ADAMS: KEY COMPONENTS IN EGFR SIGNALLING AND DEVELOPMENT

Transcription

1 ADAMS: KEY COMPONENTS IN EGFR SIGNALLING AND DEVELOPMENT Carl P. Blobel Abstract ADAM (a disintegrin and metalloprotease) proteins are membrane-anchored metalloproteases that process and shed the ectodomains of membrane-anchored growth factors, cytokines and receptors. ADAMs also have essential roles in fertilization, angiogenesis, neurogenesis, heart development and cancer. Research on ADAMs and their role in protein ectodomain shedding is emerging as a fertile ground for gathering new insights into the functional regulation of membrane proteins. METALLOPROTEASE A peptidase that depends on a coordinated metal ion (Zn 2+ ) for its catalytic mechanism. ANGIOGENESIS The process of forming new blood vessels by sprouting from pre-existing ones. EGF-LIKE DOMAIN A motif with ~50 amino acids, including six cysteine residues and a mainly β-sheet structure, found in all ErbB-binding growth factors and in extracellular matrix proteins. Arthritis and Tissue Degeneration Program and Cell Biology Program, Hospital for Special Surgery, Weill Medical College of Cornell University, 535 East 70th Street, New York, New York 10021, USA. blobelc@hss.edu doi: /nrm1548 Proteolytic processing and the release of membrane proteins can function as a post-translational switch that regulates the activity of the cleaved substrate. This process, which is referred to as protein ectodomain shedding, might activate or inactivate the substrate protein, or substantially change its functional properties 1 4.A family of membrane-anchored METALLOPRO- TEASES that are known as a disintegrin and metalloprotease (ADAM) proteins are key components in protein ectodomain shedding 3 9.In this review, I aim to highlight the most exciting recent progress in the field of ADAMs and the role of ADAMs in protein ectodomain shedding. There will be an emphasis on how ADAMs might regulate signalling through the epidermal growth factor receptor (EGFR) a tyrosine kinase receptor with important roles in development and in diseases such as cancer 10,11.Following a brief overview of ADAMs, general concepts of the effects of ectodomain shedding on the activity of substrate proteins are outlined. This leads into a discussion of the central role that ADAMs seem to have in the shedding of members of a family of membraneanchored growth factors, the ligands of the EGFR. Recent structural studies that have uncovered an unusual mechanism of EGFR dimerization (reviewed in REFS 12,13) will be placed in the context of ADAMdependent EGFR-ligand shedding in an attempt to explain the potential importance of shedding in regulating the activity of EGFR ligands. This, and other functions of ADAMs, raises questions about how ADAMs themselves are regulated, which is also briefly addressed. Finally, recent findings that implicate ADAMs in heart development and ANGIOGENESIS will be discussed. Several other aspects of ADAMs research such as their role in cell cell interactions, in the activation of the cell-surface receptor Notch, in axon guidance, in asthma, or as α-secretases in Alzheimer s disease have not been included, because they have been recently reviewed elsewhere The main goal of this review is to stimulate new interest in the area of ADAMs and to highlight their remarkable contributions to signalling through proteolysis. An overview of ADAMs A typical ADAM consists of a series of conserved and characteristic protein domains: an N-terminal signal sequence is followed by a pro-domain, a metalloprotease domain, a disintegrin domain, a cysteine-rich region, an EGF-LIKE DOMAIN,a transmembrane domain and a cytoplasmic domain (FIG. 1) (REFS 4,7,18). The first ADAMs to be recognized were the two subunits of the heterodimeric sperm protein fertilin (ADAM1 and ADAM2) Since the discovery of fertilin, many other ADAMs have been identified in various species, including Schizosaccharomyces pombe 22 (but not in Saccharomyces cerevisiae), Caenorhabditis elegans 23, Drosophila melanogaster 24 and in vertebrates (the identification numbers are assigned in the order in which ADAMs have been discovered; a frequently updated list of ADAMs that have been identified in different species 32 JANUARY 2005 VOLUME 6

2 PRO-PROTEIN CONVERTASE Member of the family of Ca 2+ - dependent, subtilisin-like serine endoproteases that are structurally related to KEX2 and furin and that cleave pro-protein substrates at the C-terminal side of doublets or clusters of basic amino acids. TRANS-GOLGI NETWORK Membranous compartment from which vesicles bud to deliver proteins and other materials to the cell surface or to the late endosomes for delivery to lysosomes. SRC-HOMOLOGY-3 (SH3) DOMAIN A protein protein interaction domain that recognizes a unique proline-rich peptide motif. This domain is found in many proteins that are involved in signal transduction and membrane cytoskeleton interactions. Metalloprotease domain Disintegrin domain Cysteine-rich region EGF-like domain Transmembrane domain Cytoplasmic domain Figure 1 ADAM domain structure. A typical ADAM (a disintegrin and metalloprotease) protein has the following domain organization: the extracellular domain contains an N-terminal metalloprotease domain, a disintegrin domain, a cysteine-rich region and an epidermal growth factor (EGF)-like domain 4,7,18 ; the cytoplasmic domain frequently contains signalling motifs such as phosphorylation sites or proline-rich regions, which bind SRC-HOMOLOGY-3 (SH3) DOMAINS. ADAMs also contain an N-terminal signal sequence and a pro-domain, which are not shown. Modified with permission from REF. 144 (1997) Elsevier. can be found on the White Laboratory web site in the online links box). Only half of the ADAMs that are known at present contain a catalytic-site consensus motif for metalloproteases HEXXH in their metalloprotease domain. Several of these have been shown to be catalytically active 25 34,which implies that other ADAMs that contain an HEXXH sequence should also possess catalytic activity. Furthermore, the catalytic activity of some, but not all, ADAMs can be inhibited by tissue inhibitors of matrix metalloproteinases (TIMPs) 35.Those ADAMs that do not contain the HEXXH sequence in their otherwise conserved metalloprotease-like domain probably do not possess catalytic activity. The