballus), pig (Sus scrofa), chicken (Gallus gallus), stickleback (Gasterosteus aculeatus), frog (Xenopus tropicalis) and tuatara (Sphenodon punctatus) MLKL into Mlkl−/− mouse dermal fibroblasts (MDFs) and MLKL−/− human U937 cells. These orthologues were chosen to maximize our sampling of phylogenetic diversity among vertebrate MLKL sequences in nature, ranging in sequence identity from 35%–85% to mouse and 36%–65% to human MLKL (Supplementary Fig. 1, Table 1). Orthologue MLKL constructs were stably introduced into these cell lines via a puromycin-selectable lentiviral vector18, from which MLKL expression could be induced using doxycycline (dox). Cells were stimulated with TNF, Smac mimetic and the Caspase inhibitor, IDN-6556/emricasan (TSI), to initiate necroptosis, as described before34,36, in the presence or absence of dox-induced orthologous MLKL gene expression. Cell death was measured by flow cytometry using propidium iodide (PI)-uptake (exemplified in Supplementary Fig. 2). Owing to sequence divergence, existing MLKL antibodies did not recognize all orthologue sequences. To enable verification of expression by western blot, chicken, stickleback, frog and tuatara MLKL were C-terminally FLAG-tagged (Supplementary Fig. 3a, b), because N-terminal tags are known to compromise the killing function of mouse and human MLKL23,25. Mouse MLKL bearing a C-terminal FLAG tag reconstituted necroptotic signaling in Mlkl−/− MDF cells (Fig. 1a), supporting the notion that C-terminal tagging does not compromise MLKL function. Among orthologues tested, only mouse and horse, and to a lesser extent pig, MLKL orthologues could reconstitute signaling in Mlkl−/− mouse fibroblasts, with death only observed when orthologue expression was induced and the necroptosis stimulus, TSI, applied. Notably, horse MLKL killed cells less potently than the mouse counterpart, and pig MLKL was only able to induce low levels of cell death (12%) upon treatment with the necroptosis stimulus, TSI. Remarkably, given the high sequence identity to mouse MLKL (86% identical; 96% similar) and the high similarity of rat and mouse RIPK3 sequences (Tables 1 and 2), rat MLKL did not kill mouse cells (Fig. 1a). Expression of MLKL orthologues in human MLKL−/− U937 cells revealed only human and pig MLKL could reconstitute necroptotic signaling in human cells (Fig. 1b). This was also surprising, because human MLKL shows greater sequence similarity to the horse protein than to pig MLKL (Table 1). These data also confirm that mouse MLKL could not substitute for human MLKL in the human necroptosis pathway, as previously inferred from immunoprecipitation studies in which human MLKL did not interact with mouse RIPK3 (ref. 37). We note that while we could verify expression of the MLKL orthologues by immunoblot (Supplementary Fig. 2a–d), it is not possible to estimate the relative abundance of each protein. Sequence divergence precludes the use of the same antibody to detect expression of each orthologue owing to epitope differences, and even where antibodies cross-react, such as for the anti-MLKL 3H1 clone, the epitopes differ between the mouse, rat, human and horse sequences. Additionally, it remains to be established if there is a threshold level of MLKL expression required for the execution of necroptotic cell death, and whether yet-to-be-identified co-effectors may differ between species and influence the kinetics of cell death upon reconstitution with orthologues. Together, our studies underscore the remarkable selectivity of RIPK3 orthologues for their cognate MLKL effectors, consistent with the idea that RIPK3–MLKL cassettes have co-evolved in response to selective pressures, such as those exerted by pathogens5.
Rat and horse MLKL exhibit conformational heterogeneity
It was of interest to understand why horse, but not rat, MLKL could reconstitute necroptotic signaling in Mlkl−/− MDFs, despite the high sequence identity of mouse and rat MLKL sequences. To obtain molecular insights into this selectivity, we crystallized the pseudokinase domains of rat (residues 179–464) and horse MLKL (residues 188–475), following their expression and purification from insect cells, and solved their crystal structures at 2.2 and 2.7 Å resolution, respectively (Fig. 2, Supplementary Table 1). Both structures exhibit typical features of the pseudokinase/kinase domain fold: a smaller, N-terminal lobe comprising principally β-strands and a larger C-lobe composed predominantly of α-helices. The disposition of the αC helix in each structure was typical of that of a conventional (catalytically active) protein kinase. The conserved Glu of the αC helix in either structure forms a salt bridge with the β3-strand Lys of the VAIK motif, which contributes to assembly of an intact regulatory (R)-spine38.
