Regulated intramembrane proteolysis – lessons from amyloid precursor protein processing


  • Stefan F. Lichtenthaler,

    1. DZNE – German Center for Neurodegenerative Diseases, Adolf-Butenandt-Institute, Biochemistry, Ludwig-Maximilians-University, Munich, Germany
    Search for more papers by this author
  • Christian Haass,

    1. DZNE – German Center for Neurodegenerative Diseases, Adolf-Butenandt-Institute, Biochemistry, Ludwig-Maximilians-University, Munich, Germany
    Search for more papers by this author
  • Harald Steiner

    1. DZNE – German Center for Neurodegenerative Diseases, Adolf-Butenandt-Institute, Biochemistry, Ludwig-Maximilians-University, Munich, Germany
    Search for more papers by this author

Address correspondence and reprint requests to Christian Haass, DZNE – German Center for Neurodegenerative Diseases, Adolf-Butenandt-Institute, Biochemistry, Ludwig-Maximilians-University, 80336 Munich, Germany. E-mail:


J. Neurochem. (2011) 117, 779–796.


Regulated intramembrane proteolysis (RIP) controls the communication between cells and the extracellular environment. RIP is essential in the nervous system, but also in other tissues. In the RIP process, a membrane protein typically undergoes two consecutive cleavages. The first one results in the shedding of its ectodomain. The second one occurs within its transmembrane domain, resulting in secretion of a small peptide and the release of the intracellular domain into the cytosol. The proteolytic cleavage fragments act as versatile signaling molecules or are further degraded. An increasing number of membrane proteins undergo RIP. These include growth factors, cytokines, cell adhesion proteins, receptors, viral proteins and signal peptides. A dysregulation of RIP is found in diseases, such as leukemia and Alzheimer’s disease. One of the first RIP substrates discovered was the amyloid precursor protein (APP). RIP processing of APP controls the generation of the amyloid β-peptide, which is believed to cause Alzheimer’s disease. Focusing on APP as the best-studied RIP substrate, this review describes the function and mechanism of the APP RIP proteases with the goal to elucidate cellular mechanisms and common principles of the RIP process in general.

Abbreviations used:

amyloid β-peptide


Alzheimer’s disease


a disintegrin and metalloprotease


APP intracellular domain


amyloid precursor protein


β-site APP cleaving enzyme


C-terminal fragment


epidermal growth factor


endoplasmic reticulum


intracellular domain


intramembrane-cleaving protease






N-terminal fragment


P: proline, A: alanine, L: leucine




regulated intramembrane proteolysis


tissue inhibitors of metalloprotease


transmembrane domain

Cells have developed a variety of mechanisms to communicate with their environment and to adapt to changes in the extracellular space. One such mechanism is regulated intramembrane proteolysis (RIP) (Brown et al. 2000; Lichtenthaler and Steiner 2007), which controls the activity of membrane proteins and occurs in all organisms studied to date. RIP is required for signal transduction and affects diverse cellular processes, such as cell differentiation, transcriptional regulation, axon guidance, neurite outgrowth, cell adhesion, lipid metabolism, cellular stress responses and the degradation of transmembrane protein fragments. In animals, RIP is essential for a variety of physiological processes, such as embryonic development, the normal functioning of the immune system and the nervous system. The RIP process appears to be tightly regulated, and a deregulation of RIP is associated with diseases, such as cancer and Alzheimer’s disease (AD).

In many cases, RIP starts with an initial cleavage of a single-span membrane protein substrate with type I or II orientation (Fig. 1). This cleavage, which is referred to as shedding, occurs within the ectodomain at a peptide bond close to the transmembrane domain (TMD), either constitutively or in response to ligand binding. Shedding results in the release of the ectodomain into the lumen of vesicles or into the extracellular milieu and in the generation of a membrane-bound stub, which then undergoes a second cleavage within its TMD, the final intramembrane cut. As a result, a small peptide is secreted into the vesicle lumen or the extracellular space, whereas the intracellular domain (ICD) is released into the cytosol. RIP occurs at/within the membrane of different cellular compartments, ranging from the endoplasmic reticulum to the Golgi and the plasma membrane as well as to the endosome.

Figure 1.

 Schematic representation of regulated intramembrane proteolysis (RIP). A membrane protein undergoes two consecutive proteolytic cleavages by two distinct proteases (shown in blue). First, a membrane-bound protease cleaves close to the transmembrane domain, resulting in shedding of the ectodomain. The remaining membrane-bound fragment is further cleaved by an intramembrane protease. A small peptide is secreted into the lumenal/extracellular space and the intracellular domain (ICD) is released into the cytosol, where it may either be further degraded, or participate in cytosolic signaling reactions or translocate to the nucleus and stimulate transcriptional activation of target genes. RIP substrates are mostly single-span membrane proteins. The membrane is represented as a grey box.

The number of transmembrane protein substrates undergoing RIP is steadily increasing, implicating RIP in an even wider range of biological processes. By now, over 60 substrates have been identified for the RIP proteases, including growth factors, cytokines, receptors, cell adhesion proteins, signal peptides and viral proteins (Edwards et al. 2008; Fluhrer et al. 2009; Freeman 2009; McCarthy et al. 2009; Willem et al. 2009).

The soluble protein fragments generated during the RIP process – in particular the ectodomain and the ICD, serve as signaling molecules or are further degraded. For example, shedding of a growth factor releases the ectodomain, which acts as a signaling molecule and activates target cells. For other RIP substrates, such as the cell surface receptor Notch and the cell adhesion protein CD44, it is the ICD, which acts as a signaling molecule, by translocating to the nucleus and stimulating the transcription of target genes (Selkoe and Kopan 2003; Nagano and Saya 2004). An additional example is the cytokine tumor necrosis factor α (TNFα), where both the ectodomain and the ICD activate signal transduction, but in different cells (Black et al. 1997; Fluhrer et al. 2006; Friedmann et al. 2006). Some ICDs do not translocate to the nucleus, but act as cytosolic signaling molecules (Georgakopoulos et al. 2006; Inoue et al. 2009). Given that ICDs are typically short-lived, some of them may not serve a signaling function and may be further degraded. In fact, intramembrane proteolysis may be an elegant way to remove the membrane-bound stubs, which remain after shedding. Thus, intramembrane proteolysis has two fundamentally different functions, namely signaling and degradation of membrane proteins.

The initial shedding is mediated by proteases referred to as sheddases, which are mostly members of the ADAM family (a disintegrin and metalloprotease) and the aspartyl proteases BACE1 and BACE2 (β-site APP cleaving enzymes). In addition, soluble and membrane-bound matrix metalloproteases mediate the shedding of membrane proteins, but it remains to be determined, which of these cleavages are followed by intramembrane proteolysis. The intramembrane cleavage is carried out by intramembrane-cleaving proteases (I-CLiPs) belonging to three distinct protease families, the GxGD-type aspartyl proteases (G: glycine, x: any amino acid, D: aspartic acid), the S2P-metalloproteases and the rhomboid serine proteases. These are multi-span integral membrane proteins, having their active sites located in their hydrophobic regions, typically in the predicted transmembrane domains. The prediction was validated by the first crystal structure of a rhomboid protease from Escherichia coli (Wang et al. 2006b; Wu et al. 2006).

