Nature Chemical Biology 1, 104-111 (2005) doi: 10.1038/nchembio712 RNA flexibility in the dimerization domain of a gamma retrovirus Christopher S Badorrek1 and Kevin M Weeks1 Retroviruses are the causative agents of serious diseases, such as acquired immunodeficiency syndromes and several cancers, and are also useful gene therapy vectors. Retroviruses contain two sense-strand RNA genomes, which become linked at their 5' ends to form an RNA dimer. Understanding the molecular basis for dimerization may yield new approaches for controlling viral infectivity. Because this RNA domain is highly conserved within retrovirus groups, it has not been possible to define a consensus structure for the 5' dimerization domain by comparative sequence analysis. Here, we defined a 170-nucleotide minimal dimerization active sequence (MiDAS) for a representative gamma retrovirus, the Moloney murine sarcoma virus, by stringent competitive dimerization. We then analyzed the structure at every nucleotide in the MiDAS monomeric starting state with quantitative selective 2'-hydroxyl acylation analyzed by primer extension (SHAPE) chemistry. Notably, SHAPE analysis demonstrated that the RNA monomer contains an extensive flexible domain spanning 50 nucleotides. These findings support a structural model in which RNA flexibility directly facilitates retroviral genome dimerization by reducing the energetic cost of disrupting pre-existing base pairings in the monomer. Retroviruses selectively package two sense-strand RNA genomes in the infectious viral particle. The RNA genomes are linked together near their 5' ends1 by a precise, but poorly understood, set of noncovalent interactions. These interactions probably involve a mixture of base pairing and tertiary interactions. The structure of this RNA 'dimer' is important in several stages of the retroviral infectivity cycle, including RNA encapsidation into nascent viral particles2, 3, 4, 5 and recombination during reverse transcription3, 6, 7. Retroviruses are valuable biotechnology tools (as gene therapy vectors) and are also the causative agents of serious diseases, including acquired immunodeficiency syndromes and several cancers. Understanding the mechanism of retroviral dimerization at a molecular level thus represents an important opportunity both to enhance vector function and to disrupt the infectivity of pathogenic viruses. Because the retroviral dimerization sequences within a virus family are often very similar8, 9, 10, 11, a consensus secondary structure cannot be inferred by phylogenetic covariation analysis, which is the most robust method to determine the secondary structure for a large RNA12, 13, 14. Secondary structure models for retroviral RNA dimerization domains are still provisional and probably only partially encompass the biological function of these RNAs. Determining the biologically relevant structure of the dimerization domain for any retroviral RNA thus represents a common problem in biology. The challenge is to understand an RNA secondary structure in enough detail to be able to formulate hypotheses about biological function, even though only one or a few highly similar sequences are known. Algorithms for predicting an RNA secondary structure from a single sequence identify roughly 50−70% of known helices correctly15, 16. However, prediction accuracy for a single RNA or for any helix within a larger structure is not known in advance. Incorrect prediction of even a few helices in a functionally important region makes it difficult or impossible to develop robust biological models. Among the gamma retroviruses, several sequences have been consistently proposed as important for dimerization of the RNA genome8, 9, 10 (summarized in Fig. 1a). PAL1 (also known as the 204−227 stem-loop17, DIS1 (ref. 18) or SL-B' (ref. 5)) and PAL2 (also known as DIS2 (ref. 18), SL-B19 or H1 (ref. 20)) are postulated to form hairpin loops (in green, Fig. 1). PAL1 and PAL2 span self-complementary ('palindromic') sequences and are conventionally proposed to interact through loop-loop interactions with PAL1 and PAL2 sequences from a second RNA, eventually forming extended duplexes in the dimer10, 17, 18, 19. Highly conserved GACG tetraloops9 (Fig. 1a) in stem-loops 1 and 2 (SL1 and SL2, also known as SL-C19 or H2 (ref. 20) and SL-D or H3, respectively), have the potential to form stable loop-loop interactions through cross-loop G-C base pairs19, 21 and seem to be important for packaging via interactions with the viral Gag protein22. Figure 1: 5'-untranslated region of MuSV. Conserved sequences and conventionally proposed secondary structures are illustrated schematically. The 5' genomic RNA cap is position 1. (a) 5' and 3' truncation mutants are shown in blue and red, respectively. (b) PAL2 mutant. Full figure and legend (7K) Figures, schemes & tables index We developed a generalizable approach for obtaining a well-constrained secondary structure for a retroviral dimerization domain and for many other classes of RNA. We first used competition experiments to rigorously define a contiguous minimal dimerization active sequence (MiDAS) for a representative gamma retrovirus, the Moloney murine sarcoma virus (MuSV). We then used a new chemical approach, selective 2'-hydroxyl acylation analyzed by primer extension (SHAPE)23, 24, to obtain comprehensive, quantitative, nucleotide-resolution and model-independent constraints for the secondary structure of the monomeric starting state of the retroviral dimerization domain. The results of these experiments emphasize that existing structural models for the dimerization domain in gamma retroviruses require reinterpretation. Most notably, a large region in the dimerization domain is conformationally flexible, which may facilitate retroviral RNA dimerization by decreasing the energetic cost of disrupting base pairing or other interactions in the monomer before the formation of functional structures specific to the dimer. Top of page Results Rigorous definition of a MiDAS In vitro studies using synthetic RNA transcripts have been essential for identifying candidate structures that contribute to dimerization in the gamma retroviruses8, 10, 17, 18, 19, 20, 25, 26, 27, 28, 29 . A significant challenge in interpreting these experiments is that most RNAs containing a stem-loop structure will dimerize if dimerization reactions are performed under conditions of sufficiently high RNA concentration or ionic strength30, 31. In exploratory experiments, we identified a roughly physiological ion environment—50 mM HEPES, 200 mM potassium acetate, and 5 mM MgCl2 (pH 7.5)—that yields well- behaved single-conformation monomer and dimer complexes for an RNA spanning most of the MuSV 5' untranslated region (Fig. 1a). This RNA spans all structures previously proposed to participate in dimerization in the gamma retroviruses. We used this RNA to impose a functional threshold in competitive dimerization experiments that, by design, strongly discriminates against promiscuous self-dimerization (Fig. 2). Figure 2: Competitive-dimerization assay for stringent definition of RNA structures essential for dimerization. Assay uses radiolabeled mutant RNA (mut; in red, with asterisk) and unlabeled full- length RNA (FL; gray). Only species containing the mutant RNA are visualized in the nondenaturing gel. Full figure and legend (14K) Figures, schemes & tables index We constructed an extensive series of viral sequences containing systematic truncations from their 5' and 3' ends (blue and red arrows, respectively, Fig. 1a). Competitive dimerization experiments were performed at 60 °C in the presence of the full-length transcript and visualized by the selective detection of the radiolabeled, truncated RNA variants in nondenaturing gels (Fig. 2). Both the radiolabeled mutant-mutant homodimer and mutant−full-length heterodimer are visualized directly. Full-length RNA homodimers also form but are not radiolabeled and thus are not observed. We scored as structurally competent dimerization active sequences only those mutant RNAs that quantitatively competed with homodimerization by the (unlabeled) full-length RNA. RNAs that only homodimerized or only heterodimerized (see middle two lanes in Fig. 2) were scored as structurally deficient. A minimal sequence active in dimerization Truncation mutants were identified by the 5' or 3' nucleotide (nt) at which the mutant sequence terminates (Fig. 1a). Competitive dimerization experiments were performed with 1.5 nM radiolabeled, truncated RNA and 1, 5 or 15 nM full-length RNA (Fig. 3). Markers for the mutant monomer and for the mutant homodimer were obtained by omitting the heating step or omitting FL RNA, respectively. Figure 3: The MiDAS for MuSV defined by competitive dimerization. (a,b) 3'-end truncations in the native and PAL2 contexts. (c,d) 5'-end truncations. (e) Schematic structure for the MiDAS in the context of the conventional secondary structure. (f) Efficient dimerization of the MiDAS in both native and PAL2 contexts. FL, full-length; M, mutant monomer; HED, mutant-FL heterodimer; HOD, mutant- mutant homodimer. Full figure and legend (51K) Figures, schemes & tables index The progressively larger 3' truncations, 3'-479, 3'-419 and 3'-374 (Fig. 3a), yielded native-like RNAs that both homo- and heterodimerized efficiently with the full-length RNA. In contrast, the 3'-339 truncation heterodimerized efficiently with the full-length RNA, but formed almost no homodimer (3'-339 panel, Fig. 3a). Thus, the 3'-339 RNA is deficient in dimerization in a way that can be rescued by the full-length RNA. This RNA also forms several monomeric conformations. Similarly, the 3'-354 truncation forms heterogeneous monomers and also forms heterodimers inefficiently (Supplementary Fig. 1 online). The 3' boundary for the MiDAS lies between nt 354 and 374; the largest fully functional deletion spans position 374 (3'-374 panel, Fig. 3a). Truncations from the 5' end through position 276 yield RNAs that correctly homo- and heterodimerized (Fig. 3c). In contrast, truncation through position 295 yields an RNA that homodimerized well but was incompetent at forming heterodimers with the full- length RNA (5'-295 panel in Fig. 3c). Further truncation through 5'-325 yields an RNA that formed neither homo- nor heterodimers. The 5' boundary for the minimum dimerization active sequence was thus set at position 276. The behavior of the 5'-295 mutant illustrates the stringency of the competitive dimerization assay. Although this RNA would have been scored as dimerization competent under less stringent conditions, it clearly lacks key elements required to dimerize competitively with the full-length RNA. Deletion of PAL2 unmasks the contribution of PAL1 The minimal dimerization active region defined by this initial analysis includes the PAL2 sequence, which several groups8, 18, 19, 20, 25, 26, 27, 32 have proposed plays a role in dimerization. To explore whether any other accessory sequences contribute to dimerization but are masked by PAL2, we compromised PAL2 by removing nt 283−294 ( PAL2 mutant, Fig. 1b) and retested the panel of 5' and 3' truncations by competition with a full-length RNA also harboring the PAL deletion. All PAL2 3' truncations through nt 374 formed both homo- and heterodimers efficiently (Fig. 3b); this indicated that the 3' boundary of the MiDAS remains at position 374. When the 5' series was analyzed, truncations through nt 205 yielded fully functional RNAs. In contrast, both homo- and heterodimer formation was significantly impaired when the RNA was truncated through 5'-235 in the PAL2 context (5'-235 panel, Fig. 3d). In addition, time-resolved dimerization experiments showed that a construct spanning the PAL2 through SL2 sequences dimerized only to about 80% (at concentrations up to 50 nM). In contrast, RNAs that also included the additional 5' 205−275 sequences dimerized to completion (Fig. 3f and data not shown). We concluded that the role of sequences containing the PAL1 region (nt 205−217) is partially masked if PAL2 is present, and we assigned the 5' boundary of the minimal dimerization domain to position 205 (Fig. 3d). A minimal dimerization domain We constructed a minimal RNA spanning MuSV sequences from 205 to 374 and tested the ability of this RNA to function in the competitive dimerization assay, both in the native PAL2 and PAL2 contexts (Fig. 3f). Both RNAs formed single-conformation monomers, homodimers and heterodimers identical to those observed for native-like truncations. We infer that nt 205−374 span the minimal dimerization domain for MuSV and, potentially, for most gamma retroviruses. Our MiDAS (Fig. 3e) incorporates RNA elements that have been proposed to either contribute or be a primary determinant for retroviral dimerization8, 10, 17, 18, 19, 20, 21, 25, 33. The competitive dimerization assay supports proposals17, 18 that the role of the PAL1 sequence may be especially important if the PAL2 sequence is compromised. Unique to the competitive dimerization assay is the strong inference that the MiDAS spans all RNA sequences stringently required for dimerization. The dimerization sequences defined here in vitro correspond closely to those from analogous experiments designed to define a MiDAS in vivo: an RNA spanning positions 215 to 404 from the Moloney murine leukemia virus (MuLV) is sufficient to increase packaging of a nonviral RNA, in dimeric form, by 50-fold5. To the extent that other viral components such as Gag or the nucleocapsid protein augment, but do not fundamentally alter, an RNA-centered process, the MiDAS represents a rigorously evaluated minimal domain for retroviral dimerization. Domain structure analyzed by RNA SHAPE chemistry Our laboratory has recently developed a single-nucleotide-resolution approach to examine, quantitatively, the local environment at every nucleotide in an RNA23, 24. RNA ribose 2'-hydroxyl groups react with N-methylisatoic anhydride (NMIA, Compound 1) to form the nucleotide 2'-ester. 