Phosphatidylserine exposure is required for ADAM17 sheddase function

ADAM17, a prominent member of the ‘Disintegrin and Metalloproteinase' (ADAM) family, controls vital cellular functions through cleavage of transmembrane substrates. Here we present evidence that surface exposure of phosphatidylserine (PS) is pivotal for ADAM17 to exert sheddase activity. PS exposure is tightly coupled to substrate shedding provoked by diverse ADAM17 activators. PS dependency is demonstrated in the following: (a) in Raji cells undergoing apoptosis; (b) in mutant PSA-3 cells with manipulatable PS content; and (c) in Scott syndrome lymphocytes genetically defunct in their capacity to externalize PS in response to intracellular Ca2+ elevation. Soluble phosphorylserine but not phosphorylcholine inhibits substrate cleavage. The isolated membrane proximal domain (MPD) of ADAM17 binds to PS but not to phosphatidylcholine liposomes. A cationic PS-binding motif is identified in this domain, replacement of which abrogates liposome-binding and renders the protease incapable of cleaving its substrates in cells. We speculate that surface-exposed PS directs the protease to its targets where it then executes its shedding function.

T he evolutionarily conserved ADAM (a disintegrin and metalloprotease) family of cell-bound proteinases regulates a plethora of biological functions via cleavage of transmembrane substrates 1 . ADAM17, originally discovered as the TNF-a-converting (cleaving) enzyme, has emerged as a pre-eminent member of this family [2][3][4] . More than 80 ADAM17 targets have been identified to date, prominent among which are cytokines, cell surface receptors and adhesion molecules 5 . Although several 'preferential' substrates such as L-selectin 4,6-8 or tumour necrosis factor receptor type 1 (TNFR1) 9,10 exist, some others are shared with the closely related protease ADAM10 (ref. 11). Substrate cleavage sites contain no consensus sequence and, remarkably, they have only one feature in common: they are located very near to the surface of the lipid bilayer 12 .
ADAM17 regulates cell growth through the liberation of epidermal growth factor receptor (EGFR) ligands and subsequent activation of ERK1/2 signalling [13][14][15] . Although ADAM17 is the major sheddase of transforming growth factor-a (TGF-a), amphiregulin, epiregulin and heparin-binding EGF, ADAM10 is predominantly involved in the release of EGF and betacellulin 14 .
In the mouse, deletion of ADAM17 leads to embryonic lethality due to uncorrectable developmental defects ensuing from disrupted EGFR signalling 4 . In humans, ADAM17 deficiency results in severe inflammatory skin and bowel disease, underlining its important role for epithelial cell homeostasis 16,17 .
Regulation of ADAM function occurs at many levels. Exit from the endoplasmic reticulum requires interaction with inactive members of the Rhomboid family of intramembrane serine proteinases (iRhom1 and iRhom2) 18,19 . During ADAM17 maturation, the prodomain is removed by pro-protein convertases such as furin 20 . A sorting protein named phosphofurin acidic cluster sorting protein 2 reportedly regulates ADAM17 trafficking and diverts the enzyme away from degradation pathways 21 . On the cell surface, sheddase activity can be rapidly induced by remarkably diverse stimuli including protein kinase C (PKC) activators, purine 2 (P2) receptor agonists, fibroblast growth factor 7 (FGF7), Ca 2 þ ionophores and membrane perturbants [22][23][24][25] . Inhibition of the responsible intracellular signalling cascades abolishes these effects. How the very heterogeneous pathways triggered by these agents should convene to activate one and the same protease at the extracellular membrane surface has remained a fascinating enigma in cell biology. The fact that the cytoplasmic domain of ADAM17 is not required for sheddase activation 9,[22][23][24]26 renders the mystery yet more perplexing.
Independent of any stimuli, ADAM17 sheddase activity is enhanced in cells undergoing apoptosis 27 . A hallmark of these cells is the breakdown of phospholipid asymmetry with a remarkable increase of phosphatidylserine (PS) exposure in the external membrane leaflet 28 . Two major processes apparently contribute to this phenomenon. First, PS and phosphatidylethanolamine passively translocating to the external leaflet are normally returned to the inner leaflet through the action of ATP-dependent 'flippases'. Second, events such as Ca 2 þ elevation and apoptosis are accompanied by the activation of scramblases, proteins that nonspecifically and bidirectionally translocate phospholipids between the outer and inner leaflets of the plasma membrane 29 . The molecules that flip and scramble have proved very difficult to characterize. In recent times, two membrane proteins have been identified as scramblases: transmembrane protein 16F (TMEM16F) 30 and Xk-related protein 8 (Xkr8) 31 . Apoptosis is accompanied by caspase-dependent Xkr8 activation and irreversible flippase inactivation. In contrast, TMEM16F is responsible for reversible PS exposure that occurs in response to cytoplasmic Ca 2 þ elevation 28 . PS externalization has generally been considered to mainly serve as an 'eat-me' signal for apoptotic cells. However, it is readily conceivable that the anionic phospholipid may itself regulate cellular functions. Elliott et al. 32 showed that rapid PS translocation and release of the ADAM17 substrate L-selectin occurs in cells stimulated with P2 receptor agonists independent of any apoptotic events.
Transient PS exposure has been reported not only in the context of purinergic receptor stimulation, but also in conjunction with PKC activation 33 and cytosolic Ca 2 þ elevation 30 . It therefore occurred to us that translocation of PS to the outer membrane leaflet might represent the penultimate event common to all routes of ADAM17 activation.