Generally, intramembrane proteolysis is preceded by ectodomain shedding. In fact, most I-CLiPs do not cleave the full-length proteins, but require the bulk of their ectodomain to be removed for efficient cleavage (Struhl and Adachi 2000). Typically, I-CLiP substrates contain about 30 or less amino acids of the ectodomain. The exception to this rule are the rhomboid I-CLiPs, which medidate intramembrane proteolysis without prior shedding (reviewed in Freeman 2009). Conversely, there are examples, where shedding can occur without subsequent I-CLiP cleavage. This is specifically the case for glycosyl phosphatidyl inositol (GPI)-anchored proteins, such as the prion protein, which lack a typical transmembrane domain. In addition, a large number of single span membrane proteins of type I or type II orientation undergo shedding (Faca et al. 2008), but it remains to be determined, if all of them are further subject to I-CLiP cleavage. Recent evidence suggests that at least some I-CLiPs efficiently differentiate even between highly similar membrane proteins (Martin et al. 2009).

The RIP process itself is tightly regulated. For example, too much or too little Notch signaling by the RIP process leads to developmental defects. Likewise, too much Notch signaling is observed in certain leukemias (Ferrando 2009). In addition, too much RIP of amyloid precursor protein (APP) by β-secretase leads to increased amyloid β-peptide (Aβ) levels and an earlier onset of AD in a large Swedish family (Hardy and Selkoe 2002). Several mechanisms controlling the RIP process have been identified, but many more need to be uncovered and may hold the potential for therapeutic intervention in diseases. The regulation of the RIP process mostly occurs at the shedding level and was recently reviewed (Hayashida et al. 2010).

This review focuses on the proteolytic mechanisms associated with the RIP processing of APP and the generation of the Aβ, with the goal to elucidate cellular mechanisms and common principles of RIP.


APP was one of the first proteins known to undergo RIP. Although the biological function of APP is not yet clear, it is the precursor of Aβ, which is the culprit in AD and forms neurotoxic aggregates in the brain (reviewed in Hardy and Selkoe 2002; Haass and Selkoe 2007). APP is constitutively processed by two RIP pathways, one leading to Αβ generation, and the other one preventing it (Fig. 2). Three protease activities are involved, which were initially referred to as α-secretase, β-secretase and γ-secretase at a time when their identity was not yet resolved (reviewed in Haass 2004). These names are still in use, not only for the APP substrate, but also for other proteins being cleaved by the same or similar proteases. In the Αβ generating pathway, APP is shed by the β-secretase, which is the aspartyl protease BACE1. This cleavage occurs at the N-terminus of the Αβ domain and is followed by γ-secretase cleavage within its TMD, leading to Αβ release. The cleavage of γ-secretase is heterogeneous and gives rise to Aβ species of 37–43 amino acids in length. The major product generated is Aβ40. The minor Aβ42 species, although only generated at ∼ 10% of total Aβ, is believed to be causative for AD. Mutations in APP at or close to the cleavage sites of the three secretases are associated with rare familial forms of AD and lead to either increased levels of total Aβ (mutations at the β-secretase site), mutant Aβ variants with enhanced aggregation properties (mutations at the α-secretase site) or in an increase of the Aβ42/40 ratio (mutations at the γ-secretase site).

Figure 2.

 RIP processing of APP by α-, β- and γ-secretase. The type I membrane protein APP is proteolytically cleaved in two competing RIP pathways. Sequential APP cleavage by β-secretase and γ-secretase is referred to as the amyloidogenic pathway and generates Aβ. β-Secretase cleavage occurs within the ectodomain of APP close to the transmembrane domain. This results in the soluble APP ectodomain (APPsβ) and the formation of the membrane-bound C-terminal fragment C99 (C-terminal 99 amino acids of APP). Subsequently, γ-secretase cleaves C99, leading to Aβ secretion and the formation of the APP intracellular domain (AICD). In the alternative, non-amyloidogenic pathway, APP is first cleaved by the metalloprotease α-secretase. This cleavage yields the soluble APP ectodomain APPsα and a C-terminal fragment (C83), which is further processed by γ-secretase, resulting in the secreted p3 peptide and AICD.

A number of the current approaches to treat AD are based on the central role of Aβ in the disease process. Therapeutic targeting of β- and γ-secretase is an obvious and major approach to treat AD and potent inhibitors and/or modulators of these secretases, in particular of γ-secretase, have been developed and are in clinical trials. As an in depth account of this topic will be beyond the scope of this review, we refer here the reader to other excellent and extensive recent reviews (Tomita 2009; Citron 2010).

The corresponding counterpart of Aβ, the APP intracellular domain (AICD) is released by γ-secretase into the cytosol, where it is rapidly degraded. A potential signaling function of AICD is controversially discussed. One study suggested that AICD acts as a transcriptional activator and translocates to the nucleus after γ-secretase cleavage (Cao and Sudhof 2001). However, another study found that only the membrane-tethered AICD, but not the free AICD acts as an indirect transcriptional activator by activating the cytosolic protein FE65, which in turn stimulates transcription of target genes (Cao and Sudhof 2004). Yet another study reported that AICD did not act as a transcriptional activator and that FE65 transactivates different promoters, even in the absence of AICD (Hebert et al. 2006).

In the alternative RIP pathway, APP shedding occurs by α-secretase in the middle of the Αβ domain, thus precluding Αβ generation. Subsequent γ-secretase cleavage releases a truncated Αβ peptide (p3) into the supernatant, which apparently does not contribute to AD pathogenesis (Dulin et al. 2008). The identity of the constitutively cleaving α-secretase was only recently finally established as ADAM10 (Jorissen et al. 2010; Kuhn et al. 2010), which is in agreement with previous studies showing that ADAM10 can act as an α-secretase (Lammich et al. 1999, Postina et al. 2004). α-Secretase cleavage of APP may be stimulated above its constitutive level and is then referred to as regulated α-secretase cleavage. Depending on the stimulus investigated, this increase can not only be mediated by ADAM10 (Kojro et al. 2006), but also by ADAM17 (Buxbaum et al. 1998; Kuhn et al. 2010) and potentially other metalloproteases (Delarasse et al. 2011). An increase in α-secretase cleavage is typically accompanied by a reduction of Aβ generation, indicating that α- and β-secretase compete for APP as a substrate, at least when APP α-secretase cleavage is increased above its constitutive level (Skovronsky et al. 2000). A more detailed description of α-secretase in APP cleavage and Aβ generation is found in a recent review (Lichtenthaler 2011). The APP ectodomain released upon α-secretase cleavage was found to rescue the mild phenotypes of APP-deficient mice, suggesting that the ectodomain of APP is sufficient to mediate the function of APP, at least for the phenotypes investigated (Ring et al. 2007). In fact, the APP ectodomain released through α-secretase has neurotrophic and neuroprotective properties (Furukawa et al. 1996; Meziane et al. 1998), but the underlying molecular mechanisms and a potential receptor for APP remain to be identified. Conversely, the APP ectodomain released through β-secretase cleavage appears to have a proapoptotic function, at least during early development, by binding and activating the death receptor 6 on neurons (Nikolaev et al. 2009).