pro-domain of catalytically active ADAMs is thought to function as an intramolecular chaperone 4,7,29. Once the ADAM is properly folded, the pro-domain keeps the enzyme inactive until it is removed by a furintype PRO-PROTEIN CONVERTASE or by autocatalytic removal in the TRANS-GOLGI NETWORK 29,32,36 39.The disintegrin domain derives its name from a high sequence similarity to snake-venom disintegrins 40.These short, soluble proteins, many of which contain an Arg Gly Asp (RGD) integrin-binding consensus motif, bind to the platelet integrin gpiibiiia and thereby function as potent competitive inhibitors of platelet aggregation 40,41. With the exception of human ADAM15 (REFS 42,43), none of the ADAMs that are known at present contain a corresponding RGD sequence. Nevertheless, several studies have implicated ADAMs in cell cell interactions (reviewed in more detail in REF. 14). The disintegrin domain and cysteine-rich region can also have a role in substrate targeting 44,45 and can facilitate the removal of the pro-domain from the catalytic domain 37.The latter implies that the catalytic domain interacts with the disintegrin domain and/or the cysteine-rich region. Finally, the cytoplasmic domain of ADAMs frequently contains signalling motifs, such as phosphorylation sites or proline-rich regions, which bind Src-homology-3 (SH3) domains 7,46. Several ADAMs are mainly or exclusively expressed in the male reproductive system of mammals, and therefore probably function predominantly in spermatogenesis or fertilization (REFS 47 50;see also the White Laboratory web site). This review mainly focuses on ADAMs that are widely, or even ubiquitously, expressed and that contain a catalytic-site consensus sequence (BOX 1). Ofthese, ADAM17 (tumour-necrosis factor (TNF)α-converting enzyme (TACE)) and ADAM10 (Kuzbanian) are best characterized at present. ADAM17 is discussed in more detail here than other ADAMs because of its key role in activating the EGFR. Little is currently known about the function of most ADAMs that lack a catalytic site. However, studies of ADAM2 and ADAM3, which are essential for fertilization 47 49,51, and ADAM23,which is crucial for brain development 52, indicate that studying other non-catalytic ADAMs might be just as rewarding as studying ADAM2, ADAM3 and ADAM23, or the catalytically active ADAMs that are reviewed here. Protein ectodomain shedding One of the most interesting and best characterized functions of catalytically active ADAMs is their role in protein ectodomain shedding the proteolytic release of the ectodomain of a membrane protein that is usually triggered by a cut adjacent to the plasma membrane 2,53 (see FIG. 2a). Ectodomain shedding affects many structurally and functionally diverse molecules, such as the pro-inflammatory cytokine TNFα,all EGFR ligands, receptors such as TNF receptor-i and -II, ErbB2, ErbB4, and a number of other proteins such as Delta, the amyloid precursor protein and L-selectin 3 5,7,8.Indeed, 2 4% of the proteins on the cell surface are subjected to ectodomain shedding 54. The functional consequences of protein ectodomain shedding. Recent studies have shown that ectodomain shedding has an essential role in regulating the function of several substrate proteins (see below). In principle, the consequences of ectodomain shedding can vary, depending on the function of the substrate protein (FIG. 2b). The most obvious consequence of ectodomain shedding is to allow a membrane-tethered growth factor or cytokine to participate in paracrine signalling that is, to function at a distance from the site of cleavage, or perhaps to enter the bloodstream 1.In the absence of shedding, a membrane-anchored ligand (such as TNFα or an EGFR ligand) could theoretically only engage a receptor on the same cell (autocrine signalling) or on an immediately adjacent cell (juxtacrine signalling) 55. Remarkably, however, with regard to ligands of the EGFR, there is good evidence that even juxtacrine and NATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 6 JANUARY

3 Box 1 Knockout mice for ADAMs that are candidate sheddases ADAM12 ADAM19 ADAM33 ADAM8 ADAM28 Consequences of targeted gene inactivation 30% embryonic lethality, defects in brown adipose tissue % postnatal lethality 1 3 days after birth, defects in cardiovascular morphogenesis 101,102. Viable and fertile, no major histopathological defects (S. Umland, Schering Plough, New Jersey, USA; personal communication). Viable and fertile, no evident spontaneous pathological or developmental phenotypes 138. N/A Nucleotide substitutions ADAM15 ADAM9 ADAM10 (KUZ) ADAM17 (TACE) Schizosaccharomyces pombe ADAM Viable, fertile, no evident spontaneous pathological or developmental phenotypes, but defects in pathological NEOVASCULARIZATION 113. Viable, fertile, no evident pathological or developmental phenotypes 139. Early embryonic lethality (embryonic day 9.5), probably owing to defects in angiogenesis; resembles Notch1 knockout 59. Perinatal lethal, probably owing to defects in heart development; phenotype resembles that of mice that are knockout for EGFR, TGFα, HB-EGF or amphiregulin 70 72,74. Defect in formation of the spore envelope, but this function does not depend on the catalytic activity 22. The consequences of targeted deletions of various catalytically active mouse ADAM (a disintegrin and metalloprotease) proteins that are considered to be candidate protein ectodomain sheddases that is, ADAMs that contain a membrane-anchored metalloprotease domain with a catalytic-site consensus sequence (HEXXH), and that are expressed in somatic tissues other than the male reproductive tract are summarized above (right). The phylogenetic relationship of the ectodomains of these ADAMs and an ADAM from Schizosaccharomyces pombe 22, as determined using the DNA Star Clustal alignment program, is indicated (above, left). Some mice that lack multiple ADAMs have been generated. Specifically, double-knockout mice that lack ADAM9 and ADAM15 and triple-knockout mice that lack ADAM9, ADAM12 and ADAM15 are viable and fertile with no spontaneously developing pathological phenotypes 67.Furthermore, quadruple-knockout mice that lack ADAM9, ADAM12, ADAM15 and ADAM17 resemble Adam17 / mice, which argues against significant compensatory or redundant roles in development among these ADAMs 67.Also, targeted