2'-Hydroxyl reactivity is gated by whether or not a given nucleotide is constrained by base pairing or tertiary interactions23, 34. Flexible nucleotides react preferentially because they are better able to reach a conformation that facilitates nucleophilic attack of the 2'-hydroxyl on NMIA (Scheme 1). Formation of the bulky 2'- O-adduct is readily detected as a stop to reverse transcriptase−mediated primer extension: the complete experiment involves selective 2'-hydroxyl acylation followed by primer extension. Scheme 1: Structure-selective reaction of RNA 2'-hydroxyl groups with NMIA (1). Full scheme and legend (3K) Figures, schemes & tables index SHAPE experiments were performed in the context of a MiDAS RNA with 30- and 5-nt native sequence extensions, respectively, at the 5' and 3' ends to facilitate analysis of the entire domain by primer extension. We also appended an RNA cassette23 to the 3' end that contains an efficient DNA primer binding site. Refolded MiDAS RNA was treated with NMIA, and sites of 2'-O-adduct formation were detected by primer extension, resolved on sequencing gels (Fig. 4a). The monomeric state of the RNA was confirmed by native gel analysis. Comparison of reactions performed in the presence of NMIA with reactions omitting the reagent reveals selective formation of 2'-O-adducts at a subset of sites in the RNA (compare + and - NMIA lanes in the MiDAS panel, Fig. 4). Individual band intensities were integrated35 and absolute reactivities were computed for every position in the MiDAS RNA construct. Figure 4: SHAPE analysis of the MuSV MiDAS RNA and of the PALSTB and 289−300 mutants. (a) 2'-O-adduct formation visualized in a sequencing gel. Reactions were performed in the presence (+) and absence (-) of NMIA. Sequencing lanes (seq) showing guanosine positions were generated by dideoxy nucleotide incorporation during primer extension. Extension products in the dideoxy sequencing ladder are exactly 1 nt longer than those in the corresponding NMIA lane; nucleotide positions are labeled with respect to NMIA lanes. (b) Sequences of (left to right) the native RNA and of the PALSTB and 289−300 mutants, drawn in the context of the conventional structure for PAL2. Full figure and legend (76K) Figures, schemes & tables index The nucleotide-resolution SHAPE experiment provides a large number of constraints that must be accommodated in any secondary structure prediction for the MiDAS RNA. We screened secondary structures for the MiDAS region (residues 205−374) by submitting positions whose calculated reactivity was at least 25% of the strongest observed reactivities (47 nt total) as chemical modification constraints to the RNAStructure program15. The quantitative data are shown superimposed on a secondary structure consistent with the entire body of SHAPE reactivity in Figure 5. Figure 5: Secondary structure model of the MuSV MiDAS RNA. Left, quantitative NMIA reactivities, minus background, are superimposed as columns at each nucleotide position. Right shows the same secondary structure, but with residues labeled explicitly. Full figure and legend (26K) Figures, schemes & tables index Residues with high and moderate reactivity (red and orange, Fig. 5) toward NMIA are located in single-stranded loops and connecting structures. Positions with low or undetectable reactivity (blue and black, Fig. 5) lie largely in base-paired helices. Because SHAPE is sensitive to any interaction that constrains a nucleotide23, including noncanonical interactions, reactive positions should fall cleanly in flexible RNA structures, whereas some unreactive nucleotides may reflect tertiary structure constraints that are yet to be defined at this stage of analysis. As expected23, 24, the 3' RNA structure cassette (Fig. 5, left) has a reactivity pattern exactly consistent with its designed fold, indicating that this appended structure does not interfere with folding of the MiDAS RNA. SHAPE analysis (Figs. 4 and 5) strongly supports secondary structures for the PAL1 loop, SL2 and the upper portion of SL1 that are consistent with earlier proposals. In contrast, the PAL2 sequence (positions 283−303; see Fig. 3e for comparison), which has been almost universally assumed to form a stable stem-loop structure, is highly reactive toward NMIA. Moreover, the reactive PAL2 sequence resides in the middle of a larger flexible domain in which most nucleotides are reactive by SHAPE chemistry (Fig. 5). PAL2 is unstructured in the MiDAS monomer Because the observed structure is significantly different from conventional models for the dimerization domain, we analyzed the structure of two MiDAS RNAs carrying instructive mutations in PAL2 (Fig. 4b). The first mutant, PALSTB (PAL stabilization), was designed to stabilize PAL2 in the conventional stem-loop structure by increasing the G-C base-pair content at flanking helix positions (circled positions, Fig. 4b). Inspection of the SHAPE data shows that stabilizing the PAL2 duplex has the desired effect. Nucleotides located in the PAL2 loop are strongly reactive, whereas base-paired positions in the stem are now much less reactive than in the native sequence (compare the MiDAS and PALSTB lanes, Fig. 4a). The experimental SHAPE reactivity data for the PALSTB mutant was subtracted from that for the native MiDAS RNA to create a quantitative difference map for every position in the PALSTB RNA (Fig. 6a). In the difference map, residues that are more reactive or more constrained in the mutant relative to the native MiDAS sequence are reported as positive and negative amplitudes, respectively (red and blue, Fig. 6). If PAL2 already existed as a hairpin in the monomeric native state, stabilizing this stem should have a minimal effect on global MiDAS RNA structure. In strong contrast to this expectation, stabilizing the PAL2 sequence as a stem-loop causes large changes to the SHAPE reactivity in the MiDAS domain. Figure 6: Quantitative difference maps for the effects of mutations on MiDAS structure. (a) PALSTB mutation. (b) 289−300 mutation. Vertical bars show absolute NMIA reactivities at each base position for the mutant RNA minus reactivity of the native RNA. Positive (red) and negative (blue) differences indicate increased versus reduced reactivity, respectively, in the mutants relative to the native MiDAS. Vertical scales are the same in upper and lower panels; the overall shorter bars in the lower panel reflect the more modest structural perturbation introduced by the 289−300 mutant. Full figure and legend (24K) Figures, schemes & tables index Consistent with the design of this mutant, nucleotides in the loop at the apex of the PAL2 stem in the PALSTB mutant show an increase in reactivity, and nucleotides in the stabilized stem are much less reactive than in the native MiDAS sequence (see PAL2, Fig. 6a). More significantly, the PALSTB mutant shows very large changes in global structure that extend almost the entire length of the RNA and up to 80 nt away (see especially nt 252−268 in the flexible domain and nt 374−382 between SL2 and the 3' end of the RNA, Fig. 6a). The peaks shown in the difference map are plotted on a scale comparable to that used in Fig. 5. Thus, the large positive peaks centered at positions 255, 312, 365 and 380 represent significant enhancements in absolute local nucleotide flexibility in these regions (for example, compare the 255 region in Fig. 4 with the difference map in Fig. 6a). Our second mutant, 289−300, was designed to delete a large region within PAL2 that is flexible as judged by SHAPE chemistry (mutant is shown in Fig. 4b; flexible region is labeled in Fig. 5). If the conventional stem-loop model for PAL2 were correct, this 12-nt deletion should have a strong effect on global MiDAS structure. On the other hand, if PAL2 is unstructured in the native monomer state, as indicated by SHAPE chemistry, then this extensive deletion may have only a minimal effect on global MiDAS structure. A difference map for the 289−300 mutant shows that this deletion, in fact, introduces very modest changes to local nucleotide flexibility in the MiDAS RNA and induces virtually no significant structural change over large regions of the sequence (Fig. 6b). The most significant effect is an increase in SHAPE reactivity in the (already flexible) 267 region of the flexible domain and a decrease in reactivity at the bulge in SL1 at nt 340. Thus, direct analysis of the MiDAS RNA (Fig. 5) and differential analysis of two mutants (Fig. 6) both strongly support a new model for the dimerization domain of MuSV in which the PAL2 sequence resides in an extensive flexible domain. Top of page Discussion Our model for the minimal dimerization active structure of MuSV makes use of two innovations that are generalizable to any RNA structure prediction problem. First, we defined a minimal sequence for dimerization using an assay that requires the simplified RNA to functionally compete with a native-like sequence (Fig. 2). Second, SHAPE chemistry quantitatively interrogates every nucleotide in an RNA (Scheme 1), which means that secondary structure models can be evaluated with much greater confidence than when traditional chemical and enzymatic reagents are used. Although the MiDAS secondary structure (Fig. 5) proposed here differs significantly from conventional models, this structure is consistent with two earlier sets of experimental information for dimerization domains in gamma retroviruses. The sequence of MuLV is almost identical to that of MuSV. The nucleotide resolution SHAPE information strongly supports the original MuLV model8 for SL2 and the upper portion of SL1. In contrast, SHAPE does not support the earlier proposal that PAL2 forms a stable stem-loop structure. However, superposition of the chemical mapping information for MuLV on our secondary structure for MuSV shows that the previous information is exactly consistent with the present MiDAS proposal (Supplementary Fig. 2 online). In particular, the PAL2 sequence is strongly reactive toward conventional single- strand−selective chemical reagents8 and is thus consistent with the idea that PAL2 lies in a flexible domain. Structure-mapping studies on the Harvey sarcoma virus (HaSV) also emphasize the importance of SL1- and SL2-like structures in the dimerization domain28. HaSV does not contain a PAL2 sequence9, 28, but the HaSV RNA can be folded into a secondary structure that is similar to that of MuSV and for which RNase-based cleavage data strongly support formation of an internal flexible domain (Supplementary Fig. 2 online). Many sites in the HaSV domain are cleaved by both single- and double-strand−selective RNases28 (Supplementary Fig. 2 online). We infer that the HaSV RNA probably contains a flexible domain in which portions of the structure are alternately both paired and flexible in distinct conformations. That the MuLV and HaSV RNAs fold to similar monomeric starting structures provides a structural basis for the observation that these viruses readily heterodimerize28, presumably via PAL1. We folded the flexible domain, including its anchoring helix (spanning positions 231−315, Fig. 5), subject to the requirement that the 27 positions with high and moderate reactivities be single-stranded. The lowest-energy structure, which is compatible with all of the SHAPE information (Fig. 5), has a total calculated15 folding free energy of only - 10.5 kcal/mol. This single low-energy structure spans 84 nt and thus has a net