In the following section, we will describe the function and mechanism of the APP proteases and their homologs, specifically the ADAM proteases ADAM10 and ADAM17, the BACE proteases, and the GxGD-type protease γ-secretase.

The RIP proteases: sheddases

ADAM proteases

In the RIP process, shedding occurs at a short, but relatively fixed distance from the membrane. Thus, it is not surprising that the ADAM and BACE proteases are themselves transmembrane proteins having their active site in their luminal domains. Over 30 different ADAMs have been identified in metazoans (reviewed in Edwards et al. 2008; Reiss and Saftig 2009). Many of them are proteolytically active and contain the typical amino acid consensus motif (HExGHxxGxxHD) required for zinc binding in the active site. Other ADAM members lack this motif and are assumed to be involved predominantly in cell adhesion. The human genome contains 21 ADAM family members, of which 12 (ADAM8, 9, 10, 12, 15, 17, 19, 20, 21, 28, 30, 33) are proteolytically active. The other ones lack critical amino acids in their active site, which would be required for proteolytic activity. It remains to be tested whether the inactive ADAMs additionally are able to inhibit the active ADAMs and thereby act as inhibitory or inactive ADAMs, similar to what has been suggested for the inactive rhomboid intramembrane proteases (Lemberg and Freeman 2007).

ADAM proteases are type I membrane proteins of about 750–900 amino acids in length. They consist of an N-terminal signal peptide, followed by a prodomain, the zinc-binding metalloprotease domain, a cysteine-rich domain, an epidermal growth factor (EGF) domain, the transmembrane domain and a cytoplasmic domain, which in many cases is proline-rich. An EGF domain is not found in ADAM10. A more detailed description of the ADAM protease domains, their structure and specific functions can be found in recent reviews (Edwards et al. 2008; Reiss and Saftig 2009). During transport through the secretory pathway the ADAMs are complex N-glycosylated. The prodomain is removed by proprotein convertases, such as furin, resulting in the active protease, which mediates proteolysis in the late compartments of the secretory pathway and at the plasma membrane. Importantly, ADAM protease-mediated shedding can occur for substrates on the same cell surface, but can also happen in trans, meaning that an ADAM protease on one cell surface can cleave a substrate residing on the neighboring cell surface. This has been shown for ADAM10 in ephrin-Eph receptor signaling (Janes et al. 2005) and may be a more generally occurring mechanism controlling shedding and the RIP process. Interestingly, ADAM10 itself is also subject to RIP. It is shed by ADAM9 and 15, followed by γ -secretase intramembrane proteolysis and translocation of the ADAM10 ICD into the nucleus, where it is found in nuclear speckles, thought to be involved in gene regulation (Parkin and Harris 2009; Tousseyn et al. 2009).

Among the known RIP substrates, the sheddase is often ADAM10, but in several cases also ADAM17. However, a generalization of this statement is difficult, given that for many RIP substrates the sheddase is not yet known and that many more RIP substrates are likely to be identified. Thus, it is well possible that other ADAMs or metalloproteases turn out to be additional major contributors to the RIP process.

The major substrates for ADAM10 and 17 during development have been identified in the corresponding protease-deficient mice. ADAM17-deficient mice show perinatal lethality and phenotypic changes, such as open eyelids, skin abnormalities and epithelial maturation defects in multiple organs (Peschon et al. 1998). These changes resemble the phenotype of mice lacking the EGF receptor or some of its ligands, such as transforming growth factor α (TGFα), which demonstrated that EGF receptor ligands are major ADAM17 substrates during development. ADAM10-deficient mice show a loss-of-Notch signaling phenotype and die during embryonic development at day 9.5 (Hartmann et al. 2002). Notch is a cell surface receptor that requires RIP for its signal transduction to control cellular differentiation. Ligand-binding induces Notch shedding, followed by γ-secretase mediated intramembrane proteolysis. This releases the Notch intracellular domain into the nucleus, where it acts as a transcriptional activator of Notch downstream target genes, such as hairy and enhancer of split (for review see Fortini and Bilder 2009; Kopan and Ilagan 2009). Recent work established ADAM10 as the relevant protease for the physiological ligand-induced Notch1 shedding and signaling, in agreement with the lack of Notch signaling in ADAM10 knock-out mice. However, under ligand-independent conditions, as it occurs for certain leukemia-linked Notch mutations, Notch cleavage is mediated by ADAM17 (Bozkulak and Weinmaster 2009; van Tetering et al. 2009). Because of the early lethality of ADAM10 and 17-deficient mice, conditional knock-out mice are currently being used to decipher further substrates as well as physiological and pathophysiological functions of both proteases (Horiuchi et al. 2007; Jorissen et al. 2010).

ADAM10 and 17 have a similar, but not identical, broad substrate specificity and can cleave peptides in vitro at the same peptide bonds, often after a positively charged amino acid at the P1 position (Caescu et al. 2009). This is in contrast to the more pronounced specificity in vivo. For example, in cells and in animals ADAM10, but not ADAM17 is the major sheddase for APP and Notch, as described above. This remarkable specificity in cells suggests the existence of as yet unknown factors, which control the protease specificity in the cellular environment. Generally, ADAM proteases cleave at a short distance from the membrane. For APP the principal determinant of the ADAM10-mediated α-secretase cleavage appears to be an α-helical conformation and the distance of 12 amino acids from the membrane (Sisodia 1992). Interestingly, for APP the initial α-secretase cleavage seems to be followed by a carboxypeptidase cleavage removing the C-terminal, positively charged lysine (Esch et al. 1990; Kuhn et al. 2010). While the functional relevance of this C-terminal truncation is not yet clear, it is interesting to note that the maturation of several prohormones also requires the removal of a C-terminal Lys or Arg by a carboxypeptidase (Steiner 1998).

Tissue inhibitors of metalloproteases (TIMPs) are soluble, secreted proteins, which physiologically inhibit ADAM proteases and matrix metalloproteases (MMPs) by binding to the metalloprotease domains. They show greater selectivity to ADAMs than to matrix metalloproteases (MMPs). TIMP1 inhibits ADAM10, but not ADAM17, whereas TIMP3 efficiently blocks ADAM17 and with a lower efficiency also ADAM10 (Amour et al. 1998, 2000). Small molecule inhibitors with selectivity to either ADAM10 or 17 have been developed (Ludwig et al. 2005). Therapeutically, such molecules are interesting for different diseases, such as inhibition of tumor necrosis factor α (TNFα) shedding in rheumatoid arthritis or for blocking RIP of Notch in leukemias. Clinical trials with ADAM10 and 17 inhibitors are ongoing, but several previous trials were halted because of mechanism-based liver toxicity.