deletions of ADAMs that lack a catalytic-site consensus sequence (HEXXH) have shown an essential role for ADAM2 and ADAM3 in fertilization 47,49,51,and for ADAM23 in brain development 52. EGFR, epidermal growth factor receptor; HB-EGF, heparin-binding EGF; KUZ, Kuzbanian; TACE, tumour-necrosis factor-α-converting enzyme; TGFα,transforming growth factor α. NEOVASCULARIZATION De novo stimulation of new blood supplies to a growing tumour. autocrine signalling might be regulated by ectodomain shedding (REFS 56,57; discussed in more detail below). The ectodomains of receptors might also be released by shedding, which could thereby inactivate the receptor. Solubilized unoccupied receptors could also potentially function as decoys that sequester soluble ligands. An example of a disease that is caused by a defect in receptor shedding is TNF-receptor-associated periodic febrile syndrome (TRAPS) 58.Patients that suffer from TRAPS have dominant mutations in the cleavage site of the p55 TNF receptor (TNFR1) that prevent or reduce its shedding and downregulation. Therefore, the TNFR1 accumulates on the cell surface, which, in turn, increases the susceptibility to TNFα and results in recurring fevers through increased inflammatory responses. On the other hand, ectodomain shedding can also activate receptors. An excellent example of this is Notch 59,60, for which membrane-proximal ectodomain processing leads to a second, presenilin-dependent cleavage within the transmembrane domain. This socalled regulated intramembrane proteolysis (RIP) releases the cytoplasmic domain from its membrane anchor, and allows it to enter the nucleus and participate in the transcriptional regulation of specific target genes. Recently, receptor ligands such as heparin-binding (HB)-EGF and neuregulin have also been found to be subjected to RIP, and the release of their cytoplasmic domain is thought to have important functional consequences 64,65.A membrane-anchored ligand might therefore also function as a counter-receptor by signalling back to the cell on which it resides 1. The examples mentioned above highlight the potential relevance of ectodomain shedding as a post-translational regulator of membrane proteins, and provide a conceptual framework for considering the potential functional consequences of ectodomain shedding for other membrane proteins. At the same time, these examples also help to emphasize that the consequences of ectodomain shedding might be different for individual substrate proteins, and will therefore need to be experimentally established on a case-by-case basis. ADAMs as molecular signalling switches All EGFR ligands are made as membrane-anchored precursors that can be proteolytically released from cells (reviewed in REF. 66). Over the past few years, proteolytic 34 JANUARY 2005 VOLUME 6

4 a Membrane protein Soluble ectodomain b Autocrine signalling Ligand Paracrine signalling Receptor Sequestration ADAM (or other protease) Plasma membrane Nucleus Juxtacrine signalling Figure 2 Protein ectodomain shedding. a A schematic representation of an ADAM (a disintegrin and metalloprotease) protein or other protease that is engaged in membrane-proximal cleavage of a membrane protein, which results in the release of its soluble ectodomain. Structurally and functionally diverse molecules are subjected to ectodomain shedding. Among these are: cytokines and growth factors (such as tumour-necrosis factor α (TNFα), transforming growth factor α (TGFα), heparin-binding epidermal growth factor (HB-EGF)); receptors (such as TNF receptor-i and -II (TNFRI and TNFRII) and ErbB4); and other molecules (such as Delta, L-selectin, fractalkine, amyloid precursor protein (APP) and angiotensin-converting enzyme (ACE)). b A receptor ligand pair is used to illustrate possible roles of ectodomain shedding. In the absence of shedding, a membrane-anchored ligand might only engage its receptor in a juxtacrine 55 or autocrine fashion (although there might be impediments to autocrine receptor stimulation, such as improper orientation of the ligand and receptor). However, to reach a receptor at a distance and to participate in paracrine signalling, a membrane-anchored ligand must be shed 1. Receptors might also be shed, which could result in their activation or inactivation. Signalling through Notch is a prime example of a role for proteolysis in activating a receptor 60. A membrane-proximal cleavage by an ADAM triggers a second (presenilin-dependent) cleavage, which is referred to as regulated intramembrane proteolysis. This, in turn, activates downstream targets of the receptor Shedding might also produce a soluble decoy receptor that could sequester a ligand. G-PROTEIN-COUPLED RECEPTOR A seven-helix membranespanning cell-surface receptor that signals through heterotrimeric GTP-binding and -hydrolysing G-proteins to stimulate or inhibit the activity of a downstream enzyme. ENDOCARDIAL CUSHION Discrete cushion-like swelling that forms in the developing heart and that gives rise to mature heart valves and to the membranous part of the ventricular septum. The ventricular septum is a wall that separates the left and right ventricles of the heart. processing of some EGFR ligands has emerged as a key regulatory switch in EGFR signalling. ADAMs have been implicated in the shedding of six out of the seven known EGFR ligands (transforming growth factor (TGF)α, EGF, HB-EGF, betacellulin, epiregulin and amphiregulin (see REF. 67 and references therein)). This raises interesting questions about how general the role of ADAMs as regulators of EGFR-dependent signalling pathways might be. Proteolysis in the regulation of EGFR signalling. One of the first clues that proteolysis affects EGFR signalling emerged from a study in which EGF was overexpressed in cells as a soluble protein that is, without its membrane anchor 56.Whereas signalling that is elicited by overexpressed EGF with a membrane anchor could be blocked by an anti-egfr antibody, stimulation of the EGFR by overexpressed soluble EGF could not be inhibited by this antibody. The authors concluded that the membrane anchor of EGF somehow prevents intracrine stimulation of the EGFR during biosynthesis in the secretory pathway of the EGF-expressing cells. A separate study showed that EGFR-dependent migration and proliferation of a breast-cancer