stability comparable to that of a simple stem-loop structure containing roughly three base pairs. Moreover, although the entire flexible domain from nt 249 through 294 contains no instances in which there are more than two strongly constrained nucleotides in a row (black positions, Fig. 5), individual nucleotides vary significantly in their 2'-hydroxyl reactivity. We therefore evaluated the alternative hypothesis that several more stable structures might be compatible with the SHAPE information. We submitted the 12 most highly reactive sites (red, Fig. 5) as chemical modification constraints for RNA structure prediction15. Four structures have calculated free energies within 10% of the most stable structure. Three of these structures have distinctive folds (Fig. 7), whereas the fourth (not shown) contains elements of the other structures. Absolute SHAPE reactivities were superimposed on the three most distinctive structures (Fig. 7). These intensity data are colored according to the fraction of structures in which they are single-stranded (always paired and always single-stranded are black and white, respectively). Each structure (A, B or C) is only partially consistent with the SHAPE data. Each of these structures, however, has a calculated folding free energy of approximately -17 kcal/mol, and thus is significantly more stable than the single consensus structure that incorporates all of the flexibility information. Figure 7: Structural model for overall flexibility in the 231−315 domain. Column heights indicate absolute NMIA reactivities (and are the same as reported in Fig. 5). Column colors illustrate the fraction of structures in which a position is single- stranded in this ensemble of three representative structures. Full figure and legend (27K) Figures, schemes & tables index This semiquantitative analysis, in which the large universe of possible structures is approximated by three low-energy structures (Fig. 7), supports a general model for RNA folding in which net, long-range flexibility in RNA can reflect contributions from several stable structures of similar energies. Models for the genomic RNA retroviral dimerization domain, in which PAL2 forms a stable stem-loop structure (Figs. 1a and 4b), have guided the gamma retrovirus field for over a decade. However, the model-independent SHAPE intensity information (Figs. 4a and 5) emphasizes that existing structural models merit careful reinterpretation. For example, many mechanistic analyses of retroviral dimerization have used simplified RNAs or RNAs in which PAL2 was mutated to enforce the conventional structure (Fig. 1a) for the RNA. The RNA structure in such mutants should be roughly similar to that of the PALSTB mutant, which was also designed to artificially reinforce the conventional structure for PAL2. The PALSTB mutant yielded extensive changes to the global structure of the MiDAS, including at residues up to 80 nt distant from PAL2 (Fig. 6a). Thus, the structure of the retroviral dimerization domain with a native sequence can be quite different from that of RNAs containing mutations in PAL2 or that are shorter than the MiDAS. Moreover, the significant global changes that occur in the MiDAS domain upon introduction of mutations in PAL2 emphasize that the flexible domain communicates via long-range interactions with other regions of the RNA. Because PAL2 (and PAL1) sequences are self-complementary, an attractive model for the noncovalent interactions that stabilize the retroviral dimer is for these sequences to form extended duplexes in the dimer10, 17, 18, 19. Earlier models proposing that PAL2 initially exists as a stable stem-loop recognized that it might be energetically costly to disrupt the extensive pre-existing base pairing in this structure. These models thus generally proposed that dimerization proceeds stepwise, via base pairing between nucleotides in the loops of two PAL2 stem-loop structures followed by helix extension. The nucleotide-resolution SHAPE experiments demonstrate that the PAL2 sequence instead lies in an RNA domain in which, on average, most nucleotides either are unconstrained by base pairing or are transiently in a single-stranded conformation (Fig. 7). Continuing SHAPE experiments do support formation of extended duplexes in the dimer (data not shown). Thus, formation of a flexible domain in the monomeric starting state has significant mechanistic implications for retroviral RNA genome dimerization. (i) Dimerization via PAL2, in the context of a flexible domain, is potentially much more thermodynamically favorable than previously thought, because fewer base-pair interactions in the monomer have to be disrupted to form extended duplexes between two PAL2 sequences in the dimer. (ii) Placing the PAL2 sequence in a flexible domain may also kinetically enhance retroviral RNA dimerization by lowering the activation barrier for extended duplex formation. (iii) The distinct conformations visualized for the flexible domain probably also have different dimerization activities. Retroviral dimerization could thus be modulated by interactions between these distinct conformations and other regions of the genomic RNA or retroviral proteins. Top of page Methods Retroviral RNA transcripts. DNA templates for in vitro transcription of the full-length RNA, 5' and 3' truncations, and MiDAS constructs were generated by PCR from the pLNBS26, 27 plasmid. RNA constructs were generated with T7 RNA polymerase−mediated transcription (500 l, 37 °C, 5 h) containing 80 mM HEPES (pH 7.4), 40 mM dithiothreitol (DTT), 0.01% (v/v) Triton X-100, 2 mM spermidine, 10 mM MgCl2, 2 mM each nucleoside triphosphate, 25 g of PCR-generated template, 20 U of SUPERase-In (Ambion) and 0.1 mg/ml of polymerase. Internally labeled RNAs were synthesized with 20 Ci of -[32P]ATP and unlabeled ATP at 0.5 mM. RNAs were purified by denaturing gel electrophoresis (5% polyacrylamide, 7 M urea), excised from the gel, eluted overnight into 1/2 TBE (45 mM Tris-borate, 1 mM EDTA) and concentrated by ethanol precipitation. RNAs were resuspended in 10 mM HEPES (pH 7.5) and 1 mM EDTA and stored at -20 °C. Competitive dimerization assay. Truncated RNA internally labeled with [32P] ( 1.5 nM) was incubated with unlabeled full-length RNA (at 1, 5 or 15 nM in 15 l). Reactions were heated to 90 °C to eliminate pre-existing dimers, rapidly cooled on ice, treated with 5 l 4 dimerization buffer (200 mM HEPES (pH 7.5), 800 mM potassium acetate (pH 7.5), 20 mM MgCl2 at 25 °C), incubated at 60 °C for 30 min and placed on ice. Samples (3 l) were mixed with 1 l of 30% (v/v) glycerol (containing marker dyes) and resolved by nondenaturing electrophoresis at 4 °C. Gels (5% polyacrylamide in TBE) were pre-run for 15 min before sample loading and subsequently run for 2 h at 20 W. SHAPE analysis of MuSV monomers. SHAPE experiments were performed with a MiDAS RNA that contained flanking 5' and 3' extensions of viral sequence of 30 and 5 nt, respectively; a 3' nonviral RNA cassette containing an efficient DNA primer binding site23 was appended to the 3' end. The MiDAS RNA construct (10 pmol) was heated at 90 °C for 3 min in 7.2 l of water, cooled on ice, treated with 1.8 l of 5 dimerization buffer (250 mM HEPES (pH 8.0), 1 M potassium acetate (pH 7.5), 25 mM MgCl2), incubated at room temperature ( 25 °C) for 30 s and returned to ice. The RNA solution was then equilibrated at 37 °C for 5 min, treated with NMIA (1 l, 180 mM in anhydrous DMSO), allowed to react for 50 min (approximately five half lives23, 24) at 37 °C, and placed on ice. Control reactions contained DMSO without NMIA. Primer extension. Two DNA primers were used to analyze the MiDAS RNA construct. Primers were complementary to the 3' end of the RNA structure cassette (5'-GAA CCG GAC CGA AGC CCG) and to SL1 (5'-CAG AAC TCG TCA GTT CCA CCA). We performed primer extension reactions by adding modified RNA (2 l, 2 pmol) and 5'-[32P]DNA primer (1 l, 1 pmol) to 9 l water and annealing by incubation at 95 °C (30 sec), 60 °C (6 min) and 35 °C (10 min). Reverse transcription buffer (7 l; 143 mM Tris (pH 8.3), 214 mM KCl, 7.14 mM MgCl2, 1.43 mM each dNTP, 14.3 mM DTT) was added, and subsequent primer extension steps were performed exactly as described23, except that primer extension was performed at 48.5 °C. cDNA fragments were resolved on a series of 8% (w/v) polyacrylamide gels to achieve nucleotide resolution throughout the region analyzed. Accession codes. BIND identifier (http://bind.ca/): 295526. Note: Supplementary information is available on the Nature Chemical Biology website. Top of page Acknowledgments This work was supported by a grant from the US National Institutes of Health (GM64803 to K.M.W. and A. Kaplan). We are indebted to A. Kaplan, C. Gherghe and A. Rein for many helpful discussions; to E. Merino and K. Wilkinson for assistance with SHAPE chemistry; and to D. Mathews for extensive advice with the RNAStructure program. Competing interests The authors declare that they have no competing financial interests. Top of page References 1. Murti, K.G., Bondurant, M. & Tereba, A. Secondary structural features in the 70S RNAs of Moloney murine leukemia and Rous sarcoma viruses as observed by electron microscopy. J. Virol. 37, 411–419 (1981). 2. Darlix, J.L., Lapadat-Tapolski, M., de Rocquigny, H. & Roques, B.P. First glimpses at structure-function relationships of the nucleocapsid protein of retroviruses. J. Mol. Biol. 254, 523–537 (1995). | Article | 3. Paillart, J. et al. A dual role of the putative RNA dimerization initiation site of human immunodeficiency virus type 1 in genomic RNA packaging and proviral DNA synthesis. J. Virol. 70, 8348–8354 (1996). 4. Laughrea, M. et al. Mutations in the kissing-loop hairpin of human immunodeficiency virus type 1 reduce viral infectivity as well as genomic RNA packaging and proviral DNA synthesis. J. Virol. 71, 3397–3406 (1997). 5. Hibbert, C.S., Mirro, J. & Rein, A. mRNA molecules containing murine leukemia virus packaging signals are encapsidated as dimers. J. Virol. 78, 10927–10938 (2004). | Article | 6. Mikkelsen, J.G., Lund, A.H., Duch, M. & Pedersen, F.S. Recombination in the 5' leader of murine leukemia virus is accurate and influenced by sequence identity with a strong bias toward the kissing-loop dimerization region. J. Virol. 72, 6967– 6978 (1998). 7. Mikkelsen, J.G., Lund, A.H., Duch, M. & Pedersen, F.S. Mutations of the kissing- loop dimerization sequence influence the site specificity of murine leukemia virus recombination in vivo. J. Virol. 74, 600–610 (2000). | Article | 8. Tounekti, N. et al. Effect of dimerization on the conformation of the encapsidation psi domain of the Moloney murine leukemia virus. J. Mol. Biol. 223, 205–220 (1992). | Article | 9. Konings, D.A.M., Nash, M.A., Maizel, J.V. & Arlinghaus, R.B. Novel GACG- hairpin pair motif in the 5' untranslated region of type C retroviruses related to murine leukemia virus. J. Virol. 66, 632–640 (1992). 10. Paillart, J., Marquet, R., Skripkin, E., Ehresmann, C. & Ehresmann, B. Dimerization of retroviral genomic RNAs: structural and functional implications. Biochimie 78, 639–653 (1996). | Article | 11. D'Souza, V., Dey, A., Habib, D. & Summers, M.F. NMR Structure of the 101- nucleotide core encapsidation signal of the Moloney murine leukemia virus. J. Mol. Biol. 337, 427–442 (2004). | Article | 12. Michel, F. & Westhof, E. Modelling of the three-dimensional architecture of group I catalytic introns based on comparative sequence analysis. J. Mol. Biol. 216, 585–610 (1990). | Article | 13. Frank, D.N. & Pace, N.R. Ribonuclease P: unity and diversity in a tRNA processing ribozyme. Annu. Rev. Biochem. 67, 153–180 (1998). | Article | 14. Gutell, R.R., Lee, J.C. & Cannone, J.J. The accuracy of ribosomal RNA comparative structure models. Curr. Opin. Struct. Biol. 12, 301–310 (2002). | Article | 15. Mathews, D.H. et al. Incorporating chemical modification constraints into a dynamic programming algorithm for prediction of RNA secondary structure. Proc. Natl. Acad. Sci. USA 101, 7287–7292 (2004). | Article | 16. Dowell, R.D. & Eddy, S.R. Evaluation of several lightweight stochastic context- free grammars for RNA secondary structure prediction. BMC Bioinformatics 5, 71 (2004). | Article | 17. Oroudjev, E.M., Kang, P.C.E. & Kohlstaedt, L.A. An additional dimer linkage structure in Moloney murine leukemia virus RNA. J. Mol. Biol. 291, 603–613 (1999). | Article | 18. Ly, H. & Parslow, T.G. Bipartite signal for genomic RNA dimerization in Moloney murine leukemia virus. J. Virol. 