Main

In the last decade, a new mechanism for membrane-to-nucleus signalling has gained recognition. In this new mechanism, instead of propagating signals through a cascade of intermediate messengers, transmembrane receptors directly respond to stimuli by undergoing regulated intramembrane proteolysis (RIP)1,2. This process is mediated by proteases that cleave within the membrane (intramembrane cleaving proteases or iCLiPs; Box 1) and results in the release of bioactive signalling fragments. The first RIP-mediated signalling pathway to be recognized involves the activation of sterol-regulatory-element-binding protein (SREBP). When cells are deprived of sterols, SREBP travels from the endoplasmic reticulum to the Golgi, where it is sequentially cleaved by site-1 protease and site-2 protease (S2P)3. Intramembrane cleavage by S2P releases the basic-helix–loop–helix-domain-containing cytosolic fragment, which, in turn, translocates to the nucleus. There, it binds directly to its cognate sites in the promoters of genes that are involved in lipid metabolism3,4.

Another well-established example of a pathway that is mediated by RIP is Notch signalling (Fig. 1a). On ligand binding, the Notch receptor undergoes a conformational change that allows ectodomain shedding through the action of tumour-necrosis-factor-α-converting enzyme (TACE), a metalloprotease. This is followed immediately by intramembrane cleavage that is mediated by the γ-secretase complex5,6, the catalytic centre of which is in the iCLiP presenilin (Fig. 1b; Box 1). The liberated Notch intracellular domain (ICD) translocates to the nucleus, where it interacts with CSL (C-promoter-binding factor/recombination signal-sequence binding protein Jκ/Suppressor-of-Hairless/Lag1), a Rel-family transcription factor7. The binding of the Notch ICD and the recruitment of the co-activator protein Mastermind converts CSL from a transcriptional repressor to an activator5,6,8, which results in the expression of Notch target genes.

Figure 1: The γ-secretase complex and its role in the intramembrane proteolysis of Notch and amyloid precursor protein.
figure 1

a | In response to ligand binding, the Notch receptor undergoes ectodomain-shedding by metalloprotease cleavage at the site S2. This is followed by γ-secretase-mediated cleavage near the cytoplasmic end of the transmembrane domain (TMD; S3), which liberates the Notch intracellular domain (ICD). Notch is also cleaved in the middle of its TMD (S4), which releases an Nβ fragment. b | Schematic representation of the components of the γ-secretase complex. Presenilin, which is endoproteolysed during its maturation, contains the catalytic aspartyl residues. Nicastrin, APH1 and PEN2 are important for the maturation and trafficking of the complex. The precise stoichiometry of the complex remains to be determined, although recent findings indicate that presenilin might function as a dimer41. Although studies indicate that these four proteins are sufficient to generate the active enzyme, reconstitution of the protease activity using purified proteins has not been shown. Part b is reproduced with permission from Ref. 6 © (2002) Macmillan Magazines Ltd. c | The amyloid precursor protein (APP) ectodomain is initially shed by α-secretase- or β-secretase-mediated cleavage, and subsequent cleavage of several scissile bonds by γ-secretase in the middle of the TMD (only one cleavage site is shown in the figure) releases P3 or the amyloidogenic Aβ peptides, respectively. The Aβ peptides (that is, Aβ40 and Aβ42) differ from each other in their carboxy-terminal composition. γ-Secretase also cleaves near the cytoplasmic end of the TMD at a site known as S3-like or ε, and thereby releases the APP ICD.

Full size image

The γ-secretase complex has long been of particular interest to Alzheimer's disease researchers, because of its involvement in the production of Aβ peptides from amyloid precursor protein9 (APP; Fig. 1c). The picture of γ-secretase that is emerging from recent studies is that of a promiscuous enzyme that seems to cleave various type-I-transmembrane-domain proteins after they have undergone ectodomain shedding6,10. Further γ-secretase substrates are shown in Table 1 (for a fully referenced version of this table, see online supplementary information S1 (table); for reviews, see Refs 6,11). With so many other membrane proteins known to undergo ectodomain shedding, the number of identified substrates will probably grow. So, are all the ICDs that are released from a membrane tether by γ-secretase bioactive molecules that have a role in signalling? Or, could γ-secretase be providing cells with a general mechanism to remove transmembrane domains from lipid membranes? And, how do we differentiate between the possible signal-generating and degradative functions of γ-secretase (or of any protease, for that matter)?

