Inside-out Regulation of Ectodomain Cleavage of Cluster-of-Differentiation-44 (CD44) and of Neuregulin-1 Requires Substrate Dimerization*♦

Background: Intracellular domain (ICD) modifications regulate extracellular ectodomain cleavage by metalloproteases. How this inside-out signal is relayed is unknown. Results: Cleavage requires substrate homodimerization; ICD modifications likely induce a relative positional change of the dimerization partners, allowing cleavage. Conclusion: Substrate dimerization might be a general requirement for cleavage. Significance: Our results fill an important gap in understanding growth factor release by ectodomain cleavage. Ectodomain shedding of transmembrane precursor proteins generates numerous life-essential molecules, such as epidermal growth factor receptor ligands. This cleavage not only releases the regulatory growth factor, but it is also the required first step for the subsequent processing by γ-secretase and the release of gene regulatory intracellular fragments. Signaling within the cell modifies the cytoplasmic tails of substrates, a step important in starting the specific and regulated cleavage of a large number of studied substrates. Ectodomain cleavage occurs, however, on the outside of the plasma membrane and is carried out by membrane-bound metalloproteases. How the intracellular domain modification communicates with the ectodomain of the substrate to allow for cleavage to occur is unknown. Here, we show that homodimerization of a cluster-of-differentiation-44 or of pro-neuregulin-1 monomers represents an essential pre-condition for their regulated ectodomain cleavage. Both substrates are associated with their respective metalloproteases under both basal or cleavage-stimulated conditions. These interactions only turn productive by specific intracellular signal-induced intracellular domain modifications of the substrates, which in turn regulate metalloprotease access to the substrates' ectodomain and cleavage. We propose that substrate intracellular domain modification induces a relative rotation or other positional change of the dimerization partners that allow metalloprotease cleavage in the extracellular space. Our findings fill an important gap in understanding substrate-specific inside-out signal transfer along cleaved transmembrane proteins and suggest that substrate dimerization (homo- or possibly heterodimerization) might represent a general principle in ectodomain shedding.

intracellular calcium influx and the release of diacylglycerol, which activates protein kinase C (13)(14)(15). Thus, intracellular signaling pathways regulate a process (known as substrate protease accessibility) that occurs outside of the plasma membrane. Because many substrates are single-pass transmembrane proteins, this begs the following question. How can a singlepass transmembrane protein transmit and execute a structural change of the ectodomain via intracellular signal-induced ICD modification?
Signal transfer along single-pass transmembrane proteins is not well understood. In the case of receptor tyrosine kinases, signal transfer occurs from outside to inside mediated by receptor dimers or trimers that are formed in the endoplasmic reticulum prior to transport to the plasma membrane (16 -19). Extracellular ligand binding causes phosphorylation at tyrosine residues in the cytoplasmic tails of the receptors. Crystal structures of the ligand-bound and -unbound molecules show the start and end conformations of both the extracellular and intracellular domains (20,21). But how is this ligand-induced signal transfer through the membrane accomplished? An interesting early proposal assumed that the ligand triggers rotation of one receptor monomer relative to the other (16,22). Very recent NMR and molecular simulation data of the EGF receptor propose a ligand-induced change in the orientation of the transmembrane helices relative to each other, a process that affects the positioning of juxtamembrane sections of the dimer at the inner side of the plasma membrane, allowing phosphorylation and activation of the receptor (23,24).
Could such outside-in signaling be a model for signaling in the opposite direction, as needed here for the inside-out signal transfer along ADAM substrates, and similar to that previously described for integrin heterodimers (25)? One prediction of this model would be that two single-pass transmembrane molecules (including the substrate) needed to change their position relative to each other, allowing cleavage.
We have tested this prediction by exploring ectodomain cleavage of the ADAM10 substrate CD44 (an adhesion molecule and stem cell marker) and the ADAM17 substrate NRG1 (proform of the epidermal growth factor (EGF) receptor ligand neuregulin). When bound to hyaluronan, CD44 triggers a proliferation-inhibitory pathway (26 -28). However, depending on the context (29 -32), it can also promote tumor growth and metastasis (33)(34)(35)(36)(37)(38). NRG1 cleavage is essential for myelination in the nervous system but also for normal development and function of the mammary gland and the heart (39).
