A FRET biosensor for necroptosis uncovers two different modes of the release of DAMPs

Necroptosis is a regulated form of necrosis that depends on receptor-interacting protein kinase (RIPK)3 and mixed lineage kinase domain-like (MLKL). While danger-associated molecular pattern (DAMP)s are involved in various pathological conditions and released from dead cells, the underlying mechanisms are not fully understood. Here we develop a fluorescence resonance energy transfer (FRET) biosensor, termed SMART (a sensor for MLKL activation by RIPK3 based on FRET). SMART is composed of a fragment of MLKL and monitors necroptosis, but not apoptosis or necrosis. Mechanistically, SMART monitors plasma membrane translocation of oligomerized MLKL, which is induced by RIPK3 or mutational activation. SMART in combination with imaging of the release of nuclear DAMPs and Live-Cell Imaging for Secretion activity (LCI-S) reveals two different modes of the release of High Mobility Group Box 1 from necroptotic cells. Thus, SMART and LCI-S uncover novel regulation of the release of DAMPs during necroptosis.

W hile apoptosis has been considered to be a typical form of programmed or regulated cell death, recent studies have revealed novel types of regulated or programmed cell death, including necroptosis, pyroptosis, and ferroptosis 1,2 . Necroptosis is morphologically similar to necrosis and can be induced by virus infection, or death ligands including tumor necrosis factor (TNF), Fas ligand, and TRAIL when caspase activation is blocked [3][4][5] . Upon stimulation with these death ligands, sequential phosphorylation and activation of receptorinteracting protein kinase (RIPK)1 and RIPK3 are induced. Activated RIPK3 phosphorylates mixed lineage kinase domainlike (MLKL), resulting in a conformational change of MLKL. Phosphorylated MLKL forms oligomers and translocates to biological membranes, resulting in the execution of necroptosis through pore formation and membrane rupture 6,7 .
Danger-associated molecular pattern (DAMP)s are hostderived molecules released from dead cells through ruptured nuclear and cytoplasmic membranes. DAMPs include nuclear proteins such as High-Mobility Group Box (HMGB)1 and histones but also heat shock proteins, IL-1 family members, and ATP 8,9 . A previous study reported that HMGB1 is passively released from necrotic cells, but actively secreted from macrophages upon lipopolysaccharide (LPS)-stimulation 10,11 . Released HMGB1 mediates inflammation and also plays a crucial role in the development of septic shock through binding to Toll-like receptor (TLR)2 and 4, and the receptor for advanced glycan endproducts (RAGE) 10 . IL-33, a cytokine belonging to the IL-1 family, is released from the nucleus of necrotic epithelial cells and promotes T helper (T H )2-type immune responses 12 . Intriguingly, IL-33 is cleaved by caspase 3 in apoptotic cells, resulting in the inactivation of its biological activity 12 . Moreover, we recently reported that large amounts of histone H3 are released from apoptotic hepatocytes and potentially involved in the development of endothelial cell injury in the livers and lungs 13 . Together, these results suggest that DAMPs are critically involved in inflammation, immune responses, and tissue injury. Thus, blockade of the release of DAMPs or neutralization of their biological activities might be novel strategies to treat various pathological conditions. However, it is not fully understood whether these DAMPs are actively secreted from dead cells or passively released from the nucleus or the cytosol to the extracellular space through the ruptured membrane. To investigate the mechanism(s) underlying the release of DAMPs from necroptotic cells, it is crucial to monitor the activation of MLKL and subsequent releases of DAMPs in a single cell. We previously developed a Live-Cell Imaging for Secretion activity (LCI-S) platform 14 . LCI-S enables us to immediately capture a cytokine secreted from a single cell, and a secreted cytokine is visualized by total internal reflection fluorescence microscopy-based fluorescence immunoassay (TIRFM-FIA).
Fluorescence resonance energy transfer (FRET) is a technique to measure the distance between two molecules based on energy transfer from a fluorescent donor dye to a fluorescent acceptor dye within a narrow distance 15 . FRET biosensors enable us to monitor dynamic changes of interaction, phosphorylation, and conformational changes of various signaling molecules in living cells. We previously developed a FRET biosensor termed SCAT (a Sensor for Caspase Activation based on FRET). SCAT1 and SCAT3 comprise peptides containing consensus caspase 1 and caspase 3 cleavage sequences between Venus and cyan fluorescent protein (CFP), enabling us to monitor pyroptosis and apoptosis, respectively [16][17][18] . Imaging of macrophages derived from mice stably expressing SCAT1 uncovered an intimate crosstalk between interleukin (IL)-1β release and membrane rupture along with caspase 1 activation 17 . On the other hand, analysis of SCAT3 transgenic mice revealed that the removal of apoptotic cells plays a crucial role in neural tube closure during the development of murine embryos 16 . Together, these results indicate that live cell imaging of caspase activation in vitro and in vivo increases our understanding of pyroptosis and apoptosis under physiological and pathological conditions. In contrast to apoptosis or pyroptosis, it is unclear how signaling molecules crucial for necroptosis are regulated within a single cell during the execution of cell death.