Table 1 γ-Secretase substrates and their proposed biological functions
Full size table

In this article, we survey the literature that describes the γ-secretase substrates that have been identified so far, and show that it indicates that γ-secretase mediates signalling from, and the degradation of, type-I transmembrane proteins. We also discuss the experimental approaches that have been used to assess signalling functions, and evaluate the possibility that the main biological activity of γ-secretase is to function as the proteasome of the membrane (Fig. 2).

Figure 2: The 26S proteasome and the γ-secretase complex: analogous roles?
figure 2

a | The 26S proteasome is a large, multisubunit protease complex, which has an important role in maintaining cellular homeostasis by degrading redundant, misfolded and damaged proteins42. Proteins are targeted for proteasomal degradation by the covalent attachment of a polyubiquitin chain through the stepwise action of ubiquitin (Ub)-activating (E1), Ub-conjugating (E2) and Ub-protein-ligase (E3) enzymes. The 19S regulatory subcomplex recognizes polyubiquitylated proteins, removes the Ub groups, unfolds the target proteins and translocates them into the interior of the 20S core subcomplex, where they are progressively degraded into peptides. The proteasome can also generate bioactive fragments — for example, the p50 and p52 subunits of the transcription factor nuclear factor-κB are processed by the proteasome from the larger precursors. In addition, the proteasome converts the transcription factor Cubitus interruptus (Ci) from an activator (Ci150) to a repressor (Ci75) in a ligand-regulated manner. b | The γ-secretase complex might have an analogous role to the proteasome. Like the proteasome, γ-secretase is a multiprotein complex that is composed of catalytic and accessory components. γ-Secretase can cleave the transmembrane domains of a wide variety of membrane proteins that have undergone ectodomain shedding. Although γ-secretase can generate biologically active fragments from some substrates (for example, Notch), its primary function might be to facilitate the selective disposal of type-I membrane proteins (for example, E-cadherin and syndecan).

Full size image

Regulation of γ-secretase cleavage

If the released intracellular fragment of a transmembrane protein has a role in signalling, then it is probable that its intramembrane proteolysis is precisely regulated. As γ-secretase readily cleaves its substrates once their extracellular domain has been removed, the regulation probably occurs at the level of ectodomain shedding. Ectodomain shedding can be modulated by extracellular or intracellular cues in different biological contexts. Among the known γ-secretase substrates, only the intramembrane cleavage of Notch, syndecan-3 and ERBB4 has been shown to be stimulated by ligand binding (Table 1; see also online supplementary information S1 (table)). The secreted glycoprotein F-spondin has recently been shown to bind the APP ectodomain and to inhibit shedding by α- and β-secretase, but the physiological relevance of this observation is still unclear12. For some substrates, ectodomain shedding seems to be constitutive, whereas for others it can be enhanced by different stimuli. For example, E-cadherin ectodomain shedding by matrilysin (also known as matrix metalloprotease-7) is enhanced in response to injury of the lung epithelium in vivo, and this process seems to be required for cell migration and wound closure13. RIP of E-cadherin could therefore function as a sensor during the healing process. N-cadherin proteolysis can be stimulated in cell culture in response to membrane depolarization by N-methyl-D-aspartate (NMDA)-receptor agonists14, although it is yet to be determined if this occurs during neurotransmission in vivo. Ectodomain shedding of CD44, LRP (low-density-lipoprotein receptor protein), nectin-1α and p75 can be induced by treating cells with phorbol esters and/or ionomycin, which activate protein kinase C and promote calcium influx, respectively. This indicates that second-messenger systems are potential modulators of RIP in these pathways. Although ectodomain shedding might be regulated by physiological conditions for all substrates, this remains to be determined experimentally for each protein.

Activity of intracellular fragments

A more stringent criterion of signalling is that the fragment released by RIP should account for either all or part of the activity that is attributed to the substrate (that is, the intact transmembrane protein). For the S2P substrates SREBP and activating transcription factor-6 (Ref. 4), and for the γ-secretase substrate Notch, such a functional relationship between the precursor and the fragment has been established by a myriad of experiments.