We report here that ADAM enzymes are pre-associated with their respective substrates at the plasma membrane prior to stimulation of cleavage. We further show that dimerization or oligomerization of substrate monomers is a pre-condition for induced cleavage.
siRNA Sequences-siRNA SMARTpool, a mixture of four siRNAs targeting ADAM10 and ADAM17, and the "Nontarget Plus TM Pool" (control) were from Thermo Scientific Dharmacon (Rockford, IL). The oligonucleotide sequences are listed in Table 1.
Cell Lines and Transfections-NIH3T3 cells (immortalized Swiss mouse embryonic fibroblasts) were from the European (C-terminal c-Myc) wild type, the noncleavable mutant CD44-KR-Mt, or with an empty vector (V). The cells were grown at low cell density, and CD44 cleavage was stimulated by treatment with 100 ng/ml TPA for 30 min. DAPT (5 M) was added to the cells to prevent degradation of the CD44⌬E cleavage product by ␥-secretase. Control cells were treated with DMSO alone (solvent for TPA and DAPT). Subsequently, CD44 full-length (CD44fl) and the membrane-bound C-terminal cleavage product CD44⌬E were detected by c-Myc antibody. CD44fl forms a double band, most likely because of differential glycosylation. Induced CD44 cleavage occurs only in the Adam17 null MEFs but not in the cells with disruption of the Adam10 gene. B, RPM-MC cells transfected with doubly tagged CD44 (N-terminal FLAG and C-terminal c-Myc) were grown at low cell density. A, TPA treatment. Expression of ADAM10 (A10) or ADAM17 (A17) was downregulated by siRNA (to 3.8 and 1.6% as calculated from the blot by ImageJ). Nontargeting siRNA (C) was used as a control. The released ectodomain was precipitated from culture supernatant by TCA prior to SDS-PAGE. Cleaved ectodomain (solCD44E), CD44fl, and CD44⌬E were detected by FLAG and c-Myc antibodies, respectively. The efficiency of siRNA knockdowns was monitored by detection of ADAM10 and ADAM17 proteins as indicated (seen are the pro-(P) and mature (M) forms). Only ADAM10 knockdown significantly reduced basal and induced release of solCD44E and CD44⌬E. BЈ, histogram shows mean values of relative level of solCD44E Ϯ S.D. from three independent experiments. ns, p ϭ 0.031081; ****, p Ͻ 0.0001; ns, not significant. C, RPM-MC cells transfected with doubly tagged CD44 (N-terminal FLAG, C-terminal c-Myc) were grown at low cell density. For inhibition of translation the cells were pre-incubated with 50 g/ml of cycloheximide (CHX). CD44 cleavage was stimulated by treatment with 100 ng/ml TPA for 4 h. WB, Western blot.  hydroxamate-based metalloprotease inhibitors 10 M batimastat (BB94) or 15 M GM-6001 at 15-30 min prior to TPA stimulation. In addition, ␥-secretase activity was blocked by 5 M DAPT (Sigma).
Precipitation of Proteins by TCA-Deoxycholate-For detection of soluble CD44 ectodomain or neuregulin, the cells were cultured in serum-free medium. Cell culture supernatants were pre-cleared by centrifugation at 10,000 rpm for 10 min to pellet cell debris. The pre-cleared supernatant was mixed with 0.01 volume of 2% deoxycholate, vortexed, and kept on ice for 30 min. Then 0.1 volume of 100% TCA was added, and the samples were kept at 4°C overnight. The precipitate was recovered by centrifugation at 15,000 ϫ g for 15 min, rinsed twice with acetone, and re-dissolved in RIPA buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS).