76, 3135–3144 (2002). | Article | 19. D'Souza, V. et al. Identification of a high affinity nucleocapsid protein binding element within the Moloney murine leukemia virus -RNA packaging signal: implications for genome recognition. J. Mol. Biol. 314, 217–232 (2001). | Article | 20. De Tapia, M., Metzler, V., Mougel, M., Ehresmann, B. & Ehresmann, C. Dimerization of the Moloney murine leukemia virus genomic RNA: redefinition of the role of the palindromic stem-loop H1 (278–303) and new roles for stem- loops H2 (310–352) and H3 (355–374). Biochemistry 37, 6077–6085 (1998). | Article | 21. Kim, C. & Tinoco, I. A retroviral RNA kissing complex containing only two GC base pairs. Proc. Natl. Acad. Sci. USA 97, 9396–9401 (2000). | Article | 22. Rein, A., Harvin, D.P., Mirro, J., Ernst, S.M. & Gorelick, R.J. Evidence that a central domain of nuclecapsid protein is required for RNA packaging in murine leukemia virus. J. Virol. 68, 6124–6129 (1994). 23. Merino, E.J., Wilkinson, K.A., Coughlan, J.L. & Weeks, K.M. RNA structure analysis at single nucleotide resolution by selective 2'-hydroxyl acylation and primer extension (SHAPE). J. Am. Chem. Soc. 127, 4223–4231 (2005). | Article | 24. Wilkinson, K.A., Merino, E.J. & Weeks, K.M. RNA SHAPE chemistry reveals non-hierarchical interactions dominate equilibrium structural transitions in tRNAAsp transcripts. J. Am. Chem. Soc. 127, 4659–4667 (2005). | Article | 25. Girard, P.M., Bonnet-Mathoniere, B., Muriaux, D. & Paoletti, J. A short autocomplementary sequence in the 5' leader region is responsible for dimerization of MoMuLV genomic RNA. Biochemistry 34, 9785–9794 (1995). | Article | 26. Ly, H., Nierlich, D., Olsen, J. & Kaplan, A. Moloney murine sarcoma virus genomic RNAs dimerize via a two-step process: a concentration-dependent kissing-loop interaction is driven by initial contact between consecutive guanosines. J. Virol. 73, 7255–7261 (1999). 27. Ly, H., Nierlich, D., Olsen, J. & Kaplan, A. Functional characterization of the dimer linkage structure RNA of Moloney murine sarcoma virus. J. Virol. 74, 9937–9945 (2000). | Article | 28. Rasmussen, S.V., Mikkelsen, J.G. & Pedersen, F.S. Modulation of homo- and heterodimerization of Harvey sarcoma virus RNA by GACG tetraloops and point mutations in palindromic sequences. J. Mol. Biol. 323, 613–628 (2002). | Article | 29. Aagaard, L., Rasmussen, S.V., Mikkelsen, J.G. & Pedersen, F.S. Efficient replication of full-length murine leukemia viruses modified at the dimer initiation site regions. Virology 318, 360–370 (2004). | Article | 30. Holbrook, S.R., Cheong, C., Tinoco, I., Jr. & Kim, S.H. Crystal structure of an RNA double helix incorporating a track of non-Watson-Crick base pairs. Nature 353, 579–581 (1991). | Article | 31. Wild, K., Weichenrieder, O., Leonard, G.A. & Cusack, S. The 2 Å structure of helix 6 of the human signal recognition particle RNA. Struct. Fold. Des. 7, 1345– 1352 (1999). | Article | 32. D'Souza, V. & Summers, M.F. Structural basis for packaging the dimeric genome of Moloney murine leukemia virus. Nature 431, 586–590 (2004). | Article | 33. Monie, T.P. et al. Identification and visualization of the dimerization initiation site of the prototype lentivirus, Maedi Visna virus: a potential GACG tetraloop displays structural homology with the - and -retroviruses. Biochemistry 44, 294–302 (2005). | Article | 34. Chamberlin, S.I., Merino, E.J. & Weeks, K.M. Catalysis of amide synthesis by RNA phosphodiester and hydroxyl groups. Proc. Natl. Acad. Sci. USA 99, 14688– 14693 (2002). | Article | 35. Das, R., Laederach, A., Pearlman, S.M., Herschlag, D. & Altman, R.B. SAFA: Semi-automated footprinting analysis software for high-throughput quantification of nucleic acid footprinting experiments. RNA 11, 344–354 (2005). | Article | Top of page 1. Department of Chemistry, University of North Carolina, Chapel Hill, North Carolina 27599-3290, USA. 2. Email: email@example.com Correspondence to: Kevin M Weeks1 Email: firstname.lastname@example.org Conserved sequences and conventionally proposed secondary structures are illustrated schematically. The 5' genomic RNA cap is position 1. (a) 5' and 3' truncation mutants are shown in blue and red, respectively. (b) PAL2 mutant. Assay uses radiolabeled mutant RNA (mut; in red, with asterisk) and unlabeled full- length RNA (FL; gray). Only species containing the mutant RNA are visualized in the nondenaturing gel. (a,b) 3'-end truncations in the native and PAL2 contexts. (c,d) 5'-end truncations. (e) Schematic structure for the MiDAS in the context of the conventional secondary structure. (f) Efficient dimerization of the MiDAS in both native and PAL2 contexts. FL, full-length; M, mutant monomer; HED, mutant-FL heterodimer; HOD, mutant- mutant homodimer. (a) 2'-O-adduct formation visualized in a sequencing gel. Reactions were performed in the presence (+) and absence (-) of NMIA. Sequencing lanes (seq) showing guanosine positions were generated by dideoxy nucleotide incorporation during primer extension. Extension products in the dideoxy sequencing ladder are exactly 1 nt longer than those in the corresponding NMIA lane; nucleotide positions are labeled with respect to NMIA lanes. (b) Sequences of (left to right) the native RNA and of the PALSTB and 289−300 mutants, drawn in the context of the conventional structure for PAL2. Left, quantitative NMIA reactivities, minus background, are superimposed as columns at each nucleotide position. Right shows the same secondary structure, but with residues labeled explicitly. (a) PALSTB mutation. (b) 289−300 mutation. Vertical bars show absolute NMIA reactivities at each base position for the mutant RNA minus reactivity of the native RNA. Positive (red) and negative (blue) differences indicate increased versus reduced reactivity, respectively, in the mutants relative to the native MiDAS. Vertical scales are the same in upper and lower panels; the overall shorter bars in the lower panel reflect the more modest structural perturbation introduced by the 289−300 mutant. Column heights indicate absolute NMIA reactivities (and are the same as reported in Fig. 5). Column colors illustrate the fraction of structures in which a position is single- stranded in this ensemble of three representative structures. Structure-selective reaction of RNA 2'-hydroxyl groups with NMIA (1). Nature Chemical Biology 1, 112-119 (2005) doi: 10.1038/nchembio711 Chemical inhibitor of nonapoptotic cell death with therapeutic potential for ischemic brain injury Alexei Degterev1, Zhihong Huang2, Michael Boyce1, Yaqiao Li1, Prakash Jagtap3, Noboru Mizushima4, Gregory D Cuny3, Timothy J Mitchison5, Michael A Moskowitz2 and Junying Yuan1 The mechanism of apoptosis has been extensively characterized over the past decade, but little is known about alternative forms of regulated cell death. Although stimulation of the Fas/TNFR receptor family triggers a canonical 'extrinsic' apoptosis pathway, we demonstrated that in the absence of intracellular apoptotic signaling it is capable of activating a common nonapoptotic death pathway, which we term necroptosis. We showed that necroptosis is characterized by necrotic cell death morphology and activation of autophagy. We identified a specific and potent small-molecule inhibitor of necroptosis, necrostatin-1, which blocks a critical step in necroptosis. We demonstrated that necroptosis contributes to delayed mouse ischemic brain injury in vivo through a mechanism distinct from that of apoptosis and offers a new therapeutic target for stroke with an extended window for neuroprotection. Our study identifies a previously undescribed basic cell-death pathway with potentially broad relevance to human pathologies. The caspase family of cysteine proteases has an indispensable role in the signal transduction and execution of apoptosis1. Stimulation of the Fas/TNFR family of death- domain receptors (DRs) by their corresponding ligands activates a canonical apoptotic pathway that includes a sequential activation of multiple caspases2. Paradoxically, however, a growing number of studies has reported that caspase inhibition does not prevent DR-induced cell death. Furthermore, cell death induced under such conditions lacks the typical features of apoptosis and instead resembles necrosis3, 4, 5, 6. For example, FasL, TNF and TRAIL induced the necrotic cell death of Jurkat cells in the presence of the pan-caspase inhibitor zVAD.fmk6. In Jurkat cells deficient for either the adaptor molecule FADD or caspase-8, two critical upstream activators of apoptotic signaling, TNF stimulation directly leads to necrotic cell death4, 6, 7, 8. Inhibition of caspases also does not block DR agonist−induced cell death and leads to necrosis of NIH3T3 cells9, mouse embryonic fibroblasts (MEFs)10, U937 monocytic cells9, and IEC-6 (ref. 11) and HT29 (ref. 12) epithelial cells. The similar necrotic morphological features shown by multiple cell types when undergoing DR-induced cell death without caspase activation suggest the presence of a shared, alternative, nonapoptotic cell-death pathway. However, it is unclear whether these observations in disparate systems are the results of a single common mechanism, nor are there tools to distinguish such a pathway from other types of cellular demise. Furthermore, the pathophysiological relevance of the DR-induced nonapoptotic death in vivo, if any, remains completely unknown. Necrotic cell death is common in a wide variety of pathological conditions13, including stroke14. Very little attempt, however, has been made to develop therapeutics to specifically target necrosis because of the conventional notion that, unlike apoptosis, necrotic cell death is a nonregulated response to overwhelming stress. The demonstration of a common necrotic cell-death pathway activated by a classical DR signal in the absence of any cellular damage directly challenges this notion and suggests that a portion of necrotic cell death in vivo might in fact be regulated by cellular machinery. This, in turn, may provide an unprecedented opportunity to selectively target pathological necrotic cell death. Here, we used small molecules to define the nonapoptotic pathway mediated by DRs in the absence of caspase signaling. We identified necrostatin-1 (Nec-1, Compound 1), a specific and potent small-molecule inhibitor of cell death caused by DR stimulation in the presence of caspase inhibition in multiple cell types. These results provide the first direct evidence that DR signaling triggers a common alternative nonapoptotic cell-death pathway, which we term necroptosis. Using Nec-1 (see Supplementary Methods online for details of the structure and preparation of this compound and its derivatives), we demonstrated that necroptosis is a delayed component of ischemic neuronal injury and may therefore represent a promising therapeutic target for the treatment of stroke. Top of page Results Identification of necrostatin-1 Because chemical inhibitors of caspases have been instrumental in the characterization of apoptosis in mammalian systems, we expected that specific necroptotic inhibitors would be equally useful for demonstrating the existence of a common, alternative cell-death pathway in multiple cell types, for providing specific tools to distinguish necroptosis from other cell-death processes, and as potential lead molecules targeting the nonapoptotic component of pathologic cell death for therapeutic benefit. We screened a chemical library of 15,000 compounds for chemical inhibitors of the necrotic death of human monocytic U937 cells induced by TNF and zVAD.fmk15, which we used as an operational definition of necroptosis. This screen resulted in the selection of Nec-1, which efficiently blocked necroptotic death in U937 cells in a concentration-dependent fashion (Fig. 1a,b). Figure 1: Identification of Nec-1 as a necroptosis inhibitor. (a) Structure of Nec-1. (b) Inhibition of cell death of U937 cells treated with TNF and zVAD.fmk for 72 h by Nec-1 (for doses of compounds see Methods, unless otherwise indicated). Cell viability in this and subsequent panels was determined by ATP-based viability assay. Numbers represent percentages of the live cells normalized to those in the untreated control wells. Throughout, error bars represent s.d. (c) Dose-response curve (in M) of Nec-1 and inactive derivative Nec-1i cytoprotection in Jurkat cells treated with FasL-CHX-zVAD.fmk for 30 h. Numbers represent viability normalized to that of the corresponding compound-treated cells in the absence of FasL. (d) Inhibition of cell death in Jurkat cells stably expressing dimerizable FADD (Jurkat-FF) by Nec-1 after treatment with dimerizer AP20187 and zVAD.fmk for 48 h. Numbers represent viability normalized to that of the corresponding compound-treated cells in the absence of dimerizer. In addition, control Jurkat cells were also treated with AP20187 in the presence or absence of zVAD.fmk without Nec-1. (e,f,h) Inhibition of cell death in BALB/c 3T3 (e), SV40-transformed MEF (f) and L929 (h) cells by Nec-1. Cells were treated with TNF (e,f,h), zVAD.fmk (e,f) and CHX (f) for 16 (f) or 24 (e,h) h. