Deciphering the rules of programmed cell death to improve therapy of cancer and other diseases

Abstract

Apoptosis, the major form of programmed cell death in metazoan organisms, plays critical roles in normal development, tissue homeostasis and immunity, and its disturbed regulation contributes to many pathological states, including cancer, autoimmunity, infection and degenerative disorders. In vertebrates, it can be triggered either by engagement of ‘death receptors’ of the tumour necrosis factor receptor family on the cell surface or by diverse intracellular signals that act upon the Bcl-2 protein family, which controls the integrity of the mitochondrial outer membrane through the complex interactions of family members. Both pathways lead to cellular demolition by dedicated proteases termed caspases. This review discusses the groundbreaking experiments from many laboratories that have clarified cell death regulation and galvanised efforts to translate this knowledge into novel therapeutic strategies for the treatment of malignant and perhaps certain autoimmune and infectious diseases.

Introduction

Cell death programmes have evolved to meet many needs. While certain unicellular organisms can invoke cell death to match cell numbers to the nutrient supply, multicellular organisms utilise ‘altruistic’ cell suicide for diverse essential purposes (Golstein, 1998). During embryogenesis, cell death sculpts the tissues by removing superfluous cells—hollowing out ducts, for example, or deleting interdigital cells during limb formation. During neuronal development, cell death is critical for matching neuron numbers with their targets (e.g. other neurons or muscle cells). In adults, cell death counterbalances cell proliferation to maintain homeostasis in rapidly renewing tissues (e.g. the intestinal epithelium or haemopoietic tissues), and drives mammary gland involution at weaning and thymus atrophy on ageing (Kerr et al, 1972). Cell death also eliminates irreparably damaged or potentially dangerous cells, such as those undergoing neoplastic transformation or lymphocytes that target self-tissues (Strasser et al, 2000). Finally, to limit pathogen spread, both the innate and adaptive immune systems target infected cells (Trambas and Griffiths, 2003)—provoking an arms race in which pathogens evolve new ways to prevent host cell suicide while their hosts refine their weaponry for killing infected cells (Vaux et al, 1994).

The major mode of programmed cell death, apoptosis, is prominent in diverse animal species and has been intensely investigated in mammals, the fruit fly Drosophila, and the worm Caenorhabditis elegans. Although evidence for additional types of programmed cell death processes (e.g. necroptosis) is growing, they have been reviewed recently (Hotchkiss et al, 2009; Yuan and Kroemer, 2010) and will only be briefly addressed here.

Morphologically, apoptosis is characterised by chromatin condensation and shrinkage of the nucleus and cytoplasm, followed by fragmentation of the cell into plasma membrane-bound ‘apoptotic bodies’, which are quickly engulfed by nearby phagocytic cells and ultimately digested in lysosomes (Kerr et al, 1972). Removal of the cellular corpse, triggered by ‘eat me’ signals on the dying cells, is so rapid that apoptosis is not readily visible histologically, even in tissues with massive cell turnover, such as the thymus (Egerton et al, 1990). In contrast to necrosis, apoptosis does not rupture the plasma membrane, thereby minimising release of inflammatory cellular contents and the risk of inducing autoimmunity (Nagata et al, 2010).

The foundation stones for current understanding of apoptosis were laid by a remarkable confluence of pioneering genetic studies in C. elegans with discoveries emerging from mammalian cancer genetics and biochemistry. The first molecularly defined cell death regulator emerged after the breakpoint of a recurrent chromosome translocation in human follicular lymphoma revealed the previously unknown gene BCL-2 (Tsujimoto et al, 1984). In a seminal discovery, enforced Bcl-2 expression was found to render cultured haemopoietic cells refractory to cell death upon cytokine deprivation (Vaux et al, 1988) and to promote B-cell accumulation in vivo (McDonnell et al, 1989; Strasser et al, 1991b), often culminating in autoimmune disease or cancer (see below). Strikingly, CED-9, which prevents developmentally programmed cell death in the worm (Hengartner et al, 1992), proved to be the functional homologue of BCL-2 (Vaux et al, 1992; Hengartner and Horvitz, 1994), heralding a partially conserved genetic programme (Figure 1). The exciting finding that the essential worm killer protein CED-3 resembled a new type of mammalian protease (Yuan et al, 1993) revealed that cell demolition relied on dedicated aspartate-specific cysteine proteases, soon termed caspases. Subsequently, elegant biochemical studies disclosed that the proteolytic cascade often initiates on the scaffold protein APAF-1 (Zou et al, 1997), which proved to be the homologue of worm CED-4 (Yuan and Horvitz, 1992). Collectively, these and other discoveries discussed below delineated the genetic programme for controlling apoptotic cell death, with immense ramifications for biology and human health.

Figure 1.

Comparison of the pathway of programmed cell death in C. elegans with the major one in vertebrates. The worm has a homologue of Bcl-2 (CED-9), of its BH3-only apoptosis inducers (EGL-1), of the caspase-activator Apaf-1 (CED-4) and of the proteolytic caspases (CED-3). However, a striking difference is that CED-9 directly binds and inhibits CED-4, until CED-4 is displaced by EGL-1, whereas vertebrate Bcl-2 does not bind to Apaf-1 but instead prevents the activation of its pro-apoptotic siblings Bax and Bak, thereby preventing their permeabilisation of the MOM and release of cytochrome c (cyt c), an essential cofactor for Apaf-1.

This review will concentrate upon apoptosis in mammalian cells, because defects in its control contribute to many human diseases (Hotchkiss et al, 2009), particularly cancer (Cory and Adams, 2002). We first briefly describe the effectors of apoptosis, the caspases, and then focus on its major regulators, particularly the Bcl-2 protein family. We discuss the physiological function of Bcl-2 family members and the remarkable ways, as yet incompletely understood, in which their interactions flip the apoptotic switch. We briefly describe the proposed link from Bcl-2 via Beclin-1 to autophagy, another ancient cell death/survival mechanism. Finally, we discuss how apoptosis is impaired in cancer, restricting current treatment modalities, and how directly targeting the apoptotic machinery is offering new hope for improved therapy for cancer and perhaps also for certain autoimmune and infectious diseases.

Caspases: the cellular demolition crew

Regardless of the initiating death stimulus or cell type, apoptosis culminates in the fragmentation of several hundred proteins and the DNA. Aspartate-specific cysteinyl proteases termed caspases mediate the proteolysis (Timmer and Salvesen, 2007) and also activate the responsible DNAse, CAD (caspase-activated DNAse), by cleaving its chaperone/inhibitor ICAD (inhibitor of CAD) (Liu et al, 1997; Enari et al, 1998), allowing CAD to chop the chromatin into a characteristic ‘ladder’ of nucleosomes. By structure and function, caspases fall into two groups: initiator (or apical) caspases and effector (or executioner) caspases (Riedl and Salvesen, 2007; Timmer and Salvesen, 2007). The executioners, caspases-3, -6 and -7, which perform nearly all the cellular proteolysis and activate CAD, are synthesised as single-chain zymogens (catalytically inactive pro-forms) with short pro-domains. The proteolytic cascade is launched when an initiator caspase cleaves them into fragments of ∼20 (p20) and ∼10 (p10) kDa that assemble into the active tetrameric (p20₂p10₂) proteases. The initiator caspases, such as caspase-8 or -9, have long pro-domains that, following an apoptotic signal, target them to specific scaffold proteins (FADD/Mort1 for caspase-8 and Apaf-1 for caspase-9; Figure 2), where conformational changes provoke their activation (Riedl and Salvesen, 2007).