Here we present evidence in support of this hypothesis. In diverse cell models, ADAM17 sheddase activity correlates with PS exposure. Substrate cleavage is inhibited by soluble phosphorylserine but not by phosphorylcholine. A cationic PS-binding motif is identified in the membrane proximal domain (MPD) of ADAM17. When this motif is exchanged, the resulting ADAM17 mutant retains its peptidolytic activity, yet physiological sheddase function is not detectable. We advance the hypothesis that the negatively charged PS head group interacts with cationic amino acid clusters present within the MPD of ADAM17. This interaction seems to be necessary for the protease to fulfil its sheddase function.

Results
ADAM17 stimuli provoke PS exposure. The first experiments were conducted to examine whether common ADAM17 stimuli would regularly induce PS exposure in human HaCaT keratinocytes. It was found that the potent ADAM17 activators melittin 25 and Ca 2 þ ionophore ionomycin (IO) both provoked massive PS externalization that could be followed on a real-time basis with the use of pSIVA (Annexin XII). The cell-impermeable, polarity-sensitive agent is added to the culture medium. It displays fluorescence only when bound to lipid bilayers and thus faithfully reports the presence of PS that is exposed in the outer membrane leaflet 34 (Fig. 1a,b). Melittin induced rapid PS exposure spreading over the entire cell surface that reached a maximum in 5-10 min and decreased thereafter ( Fig. 1a and Supplementary Movie 1). PS exposure provoked by IO was yet more rapid and remained sustained over the time of observation ( Fig. 1b and Supplementary Movie 2). PS externalization induced by the PKC activator phorbol 12-myristate 13-acetate (PMA) and the physiologic ADAM17 activator FGF7 was less pronounced and conventional staining with fluorescent-labelled Annexin V was employed because of higher sensitivity (Fig. 1c,d and Supplementary Fig. 1). Several waves of discrete, punctate staining became detectable over an extended period of 90 min. The oscillating nature of PS exposure was particularly conspicuous with FGF7 ( Fig. 1d and Supplementary Fig. 1). The differences in kinetics and localization of surface-exposed PS possibly reflect the different signalling pathways triggered by the respective stimuli. The impression was gained that patterns of PS exposure also exhibited a degree of cell-type specificity. Stimulation of endothelial cells with IO led to an Annexinstaining pattern that appeared distinct from that observed in HaCaT keratinocytes (Supplementary Movie 3).
Inhibitors of ADAM17-mediated shedding blunt PS exposure. Shedding of EGFR ligands is differentially induced by various stimuli. For example, FGF7 induces EGFR activation in keratinocytes via cleavage of heparin-binding EGF 24 , whereas TGF-a release remains hardly detectable ( Supplementary Fig. 2). Furthermore, prolonged times are usually required for levels of ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11523 shed substrates to become measurable in keratinocytes. Thus, TGF-a release can only serve as indication for ADAM17 activation and not provide a direct correlation. By contrast, the assessment of ERK1/2 activation provides an attractive option for a faster readout. Indeed, it was found that all stimuli induced ERK1/2 phosphorylation that was inhibited by the EGFRblocking antibody Cetuximab, by the metalloprotease inhibitor marimastat (MM) and, to a large extent, by the ADAM17 inhibitory antibody D1 (ref. 35) (Supplementary Fig. 3). PMA, IO and melittin all provoke the release of TGF-a and this substrate was additionally quantified in the respective experiments.
The question followed whether inhibition of sheddase activity would be mirrored by lack of PS exposure. Cells were exposed to the stimuli in the absence or presence of respective inhibitors. As shown in Fig. 2a,b, the P2R antagonist suramin, the calcium chelator EGTA, the PKC inhibitor staurosporine and the Src-kinase inhibitor PP2 all suppressed EGFR-dependent ERK1/2 activation. Release of TGF-a was reduced accordingly. The latter was inhibited by MM and the ADAM17/ADAM10 inhibitor GW, but not by the ADAM10 inhibitor GI. Conspicuously, blunted ADAM17 activation was in every case associated with the absence of PS externalization (Fig. 2c).
Induction of substrate release is blunted in PSA-3 cells. Mammalian cells contain two PS synthases (PSS). PSS1 synthesizes PS from phosphatidylcholine (PC), while PSS2 synthesizes PS from phosphatidylethanolamine. PSA-3 cells are a mutant CHO-K1 cell line defective in PSS1 activity. PS levels decrease when the cells are cultured in the absence of exogenous PS and the reduced PS content of the cells can be normalized by culturing in the presence of ethanolamine 36 . In the next experiments,  PSA-3 cells were transfected with alkaline phosphatase (AP)-tagged TGF-a to provide a common target for ADAM17. Cells were cultured either in medium supplemented with fetal calf serum (FCS) and ethanolamine or in medium supplemented only with dialysed FCS for 72 h. Cells were then transfected and after 24 h stimulated with IO and analysed by pSIVA live-cell imaging ( Fig. 3a,b). Inhibitor experiments confirmed that ADAM17 was responsible for the IO-induced TGF-a release in these cells ( Supplementary Fig. 4). As shown in Fig. 3a,b (and Supplementary Movies 4 and 5), cells cultured in the presence of ethanolamine responded with PS exposure and abundant release of TGF-a (Fig. 3c), while cells deprived of their PS source exposed little PS and shed little TGF-a.