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . (g) Dose response of Nec-1 and Nec-1i cytoprotection of FADD-deficient Jurkat cells treated with TNF for 30 h (see c for conditions). Full figure and legend (23K) Figures, schemes & tables index Next, we used Nec-1 as a tool to examine whether the DR-induced nonapoptotic cell death observed in disparate cell types is mediated by a common mechanism. Nec-1 inhibited all published examples of necrotic cell death induced by DR activation in the presence of caspase inhibitors, including the necrotic death of (i) Jurkat cells induced by FasL, cycloheximide (CHX) and zVAD.fmk (FasL-CHX-zVAD.fmk6, Fig. 1c), (ii) Jurkat cells stably expressing FADD fused to the FKBP12-based dimerization domain (JK-FF) in the presence of chemical dimerizer AP20187 and zVAD.fmk4 (Fig. 1d), (iii) BALB/c 3T3 cells15 treated with TNF and zVAD.fmk (TNF -zVAD.fmk) or FasL and zVAD.fmk (FasL-zVAD.fmk, Fig. 1e), and (iv) MEF10 (Fig. 1f), HT29 (ref. 12), and IEC-18 and HL-60 cells treated with TNF -zVAD.fmk (data not shown). Although Nec-1 was selected in a screen with zVAD.fmk, its action is not dependent upon pharmacological inhibition of caspases. Consistent with the direct activation of necroptosis, Nec-1 prevented the death of TNF -treated FADD-deficient Jurkat cells, which are unable to activate caspases in response to DR signaling7, even in the absence of zVAD.fmk (Fig. 1g). Similarly, Nec-1 efficiently inhibited the TNF -induced necrotic death of L929 cells, which does not require exogenous caspase inhibitors5 (Fig. 1h). Overall, the ability of Nec-1 to inhibit cell death in all of these systems provides the first direct demonstration of the existence of a common necroptotic pathway mediated by DR signaling. Because the induction of necroptosis in FADD-deficient Jurkat cells does not rely on the presence of other chemicals (CHX, zVAD.fmk), we used this system to determine the effective concentration for half-maximum response (EC50) for Nec-1 of 494 125 nM (Fig. 1g). Efficient protection from necroptosis by Nec-1 was confirmed by several different assays, including forward- and side-scatter analyses of cell size (Supplementary Fig. 1 online), ATP level (Fig. 1), mitochondrial dysfunction (Fig. 2a,b and data not shown), plasma membrane permeabilization (Fig. 2a−c and data not shown) and cell proliferation (Supplementary Fig. 1). Consistent with these results, morphological analyses based on electron, fluorescent and bright-field microscopy demonstrated that Nec-1 inhibited all manifestations of necroptotic cell death (Fig. 2d−f and Supplementary Figs. 1 and 2 online). These results establish Nec-1 as a potent and efficient inhibitor of necroptosis. Figure 2: Efficient inhibition of all manifestations of necroptosis by Nec-1. (a,b) Inhibition of plasma membrane integrity loss and mitochondrial dysfunction by Nec-1. Wild-type (a) and FADD-deficient (b) Jurkat cells were treated with FasL-CHX- zVAD.fmk (a) or with TNF (b) for the indicated amounts of time. Where indicated, the cells were also treated with Nec-1 for 24 h. Cells were stained with DiOC6 or with annexin V and PI, and percentages of cells with low PI and high annexin V, high PI and high annexin V, high PI and low annexin V, and low DiOC6 (low mitochondrial membrane potential) are shown. Throughout, error bars represent s.d. (c) Demonstration of Nec-1 suppression of necroptosis from Sytox assay. Jurkat-FF cells were treated with AP20187-zVAD.fmk and Nec-1 for 48 h. Values represent the percentages of live cells normalized to the compound-treated cells lacking dimerizer. Alternatively, BALB/c 3T3 cells were treated with TNF -zVAD.fmk and 40 M Nec-1 for 24 h, or Jurkat cells were treated with FasL-CHX-zVAD.fmk and Nec-1 for 48 h. (d,e) Bright-field microscopy of the cytoprotective effect of Nec-1 in (d) Jurkat cells treated with FasL-CHX-zVAD.fmk or (e) L929 cells treated with TNF for 36 h. Bright-field images were acquired at 40 magnification. (f) Electron microscopy of the effect of Nec-1 in apoptotic or necroptotic Jurkat cells treated with FasL and CHX for 16 h or TNF for 6 h, respectively. Representative EM images are shown. Arrows indicate dead cells. Con, vehicle-only control. Full figure and legend (80K) Figures, schemes & tables index Specificity of Nec-1 To establish the specificity of Nec-1, we compared its effects on DR-induced apoptosis as compared to necroptosis, which can be readily distinguished by morphological criteria and selective dye staining. Nec-1 had no effect on FasL-CHX−induced accumulation of annexin V−positive and propidium iodide (PI)−negative cells, a result indicative of apoptosis16 (Fig. 3a). Conversely, Nec-1 efficiently inhibited the simultaneous loss of mitochondrial membrane potential and plasma membrane integrity in FADD-deficient Jurkat cells treated with TNF (Fig. 2a) or wild-type Jurkat cells treated with FasL-CHX- zVAD.fmk (Fig. 2b). The onset of apoptosis was notably faster than that of necroptosis in response to the similar stimuli (FasL-CHX and FasL-CHX-zVAD.fmk, respectively; see Figs. 2b and 3a), which might suggest that apoptosis usually masks or preempts necroptosis in this cell type because of its faster kinetics. Figure 3: Specificity of Nec-1. (a) Lack of cytoprotection by Nec-1 in apoptotic Jurkat cells treated with FasL-CHX for 24 h. Time course of cellular changes in the samples not treated with Nec-1 is also shown. Cells were stained with DiOC6 or annexin V and EGFP-PI as described in Figure 2a,b. (b,c) Selective inhibition of necroptosis but not apoptosis by Nec-1 in U937 (b) and BALB/c 3T3 (c) cells treated with TNF and CHX (b only) without (apoptosis) or with (necroptosis) zVAD.fmk for 48 h (b) or 24 h (c). Cell viability was determined by ATP assay. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . Error bars, s.d. (d) Structures of the selected Nec-1 derivatives. Full figure and legend (29K) Figures, schemes & tables index Consistent with this observation, Nec-1 had no effect on apoptotic morphology (cytoplasm condensation, chromatin marginalization, nuclear fragmentation and plasma membrane blebbing) in FasL-CHX−treated apoptotic Jurkat cells (Fig. 2f), whereas it completely inhibited the appearance of necrotic morphology (nuclear condensation, organelle swelling, early loss of plasma membrane integrity and the appearance of translucent cytosol; see Fig. 2d,f) in TNF -treated, FADD-deficient Jurkat cells7. Selective protection from necroptosis induced by DR signaling was also observed in other cell-death assays (Figs. 1d and 3b,c and data not shown). These results establish the selectivity of Nec-1 in inhibiting necroptosis and point to the divergent regulation of these two types of cell death. Consistent with this model, overexpression of Bcl-xL in Jurkat cells, which potently suppressed apoptosis, did not inhibit necroptosis (data not shown). To further establish the specificity of Nec-1, we analyzed its possible effect on other cell physiological parameters. We found that Nec-1 had no general effect in healthy cells on ATP levels, mitochondrial membrane potential, plasma membrane integrity, cell shape or size, cell cycle distribution, proliferation, global mRNA expression as reflected in Agilent mRNA chip analysis, the intracellular levels of reactive oxygen species (ROS), cell adhesion, actin and microtubule cytoskeletons, or the morphology of various cellular compartments, for example, nuclei, Golgi apparatus, endoplasmic reticulum and mitochondria (Supplementary Figs. 1 and 2). These observations suggest that Nec-1 is specific for necroptosis. To further characterize the specificity of Nec-1, we performed extensive structure-activity relationship analyses and found that most of the 93 chemical modifications of Nec-1 that we tested resulted in either substantial or complete loss of activity (data not shown). For example, elimination of the methyl group in the hydantoin moiety (Nec-1i, Compound 2; Fig. 3d and Supplementary Methods) completely abolished antinecroptotic activity (Fig. 1c,g). Of all Nec-1 modifications tested, only two types of changes preserved its antinecroptotic activity. First, the addition of a chlorine to the phenyl ring of Nec-1 (7-Cl-Nec-1, Compound 3; Fig. 3d and Supplementary Methods) resulted in an approximately 2.7-fold increase in activity (EC50 = 182 24 nM). Second, changing the sulfur in the hydantoin moiety, which is a potential metabolic liability, to oxygen did not affect the antinecroptotic activity of Nec-1 (7-Cl-O-Nec-1, Compound 4; Fig. 3d and Supplementary Methods; EC50 = 206 33 nM). Such stringent structural requirements point to the highly specific mode of necrostatin cytoprotection. Finally, we compared the activity of Nec-1 to that of other small-molecule regulators of cellular signaling. Our analyses revealed that necroptotic cell death in Jurkat and BALB/c 3T3 cells cannot be blocked by small-molecule inhibition of such factors as the mitochondrial permeability transition pore, calpains, calcium homeostasis perturbation, poly(ADP-ribose) polymerase (PARP), HtrA2/Omi, phospholipase A2 and nitric oxide synthase, or the RNAi-mediated downregulation of apoptosis-inducing factor (data not shown). Furthermore, we screened a chemical library of 489 compounds with known biological activities (BIOMOL ICCB Known Bioactives library; http://iccb.med.harvard.edu) and found that no compound could block necroptosis in all cell types, as does Nec-1 (data not shown). These results underscore the unique nature of necroptosis regulation and Nec-1 activity. Oxidative stress has been suggested as having a role in DR-induced caspase-independent cell death in some cell types, including U937 and L929 (ref. 17). However, we found that only a small fraction of TNF -treated necroptotic FADD-deficient Jurkat cells showed an increase in ROS ( 30% of dying cells; see Supplementary Fig. 1), whereas necroptosis triggered by FADD dimerization was not accompanied by oxidative stress (data not shown), as previously reported4. Consistent with this observation, necroptosis in Jurkat cells was not inhibited by the antioxidant BHA (data not shown and ref. 4), nor did any chemical from a panel of general antioxidants protect U937 cells from necroptosis (Fig. 4a), despite the reported protection by BHA in this cell type17. Conversely, Nec-1 did not block the 'classic' oxidative stress-induced necrosis caused by menadione (Fig. 4a). From these results, we concluded that oxidative stress, although important in some systems, does not play a universal role in necroptotic signaling and that Nec-1 does not act as antioxidant in inhibiting necroptosis. Figure 4: Roles of oxidative stress and autophagy in necroptosis. (a) Necroptosis is distinguishable from oxidative stress-induced cell death. U937 cells were treated with 40 ng ml-1 human TNF and zVAD.fmk for 72 h or 250 M of menadione for 24 h in the presence of various antioxidants and 100 M Nec-1. In this and subsequent panels cell viability was determined via ATP assay. Numbers represent viability normalized to that of the DMSO-treated cells in the absence of TNF and menadione. Throughout, error bars represent s.d. (b) Electron microscopic detection of autophagosomes in FADD-deficient Jurkat cells treated with TNF with or without 10mM 3MA for 12 h. Representative images are shown. (c) Induction of LC3-II during necroptosis in L929 and FADD-deficient Jurkat cells treated with TNF and Nec-1 for 8 and 24 h, respectively. Protein lysates were subjected to western blotting with antibodies to LC3 and to tubulin. The ratio of LC3-II to tubulin signals was calculated ('Ratio') and normalized to the value in the control sample. (d) Induction of LC3-II during necroptosis in BALB/c 3T3 cells treated with TNF -zVAD.fmk or FasL-zVAD.fmk and Nec-1 for 24 h. Alternatively, cells were treated with 2 M rapamycin. (e−g) No inhibition of necroptosis by suppression of autophagy in BALB/c 3T3 (e), FADD-deficient Jurkat (e), Atg5-/- MEFs (f) and beclin-1 RNAi−expressing BALB/c3T3 (g) cells. Cells were treated with TNF -zVAD.fmk and 3MA or Nec-1 for 24 or 30 (Jurkat cells) h. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . (h) Expression of beclin-1 ('Bec') and tubulin ('Tub') in the stable populations of cells in g was analyzed by western blotting. Asterisk shows nonspecific band detected by antibody to beclin. Full figure and legend (85K) Figures, schemes & tables index Activation of autophagy by necroptotic signaling Autophagy, a large-scale protein degradation and catabolic mechanism18, has been implicated in caspase-independent cell death19, although its functional role remains a subject of debate. The EM analysis of necroptotic Jurkat cells demonstrated the presence of double membrane-enclosed vesicles filled with electron-dense material (Fig. 4b), which are indicative of autophagy20. We therefore further investigated the role of autophagy in necroptosis. We used the appearance of the phosphatidylethanolamine (PE)- conjugated form of microtubule-associated protein 1 light chain 3 (LC3-II) as a marker of autophagy, as it has been shown to play an early and critical role in the formation of the autophagosomes21. Indeed, autophagy, as indicated by the increase in the levels of LC3- II, was induced in a number of necroptotic systems, including FADD-deficient Jurkat cells and L929 cells treated with TNF (Fig. 4c), BALB/c 3T3 cells treated with TNF - zVAD.fmk or FasL-zVAD.fmk (Fig. 4d), and U937 cells treated with TNF -zVAD.fmk (data not shown). Whereas the production of LC3-II results from a multistep process, including the cleavage of the LC3 to the LC3-I form followed by its conjugation to PE21, we did not detect the intermediate LC3-I species in every cell type examined; this suggests that it may be rapidly processed into LC3-II in some contexts. However, necroptosis proceeded normally in the presence of 3-methyladenine (3MA), an inhibitor of autophagy (Fig. 4e and data not shown), in autophagy-deficient Atg5-/- MEF cells22 (Fig. 4f) and in cells where the critical autophagic factor beclin-1 (ref. 23) was knocked down by RNAi (Fig. 4g,h). These results establish autophagy as a common downstream consequence of necroptosis, rather than a contributing factor to necroptotic cell death. On the other hand, although inhibition of autophagy had no effect on the final demise of Jurkat cells, it resulted in the marked accumulation of electron-dense cytoplasmic material (Fig. 4b), indicating a failure to remove cellular debris. Notably, whereas the treatment with Nec-1 efficiently blocked the necroptotic LC3-II induction as well as the formation of autophagic vesicles (Figs. 2f and 4c,d), it had no effect on the LC3-II induction caused by rapamycin, a classical inducer of autophagy (Fig. 4d), a result indicating that Nec-1 inhibits a necroptotic signaling step upstream of autophagy but does not inhibit autophagy per se. Inhibition of RIP-induced necroptosis by necrostatin Previous analyses6 have suggested that the kinase activity of the DR-interacting protein RIP serves as a bifurcation point separating necroptosis from other DR-dependent pathways. Indeed, dimerized full-length RIP or its kinase domain alone is sufficient to induce kinase-dependent necroptotic cell death6, which was inhibited by Nec-1 (Fig. 5); this result confirmed that Nec-1 specifically affects the necroptotic branch of DR signaling. Taken together, our results demonstrate that Nec-1 targets a critical common necroptotic step downstream of DRs but upstream of a number of execution events, including mitochondrial dysfunction, loss of plasma membrane integrity and autophagic clearance of cellular debris. Figure 5: Nec-1 inhibits RIP kinase−induced necroptosis. FADD-deficient Jurkat cells were transiently electroporated with pEGFP and vectors encoding the FKBP12-based dimerization domain alone (pFpk) or fused to RIP (pFR), a kinase-inactive K45M mutant of RIP (pFR K45M), and kinase domain of RIP (pFR-KD) and treated 6 h later with dimerizer AP1510 without or with zVAD.fmk and Nec-1 for 48 h. Percentage of PI-negative, GFP-positive cells ('Viability, %') was determined by FACS. Numbers represent viability normalized to that of the compound-treated cells in the absence of dimerizer. Error bars, s.d. Full figure and legend (15K) Figures, schemes & tables index Necroptosis contributes to ischemic neuronal injury The stringent specificity of Nec-1 in inhibiting necroptosis prompted us to use it to explore the previously unknown role of necroptosis in vivo. Neuronal cell death caused by ischemic injury is known to contain a substantial nonapoptotic component14, 24, and the involvement of DRs in the ischemic brain injury has been suggested25, 26, 27. Therefore, we hypothesized that ischemic brain injury may create conditions that are nonoptimal for apoptosis but suitable for necroptosis. To examine the involvement of necroptosis in ischemic brain injury, we determined the effect of Nec-1 on ischemic damage resulting from middle cerebral artery occlusion (MCAO) in mice. Intracerebroventricular administration of Nec-1 significantly (P < 0.05) reduced the infarct volume after MCAO, which suggested that necroptosis could be involved in this form of pathologic death (Fig. 6a). For more detailed analyses, we switched to 7-Cl-Nec-1, which showed greater activity in vitro. 7-Cl-Nec-1 also provided a significant (P < 0.05) and dose-dependent reduction in the infarct volume and a proportionate improvement in the neurological score after MCAO (Fig. 6b). Figure 6: Inhibition of in vivo ischemic injury by Nec-1. (a) Reduction in infarct volume by administration of Nec-1 but not of Nec-1i. The experiment was performed as described in Methods; n 5 for each treatment group. Throughout, error bars represent s.e.m. In all panels, **P < 0.05 as compared to vehicle. (b) Dose-dependent reduction in infarct volume and improvement in neurological scores upon pre- or postocclusion (Methods) delivery of 7-Cl-Nec-1. Stock concentrations of the injected compounds are shown.; n 6 for each treatment group. Inset, animal behavior was assessed by an investigator unaware of the treatment group before animals were killed and scored from 3 (worst) to 0 (no defects); n = 4−5 in each treatment group. (c) 7- Cl-Nec-1 does not inhibit caspase-3 activation during brain ischemia. The zVAD.fmk or 7-Cl-Nec-1 were delivered under preocclusion delivery conditions. Brains were harvested 3 h after reperfusion and proteins from the infracted (+) or control (-) hemispheres from the same animal were subjected to western blotting using antibodies to active caspase-3 or to tubulin. The ratio of caspase-3 to tubulin signals was calculated and normalized to the values in the corresponding control hemispheres. (d) 7-Cl-O-Nec-1 has activity comparable to that of 7-Cl-Nec-1 in vivo. Mice were injected with 7-Cl-Nec-1 or 7-Cl-O- Nec-1 at 4 and 6 h postocclusion; n 10 in each treatment group. (e) Establishing the therapeutic time window of 7-Cl-Nec-1 in MCAO model by injecting compound either 5 min before or immediately after 2-h MCAO at 2, 4 or 6 h postocclusion; n 10 in each treatment group. (f) Lack of neuroprotection by zVAD.fmk upon administration 6 h postocclusion. The delivery of compounds was performed at 2 h, 4 h or 6 h after the onset occlusion; n 8 in each treatment group. P values are shown. (g) Delayed induction of LC3-II during brain ischemia in vivo. Brains were harvested at indicated time points after occlusion and subjected to western blotting with antibodies to LC3 and to tubulin. LC3- II/tubulin ratio is shown. (h) Inhibition of late induction of LC3-II by delayed administration of 7-Cl-Nec-1 at 4 and 6 h after occlusion. Brains were collected 8 h after occlusion and analyzed as in g. Full figure and legend (73K) Figures, schemes & tables index We next examined the specificity of 7-Cl-Nec-1 activity in vivo. Consistent with the specificity of Nec-1 in blocking necroptosis but not apoptosis in vitro, treatment with 7- Cl-Nec-1 did not block caspase-3 activation during ischemic brain injury, whereas zVAD.fmk did (Fig. 6c). Furthermore, coadministration of zVAD.fmk and 7-Cl-Nec-1 resulted in a significant (P < 0.05) additive effect (Supplementary Fig. 3 online), though extensive optimization would be required to assess the full extent of neuroprotection by the combination treatment. In addition, 7-Cl-Nec-1 did not affect blood oxygen and CO2 levels, body temperature or cerebral blood flow (data not shown), which indicated that it does not prevent ischemic cell death through a nonspecific effect on general physiology. To confirm the mode of action of Nec-1 in vivo, we performed a structure-activity relationship analysis of its protection against ischemic brain injury. First, Nec-1i, an inactive derivative of Nec-1 that lacks a single methyl group (Fig. 3d), did not significantly affect infarct volume (Fig. 6a). Second, 7-Cl-O-Nec-1 (Fig. 3d), possessing antinecroptotic activity in vitro similar to that of 7-Cl-Nec-1 (Fig. 3d), showed activity indistinguishable from that of 7-Cl-Nec-1 in vivo (Fig. 6d). These data demonstrate a strict correlation between the inhibition of necroptosis in vitro and the anti-ischemic activity of 7-Cl-Nec-1 in vivo, providing strong support for our hypothesis that neuroprotection by Nec-1 is accomplished through the inhibition of necroptosis. Notably, the protective effect of 7-Cl-Nec-1 was readily detectable even when the compound was administered 6 h after the onset of injury (Fig. 6e), at which point the administration of zVAD.fmk no longer reduces infarct volume28 (Fig. 6f). We reasoned that this extended time window of neuroprotection by 7-Cl-Nec-1 in vivo might reflect a delayed induction of necroptosis during ischemic brain injury. To verify this hypothesis, we analyzed the induction of LC3-II during MCAO, as our in vitro analyses suggested that this event is associated with necroptosis. We observed that although LC3-II was clearly induced after ischemic brain injury, it did not reach the maximal level until 8 h postocclusion (Fig. 6g). Furthermore, delayed injection of 7-Cl-Nec-1 at 4 and 6 h postocclusion still efficiently blocked the LC3-II increase at the 8-h time point (Fig. 6h), confirming that the late induction of LC3-II in vivo indeed reflects the delayed activation of necroptosis. Top of page Discussion Nec-1 provides a powerful tool for characterizing the role of necroptosis in vitro and in vivo. We showed that necroptosis represents a common, alternative cell-death pathway triggered by DR signaling in multiple cell types and established its physiological relevance in vivo as a delayed component of ischemic brain injury. We showed that, in addition to the early and simultaneous loss of cytoplasmic membrane integrity and mitochondrial membrane potential, which had been observed previously4, 5, 6, necroptosis also involves the activation of autophagy, probably as a self-clearance mechanism. Our study extends the conventional use of small molecules in biology, which aims to characterize specific functions of their direct targets, to a broader level of defining new biological pathways. Our data suggest that although necroptosis shares some features with all three major established mechanisms of cell death, that is, apoptosis, necrosis or type 2 autophagic death29, it represents a previously unknown and distinct form of cell death. Although the same death receptor agonist may trigger necroptosis and apoptosis, necroptosis can be differentiated from apoptosis both morphologically and functionally. Furthermore, although necroptosis shares some morphological features of necrosis, it is distinguishable from the latter by being an active cell death triggered by DR signaling, rather than nonspecific cellular injury. Autophagy has a dual role as an intrinsic cell-death mechanism under some circumstances23 and as a cell-survival pathway under others30. We showed that although autophagy is commonly induced by necroptosis, this mechanism is not critical for cellular demise. Thus, autophagy is a cellular response to necroptosis, rather than a part of the death execution mechanism per se. Furthermore, Nec-1 blocks necroptosis but not rapamycin-induced autophagy (Fig. 4d), which indicates that the two processes are independently regulated. Although we cannot exclude the possibility that in some instances autophagy might contribute more actively to the execution of necroptosis, our data clearly show that, in general, necroptosis is functionally distinguishable from autophagy and does not depend on autophagy for cell-death execution. Necroptosis does not appear to involve any previously described factors implicated in caspase-independent cell death, and Nec-1 clearly shows a unique ability in inhibiting necroptotic death as compared with other known biologically active molecules. Although definitive confirmation and full characterization of the necroptosis pathway must await the identification of the Nec-1 binding target(s), which is a subject of ongoing investigation, the unique selectivity of Nec-1 in blocking necroptosis suggests that its target may be a factor previously unknown as a regulator of nonapoptotic cell death. Our results demonstrate the role of necroptosis as a mechanism