Figure 2.

The two major pathways to caspase activation in vertebrates: the death receptor or extrinsic pathway, engaged by the indicated members of the TNF receptor family on the cell surface, and the Bcl-2-regulated mitochondrial or intrinsic pathway. The death receptors lead, via the adaptor FADD (with help by TRADD in certain death receptors), to activation of caspase-8, which then activates the effector caspases-3, -6 and -7. Caspase-8 also processes the BH3-only protein Bid, and the truncated Bid (tBid) can then activate the Bcl-2-regulated pathway. Upon MOMP, that pathway leads to effector caspase activation via Apaf-1 and caspase-9. The cytosolic E3 ubiquitin ligase XIAP can inhibit caspases-3 and -7 (and perhaps caspase-9), but that inhibition is blocked by SMAC/DIABLO when it is released from mitochondria. E3 ubiquitin ligases cIAP1 and cIAP2 work instead in part by preventing formation of the pro-apoptotic signalling complex from TNF-R1 and by regulating pro-survival NF-κB survival pathways.

Some caspases have non-apoptotic functions. Caspase-1 and its adaptors, which process pro-IL-1β and IL-18, are critical for inflammatory responses (Martinon et al, 2002). Perplexingly, caspase-8 and its adaptor FADD are essential not only for the apoptosis induced by ‘death receptors’ (see below) but also for blood vessel development, macrophage differentiation and the proliferation of certain cell types (Newton et al, 1998; Varfolomeev et al, 1998; Zhang et al, 1998). To elicit apoptosis, caspase-8 must be processed to the p20₂p10₂ tetramer, but the non-apoptotic functions require only its non-processed activated form (Kang et al, 2008). That form can prevent the RIP1 and RIP3 kinases from provoking necroptosis (programmed necrosis) (Kaiser et al, 2011; Oberst et al, 2011; Zhang et al, 2011).

Two distinct but convergent pathways to caspase activation

Vertebrates have two distinct apoptosis signalling pathways (Strasser et al, 1995) that ultimately converge upon effector caspase activation (Figure 2). The death receptor (or extrinsic) pathway is engaged on the plasma membrane by ligation of members of the tumour necrosis factor (TNF) receptor super-family containing intracellular ‘death domains’, such as Fas and TNF-R1. As reviewed elsewhere (Strasser et al, 2009), these receptors trigger apoptosis by forming a ‘death-inducing signalling complex’ (Kischkel et al, 1995), within which the FADD adaptor protein, assisted in certain death receptors (e.g. TNF-R1) by the adaptor TRADD, recruits and activates caspase-8 (and caspase-10 in humans and certain other species but not mouse). In the Bcl-2-regulated mitochondrial (or intrinsic) pathway, developmental cues and diverse cytotoxic insults, including growth factor deprivation and exposure to DNA damage or cancer therapeutics, initiate apoptosis by activating pro-apoptotic members of the Bcl-2 protein family (see below). Their action leads to the release from the mitochondrial intermembrane space not only of cytochrome c, which triggers APAF-1-mediated activation of caspase-9, but also of other apoptogenic proteins, such as SMAC/DIABLO, which prevents the Inhibitor of Apoptosis protein XIAP from inhibiting its caspase targets (Green and Kroemer, 2004; Jost et al, 2009).

Experiments with genetically modified mice revealed that FADD and caspase-8 are essential for death receptor-induced apoptosis but dispensable for apoptosis initiated by the mitochondrial pathway (Newton et al, 1998; Varfolomeev et al, 1998; Zhang et al, 1998). Conversely, cells lacking caspase-9 or its activator APAF-1 have defects in Bcl-2-regulated apoptosis but not death receptor-induced killing (Hakem et al, 1998; Kuida et al, 1998). Interestingly, however, although loss or inhibition of caspase-8 allows long-term (clonogenic) survival of cells stimulated through a death receptor (Smith et al, 1996; Longthorne and Williams, 1997; Varfolomeev et al, 1998; Salmena et al, 2003), loss of caspase-9 or APAF-1 does not allow long-term cell survival if the mitochondrial pathway is engaged (Marsden et al, 2002, 2004, 2006; Ekert et al, 2004; van Delft et al, 2009). This is because mitochondrial outer membrane permeabilisation (MOMP) by activated Bax and Bak (see below) commits the cell to die (Green and Kroemer, 2004). Even in the absence of caspase activity, the reduced respiration following cytochrome c release soon triggers a backup cell death and clearance programme (Goldstein et al, 2000).

The Bcl-2 family: the cellular life/death switch

The vertebrate Bcl-2 family contains three functional subgroups (Figure 3A). Bcl-2 and its closest homologues (Bcl-x_L, Bcl-w, Mcl-1, A1/Bfl1 and, in humans, Bcl-B), which contain four conserved sequence motifs (called Bcl-2 homology (BH) domains), all promote cell survival. The pro-apoptotic effectors Bax, Bak and the much less studied Bok share extensive similarity with their pro-survival relatives, including structural features of all four BH regions (Kvansakul et al, 2008). Despite this similarity, once activated, Bax and Bak damage rather than protect mitochondria, and either protein suffices for MOMP, indicative of functional redundancy (Lindsten et al, 2000). Lastly, the apoptosis initiators, the BH3-only proteins (which include Bad, Bik, Hrk, Bid, Bim, Bmf, Noxa and Puma) share with each other and the Bcl-2 family at large only the ∼26-residue BH3 domain. This amphipathic α-helix allows them to engage and inactivate their pro-survival relatives (Sattler et al, 1997) and perhaps also to transiently bind and activate Bax and Bak (Letai et al, 2002; Gavathiotis et al, 2008) (see below). The BH3-only proteins are activated by distinct cytotoxic stimuli in various ways, including enhanced transcription and post-translational modifications (Puthalakath and Strasser, 2002).

Figure 3.

(A) The three functional subgroups of the Bcl-2 family. Sequences most homologous to Bcl-2 (BH domains) and α-helical regions are indicated. The BH3-only proteins share sequence homology only within the BH3 domain, which mediates association between family members. Bid has a defined 3D structure but the others are relatively unstructured. The pro-survival group, which shares four regions of sequence homology, includes Bcl-B in humans but its mouse homologue (Boo) appears to be inactive due to a mutation of essential residues in BH1. The pro-apoptotic Bax/Bak group, which includes the little studied Bok, is remarkably similar in sequence and structure to the pro-survival group, including an α-helix resembling BH4 near the N-terminus (Kvansakul et al, 2008). Most family members contain a C-terminal hydrophobic trans-membrane (TM) region, which mediates their targeting and anchoring to the MOM and/or the ER, either constitutively (e.g. Bcl-2 or Bak) or after an apoptotic stimulus (e.g. Bcl-x_L or Bax). (B,C) Predominant interactions within the Bcl-2 family, including those of BH3-only proteins with their pro-survival relatives (B) and the major interactions of the latter with Bax and Bak (C).

While most BH3-only proteins are unstructured prior to engaging pro-survival proteins (Hinds et al, 2007), Bid forms an α-helical bundle resembling Bax or Bcl-2 (Chou et al, 1999; McDonnell et al, 1999), despite the lack of sequence homology except for the BH3 domain (Youle and Strasser, 2008). Bid can link the death receptor and Bcl-2-regulated pathways because its cleavage by caspase-8 generates an active C-terminal segment termed tBid, which promotes Bax/Bak-mediated MOMP (Figure 2). This tBid-activated amplification mechanism is essential for death receptor-induced killing in so-called type 2 cells, such as hepatocytes, but dispensable in type 1 cells, such as thymocytes (Scaffidi et al, 1998; Yin et al, 1999; Kaufmann et al, 2007).