PS exposure is required for shedding accompanying apoptosis. PS exposure is the hallmark of apoptosis and apoptosis is a trigger of ADAM-mediated substrate release 27,37,38 . However, human Raji lymphoma cells undergoing apoptosis expose minimal amounts of PS 39,40 due to low expression of the caspase-activated scramblase Xkr8 (ref. 31). A unique possibility thus arose to examine whether PS externalization and ADAM17mediated shedding were tightly coupled events. TNFR1 is predominantly cleaved by ADAM17 (ref. 10) and present on Jurkat and Raji cells. Exposure to agonistic anti-Fas antibody (Fas-Ab) resulted in PS externalization in Jurkat (Fig. 4a, left panel) but not in Raji cells (Fig. 4b, left panel), despite induction of apoptosis (Fig. 4c). Strikingly, TNFR1 shedding accompanying apoptosis occurred only in Jurkat cells (Fig. 4a, middle and right panel). The PS-positive cell population (P1) showed a nearly complete loss of the TNFR1, which was rescued by MM (Fig. 4a) and the ADAM17 inhibitory antibody D1 ( Supplementary Fig. 5). In contrast, the PS-negative cells (P2) were barely affected. These data indicate that activation of the apoptotic signalling cascade itself is not the trigger of ADAM17 activation and underline the importance of PS exposure for the shedding event.
Defective PS externalization results in reduced shedding. Blood cells from Scott syndrome patients present a counterpart to Raji cells. These cells have a missense mutation in TMEM16F, the scramblase that is responsible for rapid PS exposure in response to Ca 2 þ elevation in blood cells 30,41 . In contrast, PS exposure provoked via other pathways including Fas-triggered apoptosis remains undisturbed. The major consequence of the defect is therefore relatively restricted to platelet malfunction, which underlies the clinical manifestation of this rare bleeding disorder.
ADAM17 is a major sheddase of L-selectin on leukocytes 6-8 , which can conveniently be assessed by flow cytometry. Scott cells showed the same cell surface expression of ADAM17 as control cells, so that comparable basic conditions for assessment of shedding were given ( Supplementary Fig. 6a).
When calcium ionophore was applied to control B cells, flow cytometric analyses revealed rapid PS exposure that was parallelled by loss of L-selectin from the cell surface (Fig. 5a,b and Supplementary Fig. 6b). The temporal congruence between the two events, revealed here at the single-cell level, was impressive. L-selectin shedding was abrogated in the presence of MM and strongly reduced by the ADAM17 inhibitory antibody D1 ( Supplementary Fig. 6c). Intriguingly, Scott lymphocytes responded to the ionophore neither with PS exposure nor with shedding of the ADAM17 substrate (Fig. 5a,c). However, and in Soluble phosphorylserine suppresses sheddase function. Should surface-exposed PS be influencing ADAM17 function through interaction of the phospholipid head group with the protease, this would most probably occur at the extracellular membrane surface and might be competitively inhibitable by soluble phosphorylserine (OPS, O-phospho-L-serine). This question was addressed in different cell lines for different ADAM17 substrates. ERK1/2 phosphorylation in HaCaT keratinocytes in response to PMA stimulation was dose dependently inhibited by OPS, but not by the PC head group O-phosphocholine (OPC) (Fig. 6a). Release of TNFR1 can be induced in human umbilical vein endothelial cells (HUVECs) by diverse stimuli. Inhibitor experiments confirmed that ADAM17 is significantly involved in this process ( Supplementary Fig. 7). Stimulation of HUVECs in the presence of 10 mM OPS or OPC yielded results comparable to the previous data. OPS but not OPC suppressed shedding of the TNFR1 (Fig. 6b). A similar picture emerged for the release of APtagged TGF-a from transfected COS7 cells (Fig. 6c).
Fluorogenic peptide substrate assays were then performed in an attempt to clarify whether OPS might directly affect the enzyme. COS7 cells can be transfected to an extent that permits surface-expressed ADAM17 activity to be detected under cell culture conditions. Of note, stimulation of COS7 cells with PMA led to no increase in peptide turnover ( Supplementary Fig. 8). Thus, bona fide peptidolytic activity and substrate cleavage are distinguishable entities.
Cleavage of the fluorogenic peptide substrate was inhibited by the broad-spectrum metalloprotease inhibitor MM and the inhibitor GW. Conspicuously, however, no inhibitory effect of OPS on cleavage of the soluble substrate was observed. It followed that OPS was not directly interfering with the catalytic function of the protease (Fig. 6d).
The flexible MPD of ADAM17 binds to PS. The collective findings indicated that the release of transmembrane substrates relied on physical interaction of ADAM17 with surface-exposed PS. It appeared likely to be that the site for such an interaction would be located close to the membrane and the MPD of ADAM17 (Fig. 7a) presented itself as the most probable candidate. Recombinant MPD was prepared and employed in experiments using isothermal titration calorimetry (ITC). In contrast to the control (Supplementary Fig. 9), thermic reactions indicative of protein-lipid interaction were indeed detected when the MPD was titrated into buffer solutions containing PS but not PC liposomes (Fig. 7b).