cell line could be blocked by a metalloprotease inhibitor 68, and that this was reversible by exogenous soluble EGF. Finally, several studies have documented that crosstalk between G-PROTEIN-COUPLED RECEPTORS (GPCRs) and EGFR requires the activation of EGFR ligands by metalloproteases, including ADAM10, ADAM12 and ADAM17 (reviewed in REF. 69;see BOX 2). Striking evidence for the in vivo relevance of ectodomain shedding in EGFR signalling during development came from the analysis of Adam17 / mice 70. Remarkably, these animals resemble mice that lack TGFα (or the EGFR) as they have defects in the maturation and morphogenesis of epithelial structures, including a failure to undergo eyelid fusion 70.Furthermore, Adam17 / cells are deficient in TGFα shedding 70. Evidently, although the EGFR and pro-tgfα are present in Adam17 / mice, this ligand receptor pair does not function properly in the absence of ADAM17. More recent studies have uncovered additional defects in Adam17 / mice that might also result from a lack of EGFR-ligand processing. These include defects in branching morphogenesis of the lung 71, thickened and misshapen heart valves that resemble those of mice lacking HB-EGF 67,72 74, and defects in branching morphogenesis during development of the mammary gland (Z.Werb,personal communication) that are similar to those seen in mice that lack amphiregulin 75.Taken together, these results provide compelling evidence for a crucial role for shedding in regulating the function of the EGFR and at least three of its ligands, TGFα, HB-EGF and amphiregulin, during development. A central role for ADAM17 in paracrine EGFR signalling? In the case of paracrine stimulation of the EGFR (see FIG. 2b), the reason for the key role of ADAM17 is clear a membrane-tethered growth factor simply cannot activate the EGFR from a distance. The role of HB-EGF in heart development can be used as an example of paracrine EGFR signalling in vivo.in the developing heart, HB-EGF is highly expressed in a monolayer of endocardial cells that overlie the ENDOCARDIAL CUSHION 72,73. However, Hb-egf / mice have severe defects in morphogenesis of the endocardial cushion, which results in enlarged and misshapen heart valves 72,73.As mentioned NATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 6 JANUARY

5 Box 2 Crosstalk between GPCRs and EGFR GPCR βγ GRAM-POSITIVE BACTERIA The cell walls of these bacteria retain a basic blue dye during the Gram-stain procedure. These cell walls are relatively thick (15 80 nm) and consist of a network of peptidoglycans. α ADAM Plasma membrane Ligand release HB-EGF EGFR Receptor dimerization and signalling Evaluation of the role of ADAM (a disintegrin and metalloprotease) proteins in the crosstalk that occurs between G-protein-coupled receptors (GPCRs) and the receptors for epidermal growth factor (EGF) is a fertile area of research. The initial discovery that GPCR EGFR crosstalk involved metalloprotease-dependent shedding of EGFR ligands was reported in 1999 (REF. 140). The term triple-membrane-passing signal (TMPS) was coined to describe this unexpected means of crosstalk, in which a GPCR activates an ADAM, which, in turn, releases an EGFR ligand (such as heparin-binding EGF (HB- EGF); see the figure) to activate the EGFR 69.Numerous different cases of TMPS have been described, and they involve distinct GPCRs and ADAMs (see REF. 69 and references therein). It is becoming clear that the outcome of GPCR EGFR crosstalk might be different depending on which GPCR is activated and the signalling context of the recipient cell. Functional consequences of crosstalk between GPCRs and the EGFR that might be medically significant include cardiac hypertrophy 90,mucus production in the lung that is stimulated by GRAM-POSITIVE BACTERIA 92, and GPCR-dependent motility and proliferation of cancer cells 96. previously, Adam17 / mice have similar defects in heart development to Hb-egf / mice 72.ADAM17 has been implicated in HB-EGF shedding in cell-based assays 67,76,77,which implies that the heart defect of Adam17 / mice is most likely caused by a defect in the release of HB-EGF from endocardial cells. The notion that the heart defects in Adam17 / mice are caused by a lack of HB-EGF processing is further corroborated by the finding that knock-in mice with an uncleavable form of HB-EGF also have heart defects that resemble those seen in Adam17 / or Hb-egf / mice 78.Taken together, these three lines of evidence strongly imply that proteolysis in general and that mediated by ADAM17 in particular is essential for paracrine HB- EGF signalling during heart development. In this context, it should be noted that no increase in valve thickness is seen in mice that lack other candidate HB-EGF sheddases (ADAM9, ADAM12 and ADAM15), which argues against an important role for these ADAMs in activating HB-EGF in the endocardium 67. An intriguing question raised by the phenotype of Hb-egf / mice is how a lack of this growth factor can result in the increased proliferation of structures that are derived from the endocardial cushion. This issue was elegantly addressed by L. Jackson and colleagues, who provided evidence for an unexpected inhibitory role for the EGFR in proliferation of the endocardial cushion 72.Apparently, activation of the EGFR inhibits signalling through the bone morphogenetic protein (BMP) Smad1/5/8 pathway, which would otherwise stimulate growth of the endocardial cushion. Therefore, the growth factor HB-EGF might actually block proliferation of cells in the endocardial cushion once it is released by ADAM17 (REF. 72). A requirement for shedding in juxtacrine EGFR signalling? Although it is intuitively clear why shedding is required for paracrine EGFR signalling, whether or not shedding is also important in juxtacrine EGFR signalling is more controversial. On the one hand, several studies have shown that juxtacrine EGFR signalling does not require EGFR-ligand shedding. For example, cells that express uncleavable, mutant forms of TGFα can trigger EGFR phosphorylation in EGFR-expressing cells that are cocultured above them 79,80, can induce transformation of cells 81 and might even enhance EGFR signalling 82. Furthermore, cells that express HB-EGF or TGFα, but that are prevented from shedding these ligands by chemical fixation, can stimulate EGFR signalling