To investigate the mechanisms underlying the release of DAMPs from necroptotic cells, we developed a FRET biosensor termed SMART (a Sensor for MLKL Activation by RIPK3 based on FRET). SMART monitored necroptosis, but not apoptosis or necrosis. An increase in the FRET/CFP ratio of SMART depended on endogenous RIPK3 and MLKL and correlated with phosphorylation of MLKL and RIPK3. More importantly, SMART monitored plasma membrane translocation of oligomerized MLKL. SMART in combination with imaging of the release of nuclear HMGB1 and LCI-S revealed that the release of HMGB1 was tightly correlated with the increase in the FRET/CFP ratio of SMART. We also found that there were two different modes of the release of HMGB1 from necroptotic cells. Thus, SMART and LCI-S uncover novel regulation of the release of DAMPs during necroptosis.

Development of a FRET biosensor that monitors necroptosis.
To develop a FRET biosensor that monitors necroptosis, we focused on the conformational change of MLKL. Murine MLKL (mMLKL) is composed of an N-terminal four-helix bundle domain comprising 4 α helices and a C-terminal kinase-like (KL) domain. The KL domain of mMLKL is composed of 9 α helices and 7 β strands (Fig. 1a, Supplementary Fig. 1a). Crystal structures of the KL domain of MLKL show that it adopts an inactive conformation through a hydrogen bond between K219 and Q343 thereby preventing spontaneous oligomer formation 19 . RIPK3dependent phosphorylation of serines and threonine in α4 helix promotes MLKL plasma membrane translocation and oligomerization, resulting in membrane permeabilization. Since the conformational change in the KL domain of mMLKL might increase the FRET efficiency, we designed intramolecular FRET biosensors comprising fragments of the KL domain between enhanced cyan fluorescent protein (ECFP) and modified yellow fluorescent protein (YPet) serving as FRET donor and acceptor, respectively. We inserted α1 to α9 (α19), α1 to α4 (α14), and α5 to α9 (α59) helices of the KL domain into a FRET backbone vector (Fig. 1a). A murine fibrosarcoma cell line, L929 cells undergo necroptosis when treated with TNF and a caspase inhibitor, z-VAD-fmk (TZ) 20 . To test our FRET biosensors, we transfected L929 cells with them and stimulated the cells with TZ. Transfection of α19 blocked TZinduced necroptosis, and α19 showed a marginal increase in the FRET/CFP ratio in L929 cells ( Supplementary Fig. 1b, c). In sharp contrast, α14, but not α59 showed a dramatic increase in the FRET/CFP ratio in cells undergoing necroptosis ( Supplementary  Fig. 1c). Maximum changes of the FRET/CFP ratio of α14 were significantly higher than those of α59 (Fig. 1b). Cotransfection experiments revealed that three biosensors interacted with RIPK3, and MLKL in the absence or presence of RIPK3 and without induction of necroptosis ( Fig. 1c-e), suggesting that the interaction of FRET biosensors with RIPK3 or MLKL is not sufficient for FRET induction (Fig. 1d). Notably, Flag-RIPK3 was efficiently coimmunoprecipitated with Myc-MLKL in the presence of α19 (Fig. 1e), indicating that α19 did not block the binding of MLKL to RIPK3. Thus, we surmised that α19 might block the oligomerization of endogenous MLKL, thereby inhibiting TZ-induced necroptosis.
Substitution of serines and threonine in the α4 helix of mMLKL (S345, S347, T349, and S352) with alanine (3ST4A for short) blocks TNF-induced necroptosis by preventing MLKL phosphorylation and oligomer formation 21,22 . Unexpectedly, TZ stimulation increased the FRET/CFP ratio in cells expressing α14 mutants 3ST4A, S228A, or S248A (Fig. 1f, g, Supplementary  Fig. 1d), and these mutants bound to RIPK3 (Fig. 1h). Thus, while phosphorylation of MLKL is required for oligomerization of MLKL, phosphorylation of α14 is not necessary for FRET induction. Nevertheless, a RIPK1 inhibitor, Necrostatin-1 (Nec-1) or a RIPK3 inhibitor, GSK′872 blocked necroptosis and abolished an increase in the FRET/CFP ratio in cells expressing α14 upon TZ stimulation (Fig. 1i, j). Taken together these data paradoxically indicate that phosphorylation of α14 is not required for FRET activity yet kinase activities of RIPK1 and RIPK3 are nevertheless required for α14 FRET activation. Further refinement of α14 to generate SMART. Although α14 monitored necroptosis, transient transfection of α14 blocked cell proliferation. To investigate the mechanisms how α14 monitors necroptosis and circumvent this drawback, we generated a further series of α14 mutants. To test whether binding of α14 to RIPK3 is required for monitoring necroptosis, we first mutated phenylalanine at position 234 of MLKL, that is critical for RIPK3 binding in vitro, to glutamic acid (F234E) (Fig. 2a). Unexpectedly, an F234E mutant of α14 still interacted with RIPK3 and showed the increase in the FRET/CFP ratio upon TZ stimulation (Fig. 2b, c). We next replaced amino acids at the indicated fragments with a set of four flexible Ser-Ala-Gly-Gly (SAGG) repeats to maintain the same spacing between Ypet and ECFP (Fig. 2a, Supplementary Fig. 2). The TZ-induced increase in the FRET/CFP ratio was partially or completely abolished in cells expressing Δα2α3, Δα3, or Δabc. In sharp contrast, TZ increased the FRET/CFP ratio of Δa and Δab comparable to that of α14 (Fig. 2b, Supplementary  Fig. 3a, b). Notably, Δα2α3 did not interact with RIPK3 or show the increase in the FRET/CFP ratio (Fig. 2b, c). Three biosensors including α19, α59, and Δabc bound to RIPK3, but did not show the increase in the FRET/CFP ratio (Fig. 2c), suggesting that the interaction of RIPK3 with the FRET biosensors is prerequisite, but not sufficient for monitoring necroptosis. Among FRET biosensors showing the increase in the FRET/CFP ratio upon TZ stimulation, we were only able to obtain cells stably expressing Δab. Thus, this construct is henceforth referred to as SMART (a Sensor for MLKL activation by RIPK3 based on FRET). Cotransfection experiments revealed that SMART and a mutant of SMART where all putatively phosphorylated serines and threonine were replaced with alanine (4ST5A), still interacted with RIPK3 in the absence or presence of GSK'872 ( Fig. 2d). Thus, neither the kinase activity of RIPK3 nor the phosphorylation of SMART is required for their binding.