The activity of ADAM proteases can be increased above its constitutive level by diverse stimuli (reviewed in Hayashida et al. 2010). This includes growth factors, cytokines, protein kinase C (PKC) activators, G-protein coupled receptors, calcium ionophores and other natural or experimental stimuli. In many cases, the exact underlying mechanisms are not yet fully elucidated. Interestingly, a recent study showed that phorbol-12-myristate-13-acetate (PMA) leads to the isomerization of disulfide bonds in ADAM17, thereby inducing a structural change and enhanced catalytic activity (Willems et al. 2010). The concept of a structural change induced by activation is also supported by a recent study demonstrating that a tight-binding hydroxamate inhibitor only binds and blocks ADAM17 in stimulated, but not in quiescent cells (Le Gall et al. 2010). Another recent study demonstrated that activated p38 MAPK phosphorylates the cytoplasmic domain of ADAM17 and that this is required for ADAM17-mediated ectodomain shedding of transforming growth factor α (TGFα) family ligands (Xu and Derynck 2010).

Apart from signaling pathways intracellular protein trafficking is increasingly appreciated as a mechanism to control the RIP process. The involvement of protein trafficking comes from the fact that the RIP proteases and the RIP substrates are membrane proteins. In contrast to soluble proteases and their substrates, the membrane proteins cannot freely diffuse, but must be present in close proximity within the same cellular compartment for proteolysis to occur. As a consequence, the amount of shedding and intramembrane proteolysis depends on the intracellular transport of substrate and protease. Details have been reviewed recently for the RIP process in general (Sannerud and Annaert 2009) or at the example of the α-secretase cleavage of APP (Lichtenthaler 2011).

BACE proteases

Not only metalloproteases, but also the aspartyl proteases BACE1 and BACE2 mediate ectodomain shedding. BACE1 was identified as the sole APP β-secretase, which catalyzes the first step of Αβ generation (Hussain et al. 1999; Sinha et al. 1999; Vassar et al. 1999; Yan et al. 1999; Lin et al. 2000). Because BACE1 is a major drug target for AD, this protease has been intensively studied since its discovery in 1999. Little is known about BACE2, which is a close homolog of BACE1 (Saunders et al. 1999; Yan et al. 1999; Solans et al. 2000).

BACE1 is a glycosylated type I membrane protein with 501 amino acids. The N-terminal signal peptide is followed by a prodomain, a catalytic domain comprising the two catalytic aspartic acid residues [amino acids 93–96 (DTGS in the single letter amino acid code) and amino acids 289–292 (DSGT in the single letter amino acid code)] characteristic for aspartyl proteases, a transmembrane domain and a short cytoplasmic tail. The crystal structure of the BACE1 ectodomain shows a conserved general folding of aspartyl proteases (Hong et al. 2000). Upon trafficking through the secretory pathway, the propeptide is cleaved by furin, leading to the active (mature) BACE1 protease (Bennett et al. 2000; Capell et al. 2000; Huse et al. 2000; Creemers et al. 2001). BACE1 forms dimers, which show a higher activity than the monomers (Schmechel et al. 2004; Westmeyer et al. 2004). BACE1 is ubiquitously expressed, with higher expression levels observed in brain and in neurons (Vassar et al. 1999), which explains why neurons are particularly vulnerable in AD because of the increased BACE1-mediated Αβ generation. BACE1 expression in mice is very high in the nervous system right after birth and then decreases to very low levels (Willem et al. 2006). BACE1 localizes to the Golgi, the trans-Golgi network and to the endosomes (Vassar et al. 1999; Capell et al. 2000; Huse et al. 2000). In vivo BACE1 seems to be specifically active within acidic cellular compartments such as the late Golgi and specifically endosomes and lysosomes. BACE1 activity has also been observed in other compartments, however, this seems to occur only upon over-expression of the protease (Huse et al. 2002).

BACE1 is a major drug target for AD, as its inhibition lowers Αβ generation. Potent BACE1 inhibitors have been developed, but mostly do not reach sufficiently high concentrations in the brain. Nevertheless, a first inhibitor is currently being tested in a phase II clinical trial for AD (Vassar et al. 2009). Furthermore, new generation BACE1 inhibitors with significantly improved pharmacokinetics have been developed recently, and may soon be tested in patients. Other preclinical strategies to inhibit BACE1 activity consist of targeting BACE1 with antibodies and with modified drugs, which are specifically delivered to endosomes, where BACE1 cleaves APP (Chang et al. 2007; Rajendran et al. 2008; Mitterreiter et al. 2010). Some of these drugs indirectly inhibit BACE1, by locally raising the membrane-proximal pH in the endosome to levels above the pH-optimum of 4.5 for BACE1.

In addition to APP, several other BACE1 substrates have been identified over the past few years, namely neuregulin-1, the P-selectin glycoprotein ligand-1, the amyloid precursor like proteins 1 and 2, the sialyltransferase ST6Gal I, β-subunits of voltage-gated sodium channels, the interleukin-1 receptor 2, and the low density lipoprotein receptor-related protein (Kitazume et al. 2002; Lichtenthaler et al. 2003; Li and Sudhof 2004; von Arnim et al. 2005; Wong et al. 2005; Willem et al. 2006; Kuhn et al. 2007). This demonstrates, that BACE1 has a broader role in the RIP process. However, these substrates should be discussed with some care, since only in a few cases, BACE1 deficiency could be directly related to functional consequences associated with the respective substrate. Moreover, a number of potential substrates were identified only upon their over-expression or over-expression of BACE1. So far clear phenotypic consequences of a BACE1 knockout could only be related to cleavage of neuregulin-1 type III (NRG1). BACE1 knockout mice show a prominent hypomyelination in the peripheral nervous system during postnatal development, which results from reduced processing of NRG1 (Hu et al. 2006; Willem et al. 2006). Interestingly, myelination of the peripheral nervous system occurs postnatally, which coincides with the significant up-regulation of BACE1 expression. NRG1 is believed to be a hairpin protein with two transmembrane domains located in the plasma membrane of axons. Signaling to its receptors (ErbB2/ErbB3) on Schwann cells is known to require endoproteolysis of the luminal/extracellular loop. This cleavage may specifically be required to allow access to the EGF domain. Indeed, unprocessed NRG1 accumulates in BACE1 knockout animals, demonstrating that NRG1 is currently the only in vivo verified BACE1 substrate with a physiological correlate. The remaining myelination observed in the PNS of homozygous BACE1 knockout mice is because of another sheddase probably cleaving in the vicinity of BACE1. The nature of this sheddase is currently unknown, however, it is likely that one of the ADAM proteases may be involved. NRG1 has also been linked to schizophrenia, and, indeed BACE1-deficient mice show schizophrenia-like behavioral changes (Savonenko et al. 2008). If the latter is related to the developmental phenotype remains however speculative. Additional functions of BACE1 in the CNS, such as epileptic seizures, are less well understood (Hitt et al. 2010; Hu et al. 2010).