in live cells that are cultured above them 83,84. On the other hand, a recent study by M. Borrell-Pages and colleagues indicates that shedding can be essential for juxtacrine EGFR stimulation by TGFα at least, under the conditions that were used in their study 57.One key experiment showed that juxtacrine EGFR signalling in EGFRexpressing A431 cells that were cultured on top of TGFα- expressing cells could be blocked by the metalloprotease inhibitor batimastat. Importantly, binding of A431 cells to the TGFα-expressing cells was increased in the presence of batimastat, and could be inhibited by function-blocking antibodies against the EGFR, which corroborated that uncleaved TGFα can bind to the EGFR. Other approaches to abolish or reduce the release of soluble TGFα in otherwise similar experiments have included its expression in ADAM17-deficient Chinese hamster ovary (CHO) cells (which only release minute amounts of TGFα), and the expression of an uncleavable TGFα mutant in wild-type CHO cells. When TGFα-expressing wild-type CHO cells were injected into nude mice, the resulting tumours grew faster than those derived from TGFα-expressing ADAM17-deficient CHO cells, or from CHO cells that express uncleavable forms of TGFα.Taken together, these results indicate that proteolytic processing of TGFα can significantly enhance EGFR signalling, even under juxtacrine conditions. Can these findings be reconciled with previous studies that also convincingly showed that juxtacrine EGFR signalling does not require shedding of its ligands, or might even be enhanced by a membrane-anchored ligand? Possibly, if the highly unusual mode of EGFR dimerization in the context of ligand ectodomain shedding is considered. The crystal structures of the soluble EGFR ectodomain with or without a bound soluble ligand (TGFα or EGF) have been solved 12,13,85 87.These structures show that the EGFR dimerizes through a peptide loop that is hidden in the unoccupied receptor but 36 JANUARY 2005 VOLUME 6

6 TETRASPANIN FAMILY The tetraspanin family contains proteins that span the membrane four times with two exoplasmic loops, and that can be found at the cell surface. Whereas some are highly restricted to specific tissues, others are widely distributed. Members of this family have been implicated in cell activation and proliferation, adhesion, motility, differentiation and cancer. that is exposed by a dramatic conformational change on ligand binding (FIG. 3a). Interestingly, this exposed dimerization loop protrudes from the side opposite to that bound by the ligand. It is this mode of dimerization that distinguishes the EGFR from most other tyrosine kinase receptors, which dimerize or oligomerize by binding to their ligands 12,13,85,86 (FIG. 4).These structural studies indicate that several steps are required for EGFR dimerization and signalling. First, the EGFR must engage a ligand, which results in the exposure of the EGFR dimerization loop (FIG. 3a). Then, one occupied receptor must interact with another occupied receptor with its dimerization loop exposed. Two separate ligand-binding events must therefore be followed by dimerization of the occupied receptors to form a signalling dimer (FIG. 3b).It is tempting to hypothesize that membrane-tethered EGFR ligands would impede diffusion of the ligand receptor pairs, and therefore also their dimerization (FIG. 3c).At limiting concentrations of ligand, the bound ligand might also dissociate from its receptor before a second occupied receptor is encountered. However, cutting the tether that attaches the ligand to the membrane would remove this impediment to receptor diffusion, and would thereby increase the chances of dimerization (FIGS 3d,e). On the other hand, overexpressing uncleavable ligands would also result in increased ligand density, which, in turn, would facilitate dimerization of two occupied receptors and therefore result in EGFR signalling (FIG. 3f).Indeed, overexpressing uncleavable forms of TGFα or HB-EGF might even enhance juxtacrine signalling 79,80,83,84 because the uncleaved ligand might prevent endocytosis and downregulation of the receptor 82.This model provides a testable hypothesis both for how juxtacrine signalling might occur or even be enhanced by overexpressed uncleaved ligands, and for why shedding might have a unique role in facilitating juxtacrine EGFR signalling at lower, and perhaps more physiological, concentrations of ligand. Other factors, such as the TETRASPANIN CD9,are also known to enhance juxtacrine EGFR signalling 83,84. CD9 has been shown to interact with HB-EGF, amphiregulin and TGFα, and its overexpression could therefore lead to clustering of these EGFR ligands in tetraspanin webs. The resulting local increase in the density of EGFR ligands might also facilitate juxtacrine EGFR signalling. Although the exact mechanism that underlies the effects of CD9 on EGFR signalling remains to be determined, CD9 or other a Dimerization loop (hidden) b TGFα Dimerization loop (exposed) Dimerization loop EGFR Plasma membrane Receptor activation c Ligand-expressing cell d e f Ligand cleavage Receptor-expressing cell Receptor activation Receptor activation Receptor activation Figure 3 EGFR dimerization and shedding. The epidermal growth factor receptor (EGFR) has an unusual method of dimerization. Its ligands are monomeric, and are therefore not directly responsible for dimerization. Instead, the receptor dimerizes through a loop that is only exposed once the ligand has docked 12,13,85,86,141. Parts a and b show EGFR dimerization that is induced by a soluble ligand. A possible explanation for the crucial role of shedding has emerged in the context of membrane-anchored substrates (c e): a receptor with a bound ligand might be less mobile if the ligand is still tethered to an adjacent cell (c) than if the ligand has been cleaved (d,e). Alternatively, overexpression of EGFR ligands could allow juxtacrine signalling, even by uncleavable ligands 55,79 81 (f). Furthermore, an uncleavable EGFR ligand might even enhance signalling if its expression is high enough to allow EGFR dimerization, because it could also prevent internalization and downregulation of the EGFR 82. This model does not explain how fixed ligands might induce EGFR signalling 83, unless, for example, fixation results in clustered ligands, or some mobility of overexpressed ligands remains after fixation. This model also does not depict the N-terminal processing of EGFR ligands, which is similarly important for their function 81. TGFα, transforming growth factor α. NATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 6 JANUARY