SMART does not monitor apoptosis or necrosis. To characterize the specificity of SMART further, we stably expressed SMART in L929 cells. Consistent with transient transfection assays, the FRET/CFP ratio rapidly increased in L929-SMART cells treated with TZ, but not TZ and GSK'872 (TZG) (Fig. 3a-c, Supplementary Movie 1, 2). We next treated L929-SMART cells with TG to induce apoptosis visualized by the rapid appearance of the cleaved form caspase 3 (Fig. 3d). Under these experimental conditions, the FRET/CFP ratio was not increased in TGstimulated L929-SMART cells (Fig. 3c, e, Supplementary Movie 3). To test whether SMART monitors RIPK3-independent necrosis, we treated cells with an uncoupler, carbonyl cyanide mchlorophenyl hydrazone (CCCP). As expected, CCCP-induced cell death was not blocked by either zVAD or GSK'872 ( Fig. 3f), suggesting that these cells did not die by apoptosis or necroptosis. The FRET/CFP ratio was not increased in CCCP-stimulated L929-SMART cells and, on the contrary, rapidly decreased over time even prior to SYTOX positive staining indicating membrane permeabilization (Fig. 3c, g, Supplementary Movie 4). Together, SMART is a FRET biosensor that monitors necroptosis, but not apoptosis or necrosis.

SMART monitors necroptosis in MEFs and aMoC1 cells.
To test whether SMART might be used generically to monitor necroptosis, we stably expressed SMART in murine embryonic fibroblast (MEF)s and a murine colonic epithelial cell line, aMoC1 cells. When these cells were treated with the combination of TNF, a SMAC mimetic, BV6, and zVAD (TBZ), they underwent necroptosis 23 (Supplementary Fig. 4a). As expected, TBZ increased the FRET/CFP ratio of SMART in both MEFs and aMoC1 cells (Supplementary Fig. 4b, c, Supplementary Movie 5, 6).
In sharp contrast, TBG induced apoptosis, but did not increase the FRET/CFP ratio of SMART in both MEFs and aMoC1 cells ( Supplementary Fig. 4d, e). Thus, SMART monitors necroptosis, but not apoptosis in MEFs and aMoC1 cells as well as L929 cells. ΔRatio   Fig. 2 Interaction of α14 with RIPK3 is required for monitoring necroptosis. a Diagram of a series of α14 mutants. F234E; phenylalanine at 234 is substituted with glutamic acid, 4ST5A; S228, S345, T347, S349, and S352 were substituted with alanines. Zigzag lines indicate the regions that were replaced with SAGG repeats. Red boxes indicate each α helix. b L929 cells were transiently transfected with the indicated mutants and then stimulated with TZ, and the FRET/CFP ratio was analyzed as in Fig. 1b. Maximum changes of the FRET/CFP ratio. Each dot indicates individual cell (n = 10-14 cells). Statistical significance was determined using the one-way ANOVA test. *P < 0.05, ***P < 0.001, ns, not significant. c HEK293T cells were transiently transfected with the indicated mutants along with FLAG-RIPK3. Cell lysates were immunoprecipitated and analyzed as in Fig. 1c. Results are representative of two independent experiments. d GSK'872 treatment does not affect the binding of SMART to RIPK3. HEK293T cells were transiently transfected with SMART or 4ST5A along with FLAG-RIPK3. At 24 h after transfection, cells were untreated or treated with GSK'872 for 4 h, and cells were subjected to immunoprecipitation and analyzed as described in Fig. 1c. Results are representative of two independent experiments, or pooled results of two independent experiments (b). Error bars indicate s.e.m.
Monitoring of necroptosis by SMART depends on RIPK3 and MLKL. To further investigate the mechanisms how SMART monitors necroptosis, we knocked down the expression of Ripk3 or Mlkl in L929-SMART cells. Treatment of cells with Ripk3 or Mlkl, but not control siRNA blocked TZ-induced LDH release (Fig. 4a, b). As expected, knockdown of Ripk3 abolished TZinduced increase in the FRET/CFP ratio of SMART (Fig. 4c,  Supplementary Fig. 5). TZ-and TBZ-induced increase in the . Together these data support our hypothesis that SMART monitors the oligomerization of MLKL and/or subsequent plasma membrane translocation of oligomerized MLKL even in the absence of TNF stimulation.