The Bcl-2 family can be regarded as a tripartite switch that sets the threshold for commitment to cell death, primarily by interactions within the family. The pro-survival members can bind with high affinity to members of both the Bax/Bak-like and the BH3-only subgroups, via association of the BH3 domain of the pro-apoptotic proteins with a hydrophobic groove on the surface of the pro-survival proteins (Sattler et al, 1997). These interactions, however, exhibit specificity (Figure 3B, C). The affinities of BH3-only proteins for the pro-survival proteins differ markedly (Chen et al, 2005; Kuwana et al, 2005): Bim, Puma and perhaps tBid bind all with high affinity, whereas other BH3-only proteins show more selectivity (Figure 3B). Most strikingly, Bad binds Bcl-2, Bcl-x_L and Bcl-w but not Mcl-1 or A1, whereas Noxa interacts strongly only with Mcl-1 and A1. Accordingly, enforced expression of either Bim or Puma potently kills cells, whereas Bad and Noxa can efficiently induce cell death only if coexpressed (Chen et al, 2005). Bax and Bak also differ in their interaction with the pro-survival proteins (Figure 3C). Bak can be bound tightly by Bcl-x_L and Mcl-1 but only poorly by Bcl-2 (Willis et al, 2005), whereas all the pro-survival proteins probably can constrain Bax activity (Willis et al, 2007).

How the Bcl-2 apoptotic switch is flipped

Two distinct, albeit not mutually exclusive, models have been proposed to describe how the interplay between the three Bcl-2 factions activates Bax and Bak and hence produces MOMP (Adams and Cory, 2007; Chipuk and Green, 2008). The direct activation model (Figure 4A) posits that certain BH3-only proteins (termed ‘activators’), namely Bim, tBid and probably Puma (Letai et al, 2002; Kuwana et al, 2005; Kim et al, 2009), must transiently bind and activate Bax and Bak, whereas the other BH3-only proteins, termed ‘sensitisers’ (e.g. Bad, Noxa), can only bind their pro-survival relatives (‘Bcl-2 et al’ in Figure 4A). In this model, the pro-survival Bcl-2 proteins function by sequestering the ‘activators’, and apoptosis proceeds when ‘sensitisers’ displace the activators from the pro-survival proteins, allowing them to bind and activate Bax and Bak. The activator BH3-only proteins were widely assumed to target a hydrophobic surface groove on Bax and Bak resembling that on the pro-survival proteins, but they may instead (or in addition) bind to a proposed distal site involving Bax α-helices 1 and 6 (Gavathiotis et al, 2008, 2010; Kim et al, 2009). The binding of certain BH3 domains to Bax or Bak has been described as ‘hit and run’ (transient and low affinity) and this so-called ‘rear site’ is not yet well defined. A very recent study provides strong support for direct activation of Bak by Bim, tBid and, surprisingly, Noxa, but they were clearly shown to bind to the ‘front site’ of Bak (Dai et al, 2011). Perhaps both sites can be used, one for promoting initial activation of Bax/Bak and the other for recruiting additional Bax (Bak) molecules (see below).

Figure 4.

Models for how the BH3-only proteins activate Bax and Bak. (A) In the direct activation model, the activator BH3-only proteins (Bim, tBid and probably Puma), via their BH3 domain (red triangle), can directly engage and activate Bax (or Bak), but in healthy cells the pro-survival family members (‘Bcl-2 et al’) prevent this by sequestering the BH3-only activators. After an apoptotic signal, the sensitiser BH3-only proteins (e.g. Bad, Noxa, Bik) free the activators to target Bax or Bak. Inactive cytosolic Bax has its BH3 domain buried and its C-terminal hydrophobic helix (α9) lies in its surface groove, but during activation that helix is freed and can target Bax to the membrane. (B) In the indirect activation model, the BH3-only proteins need only target their pro-survival relatives, which primarily prevent activation of Bax and Bak by sequestering any Bax or Bak that becomes active (‘primed’) by exposure of its BH3 domain (red triangle). (C) The priming–capture–displacement model proposed here incorporates features of both direct and indirect activation. In it, any Bax or Bak that becomes primed, either spontaneously or by BH3-only proteins or other potential signals (e.g. phosphorylation), is immediately captured by a pro-survival relative, until the primed Bax or Bak is displaced by further activation of BH3-only proteins (e.g. Bad or Bim). The displaced Bax or Bak can then form dimers and higher oligomers (see Figure 5).

The indirect activation model (Willis et al, 2007) (Figure 4B) posits that, probably even in healthy cells, some Bax and Bak molecules (perhaps spontaneously) assume a ‘primed conformation’, that is, one in which their BH3 domain is exposed, and that pro-survival family members prevent apoptosis by binding to this domain, thereby preventing Bax and Bak from oligomerising (see further below). In this model, the primary role of all BH3-only proteins is to bind to the pro-survival proteins, and apoptosis can only proceed when all the relevant pro-survival proteins have been neutralised, thereby liberating the ‘primed’ Bax or Bak to oligomerise and cause MOMP. Consistent with this model, Bax molecules with BH3 mutations that prevent their sequestration by pro-survival relatives provoke unrestrained apoptosis (Fletcher et al, 2008; Czabotar et al, 2011).

The direct and indirect models were recently interrogated in vivo using gene-targeted mice in which the Bim BH3 domain (binding all pro-survival Bcl-2 proteins and possibly Bax) was replaced in situ with that of Bad, Noxa or Puma, creating alleles encoding, respectively, Bim^Bad (binding only Bcl-2, Bcl-x_L, Bcl-w), Bim^Noxa (binding only Mcl-1, A1) and Bim^Puma (binding all pro-survival Bcl-2-like proteins but not Bax). The results showed that, for optimal cell death, Bim must be able to bind all anti-apoptotic Bcl-2 family members, as in the indirect model, but that its interaction with Bax may contribute to cell killing, suggesting that physiological cell death may follow both models (Merino et al, 2009). The two models could represent alternative paths to cell death or different stages of a single pathway. For example, we propose in the priming–capture–displacement model (Figure 4C) that direct activation by BH3-only proteins is at least one of the ways that ‘primed’ Bax and Bak are generated, but that pro-survival relatives immediately capture this ‘primed form’ of Bax and Bak, until BH3-only proteins displace it, as in the indirect activation model.

Interestingly, the pro-survival family members may regulate Bax and Bak in multiple ways. According to a recent report, Bax in healthy cells spontaneously translocates from the cytosol to associate peripherally with mitochondria, but Bcl-x_L then binds and ‘retrotranslocates’ Bax back into the cytosol, whereupon this heterodimer disassociates (Edlich et al, 2011). Also, the ‘embedded together model’ posits that Bax and Bcl-2 interact only after both have rearranged to bury their α5 and α6 helices as well as their terminal α9 helix within the MOM (Leber et al, 2007) (see below). Furthermore, in elegant simplified systems using recombinant proteins with liposomes or isolated mitochondria, tBid rapidly bound to the membrane and recruited Bax, but its recruitment was opposed by Bcl-x_L, which could sequester both tBid and Bax, until addition of Bad displaced Bax from Bcl-x_L (Lovell et al, 2008).