Although several of the above listed BACE1 substrates are also cleaved by metalloproteases, such as ADAM proteases, and apparently such a cleavage may allow the remaining myelination observed in BACE1 knockout mice, both cleavages may also have fundamentally different functional consequences. In case of APP, the cleavage sites of BACE1 and ADAM10 are 16 amino acids apart. Recent evidence suggests that BACE1 cleavage generates a secreted APP ectodomain, which may act proapoptotically (Nikolaev et al. 2009), whereas the secreted ectodomain generated by ADAM10 appears to have a neurotrophic and neuroprotective function (Furukawa et al. 1996; Meziane et al. 1998). Additionally, BACE1 leads to Αβ generation, whereas ADAM10 cleavage occurs within the Αβ domain and thus prevents Αβ generation. In contrast to the ADAM proteases, which seem to be less dependent on specific amino acid motifs around the cleavage site, BACE1 has a more pronounced substrate specificity and prefers a leucine at the P1 position (Citron et al. 1995; Gruninger-Leitch et al. 2002). For APP the BACE1 cleavage site is further away from the membrane than the ADAM10 cleavage site, but this is not the case for all membrane proteins undergoing shedding by an ADAM protease and by BACE1. For example, for the cell adhesion protein P-selectin glycoprotein ligand-1 the cleavage site by BACE1 is closer to the membrane than the cleavage site by an α-secretase like protease (Lichtenthaler et al. 2003).

Expression of BACE1 is controlled by multiple mechanisms. Several transcription factors regulate BACE1 expression, such as nuclear factor kappa B (NFκB), peroxisome proliferator-activated receptor-γ (PPARγ) and Yin Yang-1 (YY1) (reviewed in Rossner et al. 2006). At the translational level three different mechanisms were identified to control BACE1 expression and affect the RIP-mediated Αβ generation. The 5′ untranslated region of the BACE1 mRNA strongly suppresses BACE1 translation (De Pietri Tonelli et al. 2004; Lammich et al. 2004; Rogers et al. 2004). Energy deprivation, which occurs in AD brains, is able to relieve the translational suppression, resulting in increased BACE1 levels and enhanced Αβ generation in mice (O’Connor et al. 2008). A second mechanism is a naturally occurring antisense RNA, which binds to the open reading frame of the BACE1 mRNA and increases BACE1 translation (Faghihi et al. 2008). Finally, microRNAs inhibit BACE1 mRNA translation (Hebert et al. 2008; Wang et al. 2008). All three mechanisms appear to be altered in AD, providing an explanation for the 2- to 5-fold increase in BACE1 protein levels observed in AD brains.

The RIP proteases: intramembrane proteolysis by γ-secretase

Subunit composition and assembly of γ-secretase

Following ectodomain shedding, the membrane-retained stubs become substrates for the subsequent intramembrane cleavages, which are catalyzed by γ-secretase for type I membrane protein stubs or by signal peptide peptidase-like (SPPL) proteases for those with type II membrane orientation. In the following, we will focus on the biology of γ-secretase (Fig. 3). For an in depth comparative discussion of our current knowledge of γ-secretase and other GxGD-type I-CLiPs see recent reviews on this topic (Fluhrer et al. 2009; Golde et al. 2009). γ-Secretase is a protein complex (Seeger et al. 1997; Capell et al. 1998; Thinakaran et al. 1998; Yu et al. 1998; Li et al. 2000a) composed of its catalytic subunit presenilin (PS) (Steiner et al. 1999b; Wolfe et al. 1999; Esler et al. 2000; Kimberly et al. 2000; Li et al. 2000b; Seiffert et al. 2000) and three other proteins nicastrin (NCT), anterior pharynx-defective-1 (APH-1) and presenilin enhancer-2 (PEN-2) (Yu et al. 2000; Lee et al. 2002; Steiner et al. 2002; Edbauer et al. 2003; Kimberly et al. 2003; Takasugi et al. 2003). Reconstitution experiments in yeast have shown that these four proteins are necessary and sufficient to reconstitute γ-secretase activity (Edbauer et al. 2003). Interestingly, recent cell-free reconstitution experiments showed that PS recombinantly expressed in E. coli can apparently adopt a conformation with low intrinsic γ-secretase activity when reconstituted into liposomes further proving that PS is the catalytic subunit of the γ-secretase complex (Ahn et al. 2010). However, it should be noted that PS is proteolytically not active in any eukaryotic cell type and animal system investigated unless APH-1, NCT, and PEN-2 are coexpressed.

Figure 3.

 Schematic representation of γ-secretase. γ-Secretase is a membrane protein complex comprised of the catalytic subunit presenilin (PS) and three other membrane proteins nicastrin (NCT), APH-1 and PEN-2. PS is cleaved by autoproteolysis in the large cytoplasmic loop into N- and C-terminal fragments (NTF, CTF). Transmembrane domains 6 and 7 of PS, which contain the active sites aspartate residues (D) are highlighted in green. Because of the presence of two PS homologs (PS1 and PS2) and two APH-1 homologs {APH-1a [existing as short (APH-1aS) and long (APH-1aL) splice variant] and APH-1b}, which do not co-exist in the same γ-secretase complex, in man up to six γ-secretase complexes exist. In the lower panel, the subunit arrangement of the γ-secretase complex is depicted as a top-view.

The γ-secretase subunits are all integral membrane proteins. NCT, the largest subunit of the complex, is a single pass membrane glycoprotein with type I membrane orientation (Yu et al. 2000). PS spans the membrane nine (Henricson et al. 2005; Laudon et al. 2005; Oh and Turner 2005a,b; Spasic et al. 2006) and APH-1 seven times (Fortna et al. 2004), respectively, while PEN-2, the smallest subunit, is a double-pass membrane protein (Crystal et al. 2003). Mammalian cells contain two homologues of PS (PS1 and PS2) as well as of APH-1 (APH-1a and APH-1b) (Francis et al. 2002). APH-1a exists as a short and a long splice variant, APH-1aS and APH-1aL, respectively (Lee et al. 2002). A close homologue of APH-1b exists in rodents (APH-1c) (Hebert et al. 2004). Co-immunoprecipitation analysis has demonstrated that PS1 and PS2 do not coexist in the same γ-secretase complex (Yu et al. 1998; Steiner et al. 2002). Likewise, the APH-1a splice variants and APH-1b (and APH-1c in rodents) are not found in the same γ-secretase complex (Hebert et al. 2004; Shirotani et al. 2004a). Thus, in man, dependent on the relative expression of the individual subunits principally up to six different γ-secretase complexes can exist. Whether these differ in their proteolytic properties, specifically in their ability to produce Aβ42 is controversial (Shirotani et al. 2007; Serneels et al. 2009). The native molecular weight of γ-secretase, ranging from 250 kDa to 2 MDa depending on the method used for analysis, is still under debate (Li et al. 2000a; Edbauer et al. 2002; Kimberly et al. 2003; Osenkowski et al. 2009). The stoichiometry of the γ-secretase subunits has recently been determined to be 1 : 1 : 1 : 1 (Sato et al. 2007) in agreement with the sum of the molecular weight of its four subunits. However, also evidence for dimeric PS has been obtained (Schroeter et al. 2003), which would be compatible with an overall 2 : 2 : 2 : 2 stoichiometry of the subunits and consistent with a molecular weight of ∼ 500 kDa, which is observed by most studies by native gel electrophoresis (Edbauer et al. 2002; Farmery et al. 2003; Li et al. 2003; Nyabi et al. 2003). Finally, several γ-secretase interactors have been described which may transiently interact with the enzyme (Wakabayashi et al. 2009; Winkler et al. 2009) to regulate its activity such as, for example transmembrane protein 21 (TMP21) (Chen et al. 2006) or the recently identified γ-secretase activating protein (He et al. 2010). None of these proteins are integral subunits of the γ-secretase complex, however.