Protein-disulfide isomerase (PDI) can induce a conformational change in the isolated MPD. The carboxy-terminal part of the non-PDI-treated MPD, also referred to as 'open' MPD, is largely unstructured and flexible, but becomes structured and inflexible due to isomerase-mediated rearrangement of disulfide bonding 42 . When the PDI-treated MPD was injected into a solution containing PS liposomes or PC liposomes in ITC experiments, no reaction could be observed (Fig. 7c), indicating that the PS-binding motif must be part of the PDI-affected flexible region ( Supplementary Fig. 10). NMR spectroscopy was used to further examine the binding of OPS to both MPD forms. 1 H-15 N-HSQC spectra revealed a specific interaction when OPS was added to the flexible open MPD (Fig. 7d, left panel) but not on addition to the PDI-treated MPD (Fig. 7d, right panel). Three-dimensional heteronuclear NMR experiments were performed, to identify the PS-interacting amino acid side chains. Resonance 1 could be identified as a lysine residue proceeded by a glycine, whereas resonance 2 was indicative of the NH resonances of an arginine side chain. The resonance of L624 was slightly shifted on the addition of OPS, indicating close proximity to the OPS-binding site. The MPD sequence contains one eligible motif, namely R625/K626/G627/K628 (Fig. 7d, left panel).

A cluster of basic amino acids mediates MPD-PS interaction.
Plasma coagulation factors and lactadherin bind with high affinity to surface-exposed PS either directly via discoidin-like domains or indirectly via Ca 2 þ bridges 43 , but none of the responsible motifs are present in the MPD. Interactions of cytosolic proteins such as Src kinase with ionic phospholipids are, in contrast, typically governed by electrostatic attraction of clusters of cationic amino acids to the membrane 44 and the possibility arose that such a situation might exist for ADAM17.
Non-mutated and MPD-3x were employed in surface acoustic wave experiments using biosensor chips coated with either PC or PS liposomes. Non-mutated MPD was found to bind to PS but not to PC liposomes (Fig. 8b). Despite the presence of several remaining positively charged residues, this PS-binding capacity was completely abolished in the open MPD-3x mutant (Fig. 8c).
To assess the relevance of this motif and neighbouring positively charged amino acids for sheddase function, ADAM17 mutants were constructed in which glycine replaced either K643/ R644 (A17-KR) or R625, K626 and K628 (A17-3x) or both motifs (ADAM17-5x). Sheddase function was assessed in murine embryonic fibroblasts that were co-transfected with AP-tagged TGF-a. Although ADAM17 is the major sheddase of TGF-a under normal conditions, ADAM10 releases this substrate in the absence of ADAM17 (ref. 7). Therefore, ADAM10/ADAM17 double-deficient cells were employed for retransfection experiments. These cells were derived from the corresponding knockout embryos that die around embryonic day 8.5 and show severe developmental defects quite similar to ADAM10 single deficient mice 45 (Supplementary Fig. 12).
As shown in Fig. 8d, shedding of TGF-a was induced by PMA, IO and melittin in cells transfected with wild-type (WT)-ADAM17 but not with the enzymatically inactive ADAM17-E/A mutant. Transfection with the ADAM17-KR mutant also completely restored shedding activity. However, shedding was almost completely absent in cells re-transfected with ADAM17-3x. This effect was not further influenced by additional deletion of the KR-motif (ADAM17-5x).
Exit from the endoplasmic reticulum and maturation of ADAM17 are dependent on interaction with proteolytically inactive members of the Rhomboid family of intramembrane ARTICLE serine proteinases, iRhom1 and iRhom2 (refs 18,19), and introduction of mutations might affect these processes. However, western blot analyses disclosed that lack of sheddase activity was not due to differences in protein expression or maturation of the ADAM17-3x mutant compared with WT-ADAM17 (Fig. 8e). Flow cytometric analysis revealed comparable cell surface expression of both constructs ( Supplementary Fig. 13). These findings were corroborated by fluorogenic peptide substrate assays in COS7 cells demonstrating comparable cell surface-associated peptidolytic activity (Fig. 8f). In particular, these data showed that shedding-incompetent ADAM17-3x retained its enzymatic activity.

Discussion
Remarkably diverse agents rapidly trigger ADAM17-mediated shedding of its substrates from the cell surface. No model exists to explain this perplexing phenomenon. Here we depart from the conventional view that enhanced sheddase function necessarily derives from a genuine increase in enzymatic activity. Instead, we propose that a final event common to and essential for all pathways is surface exposure of PS, which relays intracellular signalling events to the ectodomain of ADAM17. Cationic amino acid residues in the MPD then interact with the negatively charged PS-head group, bringing the protease into position for substrate processing (Fig. 9).
At the cytosolic face, PS subserves essential docking functions for enzymes such as Src kinase, MARCKs and PKC, which interact via clusters of cationic amino acids with the negatively charged phospholipid head group 43,46 . In a related manner, the accessory HIV-1 protein NEF associates with membranes in a biphasic process. A fast protein-lipid interaction driven by electrostatic attraction of a basic amino acid cluster (R_R_RR) to PS is followed by formation of an amphipathic helix that subsequently also becomes membrane bound 47 .
In the present study, the cationic cluster RK_K located in the MPD of ADAM17 was found to mediate its interaction with and binding to PS-containing membranes. Intriguingly, the ADAM17 stalk region CANDIS (conserved ADAM17 dynamic interaction sequence) 48 , which follows the MPD onward to the bilayer has recently been found to represent a membrane-interacting amphipathic helix that is also required for sheddase function 49 . The possible similarities in the mode of membrane interaction between HIV-1 NEF and ADAM17 are striking.