One of the great mysteries of cell death is how Bax and Bak homo-oligomerise and disrupt the MOM. Although no 3D structure of any active form of Bax or Bak is yet available, some veils have been lifted. Antibody and biochemical probes indicate that the globular structures of inactive Bax (Suzuki et al, 2000) and Bak (Moldoveanu et al, 2006) are substantially altered during their activation by several rearrangements (Kim et al, 2009; Gavathiotis et al, 2010). Notably, in an early step, Bak exposes its BH3 domain, which can then intercalate into the partial surface groove of another equivalently activated Bak molecule, forming a dimer that appears to be symmetric (Dewson et al, 2008) (Figure 5A). These novel dimers can then multimerise into larger oligomers through a separate interface involving the Bak α6 helix (Dewson et al, 2009). Bax oligomerisation is thought to proceed similarly (Westphal et al, 2011). Nevertheless, the proposed ‘rear’ site on Bax raises the possibility that Bax could instead insert an exposed BH3 domain into that site on another Bax molecule, forming ‘asymmetric’ dimers that would produce a ‘daisy chain’ of monomers (Figure 5B). This is consistent with some evidence (Shamas-Din et al, 2011) but not another study (Dewson et al, 2009). In any case, a key step in Bax and Bak oligomerisation and membrane permeabilisation may well be insertion of their α5 and α6 helices as a hairpin across the bi-layer of the MOM (Leber et al, 2007) (Figure 5B).

Figure 5.

Models for the oligomerisation of Bax and Bak. (A) In this dimer multimerisation model, upon activation by an apoptotic signal, Bak (or Bax) first extrudes its BH3 domain, which allows it to engage another ‘primed’ Bak (or Bax) molecule to form a ‘symmetric’ (face-to-face) dimer (Dewson et al, 2008). These dimers then multimerise by a different interface, such as one between α6 helices, to form the large oligomers that provoke MOMP, allowing cytochrome c (cyt c) to escape to the cytosol (Dewson et al, 2009). (B) In an alternative model (Shamas-Din et al, 2011), prompted by the proposal that certain BH3 domains can engage a novel ‘rear’ site on Bax, ‘primed’ Bak (or Bax) can engage the proposed rear site of another activated molecule to form an ‘asymmetric (face-to-back) dimer’, which could be extended by monomer recruitment. In this model, insertion of the α5–α6 hairpin as well as α9 is depicted. That insertion might also feature in the model shown in (A).

Whether Bax and Bak produce homo-oligomers of a defined size is unclear, but proteinaceous ‘pores’ large enough to allow egress of all the intermembrane proteins probably would require 8–12 mers (Green and Kroemer, 2004). However, the oligomers may instead simply create lipidic pores of undefined size by disrupting the lipid bi-layer (Green and Kroemer, 2004). To clarify these pivotal steps, there is a pressing need for the determination of 3D structures of activated forms of Bax and Bak and of their full-length heterodimers with pro-survival relatives, preferably anchored within the membrane.

An evolutionary conundrum

Given the central role of Bax/Bak-driven MOMP in vertebrate apoptosis, it is paradoxical that mitochondrial disruption plays little if any role in worms and flies. C. elegans lacks any pro-apoptotic Bax/Bak equivalent, and the worm APAF-1 homologue CED-4 does not require cytochrome c to activate the caspase CED-3 (Hengartner, 2000) (Figure 1). Instead, CED-9 directly sequesters CED-4, until EGL-1 binds to CED-9, liberating CED-4 to activate CED-3 (Hengartner, 2000). Furthermore, whereas Bcl-2 homologues have the crucial cell survival role in the worm and vertebrates, that role is dominated in Drosophila by the direct control of caspases by IAPs rather than by the two fly Bcl-2-related proteins, and mitochondrial disruption is not implicated (Dorstyn et al, 2002). Why apoptosis is tightly coupled to Bcl-2-regulated MOMP in vertebrates but not in the best-studied invertebrates remains unclear.

Physiological functions of Bcl-2 family members

Pro-survival proteins

Studies using transgenic and gene-knockout mice have greatly illuminated the biological functions of Bcl-2 family members. Demonstrating the role of the guardians, overexpression in lymphoid and other haemopoietic cells of Bcl-2 (McDonnell et al, 1989; Strasser et al, 1990b, 1991a, 1991b; Sentman et al, 1991), Bcl-x_L (Grillot et al, 1995), Mcl-1 (Zhou et al, 1998; Campbell et al, 2010) or A1 (Chuang et al, 2002) protects against diverse cytotoxic signals, both physiological (e.g. cytokine deprivation) or imposed (e.g. γ-irradiation, chemotherapeutic drugs). Notably, overexpression of pro-survival proteins in lymphocytes leads to lymphadenopathy, which, in the case of Bcl-2 and Mcl-1, can progress at low (but significant) incidence to malignant lymphoma (McDonnell and Korsmeyer, 1991; Strasser et al, 1993, #1475; Zhou et al, 2001, #9379; Egle et al, 2004b; Campbell et al, 2010). Bcl-2 (but not Mcl-1) overexpression can also provoke a fatal kidney disease akin to human systemic lupus erythematosus (SLE) (Strasser et al, 1991b; Egle et al, 2004a; Campbell et al, 2010), establishing that apoptotic defects can promote autoimmune disease.

Different cell types vary in their dependence on individual pro-survival proteins, presumably due to variable expression patterns and selectivity of interactions (Figure 3B and C). For example, bcl-2^−/− mice developed a fatal polycystic kidney disease (PKD), due to the death of renal epithelial stem/progenitor cells in the embryonic kidney; premature greying, due to death of melanocyte progenitors; and immunodeficiency, due to B- and T-cell attrition (Veis et al, 1993). Remarkably, all these degenerative disorders were eliminated by concomitant loss of Bim (in the case of PKD, even loss of a single bim allele) (Bouillet et al, 2001), demonstrating that Bim and Bcl-2 are the major mutually antagonistic regulators of the life/death switch in these cell types.

The cell types most affected by bcl-x loss include fetal erythroid progenitors, certain neuronal populations, male germ cells, immature (CD4⁺CD8⁺) thymocytes, hepatocytes and platelets (Motoyama et al, 1995; Kasai et al, 2003; Mason et al, 2007). The bcl-x^−/− mice die around E14–15 due to severe anaemia and neuronal degeneration (Motoyama et al, 1995) and bcl-x^+/−males have profoundly reduced fertility (Kasai et al, 2003). Interestingly, Bim loss restored fertility in bcl-x^+/− males and prevented the erythroid attrition caused by complete bcl-x loss but did not rescue the neurons, so bim^−/−bcl-x^−/− embryos still die around E15 (Akhtar et al, 2008). Thus, the Bim–Bcl-x_L interaction determines the fate of erythroid progenitors and male germ cells, but Bim is not the sole BH3-only protein regulating neuronal survival.

Loss of Mcl-1 has profound effects. Its complete loss caused pre-implantation embryonic lethality (Rinkenberger et al, 2000), and conditional deletion studies have shown that Mcl-1 is important for the survival of haemopoietic stem cells (Opferman et al, 2005), immature B and T lymphoid progenitors and their mature resting progeny (Opferman et al, 2003), activated germinal centre B cells (Vikstrom et al, 2010), granulocytes and activated macrophages (Steimer et al, 2009), certain neuronal populations in the embryonic brain (Arbour et al, 2008) and hepatocytes (Vick et al, 2009).