The ICDs of most γ-secretase substrates contain biologically relevant properties — that is, transcription-activation domains, protein-interaction domains, recognition sites for protein modification, enzymatic activities and/or alternative localization signals. It is therefore tempting to speculate that these ICDs have further functions in other parts of the cell. Several approaches have been used to gain insights into the possible functions of the various ICDs (Table 1; see also online supplementary information S1 (table)). Overexpression assays have shown that several of the ICDs that are released by γ-secretase cleavage have potential functions as transcriptional regulators. For example, the N-cadherin ICD can depress cAMP-responsive-element-binding protein (CREB)-dependent transcription by targeting the transcription co-activator CBP (CREB-binding protein) for degradation14. In addition, the CD44 ICD can activate promoters that contain a TPA (12-O-tetradecanoylphorbol-13-acetate)-responsive element, which thereby elevates its own expression15. Furthermore, the Jagged ICD can stimulate activator-protein-1-mediated reporter expression and this activity is inhibited by the Notch ICD16.

The APP ICD might also have a role in transcriptional regulation. It is thought to form a complex with the nuclear adaptor protein FE65 and the histone acetyltransferase TIP60 (Refs 17,18). This complex can bind to the promoter of the tetraspanin protein KAI1, both in cells that have been co-transfected with APP, FE65 and TIP60, and in mice that are overexpressing APP (Ref. 18). In both cases, the complex displaces the nuclear-co-repressor-associated complex, which leads to the activation of KAI1 transcription. It is still unclear whether the APP ICD is a stoichiometric component of the nuclear FE65–TIP60 complex under physiological conditions. If the APP ICD is required for KAI1 transcriptional activation in the nucleus, it should behave similarly to the Notch ICD, which is hardly detectable on western blots from living tissue, but is enriched by immunoprecipitation with antibodies against its DNA-binding partners. Such experiments have yet to be done for the APP ICD (see note in proof).

A transcriptional function has also been proposed for the LRP ICD, as it was found to inhibit the transcriptional activity of an APP–Gal4-binding-domain fusion construct (Ref. 19). However, whether LRP ICD can interfere with any of the targets of the FE65–TIP60 complex remains to be tested under physiological conditions. The ERBB4 ICD translocates to the nucleus after release by γ-secretase cleavage, which indicates that it might also have a transcriptional function. Blocking the cleavage of the ERBB4 receptor with γ-secretase inhibitors prevented the growth inhibition that can be induced by the ERBB4 ligand heregulin20,21. However, this assay was not able to differentiate between a requirement for the actual cleavage event (and therefore a rapid termination of a membrane-proximal event) and an activity that is mediated by the ICD itself. This could be clarified by testing whether the ERBB4 ICD alone can mimic the action of heregulin and lead to growth inhibition.

Most type-I transmembrane proteins have a 'stop-translocation' signal that contains a short stretch of Arg/Lys residues and can function as a nuclear localization signal (NLS) after γ-secretase cleavage (in the Notch ICD, NLS signals are located elsewhere). Although this is evocative, simply showing that nuclear translocation of the ICD occurs and that an impact on the transcription of some genes is seen in overexpression assays is not sufficient to assign a signalling function to these cleaved fragments. These methods can sometimes highlight pathological, rather than physiological, roles. For example, Notch can have an important and crucial role in cancer pathology. Overexpressed Notch ICD can stabilize Myc in Mv1Lu epithelial cells, which renders them resistant to transforming growth factor-β (TGFβ)-induced growth arrest. The Notch ICD does not have a similar activity in any other cell line22. Moreover, this activity of the Notch ICD requires abnormally high levels of expression and cannot be induced by a Notch ligand even in Mv1Lu cells22. So, it is possible that the ability of some ICDs to activate transcription when overexpressed is an important assessment of their pathological potential, but that it does not necessarily reflect their physiological role.

The use of proteins that mimic an ICD when the exact composition of an ICD fragment is unknown compounds this problem. As there is no consensus sequence that defines a γ-secretase cleavage site, the cleavage site cannot simply be deduced from the transmembrane sequence. For example, some studies have used APP ICDs that were predicted based on the possible sizes of the Aβ peptides (and were therefore 57–59 amino acids long). However, APP ICDs that were later isolated from cells overexpressing APP were actually shorter9,23 (49–50 amino acids long; Fig. 1c; Table 1). This emphasizes the importance of empirically determining the exact γ-secretase cleavage site(s) in these substrates. The unintended addition or elimination of amino-terminal amino acids could affect the activity that is measured in assays conducted with ICD mimicking proteins.