Protein Cross-linking and Co-immunoprecipitation (co-IP)-For co-immunoprecipitation experiments, transfected RPM-MC and NIH3T3 cells or stably infected HEK293T cells were grown in 15-cm plates. Cells were washed once in ice-cold 1ϫ PBS. For protein cross-linking before co-IP, the cells were incubated with varying concentrations of paraformaldehyde (0.85-1.8% PFA were tested). The results shown here were obtained with 1.25% PFA (10 min of incubation at room temperature under mild agitation). To quench the reaction, the cells were washed twice with 1.25 M glycine in 1ϫ PBS. The cell lysates (input) were prepared using ice-cold IP buffer containing 1ϫ complete protease inhibitor mixture (Roche Applied Science). The following IP buffers were used for cell lysis: for co-IP of CD44 with ADAM10 without cross-linking, 2% CHAPSO in 150 mM citrate, or after cross-linking with PFA, RIPA buffer (50 mM Tris-HCl, pH 8.0, 150 mM sodium chloride, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA); for co-IP of NRG with ADAM17 and for co-IP of NRG dimers, several buffers were tested, CHAPSO buffer as above and (as shown in Figs. 2E and 4C) 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 0.5% Triton X-100, 0.5% Nonidet P-40, 2 mM EDTA. For subsequent co-IP the 2% CHAPSO buffer was diluted to 1% CHAPSO in 150 mM citrate. Cell lysates were incubated on ice for 30 -60 min. After shearing DNA, the lysates were precleared by centrifugation at 15,000 ϫ g for 30 min.
Tagged CD44 or NRG1 was immunoprecipitated using 20 l of anti-FLAG 2 M2 affinity gel beads (Sigma). Alternatively, CD44, endogenous ADAM10, and GFP-tagged NRG1 were immunoprecipitated using 30 l of protein G plus gel beads (Santa Cruz Biotechnology), pre-conjugated with 2 g of CD44 5G8, 2 g of ADAM10 antibody, or GFP antibody, respectively. Immunoprecipitation was performed with slow rotation at 4°C overnight. Isotype-matched control antibodies were used as negative controls to estimate the nonspecific binding of target proteins. Immunocomplexes were recovered by centrifugation and washed four times with cold IP buffer and eluted with 2ϫ Laemmli sample buffer.
BiFC Assay-Cells were grown on coverslips placed at the bottom of 6-or 12-well plates. 16 h after transfection, cells were treated with either control (DMSO) conditions or under inhibited cleavage conditions for 30 min in the presence or absence of a stimulus (TPA) as described above. After treatments, cells were washed in 1ϫ PBS, and nonpermeabilized cells were incu-bated with 10 g/ml Texas Red wheat germ agglutinin in 1ϫ PBS to label the plasma membrane. Cells were washed three times with 1ϫ PBS and fixed with 4% PFA in cytoskeleton buffer (10 mM Pipes, pH 6.8, 100 mM NaCl, 300 mM sucrose, 3 mM MgCl 2 , 1 mM EGTA) at room temperature for 20 min. The cells were washed two times with 1ϫ PBS and permeabilized using 0.1% Triton X-100 in 1ϫ PBS (7 min at room temperature). After another three washing steps, the coverslips were mounted in Vectashield mounting media containing DAPI (Vector Laboratories, Inc. Burlingame, CA).
Microscopy and Immunofluorescence Analysis-Fluorescent photographs were generated with a Zeiss Axio Imager ApoTome microscope (Carl Zeiss AG, Jena, Germany) or a Nikon D-Eclipse C1 confocal laser-scanning microscope (Tokyo, Japan) using identical acquisition parameters for each experiment using the confocal acquisition software Nikon EZ C1 version 3.90 and the AxioVision imager analysis software.
Flow Cytometry-For quantitation of BiFC assays, a plasmid encoding mCherry was co-transfected together with BiFC constructs (as described above). BiFC plasmids were added in excess to mCherry encoding plasmid (DNA ratio 3:1), thus the mCherry positive cells should also express BiFC constructs. mCherry positive living RPM-MC cells were sorted by flow cytometry. The cells of one type were divided into two pools and treated either with DMSO (control) or 100 ng/ml TPA for 30 min in the presence of 10 M metalloprotease inhibitor, batimastat. The cells Statistical Analysis-For statistical analysis, the intensity of bands from immunoblots was quantified using ImageJ and Image-Lab software (Bio-Rad). All values on histograms are reported as mean Ϯ S.D. Statistical significance was determined by Student's tests, with a p value Ͻ 0.05 as the significance level.