We next compared the rat and horse MLKL pseudokinase domain structures to those of mouse and human MLKL previously reported18,39 (Fig. 2a–d). The mouse MLKL pseudokinase domain contains a distinguishing activation loop helix that abuts the αC helix (Fig. 2b) and displaces it from participating in interactions present within the active sites of conventional protein kinases18. In particular, in mouse MLKL, Q343 in the activation loop helix forms a hydrogen bond to K219 of the β3-strand VAIK motif, precluding engagement of the VAIK motif Lys (K219) in a conventional ion pair with the αC helix Glu (E239). This contrasts the human MLKL pseudokinase domain, which contains the conventional VAIK-Lys:αC helix Glu interaction (K230:E250) and exhibits an intact R-spine to resemble an active protein kinase structure39 (Fig. 2a). Based upon overall structural features, such as the αC helix position and the VAIK-Lys:αC helix Glu ion pair, the rat and horse MLKL pseudokinase domain structures reported herein more closely resemble the human MLKL pseudokinase, with respective RMSD of 0.962 and 0.712 Å across Cα atoms. Considering the high sequence similarity between rat and mouse MLKL, this is surprising, and the structural divergence of rat and mouse MLKL pseudokinase domains may explain why rat MLKL cannot reconstitute the necroptosis signaling pathway in mouse cells lacking Mlkl.
While grossly topologically similar to rat and human MLKL pseudokinase domain, the horse MLKL pseudokinase domain structure revealed unexpected features (Fig. 2d). In the N-lobe, we observed an unconventional helix in the β3-strand-αC helix loop that is positioned adjacent to the αC helix between S233 and I238. Additionally, although typically unstructured in protein kinase and pseudokinase structures, a substantial portion of the activation loop (residues L351-I358) was well resolved in the horse MLKL pseudokinase domain structure (Supplementary Fig. 4). This loop folded back from a conventional position adjacent to the αC helix towards the hinge and was buried within the “pseudoactive”/ATP-binding pocket. Crucially, two key residues, T356 and S357, are buried in this cleft (Fig. 2d, cyan), and these residues correspond to the residues phosphorylated by RIPK3 in human MLKL (T357 and S358), whose phosphorylation are widely considered hallmarks of MLKL activation.
Horse MLKL β3-αC loop helix facilitates mouse RIPK3 binding
To deduce how horse MLKL could communicate with mouse RIPK3 to induce its activation in Mlkl−/− mouse fibroblasts, we superimposed the horse MLKL pseudokinase domain structure upon the mouse MLKL pseudokinase domain within the previously described mouse MLKL:RIPK3 kinase domain crystal structure40 (Fig. 3a). We observed that the β3-αC loop helix, only observed in horse MLKL to date, spatially occupied the position of the mouse MLKL αC helix, leading us to hypothesize a role for this helix in mediating RIPK3 recognition. In particular, S233 of horse MLKL is structurally proximal to S228 of mouse MLKL, which forms a hydrogen bond with S89 of mouse RIPK3 in the MLKL:RIPK3 co-crystal structure40 (Fig. 3b). We also postulated that R242 of horse MLKL may recapitulate the π–π stacking interaction of F27 (mouse RIPK3) and F234 (mouse MLKL) with a cation–π stacking interaction, and therefore also contribute to the N-lobe RIPK3 binding interaction. The intervening residues in horse MLKL, R236 and S237, were also oriented to face the RIPK3 N-lobe in the structural overlay (Fig. 3b). Therefore, we generated Ala substitution mutants of the predicted RIPK3 interactors, S233, R236, S237 and R242 (Fig. 3b), within full-length horse MLKL and examined whether their expression in mouse Mlkl−/− MDF cells could induce cell death in the presence or absence of necroptotic stimuli (Supplementary Fig. 3, Fig. 3d). Each mutant could reconstitute the necroptosis pathway in Mlkl−/− mouse fibroblasts except S233A horse MLKL, while R242A horse MLKL exhibited constitutive activity in the absence of a necroptotic stimulus. These data support a role for the novel β3-αC loop helix structurally positioning specific residues of the horse MLKL N-lobe to facilitate the mouse RIPK3 interaction. We speculate that this conformation, which is a distinguishing feature of horse MLKL, precludes interaction with human RIPK3 and contributes to species specificity.
Within the C-lobe of MLKL, we recently implicated the hydrophobic residue, F384, of human MLKL in the engagement of human RIPK3 for necroptotic signaling in human U937 cells5. Here, we observed that Y385 of horse MLKL and F373 of mouse MLKL are positioned within the αEF-αF loop in analogous positions to facilitate mouse RIPK3 engagement (Fig. 3c). To test this idea, we introduced Ala substitutions of Y385 in full-length horse MLKL and F373 in full-length mouse MLKL and examined their capacity to reconstitute the necroptosis pathway when expressed in Mlkl−/− MDF cells (Fig. 3d, e). These substitutions completely abrogated necroptotic signaling, supporting a conserved function for this aromatic residue in contributing to the RIPK3 binding interface. Collectively, our mutational analysis of horse MLKL indicates that the acquisition of an atypical N-lobe helix that mimics the unusual mouse MLKL αC helix position enables selective activation of horse MLKL by mouse RIPK3. In addition, mutation of horse MLKL Y385 and mouse MLKL F373 supports the previously proposed idea5 that this aromatic residue, conserved in many MLKL orthologues, but not stickleback and frog MLKL (Supplementary Fig. 1), is essential for the C-lobe RIPK3 interaction.