γ-Secretase complex assembly and maturation occurs in a step-wise manner (for detailed reviews see Dries and Yu 2008; Spasic and Annaert 2008). The prevailing model suggests that first a dimeric subcomplex of NCT and APH-1 is formed (Kimberly et al. 2003; Morais et al. 2003; Shirotani et al. 2004b), which is capable of stabilizing PS that assembles as the third subunit (Takasugi et al. 2003). PEN-2 enters the ternary NCT-APH-1-PS complex as last component and triggers the endoproteolysis of PS into an N-terminal and C-terminal fragment (NTF, CTF) (Takasugi et al. 2003), which represents the principal cellular form of PS (Thinakaran et al. 1996). This immature complex, which is formed in the endoplasmic reticulum (ER) (Kim et al. 2004; Capell et al. 2005), is then further transported through the secretory pathway where complex glycosylation of NCT occurs in the Golgi complex. Unassembled PS1 and PEN-2 are retained within the ER via ER-retention signals (Kaether et al. 2002, 2007; Fassler et al. 2010), which are masked during assembly, finally allowing the complex to be targeted through the secretory pathway. The precise role and interactions of the assembly factor retention in endoplasmic reticulum 1 (RER1) in this process is controversial (Kaether et al. 2007; Spasic et al. 2007). Additional post-translational modifications occur for APH-1 and NCT, which become palmitoylated (Cheng et al. 2009). Upon complete maturation, the mature complex reaches its functional sites in the endosomal/lysosomal compartments of the secretory pathway and the plasma membrane.

Stepwise cleavage of GxGD protease substrates

The catalytic subunit of the γ-secretase complex, PS, is an unusual aspartyl protease, as compared, for example, to BACE. The two aspartate residues required for catalysis of the substrate peptide bond lie within the opposing TMDs 6 and 7 of the protein. The C-terminal aspartate residue is part of a non-classical GxGD signature sequence of the catalytic site (Steiner et al. 2000). This motif is now used to define the novel class of GxGD aspartylproteases (Haass and Steiner 2002), with PS, as the founding member, the related SPP and SPPLs (Weihofen et al. 2002), and the bacterial type four prepilin peptidases (LaPointe and Taylor 2000). Unlike the related SPPLs proteases, only PS is cleaved in its large cytoplasmic loop domain into its characteristic NTF and CTF. For PSs, this cleavage is considered to be associated with a biological activation of γ-secretase although exceptions exist.

The processing of γ-secretase substrates has been worked out for APP in substantial detail. The current model suggests a stepwise cleavage of the APP CTF until the APP TMD is sufficiently shortened to clear the membrane from the APP membrane stub by the release of the Αβ peptides (Fig. 4, upper panel). γ-Secretase cleaves first at the ε-site, very close to the cytoplasmic border of the membrane (Gu et al. 2001; Sastre et al. 2001; Yu et al. 2001; Weidemann et al. 2002). Cleavage at this site releases the AICD to the cytosol. This initial cut is then followed by a cleavage of the thus generated membrane-retained Αβ49 at the ζ-site to generate Αβ46 (Zhao et al. 2004), before γ-secretase cleaves this Αβ intermediate at the multiple γ-sites (Qi-Takahara et al. 2005; Zhao et al. 2005). The cleavages are all heterogeneous such that APP substrate cleavage of γ-secretase gives rise to two distinct product lines (Qi-Takahara et al. 2005), for which even the characteristic tri- and tetrapeptides have recently been identified (Takami et al. 2009). The major product line is Αβ49-Αβ46-Αβ43-Αβ40-Αβ37 from which Αβ40 is the principal end product (Qi-Takahara et al. 2005). The minor product line is Αβ48-Αβ45-Αβ42-Αβ38 giving rise to the pathogenic Αβ42 and the Αβ38 in comparable amounts. Recently, dimerisation of the APP TMD has been suggested to modulate the generation of Aβ42 (Munter et al. 2007, 2010). This interesting model suggests that APP TMD dimerisation may impose steric hinderance to γ-secretase such that the enzyme cannot proceed cleavage until the γ-38 site. As a consequence, γ-secretase pauses cleavage at the γ-42 site and thereby generates Aβ42. In contrast, loosening APP TMD dimerisation allows γ-secretase to cleave down to the γ-38 site to release Aβ38.

Figure 4.

 Stepwise substrate cleavage of GxGD proteases. Upper panel, the individual γ-secretase cleavage sites in the APP transmembrane domain (TMD) are depicted by thick (major cleavage sites) and thin (minor cleavage sites) vertical arrows. Horizontal arrows indicate the direction of the cleavages. The major product line (ε49 –γ37), which generates Aβ40 as principal Aβ species, is depicted with a thick horizontal arrow. The minor product line (ε48 –γ38), which gives rise to the pathogenic Aβ42, is depicted with a thin horizontal arrow. Glycine residues of a GxxxG dimerisation motif in the APP TMD, which has been shown to regulate the usage of the two product lines by the strength of APP TMD dimerisation, are highlighted as white letters. Middle and lower panels, similar ε-, ζ-, and γ-like cleavages have been shown to occur for PS autoproteolysis (middle panel, aa residues of the cleavage sites in PS1 are denoted by numbers) and for TNFα processed by the PS-related GxGD protease SPPL2b (lower panel). Note that these cleavages occur in opposite direction compared to APP. In all panels, the predicted TMDs or in case of PS1 the hydrophobic cleavage site domain are underlaid in gray. For details see text.