Generation of human version of SMART. As the murine α14 SMART prototype did not work in human cells ( Supplementary  Fig. 6), we generated a human version of SMART (hSMART). hSMART was composed of helices α1 to α5 of the KL domain of hMLKL, in which the a and b regions corresponding to mMLKL were replaced with SAGG repeats ( Supplementary Fig. 2c, Fig. 6a). hSMART efficiently monitored necroptosis in a human colon cancer cell line, HT29, following the induction of necroptosis (TBZ), but not apoptosis (TBG) (Fig. 6b, c, e). Execution of necroptosis by MLKL is composed of sequential two steps: oligomerization of phosphorylated MLKL and plasma membrane translocation of oligomerized MLKL. To address which steps are crucial for the increase in the FRET/CFP ratio of SMART, we stimulated HT29-SMART cells with TBZ in the absence or presence of Necrosulfonamide (NSA). NSA targets cysteine at 86 (C86) of the N-terminal domain (NTD) of human, but not murine MLKL, and inhibits necroptosis through preventing plasma membrane translocation of oligomerized MLKL 25,26 . As expected, NSA blocked TBZ-induced plasma membrane translocation of oligomerized MLKL, but not oligomerization of MLKL itself (Fig. 6f). Although hSMART does not possess C86, NSA treatment nevertheless abolished the increase in the FRET/ CFP ratio following TBZ stimulation (Fig. 6d, e). Since hSMART did not translocate to membrane fraction under these experimental conditions (Fig. 6f), direct interaction of hSMART with oligomerized MLKL is not required for the increase in the FRET/ CFP ratio of hSMART.
Two sequential steps of HMGB1 release during necroptosis.
While DAMPs, such as HMGB1 and histones, are considered to be released from dead cells due to the collapse of both the nuclear and cytoplasmic membrane integrity 9,10 , the underlying mechanisms are not fully understood. To address this issue, we investigated the kinetics of the release of various DAMPs along with the progression of cell death. To visualize nuclear DAMPs, we fused mCherry to the C-terminal portion of HMGB1 and histone H3, resulting in the generation of HMGB1-mCherry and histone H3-mCherry, respectively. We first confirmed that the kinetics of the release of HMGB1-mCherry was comparable to that of endogenous HMGB1 from L929 cells stably expressing HMGB1-mCherry cells stimulated with TZ ( Supplementary  Fig. 7a). We next transfected L929-SMART cells with HMGB1-mCherry or histone H3-mCherry. As expected, HMGB1-mCherry and histone H3-mCherry were localized in the nucleus (Fig. 8a, b). HMGB1-mCherry abruptly disappeared between 2 and 3 h after TZ stimulation and this happened before the nucleus became positive for SYTOX (Fig. 8a, c). However, in the same time-frame, the histone H3-mCherry remained constant in the nucleus even after the nucleus became positive for SYTOX (Fig. 8b). The FRET/CFP ratio of SMART gradually increased before the signals of HMGB1-mCherry disappeared (Fig. 8c). Loss of HMGB1 from the nucleus occurred quickly and corresponded with a precipitate rise in cytosolic HMGB1 levels ( Fig. 8c, d). However, this increase in cytosolic HMGB1 was short-lived reflecting rapid release of HMGB1 into the extracellular space (Fig. 8c, d, Supplementary Movie 7). These results suggest that the integrity of the nuclear membrane was already disrupted before the cytoplasmic membrane became ruptured.
Western blotting analysis showed that the release of endogenous   Fig. 7a, b). Similar findings were obtained by MEFs and aMoC1 cells ( Supplementary Fig. 7c, d).
The release of HMGB1 was relatively slow in HT29 and HaCaT cells compared to murine cell lines ( Supplementary  Fig. 7e, f). On the other hand, histone H3 was not, or marginally released into the extracellular spaces from cells died by either apoptosis or necroptosis at least under our experimental conditions ( Supplementary Fig. 7b-f).
LCI-S uncovers two different modes of HMGB1 release. To further investigate the kinetics of the release of HMGB1 and execution of necroptosis, we visualized the extracellular release of HMGB1 by LCI-S in combination with SMART at the single cell level. Our preliminary experiments revealed that commercially available antibodies against HMGB1 were not suitable for sandwich fluoro-immunoassay under physiological conditions. We plated a single cell of L929 co-expressing SMART and HMGB1-mCherry on a microfabricated-well array chip that was precoated with the anti-mCherry antibody. Simultaneous live imaging of SMART, intracellular HMGB1-mCherry, and extracellularly  HMGB1-mCherry by LCI-S further substantiated that the FRET/CFP ratio started to increase before the signals of HMGB1-mCherry disappeared in the cell (Fig. 9a, b, Supplementary Movie 8). Given that there was a time lag between the disappearance of the signals of HMGB1-mCherry in the nucleus and detection of extracellularly released HMGB1-mCherry by LCI-S (Fig. 9a, b), these results further substantiated that HMGB1-mCherry was released from the nucleus before plasma membrane ruptured. Average time from an increase in the FRET/ CFP ratio to the extracellular release of HMBG1-mCherry was approximately 43.8 min (Fig. 9c).