Bcl-w and A1 may have more restricted roles. Although Bcl-w is broadly expressed, its loss provoked noticeable defects only in adult spermatogenesis (Print et al, 1998; Ross et al, 1998) and in epithelial cells in the small intestine, which became more sensitive to DNA damage (Pritchard et al, 2000). The mouse has at least three expressed a1 genes and loss of one of them (a1a) accelerated the apoptosis of granulocytes (Hamasaki et al, 1998) and mast cells (Xiang et al, 2001). Targeting all the A1 genes might well reveal additional defects, since diverse stimuli induce A1 expression in many cell types.

Collectively, these results support our belief that the survival of most if not all cells in vertebrates relies upon one or more of the Bcl-2-like guardians. This is consistent with the view that death is the default for most cells unless they receive positive signals from other cells (Raff, 1992). Most likely, paracrine signals conveyed by growth factors and cell–cell or cell–matrix contact promote the survival of most cell types by up-regulating various anti-apoptotic Bcl-2 family members, as well as by curtailing expression and/or activity of particular BH3-only proteins (Youle and Strasser, 2008).

Bax/Bak

Studies with mice lacking either Bax or Bak or both have shown that their functions largely overlap. No abnormalities are discernible in bak^−/− mice (Lindsten et al, 2000), except a modest rise in platelets (Mason et al, 2007), and bax^−/− mice exhibit only mild lymphoid hyperplasia and male sterility, the latter because proper spermatogenesis requires that excess testicular germ cells die to achieve the proper ratio of germ to support cells (Lindsten et al, 2000). In striking contrast, most mice lacking both Bax and Bak die perinatally. Their developmental abnormalities include persisting interdigital webs and increased numbers of certain neuronal populations in the brain. A very small number of bax^−/−bak^−/− mice remain viable for several weeks and develop massive lymphadenopathy (Lindsten et al, 2000; Rathmell et al, 2002). Perhaps surprisingly, tissues widely thought to be shaped by apoptosis appear histologically normal in these animals, suggesting that their superfluous cells may be eliminated by Bok (Figure 3) or by non-apoptotic death processes. Cells from bax^−/−bak^−/− mice are highly resistant to the cytotoxic stimuli that activate the mitochondrial pathway, but, as expected, their lymphocytes (type 1 cells) remain sensitive to death receptor-induced killing (Lindsten et al, 2000; Rathmell et al, 2002).

BH3-only proteins

Death stimuli can induce multiple BH3-only proteins, which can also vary with cell type (Huang and Strasser, 2000; Puthalakath and Strasser, 2002; Youle and Strasser, 2008). Nonetheless, analysis of gene-targeted mice has revealed that individual BH3-only proteins have distinctive physiological roles. For example, Bim (O’Connor et al, 1998), one of the most prominent players, is essential in many cell types for apoptosis induced by growth factor deprivation (Bouillet et al, 1999) and for the deletion of autoreactive thymocytes (Bouillet et al, 2002) and immature B cells (Enders et al, 2003). Consequently, bim^−/− mice develop progressive lymphadenopathy and, on a mixed C57BL/6x129SV background, a fatal SLE-like autoimmune disease (Bouillet et al, 1999). Bim also drives most of the cell death induced by endoplasmic reticulum (ER) stress (Puthalakath et al, 2007) but has less importance in apoptosis induced by DNA damage (Bouillet et al, 1999), which is largely p53 driven.

The puma and noxa genes are both direct transcriptional targets of the tumour suppressor p53 (Vousden and Lane, 2007). Significantly, lymphocytes and certain other cell types from puma^−/− mice are highly resistant to γ-irradiation and DNA damage-inducing chemotherapeutic drugs (e.g. etoposide) (Jeffers et al, 2003; Villunger et al, 2003), demonstrating that Puma is the major mediator of the p53 apoptotic response in those cells. Puma also drives the apoptosis induced by several p53-independent insults, such as growth factor deprivation or treatment with glucocorticoids or phorbol ester (Jeffers et al, 2003; Villunger et al, 2003; Erlacher et al, 2005). Curiously, while Noxa loss only modestly impaired γ-radiation- or etoposide-induced apoptosis of lymphocytes and fibroblasts (Shibue et al, 2003; Villunger et al, 2003), its loss had more impact than Puma loss on UV-radiation-induced apoptosis of fibroblasts and keratinocytes (Naik et al, 2007).

Importantly, mice lacking both Puma and Noxa showed that they account for all of the pro-apoptotic activity induced by p53 in lymphocytes with damaged DNA (Michalak et al, 2008), challenging the physiological significance in that response of ‘mitochondrial p53’ (reviewed in Vousden and Lane, 2007). Indeed, mutations that ablate p53's transcriptional activation function in vivo eliminated all its biological functions, including tumour suppression (Brady et al, 2011). Puma and Bim appear to be the most potent BH3-only proteins in many contexts, and their combined loss renders many cell types more refractory to cell death stimuli than loss of either alone (Erlacher et al, 2006).

Activation of Bid requires its processing by caspase-8 or certain other proteases (e.g. caspase-3, granzyme B) (Li et al, 1998; Luo et al, 1998) (Figure 2). Accordingly, Bid is essential for the death induced by FasL or TNF in certain cell types (type 2 cells), including hepatocytes and pancreatic β cells (Yin et al, 1999; Kaufmann et al, 2007, 2009; Jost et al, 2009), but claims that Bid is needed for the responses to DNA damage, replicative stress or cell-cycle arrest have been challenged (Kaufmann et al, 2007).

Although studies in cell lines implicated Bmf in anoikis (apoptosis on detachment of adherent cells) (Puthalakath et al, 2001), endothelial cells and fibroblasts from bmf^−/− mice proved normally sensitive to this stress (Labi et al, 2008), possibly because other BH3-only proteins, notably Bim (Mailleux et al, 2007), cooperate with Bmf in anoikis. Bmf-deficient lymphocytes are, however, refractory to glucocorticoids, and older bmf^−/− mice accumulate excess B cells (Labi et al, 2008), demonstrating a role for Bmf in B-cell homeostasis.

The other BH3-only proteins seem to have more limited roles. No abnormalities appeared in bik^−/− mice or cells (Coultas et al, 2004), but males lacking both Bim and Bik were infertile due to impaired apoptosis of immature testicular progenitor cells (Coultas et al, 2005), as in bax^−/− males (see above). Although Bad reportedly is regulated by growth factor signalling, Bad-deficient cells remained sensitive to cytokine deprivation and diverse other cytotoxic insults (Kelly et al, 2010). The minor rise in platelets in bad^−/− mice (Kelly et al, 2010) probably reflects reduced blockade of its target Bcl-x_L, the major regulator of platelet survival (Mason et al, 2007). Hrk is expressed largely (perhaps exclusively) in the brain and only delayed death of certain hrk^−/− neuronal cell types in culture has been reported (Imaizumi et al, 2004; Coultas et al, 2007).