A concept emerges in which ADAM17 is envisaged to basically assume two orientations at the cell surface. In the quiescent state, MPD and CANDIS region are freely oriented in the pericellular space and the protease remains poised at a distance from its targets. Binding of the MPD and CANDIS to the lipid bilayer re-orients the protein, drawing its catalytic domain down to its substrate. Simple spatial considerations then explain why substrate cleavage by ADAM17 regularly occurs close to the membrane surface and on the same cell 12 . To date, ADAM17mediated shedding has indeed never been observed in trans.
The proposed model of ADAM17-PS interaction satisfactorily resolves another puzzle relating to the fact that systematic overexpression of the protease does not lead to enhanced shedding activity in vivo 50 . Should our idea be correct, shedding will be dependent not simply on levels of ADAM17 expression but also on the dynamics of PS exposure. ADAM17 overexpression alone can thus not be simply equated with increased shedding function. Our study does not address events that regulate maturation and trafficking of ADAM17, but focuses solely on the possible significance of PS exposure for activity of the sheddase after its positioning at the membrane surface. There, translocation of the anionic phospholipid from the inner to outer leaflet is proposed to conclude events leading to execution of shedding function. PS exposure may occur at different sites and with differing kinetics dependent on the cell type and the stimuli, and surface-exposed PS may then act in concert with intracellular events to control the action of the protease or the accessibility of the substrate. In this context, Dang et al. 51 have presented evidence that PKC-dependent phosphorylation guides lateral movement of the protease to its substrates in the membrane. Furthermore, Maretzky et al. 52 have found that interaction of iRhom1/2 with ADAM17 is sometimes required for triggering sheddase function. Perhaps such processes are responsible for substrate selection and are concluded by PS translocation at the respective sites.
Transient PS exposure in the absence of apoptosis has been described particularly for activated immune cells. Chemotactic activation of neutrophils leads to cell polarization and to PS exposure at the uropod 53 , where ADAM17 and L-selectin have also been shown to co-localize in stimulated cells 54 . It is tempting to speculate that the stimulus-dependent variation in PS exposure we have observed (Fig. 1) bears significance for the orchestration of sheddase function in general. This open question notwithstanding the idea that substrate selection rests on separate but coordinated events at the outside and inside of the cells obviates the need to postulate the existence of a single regulatory element. With regard to a possible role of PS in also attracting ADAM17 substrates, their large number and diversity render it a priori unlikely that all should have membraneproximal PS-interacting motifs. Indeed, candidate clusters of cationic amino acid residues are not present in TGF-a.
Transiently occurring PS translocation may subserve a number of previously unsuspected functions in living cells. Pathways leading to PS exposure are intricate and often have not been clearly delineated. Several families of scramblases have been identified that cause breakdown of membrane phospholipid asymmetry in nucleated cells. Fortuitously, cells carrying a mutated TMEM16F scramblase from a patient with Scott syndrome were available and provided a valuable tool in the present study. TMEM16F scramblase is activated by intracellular Ca 2 þ . This response is selectively defective in Scott syndrome, while mechanisms for PS externalization provoked via other pathways remain intact. Strikingly, IO simultaneously failed to provoke PS exposure and L-selectin shedding in Scott lymphocytes. In contrast, both events occurred on induction of apoptosis.
Use of Raji cells demonstrated that the apoptotic process itself cannot activate sheddase activity when there is no PS exposure.
Experiments with PSA-3 cells that are compromised in their PS synthesizing capacity accorded nicely with these findings. The PS content of these cells could be normalized simply by culture in the presence of ethanolamine. It thus became possible to quite specifically study the influence of PS on sheddase function. Satisfyingly, cells with reduced PS exposed little PS and this was associated with blunted ADAM17 sheddase activity that returned to normal on PS repletion. The fact that the close correlation between PS exposure and ADAM17 activation observed with the different stimuli and occurring in different cells should represent mere coincidence appeared highly unlikely. Evidence for an interplay between PS and the protease was subsequently obtained in a series of experiments. First, it was found that addition of soluble OPS but not of OPC to cells countered the shedding process. This occurred in the absence of any direct effect of OPS on peptidolytic activity, as was shown by unaltered cleavage of a soluble peptide substrate. It followed that soluble OPS most probably competed for interaction of the protease with the membrane-exposed phospholipid. The propensity of MPD to bind PS was then decovered and the responsible motif identified. On removal of the PS-binding motif, the mutated protease continued to be expressed in its mature form and exposure at the cell surface was ascertained by flow cytometry and through assessment of cellassociated peptidolytic activity using the fluorogenic peptide substrate. The shedding experiments were conducted with ADAM10/ADAM17 double-deficient cells to guarantee that any measured sheddase activity derived solely from the transfected enzyme. Despite its ability to cleave the peptide substrate, mutant ADAM17 lacking the PS-binding motif was unable to shed TGF-a.
Further work with a panel of mutants is needed to obtain deeper insight into the events that govern the interaction of ADAM17 with the membrane and with its substrates. It is noteworthy that the amount of PS residing in the extracellular membrane leaflet can vary considerably depending on cell population, differentiation and activation status 55,56 . It will naturally be of interest to investigate whether any of the present findings extend to ADAM10, the second prominent member of the sheddase family. Regulatory mechanisms that govern the function of this closest homologue to ADAM17 have remained equally elusive in the past.