Despite some limitations, overexpression assays will continue to be used to test ICD function. However, most would agree that these types of studies should be complemented by strategies that can assess physiological relevance. In the case of the Notch receptor, a crucial demonstration of the significance of proteolytic processing is that a loss-of-function phenotype can be rescued or reversed by expressing the γ-secretase-generated cytosolic fragments in cultured cells and in animals1,5,6. Clearly, such rescue experiments are more difficult when there is no clear phenotype associated with loss of substrate, as is the case for APP. As most γ-secretase substrates have only been identified recently, in vivo experiments that are designed to test the role of proteolysis in the biology of these substrates have not yet been carried out. For example, the importance of ectodomain shedding in disassembling E-cadherin-mediated adhesion during wound healing in vivo is well-established13, but roles for the γ-secretase-mediated cleavage and the released ICD in this process are yet to be shown. However, Marambaud et al.24 have identified mutations (GGG759–761AAA) in E-cadherin that abolish presenilin binding and E-cadherin ICD release. As with the processing-deficient Notch-receptor mutation (V1744G)25, if proteolysis and ICD release are significant for any part of the biology of E-cadherin, and if the GGG759–761AAA mutations only impact the interaction with presenilin, this E-cadherin mutant should mimic a loss-of-function phenotype when it is 'knocked-into' the mouse genome.

Signalling or degradation?

Because the regulation and biological relevance of intramembrane proteolysis have not yet been fully explored for most of the γ-secretase substrates, many conclusions about signalling functions remain speculative. Although it is possible that several γ-secretase substrates will release an ICD that has signalling properties, it is also possible that some will not. More studies have to be carried out to distinguish between the possibilities that membrane release is required for activity/termination of activity or for degradation. Experiments that test the role of RIP in a particular pathway must be able to distinguish between the functions of the ectodomain-shedding cleavage and the functions of the γ-secretase-mediated intramembrane cleavage, as well as between the release of associated proteins and the release of the ICD.

The γ-secretase-mediated cleavage of syndecan-3 is an example of a cleavage event that regulates the membrane association and subcellular localization of a protein26. The binding of the multidomain protein calmodulin-dependent serine kinase (CASK) to the cytosolic domain of syndecan-3 keeps it in a juxtamembrane location. Ectodomain shedding and γ-secretase cleavage of syndecan-3 releases its ICD as well as CASK, and the latter translocates to the nucleus. No transcriptional function has been ascribed to the syndecan-3 ICD itself, although CASK is known to bind the T-box transcription factor TBR1 (Refs 26,27). Similarly, on the induction of apoptosis or calcium influx, γ-secretase cleavage releases the E-cadherin ICD together with its bound proteins, α- and β-catenin24. This process completes the intracellular part of adherens-junction disassembly, but it is not thought that the released E-cadherin ICD is an actively signalling molecule.

Other possibilities can be envisioned. For membrane proteins that function mainly through ectodomain shedding, γ-secretase cleavage might just facilitate the disposal of the membrane-tethered carboxy-terminal fragments that are left after the shedding event. For membrane proteins that transmit signals from the membrane through protein cascades and second messengers, it is possible that following their activation/function, ectodomain shedding triggers their subsequent release from the membrane and their rapid degradation as a mechanism to downregulate receptor activity. In other contexts, ectodomain shedding and subsequent γ-secretase cleavage might function as a constitutive disposal mechanism that facilitates membrane-protein turnover28. A degradation model fits just as well as a signalling model with the available data for several γ-secretase substrates. For example, APP proteolysis — through the combined activities of the extracellular α- and β-secretases and the transmembrane γ-secretase — seems to be constitutive, which results in the unregulated release of Aβ/P3 peptides and the APP ICD (Fig. 1c). Failure to dispose of these fragments contributes to, or causes, Alzheimer's disease, but under physiological conditions it could all be part of normal APP turnover29. The APP ICD fragment is rapidly degraded and is yet to be identified in vivo30,31. Indeed, several other ICDs that are released by γ-secretase (for example, the ICDs of DCC (deleted in colorectal cancer), syndecan-3, nectin-1α and p75) are rapidly degraded as well. In these cases, the ICDs have been visualized with the aid of proteosome inhibitors or in membrane preparations in which the proteosome machinery is uncoupled from in vitro ICD generation. The same might be true for Delta, which is cleaved in the absence of Notch binding32. The Delta ICD could be targeted for degradation by the E3 ubiquitin ligases Neuralized or Mind-bomb. Notably, even the signalling function of the SREBP and Notch ICDs are ultimately terminated by proteasome degradation. The difference between degradation and signalling functions might therefore lie in the half-life of the ICD after its release. This might be another complicating factor in assessing the results of overexpression experiments that increase the amounts (and therefore the steady-state levels) of ICD fragments.