Beclin-1: an apparent link from Bcl-2 to autophagy

The Bcl-2 family may control not only commitment to apoptosis but also the initiation of autophagy, an evolutionarily conserved process for maintaining cell survival (e.g. when nutrients are limiting) by self-cannibalism of cellular components, including organelles (Kang et al, 2011). Beclin-1, which is essential for initiation of autophagy, was reported to bind to anti-apoptotic Bcl-2 family members (Pattingre et al, 2005) via a BH3-like domain (Maiuri et al, 2007; Oberstein et al, 2007). The association of Beclin-1 with Bcl-2, Bcl-x_L or Mcl-1 reportedly blocks its ability to activate the PI3 kinase Vps34 and trigger autophagy (Pattingre et al, 2005; Maiuri et al, 2007). Accordingly, a BH3-only protein of higher affinity (as most are), or a BH3 mimetic compound such as ABT-737 (see below), can displace Beclin-1 and allow autophagy. Beclin-1 can also be freed by several other mechanisms (Kang et al, 2011), including mono-ubiquitination or phosphorylation of its BH3 domain. Curiously, only Bcl-2 on the ER and not that on mitochondria seemed to block autophagy induction (Pattingre et al, 2005). These findings may have implications for tumourigenesis and therapy (see below).

The roles of the Bcl-2 family in tumourigenesis

The cancer connection was first made when overexpression of bcl-2, then newly discovered as the gene involved in the t[14;18] chromosomal translocation found in most human follicular lymphomas (Tsujimoto et al, 1984), was shown to prevent the death of haemopoietic cells deprived of cytokine in vitro and to cooperate with myc in immortalisation of lymphoid cells (Vaux et al, 1988) and in lymphomagenesis (Strasser et al, 1990a). This was the first identification of a gene that regulates cell survival (in any species) and the first evidence that evasion of apoptosis contributes to neoplastic transformation.

Bcl-2 transgenes linked to an Igh gene enhancer, mimicking the human t[14;18] translocation, produced plasma cell tumours and lymphocytic leukaemias but not, disappointingly, follicular lymphoma (McDonnell and Korsmeyer, 1991; Strasser et al, 1993). More recently, however, a bcl-2 transgene expressed widely in the haemopoietic compartment did provoke follicular lymphoma, preceded by a florid germinal centre hyperplasia dependent on CD4⁺ T cells, which also accumulated due to the high Bcl-2 levels (Egle et al, 2004a). These observations indicate that the bcl-2 translocation by itself probably is insufficient to cause human follicular lymphoma and that chronic T-cell stimulation may contribute. Further, the long latency of tumour development (McDonnell and Korsmeyer, 1991; Strasser et al, 1993) pointed to a need for additional oncogenic mutations.

As indicated above, myc and bcl-2 exhibit striking synergy in malignant transformation (Vaux et al, 1988; Strasser et al, 1990a). Pertinently, the tumours that develop spontaneously in bcl-2 transgenic mice often exhibit myc gene rearrangements (McDonnell and Korsmeyer, 1991; Strasser et al, 1993), and some highly aggressive human lymphomas harbour both a myc (t8;14) and a bcl-2 (t14;18) translocation (Lee et al, 1989). The basis for this synergy is that Myc overexpression causes not only increased cell proliferation, but under conditions of stress (e.g. growth factor limitation) can also promote apoptosis (Evan et al, 1992). By blocking this apoptosis, Bcl-2 allows the increased expansion of cycling cells at risk of acquiring further mutations.

Mcl-1 and Bcl-x_L are also important players in tumourigenesis. Mcl-1 overexpression predisposes mice to diverse B-cell lymphomas (Zhou et al, 2001) and, notably, haemopoietic stem/progenitor cell tumours (Campbell et al, 2010). Also, Bcl-x_L (Cheung et al, 2004; Swanson et al, 2004; Boylan et al, 2007) and Mcl-1 (Campbell et al, 2010) synergise with Myc in lymphomagenesis and plasmacytomagenesis, and all the pro-survival family members accelerate Myc-induced myeloid leukaemia (Beverly and Varmus, 2009). In humans, Mcl-1 expression is strikingly high in many cases of acute myeloid leukaemia (AML) and multiple myeloma, and diverse cancers exhibit amplified Mcl-1 or Bcl-x genes (Beroukhim et al, 2010).

Just as anti-apoptotic Bcl-2 proteins promote tumourigenesis, pro-apoptotic family members can function as tumour suppressors. Thus, Myc-induced lymphomagenesis was accelerated by loss of Bax (Eischen et al, 2001), Bim (Egle et al, 2004b), Puma (Garrison et al, 2008; Michalak et al, 2009) and Bmf or Bad (Frenzel et al, 2010), because nearly all contribute to Myc-induced apoptosis. Myc activates Puma through Arf-dependent up-regulation of p53 (reviewed in Vousden and Lane, 2007), but how Myc up-regulates the other BH3-only proteins is not yet known. Bax loss also accelerated brain tumour development in a transgenic model (Yin et al, 1997), and γ-radiation-induced thymic lymphomagenesis was enhanced by loss of Noxa (Michalak et al, 2010), Bmf (Labi et al, 2008) or both Bim and Bad (Kelly et al, 2010). Mice lacking both Bim and Puma have enhanced lymphocyte accumulation and some spontaneously develop lymphoma (Erlacher et al, 2006). However, mice lacking even both of p53's major apoptotic targets, puma and noxa, are not tumour prone (Michalak et al, 2009), probably because p53 transcriptional targets mediating additional functions (e.g. growth arrest and senescence) also contribute to its tumour suppressor activity (Brady et al, 2011).

Importantly, loss or suppression of pro-apoptotic family members is also found in human cancer. Bax frameshift mutations appear in ∼50% of colon carcinomas in patients with a DNA mismatch repair defect, due to slippage during replication of an eight G run in the human Bax gene (Rampino et al, 1997). Moreover, 17% of mantle cell lymphomas have homozygous bim deletions (Tagawa et al, 2005), and the bok and puma genes have suffered allelic deletions in diverse cancers (Beroukhim et al, 2010). In many Burkitt lymphomas, bim or puma is epigenetically silenced, often by hypermethylation (Garrison et al, 2008; Richter-Larrea et al, 2010). Notably, the bim hypermethylation correlated with lower remission and shorter survival, and histone-deacetylase inhibitors overcame the chemoresistance (Richter-Larrea et al, 2010).

Defects in Fas death receptor signalling are also tumourigenic. Mutations in Fas or Fas ligand in mice or in patients with autoimmune lymphoproliferative syndrome not only cause progressive lymphadenopathy and systemic autoimmune disease but the longer-term survivors are also predisposed to lymphoma, plasma cell tumours and histiocytic sarcoma (Watanabe-Fukunaga et al, 1992; Davidson et al, 1998; Straus et al, 2001; O’Reilly et al, 2009). Some human non-lymphoid malignancies also display Fas mutations, including 28% of bladder cancers (Lee et al, 1999).

Such findings have led to the now widespread view that evasion of apoptosis is a hallmark of cancer (Hanahan and Weinberg, 2000). However, since a relatively small proportion of human malignancies carry abnormalities that directly affect genes of the Bcl-2 or death receptor families, the survival of cells undergoing neoplastic transformation most likely often relies upon upstream oncogenic pathways that alter expression of Bcl-2 family members. For example, gene expression databases show high Bcl-2 levels in most human lymphoid malignancies, including not only follicular lymphoma (due to the t[14;18] translocation), but also B-chronic lymphocytic leukaemia (CLL) and multiple myeloma. Notably, over half of B-CLL cases exhibit deletions or mutations in micro-RNAs (miR-15a, miR-16-1) that down-regulate Bcl-2 mRNA (Iorio and Croce, 2009). Also, the deregulated NF-κB activity in many lymphomas (Lenz and Staudt, 2010) can up-regulate A1 (Grumont et al, 1999) and Bcl-x_L (Chen et al, 2000). Since loss of even a single bim allele substantially accelerates Myc-induced lymphomagenesis (Egle et al, 2004b), it will be important to examine human malignancies for changes to regulators that reduce its expression even two-fold.