A critical question is whether the present concept would stand in conflict with previous studies that have presented evidence for increased enzymatic activity of ADAM17 in PMA-treated cells. Perhaps, surprisingly, closer inspection of the published data reveals that they actually accord excellently with our model. PMA-mediated increase in cellular cleavage of peptide substrate reported by Doedens et al. 26 rose from basal constitutive levels of 2-3% to only 5-6% in stimulated cells within 60 min. The reason for this low increase in net peptide cleavage was revealed in the recent study of Wu et al. 57 . In elegant single-cell measurements, the authors demonstrated that stimulation of HepG2 cells with 100-200 nM PMA resulted in elevated cell-associated protease activity in o5% of the cells over a time span of 36 min. These numbers rose at higher concentrations of the stimulus but still did not exceed 25% at the overtly cytotoxic dose of 1 mM PMA. Given that substantially lower concentrations provoke rapid and maximal substrate shedding of ADAM17 substrates in living cells within minutes 4,8 , it follows that in the vast majority of cells the shedding process is not associated with or dependent on any genuine increase in protease activity. This conclusion is supported by our present finding that ADAM17-transfected COS7 cells display no alteration in surface-associated protease activity in response to prolonged stimulation with PMA.
A final conceptual step is now taken. We propose that bona fide ADAM17 enzyme activity and sheddase function need to be regarded as separate entities. A mechanistic explanation is provided that assigns a cardinal role of membrane asymmetry in controlling the function of the extracellularly oriented membrane protein. PS exposure can be expected to occur at differing sites and vary in duration and intensity depending on the trigger, cell type and local scramblase machinery. A unique stage could thus be set for the central sheddase of the cell to assume its multifaceted roles in biology.
Generation of MPD-3x. Cloning, bacterial expression and purification of the MPD-3x was performed by Trenzmye GmbH. The complementary DNA of human MPD (amino acid residues F581-E642 with the following changes: R625G/K626G/ K628G) was cloned into pTZ_E03 expression vector and transformed into Escherichia coli strain Rosetta 2 (DE3). The soluble expressed domain was purified in PBS by its amino-terminal His tag using nickel affinity chromatography.
Transfection and AP-substrate shedding assay. ADAM10/ADAM7 doubledeficent MEFs were transfected using Turbofect Transfection Reagent (Thermo Fisher Scientific) according to the manufacturer's instructions. Twenty-four hours after transfection, cells were washed with DMEM, which was replaced 1 h later by fresh DMEM and treated or not treated with PMA (200 ng ml À 1 , 2 h) or IO (1 mM, 30 min), or melittin (4 mM, 30 min). Thereafter, supernatants and cell lysates were collected and measured for AP activity at A405 employing the AP substrate 4-nitrophenyl phosphate (Sigma-Aldrich). Shown is the relative AP activity in the supernatant compared with the total AP activity of supernatant plus cell lysates normalized to the unstimulated A17-WT. For shedding assays, the cells were lysed in buffer containing 2.5% Triton-X, 10 mM 1.10-phenanthroline and 1 mM EDTA in water. The metalloprotease inhibitor phenanthroline inhibited autocatalysis of ADAM17 (ref. 65). The expression level of ADAM17 constructs was analysed in parallel by western blot analyses 24 h after transfection. For AP assay experiments in PSA-3 cells, cells were transfected after 72 h PS deprivation for 6 h in Ham's F12 without FCS. Thereafter, medium was changed again to the starvation medium. All AP assays and imaging experiments were performed after 96 h of PS deprivation. As starvation reduced the transfection efficiency, the stimulation effect was normalized to the non-treated cells, respectively.
Confocal live-cell imaging. Time-lapse imaging was performed with an inverted confocal microscope (Fluoview FV1000, Olympus) equipped with an environmental chamber using a UPLSAPO Â 60 oil-immersion objective (numerical aperture: 1.35) and 2 Â zoom. Before starting the experiments the chamber was equilibrated to 37°C in a humidified atmosphere. The cells were seeded in glass-bottom imaging dishes (PAA) and grown to semi-confluence. Cells were allowed to rest for at least 1 h in the microscope stage. Five minutes before acquisition started, PSIVA-IANBD was added to the cells in a final concentration of 6.25 mg ml À 1 . PSIVA-IANBD was excited at 488 nm and emission was recorded at 520 nm every 30 or 60 s. Laser and detector settings were the same for each single experiment.
Immunocytochemistry. HaCaTs were seeded on glass coverslips and grown to semi-confluence. After indicated stimulation/inhibition periods, coverslips were immediately incubated with a 1:20 solution of Annexin V-FITC in Annexin binding buffer (ABB) for 5 min in the dark at room temperature, washed twice with ABB (10 mM HEPES, 140 mM NaCl and 2.5 mM CaCl 2 pH 7.4) and fixed for 15 min with 3% paraformaldehyde. After fixation, coverslips were washed six times with PBS, once with distilled water and mounted in embedding medium. Image acquisition was performed with an inverted confocal microscope (Fluoview FV1000, Olympus) using a UPLSAPO Â 60 oil-immersion objective (numerical aperture: 1.35) and 2 Â zoom. Annexin V-FITC was excited at 488 nm and emission was recorded at 520 nm. Images were acquired with the same laser and detection settings for each experimental setup, but vary between Figs 1 and 2 because of different Annexin V-FITC lots.