A similar stepwise mode of processing as found for APP has also been shown to occur for Notch1 (Okochi et al. 2002), CD44 (Lammich et al. 2002) and APLP1 (Yanagida et al. 2009). Does that mean that stepwise endoproteolysis is a general mechanism exerted by GxGD proteases with the goal to reduce the hydrophobicity of their substrates, the TMDs? This may very well be necessary since a single cut in the middle of the TMD (at the γ-site) as originally assumed, would probably not allow sufficient and controlled release of an ICD required for nuclear signaling as such rather long ICDs may stick in the membrane because of their higher hydrophobicity. Furthermore, even if there are no functional requirements of the resulting ICDs, degradation of membrane protein stubs may not be efficient enough. Clearly multiple intramembrane cuts dramatically help to reduce membrane retention. The evolutionary importance of this mechanism becomes apparent when one takes a look at endoproteolysis of PS itself. Soon after the identification of the catalytically important aspartate residues in TMDs 6 and 7, it was suggested that PS endoproteolysis occurs via autoproteolysis, since substitution of only one of the aspartates with any other amino acid investigated was sufficient to block γ-secretase activity and endoproteolysis (Wolfe et al. 1999). Moreover, in vivo reconstitution of γ-secretase in yeast demonstrated that the assembly of a biologically active γ-secretase complex is associated with PS endoproteolysis (Edbauer et al. 2003). If PS cleaves itself it may do so by utilizing similar mechanisms of endoproteolysis, since after all PS itself would be substrate of γ-secretase. Indeed the exon 9 encoded domain, which contains the PS cleavage site is hydrophobic and may thus interact or even dive into the membrane. Interestingly, shortly after the first description of PS and its endoproteolysis, the PS cleavage site was identified (Podlisny et al. 1997). Surprisingly not only one cleavage site, but two sites were found in PS1, a major site after amino acid (aa) 298 and minor cleavage sites after aa 292/293 (Podlisny et al. 1997). Moreover, substitution of the methionine at aa 292 by a negatively charged aa fully abolished endoproteolysis, although this mutation was located relatively far away from the major cleavage site (Steiner et al. 1999a). This has been interpreted by a stepwise cleavage, whereby the cleavage is initiated at aa 292 and terminates after the cut at aa 298 thus leading to a major cleavage product (the PS1 CTF) with a N-terminus at aa 299. Very similar findings were made for PS2 (Shirotani et al. 1997, 2000; Jacobsen et al. 1999). A more detailed recent analysis finally validated this model (Fukumori et al. 2010). Surprisingly, a three aa spaced cleavage was found, which may be consistent with stepwise endoproteolysis similar to that described for γ-secretase substrates (Fig. 4, middle panel). Indeed, a first cleavage at aa 292/3 seems to initiate endoproteolysis. This cleavage is then followed by a second cut at aa 295/6 and culminates with the final cut at aa 299 (Fukumori et al. 2010). This cleavage pattern is remarkably similar to that described for the intramembrane cleavage of APP and other γ-secretase substrates. Thus, ε-, ζ-, and γ-like cleavages seem to be involved in an autoproteolytic cleavage mechanism of PS. Autoproteolysis itself is further supported by the finding that at least some FAD-associated PS mutations affect the precision of endoproteolysis (Fukumori et al. 2010). However, there is one major difference between the stepwise cleavage of γ-secretase substrates and PS itself. Whereas γ-secretase substrates are principally in type I orientation and cleaved such that the cleavage initiates at the C-terminus and proceeds to the N-terminus, for PS itself the opposite is the case, here the cleavage may proceed towards the C-terminus. Nevertheless, these findings suggest that the exon 9 encoded cleavage site may indeed plug the catalytically active center of PS and thus keep PS in its inactive ‘proform’ preventing external substrates from reaching the catalytic pore as initially suggested by inhibitor studies using exon 9-based peptides (Knappenberger et al. 2004). Upon autoproteolysis the plug is removed to subsequently allow substrate access via a lateral entry. Again, stepwise endoproteolysis may greatly facilitate sufficient cleavage and removal of a hydrophobic domain. This model may be supported by the finding that other GxGD proteases such as SPPL2b may also cleave their membrane embedded substrates by a stepwise cleavage mechanism (Fluhrer et al. 2006), very similar to that described for APP and a few other γ-secretase substrates plus PS itself (Fig. 4, lower panel). If that holds true for all SPPL proteases and their substrates remains however to be proven.

PS mutations exhibit both loss and gain of function

The ratio of the pathogenic Aβ42 species to total Aβ is of central importance for AD. Increased ratios are observed in the rare cases of familial AD and lead to an early onset of the disease (Scheuner et al. 1996). Even very subtle changes appear to be sufficient to cause the disease (Scheuner et al. 1996) by eliciting increased synaptic and cellular neurotoxicity (Kuperstein et al. 2010). Mutations in PS are responsible for the majority of these cases. More than 170 mutations have been identified in PS1, and a few were identified in PS2. These FAD mutations are located all over the PS molecule and, like the FAD mutations in the γ-secretase cleavage site domain of APP, alter the γ-secretase cleavage specificity and increase the generation of the pathogenic Aβ42 (and/or Aβ43 species). However, specifically the most aggressive mutations with a very early age of onset often result also in a partial loss of function (for an overview see Shen and Kelleher 2007). Kinetic analyses revealed that in that case the Αβ40-generating product line seems to be reduced, while the Αβ42-generating product line appeared to be unaffected (Fluhrer et al. 2008; Shimojo et al. 2008). This not only leads to a pathological shift of the Aβ42/40 ratio but may also lead to a significant loss of ICD generation. In fact, it is known since a long time already that Notch signaling appears to be severely affected by some of these mutations (Levitan et al. 1996; Baumeister et al. 1997; Song et al. 1999). If that contributes to the disease phenotype, is however unknown. Thus, many PS FAD mutants exhibit both loss and gain of function properties, which may be because of a decreased activity of stepwise APP substrate cleavage. However, it should be noted that not all mutations can be explained by this mechanism (Moehlmann et al. 2002; Heilig et al. 2010).

The molecular structure of γ-secretase and its active site

Only little information is available regarding the structure of γ-secretase. This is likely because of its nature of being a complex of four subunits, which is difficult to isolate in the high amounts needed for crystal structure determination. Most electron-microscopic studies have shown γ-secretase particles of spherical shape. However, the electron microscopy studies have so far provided only little insight, as the resolution with 12–15 Å of these studies was too low to locate the individual subunits and to safely identify molecular details (Lazarov et al. 2006; Osenkowski et al. 2009). Thus, only vague speculations on the locations of the catalytic site or the substrate binding sites were possible. Cross-linking studies have shown that the overall architecture of the individual γ-secretase complexes is very similar to each other, with close contacts of the PS NTF and CTF in the catalytic subunit (Steiner et al. 2008). In addition, the PS NTF is in close proximity to PEN-2, and the PS CTF with APH-1, which itself is in close neighborhood with NCT (Steiner et al. 2008). These data are in good agreement with data obtained in the studies addressing the subunit interactions of the γ-secretase complex by different means (Fraering et al. 2004; Kim and Sisodia 2005; Watanabe et al. 2005). Regarding structural information on γ-secretase, even more details have been obtained by cysteine-accessibility studies. These studies showed that TMD6 and 7 of PS face each other and form a hydrophilic pore or cavity (Sato et al. 2006; Tolia et al. 2006). A number of residues could be identified, which are part of this cavity including residues of the GxGD motif (Sato et al. 2006; Tolia et al. 2006). In addition, residues of the conserved PAL (P: proline, A: alanine, L: leucine) motif, which is very close to TMD6, were found to be water-accessible suggesting these are part of the cavity as well (Sato et al. 2008a; Tolia et al. 2008). Other TMDs of PS also contribute residues, which are water accessible and could thus be part of the same active site-containing cavity (Sato et al. 2008a; Tolia et al. 2008; Takagi et al. 2010). Finally, a recent NMR structural study of the isolated PS1 CTF revealed a half-helix for TMD7 with the GxGD-region in a loose random coil conformation, a full helix for TMD8 and a kinked helix for TMD9 (Sobhanifar et al. 2010). Although the detergent conditions used were not compatible with the catalytic activity of γ-secretase, the data nevertheless fit considerably well with the earlier cysteine-accessibility studies.