7 a b c TNFα TNFα TNFα TNFR TNFR Plasma membrane TNFR Receptor activation Figure 4 Juxtacrine signalling through TNFα. Tumour-necrosis factor α (TNFα) is synthesized as a trimer 142. TNF receptors (TNFRs) are therefore clustered by binding to TNFα, regardless of whether or not TNFα is anchored to the membrane. The binding of three TNFR monomers to a membrane-anchored TNFα trimer can therefore trigger a signal, as depicted in a c. The shedding of TNFα would theoretically only be required for paracrine signalling, but not for juxtacrine signalling. The different functions of soluble and membrane-anchored TNFα have been carefully dissected by Ruuls et al tetraspanins, or other yet-to-be-identified factors might therefore also be involved in determining how ligand shedding affects EGFR signalling in different cell lines. Finally, the mode of EGFR dimerization might also explain why shedding is important for autocrine signalling 56.As opposed to a soluble ligand, a membrane-tethered ligand might not, for example, be able to bind the receptor in a way that exposes the dimerization loop, or, alternatively, if binding can occur, it might misorientate the loop and prevent its interaction with another receptor. Shedding of other EGFR ligands. It is noteworthy that the amino-acid residues in TGFα and EGF that interact with the EGFR in two separately solved crystal structures are conserved in all seven EGFR ligands, as is the distance from the C-terminal receptor-binding site to the plasma membrane (FIG. 5). This indicates that shedding might regulate the juxtacrine signalling of all EGFR ligands in the manner described for TGFα.As the EGFR pathway is a validated target for anti-cancer drugs 11,88, upstream activators of EGFR ligands their sheddases and regulators of these sheddases might now enter the spotlight as potential new drug targets in the EGFR pathway 57,67.Ifspecific aspects of the model that is described above (FIG. 3) can be corroborated through further experiments, it might have important implications for the use of metalloprotease inhibitors in the blocking of EGFR signalling. Specifically, this model would indicate that blocking EGFR signalling with metalloprotease inhibitors might be beneficial when the ligand is expressed at low levels or if an ADAM is overexpressed, because this would decrease the chances of receptor dimerization. However, metalloprotease inhibitors might have the opposite effect and stimulate EGFR signalling at higher ligand concentrations. Targets for inhibition or regulation of EGFR signalling include ADAM17, which has emerged as an important sheddase of HB-EGF, TGFα, amphiregulin and epiregulin in several different cell types 67,70,76,77, and ADAM10, which is the main sheddase of EGF and betacellulin in mouse embryonic fibroblasts 67.The remaining EGFR ligand, epigen, is mainly processed by enzymes that are insensitive to metalloprotease inhibitors in mouse embryonic cells, and that remain to be identified (U. Sahin and C.P.B., unpublished observations). ADAM9, ADAM10 and ADAM12 have also been implicated in the shedding of HB-EGF, although they are probably regulated differently from ADAM17 (REFS 89 93).Furthermore, different ADAMs are apparently important for GPCR EGFR crosstalk in different tumour cell lines It will be interesting to define which ADAMs or other enzymes can release different EGFR ligands from cells when they are overexpressed or misregulated, and to evaluate individual tumours for the expression of candidate EGFR-ligand sheddases as well as EGFR ligands. It will also be valuable to determine whether there are signalling complexes that consist of different combinations of ADAMs and EGFR ligands in different cells types, and whether this can explain why similar stimuli seem to activate different ADAMs in different cells 69. Besides the EGFR, how important might shedding be for the activation of ligands of other ErbB receptors (ErbB2, ErbB3 and ErbB4)? The structure of ErbB2 indicates that homo- and heterodimers that contain this ErbB receptor might be less dependent on ligand shedding because the receptor dimerization loop of ErbB2 is always exposed and does not require ligand binding (there is no known ligand for ErbB2) 13,87,97.Similarly, in ErbB2 heterodimers, only one of the partners, such as ErbB3, needs to bind a ligand to form the signalling heterodimer 98.In this context, it is worth pointing out that several neuregulins a group of ErbB ligands that also have crucial roles in development are made as membrane-anchored precursors and are shed from the plasma membrane However, it is conceivable that the juxtacrine signalling activity of neuregulins might not require ectodomain shedding in cases where heterodimers that contain ErbB2 are activated. A lack of 38 JANUARY 2005 VOLUME 6

8 TGFα Amphiregulin HB-EGF Epiregulin EGF Betacellulin Epigen R L/M CPDSHTQYCFHGT-CRFLVQEEKPACVCHSGYVGVRCEHADLLAVVAASQKKQAITALVVVSIVALAVLIITCVLIHCCQLR 81 CTAKFQNFCIHGE-CRYIENLEVVTCNCHQDYFGERCGEKSMKTHSEDDKDLSKIAVVAVTIFVSAIILAAIGIGIVITVHLW 82 CLRKYKDYCIHGE-CRYLQEFRTPSCKCLPGYHGHRCHGLTLPVENPLYTYDHTTVLAVVAVVLSSVCLLVIVGLLMF 77 CSSDMDGYCLHGQ-CIYLVDMREKFCRCEVGYTGLRCEHFFLTVHQPLSKEYVALTVILIFLFLIITAGCIYYFCRWYK 78 CPSSYDGYCLNGGVCMHIESLDSYTCNCVIGYSGDRCQTRDLRWWELRHAGYGQKHDIMVVAVCMVALVLLLLLGMWGTYYY 82 CPKQYKHYCIHGR-CRFVVDEQTPSCICEKGYFGARCERVDLFYLQQDRGQILVVCLIVVMVVFIILVIGVCTCCHPLR 78 CLEDHNSYCINGA-CAFHHELKQAICRCFTGYTGQRCEHLTLTSYAVDSYEKYIAIGIGVGLLISAFLAVFYCYI 74 C C C C C C Juxtamembrane region Transmembrane domain Figure 5 Is shedding essential for signalling through all EGFR ligands? An alignment of the seven known ligands for the epidermal growth factor receptor (EGFR) transforming growth factor (TGF)α, amphiregulin, heparin-binding EGF (HB-EGF), epiregulin, epidermal growth factor (EGF), betacellulin and epigen shows that the residues that interact with the EGF receptor (EGFR) in TGFα and EGF are conserved (shaded blue; R and L/M indicate the consensus residues), and that the length of the membrane tethers (the juxtamembrane region next to the