Image analysis and image statistics. Image analysis was performed with ImageJ 1.47 m. Fluorescence area above background fluorescence was determined and correlated to the cell growth area. Immunocytochemistry for each group, at least three independent coverslips, each with six to eight images of different areas, were analysed. The mean fluorescence area of each coverslip was taken for statistical analysis. Groups were tested by one-way analysis of variance and Bonferroni multiple comparison post hoc test.
Flow cytometric analysis. For analysing ADAM17 cell surface expression, cells were washed with PBS and stained with anti-ADAM17 (K133, 10 mg ml À 1 ) or rabbit IgG-control antibodies (1:100) in 100 ml PBS with 3% BSA for 1 h on ice. After washing, cells were incubated with anti-rabbit Alexa Fluor 488 (1:100) for 30 min on ice. Cells were washed and resuspended in PBS.
Kinetic analysis of PS exposure and L-selectin surface expression was performed by staining cells with APC mouse anti-human L-selectin (BD Bioscience) or mouse IgG1 1:100 in 500 ml RPMI with 1% FCS for 30 min on ice. Thereafter, cells were washed twice and resuspended in 190 ml RPMI. Before commencement of the experiments, cells were warmed to 37°C for 5 min and the media was supplemented with 2 mM CaCl 2 . pSIVA-IANBD (10 ml) was then added and incubated for 3 min. The samples (20 ml each) were diluted in 500 ml ice-cold ABB and analysed by flow cytometry before and after addition of IO (2 mM) at the indicated time points for a period of 6 min.
Staining after Fas-Ab treatment (6 h Scott cells and WT B cells; 2 h Jurkats and Rajis) was performed as follows: cells were stained with Annexin V-FITC (1:100) in ABB for 10 min in the dark at room temperature, washed with ABB and fixed for 15 min with 3% paraformaldehyde on ice. Cells were subsequently washed in PBS and stained for 1 h with the corresponding antibody or control at 4°C in PBS with 1% BSA (anti-TNFR-APC or isotype control (Jurkats and Rajis), anti-L-selectin or isotype control (Scott cells and WT B cells). After staining, cells were washed and analysed by flow cytometry with FACSVerse (BD Bioscience). Data were analysed using FlowJo 8.7.3.
ADAM peptide substrate assay. COS7 cells were transfected in 24-well plates with WT-ADAM17, ADAM17-E/A or ADAM17-3x constructs or with the empty control plasmid (pcDNA3.1). Twenty-four hours after transfection, medium was changed to DMEM with 10 mM fluorogenic ADAM substrate (Enzo Life Sciences) and the cells were incubated for the indicated time points. Fluorescence in the supernatant was determined at 485 nm excitation and 530 nm emission.
NMR spectroscopy and LC-MS. The 15 N-and 13 C-isotope-labelled open and closed conformation of the MPD of ADAM17 was prepared as described previously 42 . NMR measurements were performed on a Bruker Avance 600-MHz spectrometer equipped with a z-gradient triple-resonance cryoprobe. The proteins were dissolved in 10 mM phosphate buffer pH 7.4, 0.01% NaN 3 and 7% D 2 O. OPS was added in a tenfold molar excess. Sequence-specific backbone resonance assignments of the MPD conformers were established using the following spectra: three-dimensional HNCA, HNCO, (H)C(CO)NH, H(CCO)NH, HN(CA)CO, HN(CO)CA, HN(CO)CACB, HNCACB, 15 N-edited TOCSY and 13 C-edited HCCH-TOCSY. The spectra were acquired at 300 K and referenced to the water resonance at 4.79 p.p.m. All spectra were processed with NMRPipe 66 and analysed with NMRViewJ 67 . LC-MS experiments for the determination of MPD-3x disulfide bonds were performed as described before 42 .
Enzyme-linked immunosorbent assay. TGF-a and TNFR1 enzyme-linked immunosorbent assay (both R&D) were performed according to the manufacturer's instructions. Cells were grown until confluence in 12-well plates. HaCaTs were stimulated with PMA (300 ng ml À 1 ), IO (1 mM) or melittin (1 mM) in the presence or absence of MM (10 mM), GI (2 mM) or GW (2 mM) in duplicates for 30 min. HUVECs were stimulated with PMA (200 ng ml À 1 ), IO (1 mM) or melittin (1 mM) in the presence or absence of OPS (10 mM) or OPC (10 mM) in duplicates for 1 h. Supernatants were analysed for soluble shedding products in duplicates.
Western blot analysis. Cells were washed once with PBS and lysed in lysis buffer (5 mM Tris-HCl pH 7.5, 1 mM EGTA, 250 mM saccharose and 1% Triton X-100) supplemented with cOmplete inhibitor cocktail (Roche Applied Science) and 10 mM 1,10-phenanthroline monohydrate to prevent ADAM17 autocleavage 65 . Equal amounts of protein were loaded on 10% SDS-PAGE gels. The samples were electrotransferred onto polyvinylidene difluoride membranes (Hybond-P; Amersham) and blocked overnight with 5% skim milk in Tris-buffered saline. After incubation with the indicated antibody in blocking buffer, the membranes were washed three times in TBST (Tris-buffered saline containing 0.1% Tween-20). Primary antibodies were detected using affinity-purified peroxidase (POD)-conjugated secondary antibodies (1:10,000) for 1 h at room temperature. Detection was carried out using the ECL detection system (Amersham). Signals were recorded by a luminescent image analyser (Fusion FX7 imaging system; PEQLAB Biotechnologie). To analyse the expression of pERK1/2 in comparison with total ERK1/2, western blottings were incubated in stripping reagent (100 mM 2-mercaptoethanol, 2% (w/v) SDS, 62.5 mM Tris-HCl pH 6.7) at 55°C for 30 min and re-probed with the exception of the control blot in Fig. 6a. Here, protein samples were loaded on separate western blottings and probed as indicated. Uncropped scans of the blottings are shown in Supplementary Fig. 14.