Upstream mutations can also affect the stability and therefore the activity of Bcl-2 family members. In particular, Mcl-1 is very labile, but its level is elevated in diverse tumours by loss of the FBW7 tumour suppressor, the substrate-binding component of an E3 ubiquitin ligase that can promote Mcl-1 degradation (Wertz et al, 2011). Also, in many B lymphoid tumours, Mcl-1 is elevated by increased expression of the de-ubiquitinase USP9X, which spares Mcl-1 from degradation (Schwickart et al, 2010).

Perhaps surprisingly, although Bcl-2 overexpression enhances Myc-induced lymphomagenesis (Strasser et al, 1990a) and is required for sustained growth of such lymphomas (Letai et al, 2004), loss of endogenous Bcl-2 does not impair Myc-induced lymphomagenesis (Kelly et al, 2007). Presumably, other pro-survival relatives sustain the pre-leukaemic cells while they acquire the mutations allowing transformation and progression. Pertinently, loss of even a single mcl-1 allele protected mice from developing Myc-induced AML (Xiang et al, 2010). Identifying the pro-survival family members critical for development and sustained growth of distinct tumour types should provide valuable clues for early intervention and improved treatment of malignant disease (see below).

Although evasion of cell death contributes importantly to cancer development, in certain circumstances excessive cell death actually promotes tumourigenesis. Repeated low-dose γ-irradiation in certain mouse strains causes thymic lymphoma, which arises from bone marrow-derived haemopoietic stem cells that have sustained oncogenic mutations. Remarkably, protection of the differentiated leukocytes from the γ-radiation-induced apoptosis, either by loss of Puma or by Bcl-2 overexpression, ablated lymphomagenesis by obviating the compensatory proliferation of mutated haemopoietic stem/progenitor cells (Labi et al, 2010; Michalak et al, 2010). Potential parallels in human malignancies include hepatocellular carcinoma in patients with viral- or alcoholism-induced cirrhosis, which develops from stem or progenitor cells repeatedly driven into action through cycles of cell death and regeneration (Farazi and DePinho, 2006). Pertinently, liver-specific loss of Mcl-1 in mice provoked sustained hepatic apoptosis, regeneration and ultimately carcinoma (Vick et al, 2009). Furthermore, many cancer survivors eventually develop new primary tumours provoked by their courses of radiotherapy or chemotherapy, which elicit cycles of depletion of differentiated cells replenished by recruitment of stem/progenitor cells, some of which will have sustained oncogenic mutations (Allan and Travis, 2005).

The crucial roles of the Bcl-2 family in cancer therapy

It has long been recognised that most types of tumour cells exposed to cytotoxic cancer therapies exhibit hallmarks of apoptosis and that this process could be critical for the therapeutic response (Kerr et al, 1972). Indeed, overexpression of Bcl-2, Bcl-x_L or Mcl-1 in mice protects lymphoid and myeloid cells against many therapeutic agents (e.g. γ-irradiation, glucocorticoids, etoposide) (Strasser et al, 1991a; Grillot et al, 1995; Zhou et al, 1997; Campbell et al, 2010). Verifying the essential role of the Bcl-2-regulated pathway, combined loss of Bax and Bak renders diverse cell types highly resistant to such cytotoxic insults (Lindsten et al, 2000; Rathmell et al, 2002), whereas crippling the death receptor pathway had no impact (Varfolomeev et al, 1998; Newton and Strasser, 2000; Salmena et al, 2003). Pertinently, in diverse human cancer types, elevated Bcl-2, Bcl-x_L, Mcl-1 or A1/Bfl1 correlates strongly with chemoresistance (reviewed in Cory and Adams, 2002; Youle and Strasser, 2008).

Gene-targeted mice have revealed that distinct BH3-only proteins are required to initiate apoptosis in response to different anti-cancer therapeutics (reviewed in Adams and Cory, 2007; Youle and Strasser, 2008). For DNA-damaging cancer therapeutics, the crucial role of p53 (Vousden and Lane, 2007) implicated its direct transcriptional targets puma and noxa, and indeed Puma and to a lesser extent Noxa proved critical in many cell types (Jeffers et al, 2003; Shibue et al, 2003; Villunger et al, 2003; Erlacher et al, 2005; Michalak et al, 2008). Apoptosis elicited by glucocorticoids, which are used to treat certain lymphoid malignancies, requires Bim, Puma and Bmf (Bouillet et al, 1999; Jeffers et al, 2003; Villunger et al, 2003; Erlacher et al, 2005; Labi et al, 2008), and HDAC inhibitors kill in a Bim- and Bmf-dependent manner (Labi et al, 2008). Taxol-induced cell death also relies heavily on Bim (Bouillet et al, 1999).

These findings for non-transformed cells have generally held for tumour-derived cell lines (Michalak et al, 2009; Happo et al, 2010), although, surprisingly, efficient killing of Myc-driven B lymphoid tumours was found to depend not only on Puma and Noxa but also on Bim (Happo et al, 2010). Since DNA damage of transformed cells can engage the Bcl-2-regulated apoptotic pathway through both p53-dependent and p53-independent routes (Strasser et al, 1994), Bim might be involved in the latter, as p53 does not appear to regulate it directly (Egle et al, 2004b; Kelly et al, 2010).

The discovery that certain oncogenic proteins are critical for sustained growth of a tumour (‘oncogene addiction’) has prompted development of designer drugs that target these key factors, which include several oncogenic kinases (Sharma et al, 2006). Like traditional chemotherapeutics, many designer drugs act at least in part by inducing apoptosis. Thus, the death of chronic myelogenous leukaemia (CML)-derived cell lines treated with imatinib (Gleevec), which inhibits Bcr–Abl, the kinase that causes this disease, proved to require Bim and, to a lesser extent, Bad (Kuroda et al, 2006). Bcr–Abl shutdown may activate Bim both by impairing PI3K-Akt signalling, thereby allowing FOXO3A to induce bim, and by preventing ERK-mediated phosphorylation of Bim, which can target it for ubiquitination and proteasomal degradation (reviewed in Ley et al, 2005). Additional pre-clinical studies have shown that Bim is also critical for the killing of tumour-derived cell lines by drugs that target other oncogenic kinases: EGF receptors in non-small cell lung cancer (Costa et al, 2007; Cragg et al, 2007; Gong et al, 2007), mutant B-Raf in melanoma and colon cancer (Cragg et al, 2008) and JAK2 in myeloproliferative neoplasms (Kobayashi et al, 2010). Thus, although non-apoptotic processes (e.g. senescence) may well contribute to the efficacy of these cancer therapies, activation of apoptosis via BH3-only proteins plays a crucial role (Cragg et al, 2009).

Novel anti-cancer therapeutics that directly trigger the apoptotic machinery

BH3 mimetics

The efficacy of most chemotherapeutic drugs and γ-radiation is blunted by anti-apoptotic abnormalities in the tumour cells, for example, mutant p53 or elevated Bcl-2, Bcl-x_L or Mcl-1. Hence, chemical mimetics of BH3-only proteins that can bypass these defects and directly flip the Bcl-2-regulated apoptotic switch are now avidly sought (reviewed in Lessene et al, 2008). Although a number of compounds have been proposed to function as BH3 mimetics, only ABT-737 (Oltersdorf et al, 2005) and its orally available derivative ABT-263 (navitoclax) (Tse et al, 2008), which is now undergoing clinical trials, have clearly been shown to kill cells through a Bax/Bak-dependent mechanism. The others kill bax^−/−bak^−/− cells as well as wt cells, demonstrating off-target toxicity (van Delft et al, 2006; Lessene et al, 2008).