The membrane proteasome?

All the previously mentioned possibilities imply a general role for γ-secretase (and perhaps other iCLiPs) in maintaining membrane-protein homeostasis. In this model, γ-secretase functions as the membrane proteasome (Fig. 2). It facilitates the disposal of type-I transmembrane domains by executing numerous cleavages within the short hydrophobic transmembrane sequence, which results in regulated intramembrane degradation. This would be consistent with the apparent loose specificity of γ-secretase with respect to its cleavage sites within transmembrane regions10. Moreover, the possibility that degradation is an important function of γ-secretase gains support from the observation that active γ-secretase complexes are enriched in lysosomal membranes from rat livers33.

The possibility that the γ-secretase complex is analogous to the intracellular proteasome (Fig. 2) has been mentioned before23,34,35. Like the proteasome, γ-secretase is also a multiprotein complex that is composed of catalytic and accessory proteins. Analogous to the ubiquitin conjugation that prepares target proteins for proteasomal degradation, ectodomain-shedding prepares membrane substrates for subsequent γ-secretase cleavage. Notably, the proteasome, which is well recognized as a dedicated protein-degradation machine, does not always completely degrade its substrates. In a few instances, the proteasome can specifically process a polyubiquitylated substrate to produce a biologically active protein fragment. For example, proteasome-dependent processing generates the active subunits (p50 and p52) of nuclear factor-κB from longer inactive precursors (p105 and p100, respectively)36. This is also true for the Drosophila melanogaster transcription factor Cubitus interruptus (Ci). The full-length form (Ci155) is a transcriptional activator, which is cleaved by the proteasome to generate a truncated protein (Ci75) that functions as a transcriptional repressor37,38. In fact, the proteasome can even mediate the regulated release of membrane-tethered nuclear factors in yeast, a process that has been called regulated ubiquitin/proteasome-dependent processing or RUP39. Therefore, a fraction of γ-secretase and proteasome substrates acquire other activities after their release — our job is to correctly identify the ones that do.

Conclusions

Intramembrane proteolysis is an emerging and exciting field in cell biology. In deciphering the mechanisms of Notch signalling, modern investigators benefited immensely from a century's worth of genetics on the notch locus in Drosophila melanogaster and several decades of genetics on the Notch homologues GLP-1 and LIN-12 in Caenorhabditis elegans to complement their cell-culture-based experiments. Notch-related disease processes also provided important clues. Although the assignment of any function to a proteolytic event or to a protein fragment requires rigorous confirmation, perhaps the most important contribution that has been made by the studies summarized here (and by the many more that had to be omitted because of space constraints) is in validating that some proteins can release bioactive, signalling 'offspring' through the activity of proteases. Even if only a small fraction of the proteome truly functions in sensing or signalling through RIP, what an exciting layer of complexity that would add to biology.

Another conclusion that emerges from comparing the handful of known iCLiPs is that the same familiar protease motifs that are found in intracellular proteases (serine proteases, metalloproteases and aspartyl proteases) are used for proteolysis in the lipid bilayer. Perhaps iCLiPs evolved primarily to facilitate the regulated degradation of integral membrane proteins but, with time, as some of the released fragments acquired signalling properties, evolution selected the mechanisms that allow regulated signalling by some substrates.

Finally, further studies are required before the main cellular roles that are carried out by γ-secretase can be fully understood. It is still unclear if γ-secretase represents a heterogeneous activity. It is possible that different γ-secretase complexes exist (for example, complexes that are composed of different APH1 isoforms or presenilin-1 versus presenilin-2; Fig. 1b), and these complexes might differ from each other in their localization, their selection of substrates and, therefore, their cellular function. Further challenges include determination of the full spectrum of γ-secretase substrates, genetic analyses to confirm proposed physiological functions for the cleavage products that are generated by γ-secretase, and determination of whether several other iCLiPs (including those that are yet to be discovered) can overlap with γ-secretase to collectively provide the cell with the 'proteasome of the membrane'.

Note in proof

A recent report from Cao and Sudhof argues that the APP ICD functions at the membrane to potentiate FE65 and is not participating in a nuclear complex48.