Preparation of lipid aggregates/liposomes. Aggregates/liposomes were prepared as 1 mM aqueous dispersions of the phospholipids in buffer (5 mM HEPES and 100 mM KCl pH 7.0) as follows. The lipids were dissolved in chloroform to a concentration of 1 mg ml À 1 . The solvent was evaporated under a stream of nitrogen, buffer was added and sonicated with a Branson sonicator (Branson Ultrasonics Corporation) for 1 min (1 ml solution). Subsequently, the preparation was cooled for 30 min at 4°C and two times heated for 30 min at 60°C and cooled to 4°C. Preparations were stored at 4°C overnight before measurements.
Acoustic wave biochip analyses. Gold-coated chips (S-sens K5 Biosensor Quartz Chips, SAW Instruments GmbH) were functionalized as described 68 . A functionalized chip was incubated overnight in ethanol and placed into the reader unit of the S-sens K5 readout system (SAW Instruments GmbH). Biomolecular interaction processes on the surface of the sensor chip can result in changes of phase and amplitude of the surface acoustic wave. Changes of these parameters correlate with mass loading. Measurements were performed at a continuous flow of measuring buffer (5 mM HEPES pH 7.4) of 20 ml min À 1 at 22°C. By injection of 100 ml of poly-L-lysine (50 mg ml À 1 , Fluka), a positively charged layer was formed on top of the negatively charged CM-dextran matrix. One hundred microlitres of phospholipid liposomes (300 mg ml À 1 ) (PS liposomes or PC (9) Isothermal titration calorimetry. The binding of the MPD with PC or PS liposomes was analysed by microcalorimetric measurements on an ITC200 (GE Healthcare) 69  Generation of ADAM10/ADAM17 double-deficient mice and cells. ADAM10deficent mice have been described 45 . To generate ADAM10/ADAM17 doubledeficient mice, heterozygous ADAM10 À / À mice were mated with heterozygous ADAM17-deficient mice, which were generated by gene trap as follows. Embryonic stem cell line RST495 with an intronic gene trap insertion between exon 6 and exon 7 of the Adam17 gene were obtained from BayGenomics. The embryonic stem cells were injected into the blastocoel cavity of C57Bl/6J-derived blastocysts and implanted into pseudopregnant females. Chimeric mice generated from the injections transmitted the targeted allele to their progeny. Genotyping was performed by intron spanning ADAM17 PCR 5 0 -tcatcgattttataaatacatgggccg-3 0 (fwd), 5 0 -accctgcattatcccacgacgtgttcc-3 0 (rev) and neomycin PCR 5 0 -gttgtcactgaa gcgggaagggactggctg-3 0 (fwd), 5 0 -gcgaacagttcggctggcgcgagcccctga-3 0 (rev) or b-galactosidase PCR (not shown). Heterozygous ADAM17 þ / À mice were normal and fertile. They were intercrossed to generate homozygous gene-targeted mice. ADAM17-deficient mice showed a comparable phenotype as has been described for mice with deleted ADAM17 Zn 2 þ -binding domain 4 including eye, hair and skin defects ( Supplementary Fig. 12), and embryonic lethality. MEFs were derived from dissociated E16 embryos and maintained in DMEM supplemented with FCS and antibiotics. Total RNA from MEFs was reverse transcribed and subjected to PCR amplification for Adam17 using the primers 5 0 -gtacgtcgatgcagagcaaa-3 0 (fwd) and 5 0 -aaaccagaacagacccaacg-3 0 (rev). Actin was used as control (5 0 -gacctctatgcca acacagt-3 0 (fwd) and 5 0 -agtactttgcgctcaggagga-3 0 (rev)). Protein deficiency was confirmed by western blotting using anti-ADAM17 ectodomain antibodies (Merck Millipore). Heterozygous ADAM17 þ / À mice were intercrossed with heterozygous ADAM10 þ / À mice. ADAM10/17 À / À embryos were evaluated at different stages of development. At day 9 of embryogenesis, embryos presented in a Mendelian frequency (not shown), as evaluated by PCR. No viable ADAM10/17 À / À embryos were obtained beyond E9. MEFs were derived from dissociated E9 embryos and maintained in DMEM supplemented with FCS and antibiotics. Protein deficiency was confirmed by RT-PCR (not shown) and western blotting. The functional analysis of these cells is described elsewhere 7,22 .
Statistical analysis. All values for the ectodomain shedding assays are expressed as means ± s.e.m. The s.e. values indicate the variation between mean values obtained from at least three independent experiments. Statistics were generated using one-way analysis of variance and Bonferroni multiple comparison post hoc test. P-valueso0.05 were considered statistically significant (either indicated with * or #).