ABT-737 and ABT-263 both bind with high (low nanomolar) affinity to Bcl-2, Bcl-x_L and Bcl-w but not to Mcl-1 or A1 (Oltersdorf et al, 2005; Tse et al, 2008). Accordingly, these drugs show significant single-agent efficacy in culture and in xeno-transplant mouse models against human tumour cells that express high levels of Bcl-2 (or Bcl-x_L) and low levels of Mcl-1 (Oltersdorf et al, 2005; van Delft et al, 2006), particularly lymphoid malignancies and small cell lung carcinoma (Oltersdorf et al, 2005). ABT-737 also cooperates potently with many standard chemotherapeutic drugs, such as etoposide or cyclophosphamide, and their combined action can even kill refractory tumour cells with high levels of Mcl-1 (Mason et al, 2008), often because those agents promote Mcl-1 degradation (van Delft et al, 2006) or activate BH3-only proteins that can neutralise it (Cragg et al, 2008). For example, agents that provoke mitotic arrest, such as docetaxel, lead to Mcl-1 degradation (Wertz et al, 2011). Pertinently, xenografts of human breast tumours that exhibited elevated Bcl-2 were sensitised by ABT-737 to docetaxel treatment (Oakes et al, 2011), as were lung carcinoma xenografts with elevated Bcl-x_L by navitoclax (Tan et al, 2011). Remarkably, hypoxic regions of tumours, often the most refractory to standard therapy, seem particularly vulnerable to ABT-737, probably because their Mcl-1 levels are lower (Harrison et al, 2011). Importantly, ABT-737 spares most normal cells but preferentially kills older platelets, which depend upon Bcl-x_L (Mason et al, 2007). Hence, transient thrombocytopenia has been the dose-limiting but manageable toxicity in clinical trials of ABT-263. Some normal lymphoid and myeloid populations, such as mast cells, are also relatively sensitive (Carrington et al, 2010).

Of note, ABT-737 synergises potently with drugs designed to target specific oncoproteins, including inhibitors of oncogenic kinases. Examples include synergy with the bcr-abl inhibitor imatinib in killing CML-derived cell lines (Kuroda et al, 2006) and inhibitors of mutant B-Raf signalling in cell lines derived from melanoma or colon carcinoma (Cragg et al, 2008). Thus, combining a BH3 mimetic with an inhibitor of an oncoprotein that targets only the cancer cells should be highly effective at eliminating the tumour cells while limiting collateral damage to healthy tissues (Cragg et al, 2009). Recent work indicates that anti-VEGF therapy of solid tumours destroys the tumour vasculature and abates tumour growth by up-regulating Bim in the endothelial cells (Naik et al, 2011). Thus, some BH3 mimetics may prove to be efficacious in combination therapy by killing not only the tumour cells but also their support cells.

Since BH3 mimetics may promote autophagy by freeing Beclin-1 (see above), this cell survival mechanism might limit the efficacy of BH3 mimetic therapy in some circumstances. Treatment might be improved in such cases by including well-tolerated inhibitors of autophagy, such as chloroquine, which can enhance therapy of pancreatic cancer (Yang et al, 2011).

In principle, a BH3 mimetic drug that attacks a larger subset of pro-survival proteins than ABT-737 (e.g. including Mcl-1) would be more efficacious against cancer cells. However, such agents might well damage too many normal tissues, and BH3 mimetics specific for single pro-survival targets could well have greater clinical utility. Pertinently, GDC-0199, a novel BH3 mimetic from Abbott and Genentech specific for Bcl-2, which is now entering clinical trials for lymphoid malignancies, should avoid the dose-limiting thrombocytopenia that navitoclax provokes by its inhibition of Bcl-x_L.

Once the molecular mechanisms for activation of Bax and Bak are better understood (see above), it may also be possible to design compounds that directly trigger this critical step in apoptosis signalling to enhance the killing of cancer cells or unwanted cells in other diseases, such as autoreactive lymphocytes in autoimmunity. However, whether a therapeutic window exists for such agents remains to be determined.

Death receptor agonists

Engaging the death receptor pathway also has considerable therapeutic potential. Although prohibitive liver damage probably precludes activation of Fas (Ogasawara et al, 1993; Huang et al, 1999), in vitro and xeno-transplant studies suggest that stimulation of TRAIL receptor(s) is safe and efficacious with diverse tumour types, including glioblastoma and certain haemopoietic malignancies (Gonzalvez and Ashkenazi, 2010). Hence, TRAIL and agonistic antibodies against its receptors are in early clinical trials. Since killing through this pathway in most cell types relies upon amplification via the Bcl-2-regulated pathway (see above), combination therapy with BH3 mimetics should improve efficacy.

IAP antagonists

Finally, the elevated IAP levels found in many cancers have stimulated the development of small molecule inhibitors that mimic the action of the natural IAP antagonist SMAC/DIABLO (Du et al, 2000; Verhagen et al, 2000), and these ‘SMAC mimetics’ are now in clinical trial (Straub, 2011). Their key target was expected to be XIAP, because it potently inhibits the effector caspases and perhaps caspase-9. However, cIAP1 and cIAP2, components of the TNF-R1 signalling complex, have also proved to be important targets, because their engagement reduces pro-survival signalling from this receptor and, via NF-κB activation, provokes autocrine production of TNFα, which then elicits TNF-R1-mediated apoptosis (Gaither et al, 2007; Varfolomeev et al, 2007; Vince et al, 2007). Critical questions now are which tumour types are most susceptible to IAP inhibitors and how can they best be combined with other agents? There is evidence for their synergy with a death receptor agonist and cooperation with BH3 mimetics also seems highly likely (Straub, 2011).

Autoimmune, infectious and degenerative diseases

Since defects in apoptosis contribute to the development of autoimmune diseases (Strasser et al, 1991b; Watanabe-Fukunaga et al, 1992; Bouillet et al, 1999), directly targeting the apoptotic machinery could well aid their treatment. Indeed, ABT-737 has shown considerable promise in mouse models of SLE, rheumatoid arthritis and transplant rejection (Bardwell et al, 2009; Carrington et al, 2010). Furthermore, the distinctive Bcl-2 homologues found in many viruses (Cuconati and White, 2002) suggest that virus-specific BH3 mimetics could limit viral infections, including that of viruses such as Epstein–Barr virus that contribute to human cancer. Similarly, parasites, such as schistosomes, have Bcl-2 homologues that could be targeted as a new approach to parasite elimination (Lee et al, 2011). Finally, since excess cell death is the hallmark of degenerative disorders, such as Alzheimer's disease, a deeper understanding of the underlying cell death mechanisms should eventually suggest strategies to block their progression.

Concluding remarks

Over the past 25 years, work from hundreds of laboratories has laid a solid framework for understanding the pathways to cell death, their mechanisms and physiological impact, as well as insights into the types of diseases that can be caused or exacerbated by defects in these processes. Armed with this knowledge, the field is now poised to develop novel therapeutic strategies for cancer, autoimmune diseases and other disorders.