transmembrane region, which is shaded pink) is similarly short. In light of the model proposed in FIG. 3, this indicates that shedding might affect juxtacrine signalling through all EGFR ligands, similar to TGFα. The six cysteine residues (C) of the EGF-like domain are shown. PHORBOL ESTER A polycyclic ester that is isolated from croton oil. The most common is phorbol-12- myristate-13-acetate (PMA) and 12-O-tetradecanoyl-phorbol- 13-acetate (TPA). These are potent carcinogens or tumour promoters because they mimic diacylglycerol, and thereby irreversibly activate protein kinase C. YEAST TWO-HYBRID SCREEN A technique used to test if two proteins physically interact with each other. One protein is fused to the GAL4 activation domain and the other to the GAL4 DNAbinding domain, and both fusion proteins are introduced into yeast. Expression of a GAL4-regulated reporter gene indicates that the two proteins physically interact. neuregulin shedding might even prevent the endocytosis and downregulation of heterodimers of ErbB2 with other ErbB receptors, and might therefore enhance signalling. Further studies will be necessary to learn more about neuregulin sheddases and to understand the functional consequences of neuregulin shedding. At the same time, it will be interesting to investigate what role shedding has in the regulation of ErbB2 and ErbB4, both of which are released from cells by metalloproteases The anti-erbb2 antibody Herceptin, which is used to treat breast cancer, reportedly blocks shedding of ErbB2, which raises the possibility that shedding has a role in regulating the function of ErbB2 itself 105. Finally, it is striking that ADAMs have not yet been linked to the activation of EGFR ligands in Drosophila melanogaster Instead, seven-membrane-spanning proteases called rhomboids are important for this process in these organisms. This might be an unusual example of an important signalling pathway in which select components do not share highly conserved functions between flies and mammals 110. Regulation of ADAMs The central role of ADAMs in EGFR signalling, as well as their other known functions in development and disease 26,27,60,101,102,113,114,provides a strong impetus to learn more about how these enzymes are regulated. The shedding of ADAM17 substrates, for example, can be stimulated by PHORBOL ESTERS such as TPA (12-O-tetradecanoyl phorbol-13-acetate) or PMA (phorbol-12-myristate- 13-acetate), and by the phosphatase inhibitor pervanadate 1,54,115,116. PMA treatment also increased proteolytic processing of a synthetic TNFα-cleavage-site peptide that was added to the medium of cultured cells, which implies that the activity of ADAM17 is upregulated by this phorbol ester 117.However, the mechanism that underlies this activation remains obscure 118.Interestingly, PMA can even activate a mutant form of ADAM17 that lacks its cytoplasmic domain 45. One possible explanation for this puzzling result is that ADAM17 activity might be regulated by one or more membrane-spanning proteins. Another question that is frequently raised in this context is whether stimulation of ADAM17 with phorbol esters in tissue culture has any relevance for understanding the physiological regulation of ADAM17 in vivo.in the case of EGFR-ligand shedding, at least, there is a good correlation between the requirement for ADAM17 in phorbolester-dependent shedding of TGFα, amphiregulin and HB-EGF in culture and the established role for ADAM17 in activating these ligands during development in vivo (REFS 67,70,72 and Z. Werb, personal communication). This does not necessarily indicate that the activation of protein kinase C (PKC) which is mediated by phorbol esters is important for ADAM17 activity in vivo,but instead indicates that ADAM17 is as important for PMAstimulated EGFR-ligand shedding in cultured cells as it is for the activation of these ligands in vivo.as well as PKC signalling, several other signalling pathways can affect the shedding of ADAM17 substrates, including the receptor-tyrosine-kinase-activated extracellular signal-regulated kinase (ERK)/mitogen-activated protein kinase (MAPK) pathway 119, as well as stimulation of GPCRs Furthermore, YEAST TWO-HYBRID SCREENS have identified several cytoplasmic binding partners of ADAM17, although much remains to be learned about how these proteins affect the function of ADAM17 (REFS ).In light of the key role of ADAM17 in EGFR signalling and in TNFα shedding (among other functions), it is clear that understanding how ADAM17 is regulated in cells and in an intact organism is a high priority in the field. At present, little is known about how ADAMs other than ADAM17 are regulated (for a recent review, see REF. 7). Most ADAMs contain predicted signalling motifs in their cytoplasmic domains, including phosphorylation sites and proline-rich regions that can interact with SH3 domains 7.Several interactions between ADAMs and other cytoplasmic proteins have been identified Some of these interacting proteins, such as PACSIN2 (PKC and casein-kinase substrate in neurons-2) and PACSIN3,have been implicated in regulating the function of ADAMs 123,125, although little is known about the underlying mechanism. Furthermore, although the activity of ADAM17 is strongly stimulated by phorbol esters, at least two ADAMs ADAM10 and ADAM19 have significant constitutive activity towards their substrates in cell-based assays 33,67.Constitutive shedding events that depend on ADAM10 and ADAM19 NATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 6 JANUARY