Intriguingly, we found that in some cells, HMGB1-mCherry release was stopped within 10 min, whereas in other cells, HMGB1-mCherry release lasted more than 100 min ( Fig. 9d-f, Supplementary Movie 9, 10). Thus, we surmised the release mode of HMGB1-mCherry could be divided into two groups: one is a burst-mode and the other is a sustained-mode. To compare these two modes quantitatively, we estimated the duration of the HMGB1-mCherry release in individual cells and classified them into two groups by k-means clustering. The representative duration of extracellular HMGB1 release of the burst-mode and the sustained-mode are 7.1 and 109 min, respectively (Fig. 9f). To investigate the kinetics of the extracellular release of HMGB1-mCherry in more detail, we quantified the relative intensities of intracellular (total), extracellular (TIRF), and nuclear HMGB1-mCherry in a single cell, respectively. Then, we plotted representative kinetics of signal intensities of a single cell at the indicated times before and after the extracellular release of HMGB1-mCherry (Fig. 9g). The intracellular and nuclear signals of HMGB1-mCherry drastically diminished in burst-mode cells, but remained relatively high in sustained-mode cells after the cytoplasmic membrane rupture (Fig. 9g, left vs right). Consistently, some portions of nuclear HMGB1-mCherry still existed in the nucleus of a sustained-mode cell, whereas most nuclear HMGB1-mCherry disappeared in a burst-mode cell (Epi, Fig. 9d). Together, the extent of both nuclear and cytoplasmic membrane damage induced by MLKL might critically determine whether cells release HMGB1 in a burst-mode or a sustained-mode.
CHMP4B is involved in a sustained-mode release of HMGB1. The endosomal sorting complex required for transport (ESCRT) is involved in mediating receptor sorting, membrane remodeling, and membrane scission 29 . A recent study has shown that MLKL binds to the ESCRT proteins and generates extracellular vesicle 30 . Moreover, CHMP2A and CHMP4B, components of the ESCRT-III complex, have been shown to delay or prevent TNF-induced necroptosis through shedding MLKL-containing vesicles 31 . Taken that the knockdown of Chmp2a or Chmp4b enhances TNF-induced necroptosis 31 , we surmised that the ESCRT-III proteins maintained a sustained-mode release of HMGB1 by promoting membrane repair. To test this possibility, we knocked down Chmp4b in L929-SMART/HMGB1-mCherry cells by siRNA (Fig. 10a). After TZ stimulation, we monitored HMGB1-mCherry release by LCI-S and estimated the duration of the release of HMGB1 of individual cell. Intriguingly, knockdown of Chmp4b substantially reduced the duration of the HMGB1-mCherry release compared to control siRNA-treated cells (Fig. 10b). Moreover, when we classified the assembly from both of these siRNA-treated cells into two groups based on the duration of the HMGB1-mCherry release by k-means clustering, cells that released HMGB1-mCherry via the sustained-mode were abolished in Chmp4b, but not in control siRNA-treated cell populations (Fig. 10b, Supplementary Fig. 8).
As expected, the time between the start of the release of HMGB1 and the burst of cells was shortened, and ΔFRET/CFP ratio was more rapidly increased in cells treated with Chmp4b siRNA than those with control siRNA (Fig. 10c, d). Together, these results suggest that CHMP4B contributes to maintain a sustained-mode of HMGB1 release, possibly by promoting plasma membrane repair.

Discussion
In the present study, we developed a FRET biosensor that detected necroptosis in living cells. The increase in the FRET/CFP ratio of SMART depended on RIPK3 and MLKL, and was correlated with phosphorylation of RIPK3 and MLKL, hallmarks of necroptosis. Moreover, SMART monitored plasma membrane translocation of oligomerized MLKL even in the absence of TNF stimulation. SMART monitored necroptosis, but not apoptosis or necrosis. Simultaneous live imaging of SMART and the release of nuclear DAMPs by LCI-S uncovered two different modes of the release of HMGB1 from cells undergoing necroptosis. Moreover, CHMP4B, a component of the ESCRT-III complex might determine whether a cell exhibits a burst-mode or a sustainedmode of HMGB1 release.
Many groups including us developed FRET biosensors to monitor apoptosis in living cells 16,18,[32][33][34] . Imaging of necroptosis is rather difficult, since there has been no specific polypeptide(s) that are cleaved by protease(s) activated during necroptosis. Taken that the phosphorylation of MLKL is prerequisite for necroptosis 21,22 , we assumed that a short fragment containing a MLKL-phosphorylation site (~15 amino acids) and a phosphopeptide recognition domain, such as the Forkhead-associated (FHA) domain fused by a linker peptide might be suitable for generating a FRET biosensor to monitor necroptosis. However, such fragment could not monitor necroptosis. Thus, we next Fig. 4 Monitoring of necroptosis by SMART depends on RIPK3 and MLKL. a, b Knockdown of Ripk3 or Mlkl abolishes the TZ-induced increase in the FRET/ CFP ratio of SMART. L929-SMART cells were transfected with control, Ripk3, or Mlkl siRNAs. Expression of RIPK3 or MLKL was analyzed by immunoblotting with the indicated antibodies (a). After transfection, cells were unstimulated or stimulated with TZ for 8 h. Cell viability was determined by LDH release assay (b). Results are mean ± s.d. of triplicate samples. Statistical significance was determined using the one-way ANOVA test. ***P < 0.001. c L929-SMART cells were treated with the indicated siRNAs and then stimulated with TZ. The FRET/CFP ratio was analyzed as in Fig. 1b Further study will be required to address this issue. Time-lapse imaging of the release of nuclear DAMPs and SMART uncovered differential regulation of the release of HMGB1 and histone H3 from cells undergoing necroptosis or apoptosis. HMGB1 was rapidly released from necroptotic cells just after cells became positive for SYTOX. It is currently unknown why HMGB1 bound to chromatin was rapidly released from the nucleus into the extracellular spaces. Disulfide-HMGB1, but not the reduced form of HMGB1, is released from cells during necroptosis 35 . Thus, we surmise that conformation change of HMGB1 induced by disulfide bonding might weaken the interaction of HMGB1 with chromatin, resulting in rapid release from the nucleus to the extracellular spaces