of delayed ischemic brain injury, providing the first example of active necrotic cell death in vivo. The slow-onset kinetics of necroptosis in vitro and in vivo suggest the intriguing possibility that this pathway might provide a new target for neuroprotective intervention with an extended therapeutic window. A limited therapeutic window has been one of the major factors in the failure of several antistroke drug candidates in clinical trials31. On the other hand, the therapeutic time window of 7-Cl-Nec-1 exceeds that of most other compounds found previously to work in the MCAO model, including inhibitors of PARP, a known nonapoptotic death regulator32, 33. The exact stimulus that activates necroptosis in ischemic brain injury is currently unknown. Although we do not exclude the possibility that other pro-death signals might activate necroptosis in ischemic neurons and other conditions, we speculate that the activation of DRs by FasL and TNF contributes to the activation of necroptosis during ischemic brain injury, because the induction of FasL and TNF and the activation of DRs have been shown to contribute to ischemic and traumatic brain injury26, 34. Furthermore, we hypothesize that the temporal course of upregulation of FasL, TNF and other DR ligands' expression in ischemic brain injury may contribute to the delayed induction of necroptosis in vivo, which suggests that continuous administration of Nec-1 might provide additional neuroprotective benefits. Necroptosis may function as a cellular 'backup' mechanism to ensure the elimination of damaged cells under stress conditions when apoptosis is inhibited. The slow-onset kinetics of necroptosis are consistent with this proposal. Furthermore, although our data suggest that neuronal necroptosis can occur under ischemic conditions without exogenous caspase inhibitors, it may still reflect the failure to undergo apoptosis in some parts of the ischemic brain. This may result from the development of an apoptosis-nonpermissive environment upon ischemic injury owing, for example, to insufficient cellular energy supplies, as the level of cellular ATP is a critical determinant for cells in undergoing apoptosis or necrosis35. Alternatively, it is also possible that necroptosis may function as the primary cell death mechanism in a subpopulation of neurons. Future studies are needed to distinguish among these possibilities. Finally, although our study specifically explored the role of necroptosis in ischemic brain injury as a proof-of-principle example system, the demonstration of an intrinsic necrotic cell-death mechanism suggests the broader possibility of specifically targeting necrosis as a therapeutic strategy in other contexts. Indeed, there is extensive evidence that necrotic cell death plays a prominent role in a wide range of human pathological conditions, such as myocardial infarction and acute and chronic neurodegeneration36, 37. The availability of Nec-1 offers a unique opportunity to characterize the role of necroptosis in these and other human pathologies. Top of page Methods Antibodies and reagents. FasL (used at 10 ng ml-1) and zVAD.fmk (used at 100 M) were from Alexis Biochemicals. Human or mouse (used for MEF cells) TNF (used at 10 ng ml-1) was from Cell Sciences. Sytox Green, TO-PRO-3 and DiOC6 were from Molecular Probes. Propidium iodide was from Roche. The dimerizing agents AP1510 and AP20187 (both used at 100 nM) were obtained from Ariad Pharmaceuticals. CHX (used at 1 g ml-1), antioxidants (N-acetyl cysteine (NAC), used at 2.5 mM; superoxide dismutase (SOD), 800 U ml-1; catalase (cat.), 1,400 U ml-1; vitamin E (vit. E), 5 mM), phalloidin-TRITC and all common chemicals were from Sigma. The following antibodies were used: mouse anti− -tubulin (Stressgen), rabbit anti-LC3 (ref. 20), rabbit anti−beclin-1 (Santa Cruz), rabbit anti-giantin (Covance), mouse anti-KDEL (Stressgen) and mouse anti−cytochrome c (Pharmingen). Secondary horseradish peroxidase (HRP)-conjugated antibodies were from Southern Biotech. Secondary Alexa 488−conjugated antibodies were from Molecular Probes; Cy3-conjugated antibodies were from Jackson ImmunoResearch. Preparation and use of Nec-1 and its derivatives. For details of chemical synthesis see Supplementary Methods. We used compounds at 30 M in all cellular assays unless otherwise indicated. Chemical screening. We plated U937 cells in 384-well plates at 5,000−10,000 cells per well in 40- l phenol red−free RPMI 1640 medium containing 100 M zVAD.fmk and 40 ng ml-1 human TNF using a Multidrop dispenser (Thermo Electron). We added 100 nl of the DiverSetE (5 mg ml-1 in DMSO, Chembridge) using a Seiko-based custom-built pin transfer robot (Institute of Chemistry and Cell Biology, Harvard Medical School). After 72 h, we assessed cell viability using a luminescence-based ATP assay (ATPLite-M, PerkinElmer). We also dispensed cells not treated with TNF in each plate as a positive control. We purchased Nec-1 and other preliminary positive hits (not described) from Chembridge for individual retesting. Cell viability assays. We seeded cells in 96-well plates (white plates for luminescent assays; black plates for fluorescent assays; clear plates for MTT assay) at the density of 5,000−10,000 cells per well for adherent cells or 20,000−50,000 cells per well for suspension cells in 100 l of the appropriate phenol red−free media. After incubation, we determined cell viability using one of the following methods. For the ATP assay, we used luminescence-based commercial kits (CellTiter-Glo, Promega or ATPLite-M, PerkinElmer) and analyzed luminescence using a Wallac Victor II plate reader (PerkinElmer). For Sytox assay, we incubated cells with 1 M Sytox Green reagent for 30 min at 37 °C, and then performed fluorescent reading. Subsequently, we added 5 l of 20% Triton X-100 solution into each well to produce maximal lysis and incubated cells for 1 h at 37 °C, then performed the second reading. We calculated the ratio of values (percentage of dead cells in each well) before and after Triton treatment and normalized it to the relevant controls not subjected to cytotoxic stimuli, as indicated in figure legends. For the MTT assay, we used the CellTiter 96 AQueous Non-Radioactive Cell Proliferation Assay kit (Promega). For PI exclusion assays, we added 2 g ml-1 PI into the medium and immediately analyzed samples using FACSCalibur (BD Biosciences). For PI−annexin V assay we used the ApoAlert Annexin V-EGFP Apoptosis Kit (Clontech). For DioC6 staining, we incubated cells with 40 nM DiOC6 for 30 min at 37 °C, washed once and analyzed in FACSCalibur. For ROS analysis, we incubated cells with 5 M dihydroethidium (Molecular Probes) for 30 min at 37 °C, washed once and analyzed in FACSCalibur. EM analyses were performed at the Harvard Medical School EM facility. We acquired bright-field images of the cells using an Axiovert 200 microscope (Zeiss). Transient focal cerebral ischemia in the mouse. The animal protocol was approved by the Massachusetts General Hospital Subcommittee on Animal Research Care that serves as the Institutional Animal Care and Use Committee, and all procedures were performed in accordance with the Public Health Service Policy on Humane Care and Use of Laboratory Animals. Animals were maintained in accordance with the "Guide for the Care and Use of Laboratory Animals" (National Research Council, 1996). We anesthetized spontaneously breathing adult male SV-129 mice (19−23 g; Taconic Farms) with 2% isoflurane and maintained them on 0.8−1% isoflurane in 70% N2O and 30% O2 using a Fluotec 3 vaporizer (Colonial Medical). We occluded the left MCA with an intraluminal 8-0 nylon monofilament (Ethicon) coated with a mixture of silicone resin (Xantopren, Bayer Dental) and a hardener (Elastomer Activator, Bayer Dental). The procedure lasted 15 min, and the anesthesia was discontinued. We briefly reanesthetized animals 2 h later with isoflurane, and withdrew the filament. Eighteen hours after reperfusion we divided forebrains into five coronal (2-mm) sections using a mouse brain matrix (RBM-2000C; Activational Systems), and stained the sections with 2% 2,3,5-triphenyltetrazolium chloride (Sigma). We quantified infarct areas using an image-analysis system (Bioquant IV, R & M Biometrics) and calculated infarct volume directly by adding the infarct volume in each section. For drug administration, we dissolved 7-Cl-Nec-1 or other derivatives in 4% methyl- - cyclodextrin (Sigma) solution in PBS and administered it through intracerebroventricular administration at the time points indicated. Typically, we performed two 2- l injections of 4 mM stock solution (unless otherwise indicated). For preocclusion delivery, we performed injections 5 min before the onset of 2-h MCAO occlusion and immediately after the cessation of the occlusion, at the time of the reperfusion. For postocclusion delivery, we performed injections at the time of reperfusion after 2 h of MCAO as well as 2 h after the onset of reperfusion. In the case of infusion, we infused 20 l of compound over a 30-min time period. In the case of injection 6 h after occlusion, we injected a single 4- l dose. In the case of zVAD.fmk administration, we added it to the Nec-1 formulation and administered a total dose of 160 ng. Note: Supplementary information is available on the Nature Chemical Biology website. Top of page Acknowledgments This work was supported in part by grants from the US National Institute of General Medicine (R01 GM64703) and National Institute on Aging (R37 AG012859) to J.Y., the National Institute of Neurological Disorders and Stroke (R01 NS37141-08) to M.M. and J.Y., and funding from the Harvard Center for Neurodegeneration and Repair to G.D.C. A.D. is a recipient of a National Institute on Aging Mentored Research Scientist Career Development Award and an American Health Assistance Foundation Pilot Award. We thank X. Teng for help in preparing compounds for animal testing; M. Lipinski and R. Olea-Sanchez for critical reading of the manuscript; C. Ayata for helpful suggestions with MCAO experiments; and G. Nunez, T. Jacks, J. Blenis and T. Yoshimori for providing RIP constructs, pSRP vector and mutant Jurkat cells, and anti-LC3 antibody, respectively. Competing interests The authors declare that they have no competing financial interests. Top of page References 1. Degterev, A., Boyce, M. & Yuan, J. A decade of caspases. Oncogene 22, 8543– 8567 (2003). | Article | 2. Wallach, D. et al. Tumor necrosis factor receptor and Fas signaling mechanisms. Annu. Rev. Immunol. 17, 331–367 (1999). | Article | 3. Vercammen, D. et al. Dual signaling of the Fas receptor: initiation of both apoptotic and necrotic cell death pathways. J. Exp. Med. 188, 919–930 (1998). | Article | 4. Matsumura, H. et al. Necrotic death pathway in Fas receptor signaling. J. Cell Biol. 151, 1247–1256 (2000). | Article | 5. Schulze-Osthoff, K., Krammer, P.H. & Droge, W. Divergent signalling via APO- 1/Fas and the TNF receptor, two homologous molecules involved in physiological cell death. EMBO J. 13, 4587–4596 (1994). 6. Holler, N. et al. Fas triggers an alternative, caspase-8-independent cell death pathway using the kinase RIP as effector molecule. Nat. Immunol. 1, 489–495 (2000). | Article | 7. Chan, F.K. et al. A role for tumor necrosis factor receptor-2 and receptor- interacting protein in programmed necrosis and antiviral responses. J. Biol. Chem. 278, 51613–51621 (2003). | Article | 8. Kawahara, A., Ohsawa, Y., Matsumura, H., Uchiyama, Y. & Nagata, S. Caspase- independent cell killing by Fas-associated protein with death domain. J. Cell Biol. 143, 1353–1360 (1998). | Article | 9. Khwaja, A. & Tatton, L. Resistance to the cytotoxic effects of tumor necrosis factor alpha can be overcome by inhibition of a FADD/caspase-dependent signaling pathway. J. Biol. Chem. 274, 36817–36823 (1999). | Article | 10. Lin, Y. et al. Tumor necrosis factor-induced nonapoptotic cell death requires receptor-interacting protein-mediated cellular reactive oxygen species accumulation. J. Biol. Chem. 279, 10822–10828 (2004). | Article | 11. Ruemmele, F.M., Dionne, S., Levy, E. & Seidman, E.G. TNF -induced IEC-6 cell apoptosis requires activation of ICE caspases whereas complete inhibition of the caspase cascade leads to necrotic cell death. Biochem. Biophys. Res. Commun. 260, 159–166 (1999). | Article | 12. Wilson, C.A. & Browning, J.L. Death of HT29 adenocarcinoma cells induced by TNF family receptor activation is caspase-independent and displays features of both apoptosis and necrosis. Cell Death Differ. 9, 1321–1333 (2002). | Article | 13. Nieminen, A.L. Apoptosis and necrosis in health and disease: role of mitochondria. Int. Rev. Cytol. 224, 29–55 (2003). 