Substrate recognition of γ-secretase

While the active site of γ-secretase has been clearly identified to reside in PS, the location of the substrate binding site(s) of γ-secretase is much less clear, although there is evidence that this site is separate from the active site (Esler et al. 2002) and thus representing an exosite. Accordingly, the current model suggests that γ-secretase substrates are first bound at this exosite before they get access to the active site where the peptide bond hydrolyses occur. Although evidence was presented that NCT serves as an initial substrate receptor, recognizing the free N-terminus of the substrate, by demonstrating specific NCT – APP CTF substrate interactions in various experimental settings (Shah et al. 2005), follow-up studies have yielded both additional supporting data (Dries et al. 2009) as well as further conflicting data (Chavez-Gutierrez et al. 2008; Martin et al. 2009; Zhao et al. 2010). APH-1 has recently been implicated in γ-secretase – APP CTF substrate interaction as well (Chen et al. 2010). However, also full length APP, which is not a substrate of γ-secretase has been reported to interact with APH-1, which seems inconsistent with our current understanding of γ-secretase function. Clearly, whether γ-secretase subunits with substrate receptor function exist requires further investigation. Whether initial substrate recognition occurs by NCT and/or APH-1 or not, the substrate comes in contact with a so-called substrate docking-site in PS. While the existence of this site has been clearly documented by pharmacological means including substrate-mimicking peptides, its precise location is less clear. Initial studies suggest rather broadly the PS NTF/CTF interface as docking site, which partially overlaps with the PS active site (Kornilova et al. 2005). However, both the NTF and CTF have also been mapped individually as targets of substrate-mimicking peptides (Sato et al. 2008b). Thus, each PS fragment may provide separate docking sites.

The close vicinity of docking and catalytic site implicates that alteration of certain residues in this region might have consequences on substrate processing. Indeed, mutational analysis of residues of the GxGD active site motif showed that this region is highly critical for substrate processing (Steiner et al. 2000; Yamasaki et al. 2006; Perez-Revuelta et al. 2010). The glycine residues of the motif are probably conserved to allow processing of all known substrates and are likely providing space and/or flexibility to accommodate substrates close to the catalytic aspartates (Perez-Revuelta et al. 2010). This would be consistent with the flexible structure that has been identified for the motif by cysteine-accessibility studies and the NMR structure of the PS1 CTF. Very similar observations on γ-secretase substrate processing have been made for the PAL motif, which although distant in the primary sequence is located close to the active site. Also this motif is very sensitive to mutation and only a few residues are tolerated for substrate processing (Wang et al. 2006a). Thus, the available data demonstrate the existence of strict sequence requirements at the γ-secretase active site for substrate cleavage and suggest that the GxGD and PAL motifs could be subsites of the protease, interacting with the substrate and thus being important for substrate docking. A certain degree of substrate selectivity may be provided by the stable interaction of the protease with the substrate sequence in this subsite region. Substrates that may not optimally fit in this subsite region, that may be part of the docking site, would only be poorly cleaved. Finally, not only residues at the active site domain have strict requirements for substrate processing. The juxtamembrane domain, TMD or ICD of the substrate itself have substantial influence for recognition and cleavage by γ-secretase and other GxGD proteases (Zhang et al. 2002; Ren et al. 2007; Hemming et al. 2008; Martin et al. 2009). It is unclear, however, how these substrate domains contribute to the docking process. At least the extracellular N-terminal domain may play a role for an initial interaction with NCT if one accepts a role of this subunit as substrate receptor. Apart from the substrate’s TMD itself, for which a role in substrate docking seems more obvious, the extracellular and intracellular domains surrounding the TMD may play a role in inducing a conformation of the TMD that allows its proper positioning at the docking site(s) of the enzyme prior to intramembrane cleavage.


The expression ‘regulated intramembrane proteolysis’ has been coined 10 years ago by Brown, Goldstein and coworkers (Brown et al. 2000). At that time, only five mammalian proteins were known to undergo RIP and only few intramembrane proteases had been identified. It became clear that shedding and intramembrane proteolysis are linked and constitute a completely new mechanism of signal transduction. In the meantime, many more RIP substrates and RIP proteases have been identified. In addition, the first intramembrane proteases have been crystallized. Their structure has confirmed that proteolysis can indeed take part in the membrane. In the last years, it also became clear that RIP is involved in many biomedical processes under physiological, but also pathophysiological conditions. Despite the exciting progress in this young research field, our understanding of the RIP process is partial and many questions remain open.

Several hundred membrane proteins undergo shedding (Faca et al. 2008), but in most cases we need to find out by which proteases they are processed and whether all of them further undergo intramembrane proteolysis. This will allow us to decipher the mechanisms that govern substrate recognition and substrate specificity and will also tell us how much functional overlap exists between the distinct RIP proteases. In fact, we still do not fully understand how RIP substrates are recognized. Likewise, the determinants of what makes a RIP substrate a ‘good substrate’ or ‘bad substrate’ are largely unknown. Furthermore, we also do not know much about how the lipid environment affects the intramembrane cleavage. Another open question is the biological or pathological function of RIP for a given substrate. Do most ICDs mediate signal transduction, and, if yes, by which mechanism? What are the mechanisms that regulate the RIP cascade and how can they be used for therapeutic interventions? In particular, regarding therapeutic issues such as the treatment of AD, how can we overcome the problem of known side effects associated with, for example, γ-secretase inhibition because of interference with Notch cleavage and potentially of other physiologically crucial substrates? Binding sites and mechanistic mode(s) of action of Notch-sparing inhibitors and modulators of γ-secretase are not established yet. How safe will inhibitors of β-secretase be? Will there be similar problems encountered as for γ-secretase inhibition? Finally, the field awaits the crystal structure determination of GxGD-type aspartyl proteases. While that may be not too remote for SPP and SPPLs and perhaps PS itself (Ahn et al. 2010), it will be an extremely challenging task to accomplish for the γ-secretase complex.


Research in our labs was funded by the Deutsche Forschungsgemeinschaft (SFB596), the BMBF (KNDD) and the Center of Integrative Protein Science Munich (CIPSM). The LMU excellent program supports C.H. with a research professorship. We apologize that not all previous work on RIP and APP processing could be included in this review.