in necroptotic cells. Taken that disulfide-HMGB1, but not reduced HMGB1 has a potent activity to produce inflammatory cytokines 35,36 , only biologically active disulfide-HMGB1 might be released from necroptotic cells. Analysis of the detailed kinetics of the release of HMGB1 revealed that the release of HMGB1 from the nucleus to extracellular space appeared to consist of two sequential steps: the first step is cytoplasmic translocation from the nucleus, and the second step is extracellular release from the cytoplasm. This suggests that nuclear membrane damage is induced during the execution of necroptosis before the cytoplasmic membrane becomes ruptured. Taken that oligomers of MLKL might translocate to lipid bilayers of any organelles in cells including nuclear membrane 6,7,37,38 , it is reasonable to hypothesize that nuclear membrane might be disrupted before the breakdown of the cytoplasmic membrane. LCI-S uncovered that there were two different modes of HMGB1 release, "a burst-mode" and "a sustained-mode". The ESCRT complex has been shown to be involved in plasma membrane wound repair 29,39 . Moreover, recent studies reported that components of the ESCRT-III protein complex are involved in attenuation of TNF-induced necroptosis through extrusion of MLKL-containing vesicles 30,31 . Taken that the knockdown of Chmp4b abolished cells showing a sustained-mode release of HMGB1, the balance between the extent of pore forming activity of MLKL and membrane repairing capacity of proteins of the ESCRT complex, such as CHMP4B, might determine whether cells release HMGB1 via the sustained-mode or burst-mode. It would be intrigued to investigate mRNA levels of Mlkl and Chmp4b of each cell that exhibits a sustained-mode or a burstmode.
Currently, the exact biological significances of these two different modes of the releases of DAMPs remain unclear. It is reasonable to surmise that a burst-mode and a sustained-mode cell might elicit qualitatively or quantitatively different biological responses of neighboring cells that sense DAMPs, such as macrophages and dendritic cells (DCs). Thus, it would be interesting to test whether amounts of production of inflammatory cytokines by macrophages or DCs might be different when cells were cocultured with a burst-mode cell or a sustained-mode cell. Alternatively, taken that HMGB1 also has a chemoattractant activity 40,41 , a burst-mode cell might recruit cells such as DCs or neutrophils from a distant site to inflammation sites where cells undergo necroptosis by a concentration gradient of HMGB1 like a chemokine.
RIPK3-dependent oligomerization and subsequent translocation of MLKL to the cytoplasmic and nuclear membrane are reminiscent of caspase 1-or 11-dependent cleavage of Gasdermin (GSDM) D and its subsequent translocation to the cytoplasmic membrane 38,42,43 . Very recently, two groups have reported that caspase 3-dependent cleavage of GSDME/DFNA5 mediates secondary necrosis 44,45 . So far, it is unclear whether GSDMD or GSDME translocate to the nuclear membrane and participates in the release of nuclear DAMPs. In this respect, simultaneous live imaging of nuclear DAMPs, LCI-S, and FRET biosensors that monitor cleavages of the GSDM family will address whether GSDMD or GSDME are also involved in the collapse of the nuclear membrane and the release of nuclear DAMPs.
To generate hSMART, the cDNA fragments of human MLKL containing α1 to α5 helices were amplified by RT-PCR from mRNA of HeLa cells as a template using the following primers (5′-GTTCCTCGAGCAGGCTGGCAGCATTGCAAT AGTGA-3′ and 5′-GTTCCTCGAGCAGGCTGGCAGCATTGCAATAGTGA-3′). Amplified fragments were ligated into 3536NES, resulting in the generation of Fig. 9 LCI-S uncovers two different modes of the release of HMGB1. a-c L929-SMART cells were transiently transfected with HMGB1-mCherry and then stimulated with TZ. The FRET/CFP ratio was analyzed as in Fig. 1b. Representative images of the FRET/CFP ratio (Ratio), intracellular (Epi), and extracellularly released (TIRF) HMGB1-mCherry (a) (n = 19 cells). Epi shows merged images of bright field and intracellular HMGB1-mCherry. Scale bar, 25 μm. Representative temporal relationship between MLKL activation and extracellular release of HMGB1-mCherry (b). The FRET/CFP ratio and the signals of extracellularly released HMBG1-mCherry were plotted at the indicated times after stimulation. Two vertical dotted lines at 94 and 142 min indicate initiation of an increase in the FRET/CFP ratio and extracellular release of HMGB1-mCherry, respectively. Average intervals between the initiation of the increase in the FRET/CFP ratio and extracellular release of HMGB1 (c) (n = 19 cells). d-g Two different modes of extracellular release of HMGB1 during necroptosis. L929-HMGB1-mCherry cells were stimulated with TZ, and extracellularly released and intracellular HMGB1-mCherry were analyzed every 2 min. Representative images of extracellularly released (TIRF) and intracellular (Epi) HMGB1-mCherry at the indicated times (d). Epi indicates merged images of intracellular HMGB1-mCherry and bright field. Time 0 indicates the time when the extracellular release of HMBG1-mCherry started. Scale bar, 25 μm. Representative plots of the change in intensities of extracellularly released HMGB1-mCherry in a burst-mode cell and a sustained-mode cell (e). The duration of the HMGB1-mCherry release in individual cell (f). The duration of the HMGB1-mCherry release of a single cell is plotted and clustered into two groups. Centers of each group are 109 and 7.1 min, respectively. Temporal relationship of a decrease in intensities of total and nuclear HMGB1-mCherry and an increase in extracellularly released HMGB1-mCherry (g). Intensities of each fraction of HMGB1-mCherry were calculated by epifluorescence (left vertical axis) microscopy and TIRF (right vertical axis) and plotted at the indicated times. Representative plots of a burst-mode cell (left, n = 10 cells) and a sustained-mode cell (right, n = 8 cells) in the same experiment. All results are representative of at least two independent experiments hα15. Then, oligonucleotides where the a and b regions corresponding to murine MLKL were replaced with SAGG repeats were synthesized and ligated into XhoI and StuI sites of hα15, resulting in the generation of hSMART (Supplementary  Table 1c and Supplementary Fig. 2c).