14. Lo, E.H., Dalkara, T. & Moskowitz, M.A. Mechanisms, challenges and opportunities in stroke. Nat. Rev. Neurosci. 4, 399–415 (2003). | Article | 15. Li, M. & Beg, A.A. Induction of necrotic-like cell death by tumor necrosis factor and caspase inhibitors: novel mechanism for killing virus-infected cells. J. Virol. 74, 7470–7477 (2000). | Article | 16. Kain, S.R. & Ma, J.T. Early detection of apoptosis with annexin V-enhanced green fluorescent protein. Methods Enzymol. 302, 38–43 (1999). 17. Fiers, W., Beyaert, R., Declercq, W. & Vandenabeele, P. More than one way to die: apoptosis, necrosis and reactive oxygen damage. Oncogene 18, 7719–7730 (1999). | Article | 18. Klionsky, D.J. & Emr, S.D. Autophagy as a regulated pathway of cellular degradation. Science 290, 1717–1721 (2000). | Article | 19. Gozuacik, D. & Kimchi, A. Autophagy as a cell death and tumor suppressor mechanism. Oncogene 23, 2891–2906 (2004). | Article | 20. Kabeya, Y. et al. LC3, a mammalian homologue of yeast Apg8p, is localized in autophagosome membranes after processing. EMBO J. 19, 5720–5728 (2000). | Article | 21. Kabeya, Y. et al. LC3, GABARAP and GATE16 localize to autophagosomal membrane depending on form-II formation. J. Cell Sci. 117, 2805–2812 (2004). | Article | 22. Kuma, A. et al. The role of autophagy during the early neonatal starvation period. Nature 432, 1032–1036 (2004). | Article | 23. Yu, L. et al. Regulation of an ATG7-beclin-1 program of autophagic cell death by caspase-8. Science 304, 1500–1502 (2004). | Article | 24. Gwag, B.J., Lobner, D., Koh, J.Y., Wie, M.B. & Choi, D.W. Blockade of glutamate receptors unmasks neuronal apoptosis after oxygen-glucose deprivation in vitro. Neuroscience 68, 615–619 (1995). | Article | 25. Rosenbaum, D.M. et al. Fas (CD95/APO-1) plays a role in the pathophysiology of focal cerebral ischemia. J. Neurosci. Res. 61, 686–692 (2000). 26. Martin-Villalba, A. et al. CD95 ligand (Fas-L/APO-1L) and tumor necrosis factor-related apoptosis-inducing ligand mediate ischemia-induced apoptosis in neurons. J. Neurosci. 19, 3809–3817 (1999). 27. Martin-Villalba, A. et al. Therapeutic neutralization of CD95-ligand and TNF attenuates brain damage in stroke. Cell Death Differ. 8, 679–686 (2001). | Article | 28. Endres, M. et al. Attenuation of delayed neuronal death after mild focal ischemia in mice by inhibition of the caspase family. J. Cereb. Blood Flow Metab. 18, 238– 247 (1998). | Article | 29. Kitanaka, C. & Kuchino, Y. Caspase-independent programmed cell death with necrotic morphology. Cell Death Differ. 6, 508–515 (1999). | Article | 30. Lum, J.J. et al. Growth factor regulation of autophagy and cell survival in the absence of apoptosis. Cell 120, 237–248 (2005). | Article | 31. Richard Green, A., Odergren, T. & Ashwood, T. Animal models of stroke: do they have value for discovering neuroprotective agents? Trends Pharmacol. Sci. 24, 402–408 (2003). | Article | 32. Takahashi, K. & Greenberg, J.H. The effect of reperfusion on neuroprotection using an inhibitor of poly(ADP-ribose) polymerase. Neuroreport 10, 2017–2022 (1999). 33. Abdelkarim, G.E. et al. Protective effects of PJ34, a novel, potent inhibitor of poly(ADP-ribose) polymerase (PARP) in in vitro and in vivo models of stroke. Int. J. Mol. Med. 7, 255–260 (2001). 34. Matsuyama, T. et al. Fas antigen mRNA induction in postischemic murine brain. Brain Res. 657, 342–346 (1994). | Article | 35. Eguchi, Y., Shimizu, S. & Tsujimoto, Y. Intracellular ATP levels determine cell death fate by apoptosis or necrosis. Cancer Res. 57, 1835–1840 (1997). 36. McCully, J.D., Wakiyama, H., Hsieh, Y.J., Jones, M. & Levitsky, S. Differential contribution of necrosis and apoptosis in myocardial ischemia-reperfusion injury. Am. J. Physiol. Heart Circ. Physiol. 286, H1923–H1935 (2004). | Article | 37. Yuan, J., Lipinski, M. & Degterev, A. Diversity in the mechanisms of neuronal cell death. Neuron 40, 401–413 (2003). | Article | Top of page 1. Department of Cell Biology, Harvard Medical School, 240 Longwood Avenue, Boston, Massachusetts 02115, USA. 2. Stroke and Neurovascular Regulation Laboratory, Massachusetts General Hospital, Harvard Medical School, 149 13th Street, Charlestown, Massachusetts 02129, USA. 3. Laboratory for Drug Discovery in Neurodegeneration, Harvard Center for Neurodegeneration and Repair, Brigham & Women's Hospital and Harvard Medical School, 65 Landsdowne St., Cambridge, Massachusetts 02139, USA. 4. Department of Bioregulation and Metabolism, The Tokyo Metropolitan Institute of Medical Science, 3-18-22 Honkomagome, Bunkyo-ku, Tokyo 113-8613, Japan. 5. Department of Systems Biology, Harvard Medical School, 240 Longwood Avenue, Boston, Massachusetts 02115, USA. 6. Email: email@example.com Correspondence to: Junying Yuan1 Email: firstname.lastname@example.org (a) Structure of Nec-1. (b) Inhibition of cell death of U937 cells treated with TNF and zVAD.fmk for 72 h by Nec-1 (for doses of compounds see Methods, unless otherwise indicated). Cell viability in this and subsequent panels was determined by ATP-based viability assay. Numbers represent percentages of the live cells normalized to those in the untreated control wells. Throughout, error bars represent s.d. (c) Dose-response curve (in M) of Nec-1 and inactive derivative Nec-1i cytoprotection in Jurkat cells treated with FasL-CHX-zVAD.fmk for 30 h. Numbers represent viability normalized to that of the corresponding compound-treated cells in the absence of FasL. (d) Inhibition of cell death in Jurkat cells stably expressing dimerizable FADD (Jurkat-FF) by Nec-1 after treatment with dimerizer AP20187 and zVAD.fmk for 48 h. Numbers represent viability normalized to that of the corresponding compound-treated cells in the absence of dimerizer. In addition, control Jurkat cells were also treated with AP20187 in the presence or absence of zVAD.fmk without Nec-1. (e,f,h) Inhibition of cell death in BALB/c 3T3 (e), SV40-transformed MEF (f) and L929 (h) cells by Nec-1. Cells were treated with TNF (e,f,h), zVAD.fmk (e,f) and CHX (f) for 16 (f) or 24 (e,h) h. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . (g) Dose response of Nec-1 and Nec-1i cytoprotection of FADD-deficient Jurkat cells treated with TNF for 30 h (see c for conditions). (a,b) Inhibition of plasma membrane integrity loss and mitochondrial dysfunction by Nec-1. Wild-type (a) and FADD-deficient (b) Jurkat cells were treated with FasL-CHX- zVAD.fmk (a) or with TNF (b) for the indicated amounts of time. Where indicated, the cells were also treated with Nec-1 for 24 h. Cells were stained with DiOC6 or with annexin V and PI, and percentages of cells with low PI and high annexin V, high PI and high annexin V, high PI and low annexin V, and low DiOC6 (low mitochondrial membrane potential) are shown. Throughout, error bars represent s.d. (c) Demonstration of Nec-1 suppression of necroptosis from Sytox assay. Jurkat-FF cells were treated with AP20187-zVAD.fmk and Nec-1 for 48 h. Values represent the percentages of live cells normalized to the compound-treated cells lacking dimerizer. Alternatively, BALB/c 3T3 cells were treated with TNF -zVAD.fmk and 40 M Nec-1 for 24 h, or Jurkat cells were treated with FasL-CHX-zVAD.fmk and Nec-1 for 48 h. (d,e) Bright-field microscopy of the cytoprotective effect of Nec-1 in (d) Jurkat cells treated with FasL-CHX-zVAD.fmk or (e) L929 cells treated with TNF for 36 h. Bright-field images were acquired at 40 magnification. (f) Electron microscopy of the effect of Nec-1 in apoptotic or necroptotic Jurkat cells treated with FasL and CHX for 16 h or TNF for 6 h, respectively. Representative EM images are shown. Arrows indicate dead cells. Con, vehicle-only control. (a) Lack of cytoprotection by Nec-1 in apoptotic Jurkat cells treated with FasL-CHX for 24 h. Time course of cellular changes in the samples not treated with Nec-1 is also shown. Cells were stained with DiOC6 or annexin V and EGFP-PI as described in Figure 2a,b. (b,c) Selective inhibition of necroptosis but not apoptosis by Nec-1 in U937 (b) and BALB/c 3T3 (c) cells treated with TNF and CHX (b only) without (apoptosis) or with (necroptosis) zVAD.fmk for 48 h (b) or 24 h (c). Cell viability was determined by ATP assay. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . Error bars, s.d. (d) Structures of the selected Nec-1 derivatives. (a) Necroptosis is distinguishable from oxidative stress-induced cell death. U937 cells were treated with 40 ng ml-1 human TNF and zVAD.fmk for 72 h or 250 M of menadione for 24 h in the presence of various antioxidants and 100 M Nec-1. In this and subsequent panels cell viability was determined via ATP assay. Numbers represent viability normalized to that of the DMSO-treated cells in the absence of TNF and menadione. Throughout, error bars represent s.d. (b) Electron microscopic detection of autophagosomes in FADD-deficient Jurkat cells treated with TNF with or without 10mM 3MA for 12 h. Representative images are shown. (c) Induction of LC3-II during necroptosis in L929 and FADD-deficient Jurkat cells treated with TNF and Nec-1 for 8 and 24 h, respectively. Protein lysates were subjected to western blotting with antibodies to LC3 and to tubulin. The ratio of LC3-II to tubulin signals was calculated ('Ratio') and normalized to the value in the control sample. (d) Induction of LC3-II during necroptosis in BALB/c 3T3 cells treated with TNF -zVAD.fmk or FasL-zVAD.fmk and Nec-1 for 24 h. Alternatively, cells were treated with 2 M rapamycin. (e−g) No inhibition of necroptosis by suppression of autophagy in BALB/c 3T3 (e), FADD-deficient Jurkat (e), Atg5-/- MEFs (f) and beclin-1 RNAi−expressing BALB/c3T3 (g) cells. Cells were treated with TNF -zVAD.fmk and 3MA or Nec-1 for 24 or 30 (Jurkat cells) h. Numbers represent viability normalized to that of the compound-treated cells in the absence of TNF . (h) Expression of beclin-1 ('Bec') and tubulin ('Tub') in the stable populations of cells in g was analyzed by western blotting. Asterisk shows nonspecific band detected by antibody to beclin. FADD-deficient Jurkat cells were transiently electroporated with pEGFP and vectors encoding the FKBP12-based dimerization domain alone (pFpk) or fused to RIP (pFR), a kinase-inactive K45M mutant of RIP (pFR K45M), and kinase domain of RIP (pFR-KD) and treated 6 h later with dimerizer AP1510 without or with zVAD.fmk and Nec-1 for 48 h. Percentage of PI-negative, GFP-positive cells ('Viability, %') was determined by FACS. Numbers represent viability normalized to that of the compound-treated cells in the absence of dimerizer. Error bars, s.d. (a) Reduction in infarct volume by administration of Nec-1 but not of Nec-1i. The experiment was performed as described in Methods; n 5 for each treatment group. Throughout, error bars represent s.e.m. In all panels, **P < 0.05 as compared to vehicle. (b) Dose-dependent reduction in infarct volume and improvement in neurological scores upon pre- or postocclusion (Methods) delivery of 7-Cl-Nec-1. Stock concentrations of the injected compounds are shown.; n 6 for each treatment group. Inset, animal behavior was assessed by an investigator unaware of the treatment group before animals were killed and scored from 3 (worst) to 0 (no defects); n = 4−5 in each treatment group. (c) 7- Cl-Nec-1 does not inhibit caspase-3 activation during brain ischemia. The zVAD.fmk or 7-Cl-Nec-1 were delivered under preocclusion delivery conditions. Brains were harvested 3 h after reperfusion and proteins from the infracted (+) or control (-) hemispheres from the same animal were subjected to western blotting using antibodies to active caspase-3 or to tubulin. The ratio of caspase-3 to tubulin signals was calculated and normalized to the values in the corresponding control hemispheres. (d) 7-Cl-O-Nec-1 has activity comparable to that of 7-Cl-Nec-1 in vivo. Mice were injected with 7-Cl-Nec-1 or 7-Cl-O- Nec-1 at 4 and 6 h postocclusion; n 10 in each treatment group. (e) Establishing the therapeutic time window of 7-Cl-Nec-1 in MCAO model by injecting compound either 5 min before or immediately after 2-h MCAO at 2, 4 or 6 h postocclusion; n 10 in each treatment group. (f) Lack of neuroprotection by zVAD.fmk upon administration 6 h postocclusion. The delivery of compounds was performed at 2 h, 4 h or 6 h after the onset occlusion; n 8 in each treatment group. P values are shown. (g) Delayed induction of LC3-II during brain ischemia in vivo. Brains were harvested at indicated time points after occlusion and subjected to western blotting with antibodies to LC3 and to tubulin. LC3- II/tubulin ratio is shown. (h) Inhibition of late induction of LC3-II by delayed administration of 7-Cl-Nec-1 at 4 and 6 h after occlusion. Brains were collected 8 h after occlusion and analyzed as in g.