For more than two decades, CD95 — also known as Fas and APO-1 — has been considered a killer, the capo di tutti capi of death receptors in the tumour-necrosis-factor receptor family. Interaction of this cell-surface receptor with its ligand, CD95L, or with activating antibodies induces rapid apoptotic death in many cell types, and injection of such ligands or antibodies into animals results in liver destruction and death. Moreover, deletion or mutation of CD95 causes an immune-related disease due to loss of apoptosis. In many cells, even when apoptosis is blocked, CD95 induces necrotic cell death through a different signalling pathway. So the idea that CD95 actually promotes cancer — which is generally characterized by decreased rather than increased cell death — seems ridiculous. But on page 492 of this issue, Chen and colleagues1 provide overwhelming evidence that this is indeed the case, a finding that is tantamount to a paradigm shift.

The authors use various human tumour-cell lines, primary cancer cell lines and mouse models of spontaneous and damage-induced cancer to show that reducing or abolishing CD95 compromises tumour growth without causing cell death. Furthermore, they find that tumour-derived CD95L is necessary for tumour promotion, presumably acting in a cell-intrinsic (autocrine) way. Their conclusion, that CD95L–CD95 signalling is generally important for the generation and maintenance of cancers, is therefore compelling.

Hints that CD95 has non-apoptotic functions have surfaced repeatedly in the past. (Sometimes, however, a viewpoint becomes so ingrained in our collective scientific consciousness that alternatives go unheeded.) For example, previous work has shown that activation of CD95 has a co-stimulatory effect with antigen-receptor signalling in T lymphocytes, and that neurons in which CD95 is triggered display neurite outgrowth2. In addition, Chen and colleagues1 show that ablation of CD95 in the liver slows compensatory proliferation of liver cells following partial liver resection (hepatectomy).

So how does CD95 promote growth? This receptor's only known ligand, CD95L, can be either membrane bound or cleaved to a soluble form. The membrane-bound form is required for CD95L-induced apoptosis, and animals that can express only the soluble form develop immune-system abnormalities more rapidly than do CD95L-deficient mice3. Remarkably, these animals also develop hepatic sarcoma, a form of cancer rarely seen in animals lacking CD95 or CD95L. The idea that the soluble form of CD95L might be responsible for CD95 signals that promote tumour formation is therefore attractive. Indeed, patients with cancer often have high levels of soluble CD95L in their blood.

Inflammation is an important factor in promoting cancer4. So one idea is that CD95 aids tumour growth by inducing inflammation. Expression of CD95L in abnormal locations such as pancreatic islet cells or transplanted tissues can induce a dramatic infiltration of white blood cells5 — a hallmark of inflammation. Against this idea, however, Chen et al. report that CD95L promotes tumour growth by mechanisms within the tumour, with no profound differences in inflammation between tumours expressing CD95 and those that do not express it. Furthermore, earlier work6 has shown that, unlike membrane-bound CD95L, soluble CD95L does not promote inflammation.

Activation of death receptors, including CD95, triggers several signalling pathways besides that leading to apoptosis (Fig. 1). But which of these signals might promote tumour growth? It has been proposed that activation of the transcription factor NF-κB (probably through the RIPK1 enzyme, which is recruited to the death-induced signalling complex, DISC) might account for the growth-promoting effects of CD95 activation3. But this seems unlikely, because although CD95 promotes damage-induced hepatocellular carcinoma1, this form of cancer is exacerbated in mice that lack the canonical NF-κB signalling pathway in their livers4.

Figure 1: CD95 signalling for apoptosis, and for tissue growth and tumour promotion.
figure 1

On binding to its ligand, CD95L, CD95 generates different signals, depending on the cell type and other conditions. a, For apoptosis, the adaptor protein FADD must be recruited to the exposed death domain of the activated receptor. FADD, in turn, recruits the initiator procaspase-8, which cleaves itself to form the active caspase-8 — an enzyme that promotes apoptosis. b, When cells express FLIP, apoptosis does not occur and, instead, other cell-type-dependent signalling events become manifest. These include activation of JNK by an unknown mechanism (leading to expression of EGR-1 and Fos), of NF-κB by a scaffold complex involving RIPK1, and of PI3 kinase (PI3K), mediated by Yes and p85 (not shown), and its downstream effector, AKT. The tumour-promoting functions of CD95, which Chen et al.1 describe, may involve some or all of these non-apoptotic activities.

Alternatively, Chen et al.1 suggest a role for the enzyme JNK, and increased expression of EGR-1 and Fos transcription factors downstream of it. They show that chemical inhibition of JNK retards the growth of CD95-expressing tumour cells. Moreover, CD95 activation in liver cells undergoing proliferation after partial hepatectomy (a situation that may be mechanistically similar to tumour initiation) profoundly increases levels of EGR-1 and Fos.

Unfortunately, we know almost nothing about how CD95 might activate JNK. Soluble CD95L does not trigger JNK activation7, but a deeper problem is that, in the absence of CD95, JNK can be activated in many other ways. Indeed, several other related proteins — such as tumour-necrosis factor — readily induce JNK, and it is difficult to see why CD95 would be needed for this signal to promote cancer. In other words, what is so special about CD95?

Apoptosis in response to CD95 activity is controlled by proteins that associate with the intracellular death domain of this receptor, such as the adaptor protein FADD, leading to the recruitment of DISC and activation of the enzyme caspase-8 (Fig. 1). FLIP — a caspase-like molecule without proteolytic activity — alters this signalling in favour of survival. Presumably, this is how tumour cells avoid death signalling by CD95. But there is a hint that this is not the whole story.

Animals with defective CD95 expression exhibit delayed liver-cell proliferation and liver regeneration after partial hepatectomy, whereas animals with a mutation in CD95 known as lprcg, which disrupts the structure of the death domain, show normal regeneration8. Remarkably, when lprcg-mutant mice were provided with a wild-type haematopoietic system (to prevent the immune-system disease), they developed liver tumours9. It seems, therefore, that the tumour-promoting signal that CD95 generates — perhaps normally offset by CD95-induced apoptosis — persists in this mutant.

Such a signal might come about through the recruitment and activation of the enzyme PI3 kinase. In glioblastoma cells, activated CD95 recruits both the Src-family kinase Yes and the p85 subunit of PI3 kinase, leading to PI3-kinase activation10. Yes and p85 interact with the membrane-proximal region of the CD95 death domain, which is likely to remain intact in the lprcg mutant, although this has not been tested. Thus, CD95 activation may promote the activity of the signalling molecule AKT, downstream of PI3 kinase, to induce cell proliferation.

Whatever the mechanism, a general role for autocrine CD95 signalling in promoting cancer is a stunning revelation that goes against many of the prevailing ideas of what this receptor does. There are obvious practical consequences as we explore the benefits of blocking CD95L–CD95 interactions in cancer therapy. But perhaps most of all, as we confirm the generality of this process in cancers, there is our realization that this 'wolf' — this potentially deadly tumour-promoting mechanism — has been there all along, disguised as a mechanism of cell death.