To generate stable transfectants of murine and human SMART, we inserted murine and human SMART cDNAs into a transposon vector, pT2KXIG (provided by K. Kawakami) 48 . We then transfected murine and human cells with pT2KXIG-mSMART and pT2KXIG-hSMART along with pCAGGS-T2TP encoding a transposase by electroporation using the Gene Pulser II (Bio-Rad), respectively. After electroporation, ECFP-positive cells were sorted by BD FACSAriaTM III (BD Biosciences).
Knockdown by siRNAs. L929-SMART/HMGB1-mCherry cells were transfected with control (D-001810-10-05), Ripk3 (L-049919-00-0005), Mlkl (LQ-061420-00-0002, pools of J-061420-05 and -07), Chmp4b (L-041531-01-0020) siRNAs by lipofectamine 2000 (Invitrogen). siRNAs were purchased from Dharmacon. Knockdown of the expression of RIPK3 and MLKL was analyzed by immunoblotting with the indicated antibodies at 24 h after transfection. After transfection, cells were stimulated with TZ and subjected to LDH release assay or FRET analysis. Since anti-CHMP4B antibody did not work well in L929 cells, knockdown of the expression of Chmp4b was determined by quantitative polymerase chain reaction (qPCR) using the following primers. Chmp4b-F: 5′-GGAGAAGAGTTCGACGAG GAT-3′ and Chmp4b-R: 5′-TGGTAGAGGGACTGTTTCGGG-3′. qPCR analysis was performed with the 7500 Real-Time PCR detection system with SYBR green method of the target genes together with an endogenous control, murine Hprt with 7500 SDS software (Applied Biosystems). The amounts of Chmp4b were calculated relative to those of murine Hprt with 7500 SDS software (Applied Biosystems), and relative expression of Chmp4b in cells treated with Chmp4b siRNA vs control siRNA to be 1.0.
To detect the phosphorylation of RIPK1, RIPK3, and MLKL, cells were preincubated with BV6 and zVAD for 30 min and then stimulated with TNF for the indicated times. Cells were harvested and lysed with 1× sample buffer (1% SDS, 10% sucrose, 62.5 mM Tris-HCl [pH 6.8], 2.5% β-mercaptoethanol), followed by brief sonication. Cell lysates were subjected to the SDS-PAGE. After transferring the membrane, the membrane was blocked with a blocking buffer (Cell Signaling) and then incubated with the indicated antibodies.
Uncropped images of the results of Western blotting are included in Supplementary Figures 9-13.
Blue Native PAGE. HT29-SMART cells were untreated or treated with TNF, BV6, and zVAD in the absence or presence of NSA for the indicated times. Cell fractionation was performed as previously described 52 . Briefly, cells were harvested with cold PBS and resuspended in a cell fractionation buffer (20 mM HEPES [pH 7.5], 100 mM KCl, 2.5 mM MgCl 2 , and 100 mM sucrose) containing 0.025% digitonin (14592, Cayman Chemical), 25 mM β-glycerophosphate, 1 mM sodium orthovanadate, 1 mM sodium fluoride, 1 mM PMSF, 1 μg ml −1 aprotinin, 1 μg ml −1 leupeptin, and 1 μg ml −1 pepstatin on ice for 10 min. After centrifugation, the supernatant was collected as a cytosolic fraction. The resulting pellet was resuspended with the cell fractionation buffer described above and the final concentrations of digitonin were adjusted to 1% w/v and kept on ice for 20 min. After centrifugation, the 1% digitonin soluble membrane fraction and cytosolic fraction were subjected to 4-16% Bis-Tris Native PAGE gel (BN1002BOX, ThermoFisher) and then transferred onto PVDF membrane. In parallel experiments, membrane and cytosolic fractions were subjected to reducing SDS-PAGE and then transferred onto PVDF membrane. The membranes were analyzed as described in Western blotting.
Detection of released HMGB1 and histone H3. Cells were plated onto 24 well plates. After 12 h incubation, the culture medium was changed to Opti-MEM (Thermo Scientific), cells were stimulated with the indicated agents. Culture supernatant was collected at the indicated times after stimulation, and analyzed by immunoblotting as described above.
In the former experiments (Figs. 1, 2, Supplementary Fig. 1, 3, 6), cells were incubated in growth medium and placed in a heated chamber. Imaging analysis was carried out using a fluorescence microscope (IX-81; Olympus) with a CCD camera (ORCA-R 2 , Hamamatsu) controlled by MetaMorph 7.0 Software (Molecular Devices). 440AF21 excitation (Ex) filter, 455DRLP dichroic mirror, and two emission (Em) filters (480AF30 for ECFP and 536AF26 for YPet) were used for imaging. The FRET emission ratio (FRET/CFP) was calculated by dividing Ex440/ Em536 (FRET) by Ex440/Em480 (CFP) using MetaMorph. For the detection of SYTOX Orange uptake, an additional filter set for TRITC was used.