Function of MLKL phosphorylation differs amongst orthologues
RIPK3-mediated phosphorylation of the MLKL pseudokinase domain activation loop is a recognized hallmark of necroptosis pathway activation18,19,27,41. However, the precise function it serves in MLKL activation remains unclear. In the case of mouse MLKL, activation loop phosphorylation on S345 is thought to be sufficient to convert MLKL into a killer protein18,26. By contrast, phosphomimetic or phospho-ablating mutations of the activation loop residues, T357 and S358, prevented human MLKL’s participation in necroptosis signaling, presumably by prohibiting RIPK3-mediated activation34. Here, we sought to test whether rat and horse MLKL behave like mouse MLKL, where phosphomimic mutants can trigger constitutive cell death, or like human MLKL, where mutation of phosphosites blocks killing. We introduced the S345D mutation into rat MLKL to mimic substitution in its mouse MLKL counterpart (Supplementary Fig. 3e). When inducibly expressed in Mlkl−/− MDF or MLKL−/− U937 cells, cell death was significantly elevated ~2-fold above background levels of death (Fig. 4a). These data support the idea that a phosphomimetic substitution in the rat MLKL pseudokinase domain can trigger interconversion to a pro-necroptotic state, albeit less effectively than the counterpart mutation within mouse MLKL. In horse MLKL, we introduced Ala or Glu substitutions of T356 and S357, the horse counterparts of the RIPK3 substrates in human MLKL, T357 and S35819, and found that these mutations completely abrogated the capacity of these constructs to participate in necroptotic signaling in mouse cells (Fig. 4a). The lack of activity by the Ala substitution mutant suggests that the activity of horse MLKL in mouse cells is reliant on RIPK3-mediated phosphorylation of those residues. The phosphomimetic data suggest that, unexpectedly, the horse MLKL T356E-S357E mutant behaves more similarly to the human MLKL phosphomimetic mutant, which is inactive, than the constitutively active mouse counterpart.
In our horse MLKL pseudokinase domain structure, we observed nestling of the activation loop within the pseudoactive/ATP-binding site (Fig. 4b–d). To our knowledge, such an activation loop orientation in the ATP-binding cleft has not been reported in earlier pseudokinase or kinase domain structures. More broadly, such a conformation raises the prospect that MLKL might sequester its activation loop within the ATP-binding site, and phosphorylation by RIPK3 might trigger the conformational change that underlies transition of MLKL from dormant monomer to a pro-necroptotic oligomer. Accordingly, we mutated residues in the activation loop and proximal residues in the “pseudoactive” site of horse MLKL to explore their role in necroptotic signaling. Crucially, unlike mouse MLKL18, mutation of the activation loop residue equivalent to Q343 in mouse MLKL, horse MLKL Q355, which mediates H-bonding between the activation loop and the hinge region in the horse MLKL structure, did not impact necroptotic signaling (Fig. 4e). The rotamer of T356 could not be assigned with certainty in our structure, and appeared to form a hydrogen bond with either the backbone carbonyl group of L207 or the hydroxyl group of the T208 side chain in different chains of the asymmetric unit. Mutation of T208 to Ala did not perturb the function of horse MLKL in mouse cells, which may indicate that T356 hydrogen bonds with the backbone of L207, or that the N-lobe:Q355 and T356 activation loop hydrogen bonds that stabilize the activation loop in this conformation are not essential for horse MLKL function (Fig. 4e). We further explored the molecular basis for these observations by subjecting the pseudokinase domain of horse MLKL to molecular dynamics simulations.
Unbiased simulations of the phosphorylated and dephosphorylated horse MLKL were performed starting from the crystallized conformation in which the activation loop is buried in the pseudoactive site, with the unresolved portion of the activation loop (residues 358–366) modelled using RosettaRemodel. The buried conformation can be characterized by hydrogen bonds between the C284 peptide backbone and the activation loop Q355 carboxamide (Fig. 4c, g, left). While this interaction remains stable in dephosphorylated MLKL, it is unstable in the pT356/pS357 MLKL simulations, as indicated by the high root mean square fluctuations of the activation loop residues in the phosphorylated form in comparison to the dephosphorylated form (Fig. 4f). Furthermore, analysis of simulation snapshots depicts the phosphorylated activation loop moving out of the pseudoactive site to an extended conformation in which the Q355–C284 interaction is broken (Fig. 4g). These data suggest that the addition of phosphate groups destabilizes the buried activation loop conformation and support the concept that, among some MLKL orthologues, activation loop phosphorylation may induce increased mobility to promote RIPK3 dissociation following its phosphorylation of MLKL. Further exploration of this idea in other orthologues, such as human MLKL, using molecular dynamics simulations is difficult with currently available structures, because the activation loop residues subject to phosphorylation are unresolved. Despite the experimental challenges, it remains of immense interest whether phosphorylation-induced activation loop mobility is a generalized feature of MLKL activation by RIPK3.