9 ORTHOLOGUES Functionally related genes with extensive sequence similarity, which indicates a common ancestor. The term orthologues is often used to indicate the most closely related members of larger gene families in different species. RETINOPATHY A non-inflammatory degenerative disease of the retina, commonly found as a complication of diabetes. RNA INTERFERENCE A form of post-transcriptional gene silencing in which expression or transfection of double-stranded RNA induces degradation, by nucleases, of the homologous endogenous transcripts, mimicking the effect of the reduction, or loss, of gene activity. COS CELLS Cells from the monkey CV1 cell line that have an integrated SV40 genome that lacks an origin of replication. Plasmids with an SV40 origin of replication are replicated to a high copy number when transfected. might therefore be regulated, at least in part, by the expression levels of these proteases. Also, ADAM10 has been shown to respond to the activation of GPCRs Finally, it will be interesting to determine why cholesterol removal can activate several sheddases, as has been reported for ADAM10 and ADAM17 (REFS ). ADAMs in heart development and angiogenesis At present, knockout mice for all but one of the catalytically active ADAMs that are expressed in somatic tissues have been generated (see BOX 1). An evaluation of Adam19 / mice recently uncovered similar defects in heart development to those seen in Adam17 / mice thickened, misshapen and improperly remodelled heart valves 101,102.However, despite the phenotypic similarities in the heart defects between these mice, there is no evidence for a role for ADAM19 in HB-EGF shedding 102. And although the heart defect in Adam19 / mice that were generated by two independent groups was similar, the two studies differed in their views as to whether ADAM19 is essential for shedding certain neuregulins. Further analyses will be necessary to understand the mechanism that underlies the role of ADAM19 in heart development, and to determine whether the increased proliferation of endocardial-cushion cells that occurs in the absence of ADAM19 could be explained, for example, by a lack of shedding and a downregulation of receptors such as the BMP2 receptor or ErbB2, or by yet-to-be determined functions of other domains of ADAM19, such as signalling through its cytoplasmic domain or by cell cell interactions that are mediated by its extracellular domains. As both ADAM19 and ADAM17 have essential roles in heart development in mice, it will also be interesting to determine whether mutations in their human ORTHOLOGUES might be responsible for congenital heart defects the most common congenital defects in humans. Two other interesting processes in which ADAMs have recently been implicated are angiogenesis and neovascularization. Adam10 / mice die at embryonic day 9.5 (E9.5) with malformed vessels in their yolk sacs, possibly owing to defects in Notch1 and/or Notch4 signalling 59.More recently, ADAM15 was found to have a role in pathological neovascularization in a mouse model for RETINOPATHY of prematurity, even though it is not required for developmental angiogenesis or adult homeostasis 113.The growth of implanted tumour cells was also strongly inhibited in Adam15 / mice, which is consistent with a role for ADAM15 in neovascularization. As ADAM15 is not essential for development or adult homeostasis, it might represent a good target for the design of inhibitors of pathological neovascularization. In light of the known role of ectodomain shedding as a post-translational effector of membrane proteins, it will also be interesting to determine whether ADAM10 or ADAM15, or other ADAMs, might be involved in shedding receptors or other proteins that have a role in angiogenesis and neovascularization, such as the vascular endothelial growth factor (VEGF) receptors-1 and -2 and the angiopoietin receptors TIE1 and TIE2. Conclusions and perspectives Over the past few years, functional studies of ADAMs and their roles in protein ectodomain shedding have become an increasingly stimulating and fruitful area of research. Now that the relevance of ectodomain shedding is well established for certain signalling pathways (for example, EGFR signalling, TNF-receptor signalling and Notch signalling), it is tempting to speculate that this process might have a more general role in regulating the function of other membrane proteins that are shed (as noted previously, 2 4% of the proteins on the cell surface are shed 54 ). Indeed, shedding might have evolved in part to fulfil a similar purpose to constitutive and regulated exocytosis, in this case by allowing membrane proteins to be released constitutively or on demand. It has now become feasible to identify the sheddase(s) for any membrane protein of interest using RNA INTERFERENCE (for an example, see REF. 96), cells from various ADAM-knockout mice 67,131,132,or by comparing the response of an unknown sheddase to pharmacological activators and inhibitors of shedding with the properties of known sheddases (for an example, see REF. 131). In addition to loss-of-function experiments that define which sheddase(s) are essential for processing a specific substrate in a given cell type, it will be equally important to perform gain-of-function experiments (for an example, see REF. 133). These help to establish which enzymes can process a substrate, and might therefore contribute to its shedding in vivo, especially in cells or tissues where the enzyme is highly expressed, or in pathological conditions where it is misregulated. Knowing which enzymes can process a substrate also helps in the design of experiments to address potential compensatory or redundant functions of different ADAMs. With few exceptions, shedding studies have so far been performed in garden variety cell types, such as CHO and COS CELLS, or in mouse embryonic cells. It will now be crucial to extend these studies to more specialized cell types and tissues, particularly those in which a given substrate of interest exerts its function. Furthermore, knock-in mutations of the cleavage site of the substrate can provide invaluable information on the functional consequences of shedding for any given substrate (a caveat is that it must be confirmed that shedding is indeed abolished, as mutations in cleavage sites could create new sites for other proteases; for an example, see REF. 134). Given the large number of proteins that are shed, and the comparably small number of sheddases, it is likely that every sheddase will have several substrates, as is the case for ADAM10 and ADAM17 (REFS 4 7).In any given cell or tissue, the function of a particular ADAM might then be equivalent to the combined consequences of the shedding events that it mediates. Nevertheless despite their potentially pleiotropic nature key functions for ADAMs have emerged in certain well-defined signalling pathways, such as activation of the EGFR or of Notch. Genetic systems such as knockout mice can therefore establish predominant roles for ADAMs in certain developmental or disease processes, which, in turn, might 40 JANUARY 2005 VOLUME 6