In the latter experiments (Figs. 3-8, 10, Supplementary Fig. 4, 5), imaging of FRET was collected using a DeltaVision microscope system (GE Healthcare) built on an Olympus IX-71 inverted microscope base equipped with Photometric Coolsnap HQ2 CCD camera, using 60×/NA1.516 PlanApo oil immersion lens (Olympus). For live cell imaging with FRET sensors, cells were seeded on CELLview Cell Culture Dish (Greiner Bio-One) at 37°C heat chamber with 5% CO 2 gas. A Blue Ex filter (400-454 nm), two Em filters (Blue Green, 463-487 nm for ECFP; Yellow Green, 537-559 nm for YPet), and C-Y-m polychroic mirror were used for imaging. The FRET emission ratio (FRET/CFP) was calculated by dividing Ex 436 nm/Em 560 nm (FRET) by Ex436 nm/Em 470 nm (CFP) using SoftWoRx (Applied Precision Inc.). For statistical analyses, the obtained images were analyzed by ImageJ and MetaMorph. ΔFRET/CFP ratio was calculated by subtracting the FRET/CFP ratio at time 0 from the FRET/CFP ratio at the indicated times. To detect Hoechst, UV filter set (Ex: 381-401 nm and Em: 409-456 nm) was used. To detect SYTOX Orange or mCherry, mCherry filter set (Ex 575 nm/Em 625 nm) was used for data collection.
Simultaneous LCI-S, intracellular HMGB1-mCherry, and SMART. Imaging of the release of HMGB1-mCherry by LCI-S was performed as previously described with some modifications 14 . Briefly, time-resolved measurement was performed with a completely automated inverted microscope (ECLIPSE Ti-E; Nikon, Tokyo, Japan) equipped with a high numerical aperture (NA) objective lens (CFI Apo TIRF 60× Oil, NA = 1.49, Nikon), a stage-top incubator (INUBG2TF-WSKM; Tokai Hit Co., Shizuoka, Japan) and an EM-CCD camera (ImagEM C9100-17; Hamamatsu Photonics K.K., Shizuoka, Japan). A high-pressure mercury lamp (Intensilight, Nikon) and a LED (540-600 nm, X-Cite XLED1; Excelitas technologies Corp., Waltham, MA) were used as light sources. The following sets of excitation (Ex) and emission (Em) filters and a dichroic mirror (DM) were used: For FRET, Ex: FF02-438/24-25, Em: FF01-483/32-25 (for ECFP) or FF01-542/27-25 (for Ypet), and DM: FF458-Di02-25×36; for HMGB1-mCherry, Ex: FF01-559/ 34-25, Em: FF01-630/69-25, and DM: FF585-Di01-25×36. These optical filters were purchased from Semrock (Rochester, NY). Eight to fifteen hours before observation, L929 cells stably expressing the indicated protein were plated to a PDMS-glass hybrid microwell array chip, on which anti-mCherry antibody was immobilized. Immediately before observation, the culture supernatant was replaced with a freshly prepared culture medium containing 1% BSA. Mineral oil was layered on top of the medium to prevent evaporation. We observed 40 and 56 microwells at 1.4 and 2 min intervals, respectively, to detect successively the signals of FRET, intranuclear and intracellular HMGB1-mCherry by epi-fluorescence microscopy and extracellular release of HMGB1-mCherry by TIRFM. The time course analysis of each fluorescence was performed using NIS Elements 4.6 (Nikon) and their relative intensities were calculated by ImageJ software. The steepness (k) of the HMGB1 release was estimated by fitting with the modified logistic function below by data analysis software (Origin Pro 2017, OriginLab Co., MA): where I 0 is the maximum intensity of HMGB1 release signal, τ is the time constant of the exponential decrease due to photobleaching of mCherry and release of HMGB1-mCherry from the capture antibody, k is steepness of the curve and t c is the time of the sigmoid's midpoint. Then the duration D needed for the signal reaching to 0.95 from 0.05 of the sigmoid curve was calculated from k as follows: The logarithm of durations (logD) was classified into two groups ("burst" and "sustained") by k-means clustering methods. Average of the logD was used as the representative value of each cluster.
To determine the mode of HMGB1-mCherry release, L929-SMART/HMGB1-mCherry cells were transfected with control or Chmp4b siRNAs by Lipofectamine. The cells were harvested and plated into a microwell array chip at 8 h before observation. We observed 76 microwells at 2.5 min intervals to detect successively the signal of FRET and the intracellular HMGB1-mCherry by epi-fluorescence microscopy and extracellular release of HMGB1-mCherry by TIRFM. The time course analysis of each fluorescence was performed using NIS Elements 4.6 (Nikon). To determine the mode of HMGB1-mCherry release, normalized HMGB1-mCherry TIRF signals were approximated by the above modified logistic function, and the logD of HMGB1-mCherry release were compared between control and Chmp4b knockdown cells by Mann-Whitney test. Assembly of the logD from control and Chmp4b knockdown cells were then classified into two groups (burst and sustained) by k-means clustering methods. Average of the logD was used as the representative value of each cluster.