Results
To analyze inside-out signaling for ectodomain cleavage, we focused on two ADAM substrates, CD44 (ADAM10) and NRG1 (ADAM17). N-and C-terminally double-tagged CD44 and NRG1 were transfected into RPM-MC cells, MEFs, NIH3T3, or HEK293T cells. We established the following experimental conditions and controls. Cleavage was induced using TPA (a phorbol ester, diacylglycerol mimic, and PKC activator) or, in cells carrying the AT1 receptor, angiotensin II (AngII; a GPCR ligand that induces PKC activation) and detected by measuring both the released ectodomain (solCD44E; solNRG1E) and the membrane-bound residual fragments (CD44⌬E; NRG1⌬E). To prevent loss of the membrane-bound cleavage product, we inhibited further processing by ␥-secretase using DAPT. In some cases, cleavage was prevented by the addition of an ADAM inhibitor (batimastat) and, in the case of CD44, by the introduction of an uncleavable ICD mutant, CD44-KR-Mt (the so-called lysine-arginine-rich KR domain, the binding site for ERM (ezrin, radixin, and moesin) proteins and merlin) (28). Induced cleavage of the mutant was indeed reduced significantly ( Fig. 1A; the C-terminal cleavage product CD44⌬E is shown). Furthermore, we confirmed that CD44 was an ADAM10 substrate using MEFs with disruption of the Adam10 gene (Fig. 1A) or down-regulation by siRNA (Fig. 1B); there was only marginal cleavage by ADAM17. For NRG1 cleavage, we recently described the necessity of ADAM17 (4). In some experiments, we observed increased levels of CD44fl after TPA stimulation. However, we proved that the induction of ectodomain cleavage by TPA does not involve de novo protein synthesis. Increased CD44 ectodomain cleavage was observed already after 15-20 min, which is too short for up-regulation of protein synthesis of either ADAM10 or CD44. In fact, blocking translation by cycloheximide did not interfere with TPA-stimulated CD44 ectodomain cleavage (Fig. 1C).
As a first option of a protein dimer complex that would allow intracellular signaling to regulate protease action outside the plasma membrane, we considered protease-substrate association. To explore whether substrate and protease were associated with each other, we used co-immunoprecipitation (co-IP) and BiFC, a technique that documents protein-protein interaction with great specificity (for details of this method see under "Experimental Procedures"). Indeed both pairs, ADAM10/ CD44 and ADAM17/NRG1, were associated with each other with or without TPA stimulation and independent of whether the enzyme was inhibited and/or the substrate mutated to be noncleavable. ADAM10 complemented fluorescence with either CD44 WT or the uncleavable CD44 KR mutant (CD44 KR-Mt) was observed with or without TPA stimulation ( Fig.  2A, cleavage inhibited by batimastat). Adiponectin receptor is known to dimerize and was used as a positive control. The adiponectin receptor does not interact with ADAM10. We therefore used the appropriate construct pair, ADAM10 and adiponectin receptor, as negative control in BiFC assays. Quantitation is shown by column diagram in Fig. 2AЈ. Similarly, NRG1 was associated with ADAM17 as shown by co-immunoprecipitation, irrespective of PKC activation by TPA or AngII stimulation ( Fig. 2E; cleavage inhibited by batimastat). PKC␦, which regulates TPA-induced NRG1 cleavage (4), only co-precipitated with NRG1 after TPA or AngII stimulation (Fig. 2E, lower panel; cleavage inhibited by batimastat), suggesting that the observed association of NRG1 with ADAM17 is specific and does not depend on PKC␦. ADAM10 (mainly the mature form) was co-immunoprecipitated together with CD44 WT or CD44  Fig.  1. By the use of a nonreducing gel, cysteine bridge-stabilized dimers are visualized. Cleavage was determined by detection of released ectodomains in the culture supernatant (solCD44E) and of the membrane-bound C-terminal cleavage products, which remained dimerized due to the two cysteines still present in the product (CD44⌬E dimer). Cysteine mutations decreased, whereas CD44 S291A increased basal and TPA induced dimerization and cleavage as compared with CD44 WT. The experiment was repeated three times. B, CD44 dimers are preferentially cleaved by ADAM10. Dimers, but not monomers, disappear upon induced cleavage by ADAM10. Cells, treatments, and cleavage detection were as in Fig. 1B. Where indicated, expression of ADAM10 (A10) or ADAM17 (A17) was down-regulated by siRNA. Nontargeting siRNA C was used as a control. ADAM protein knockdown (not shown) was monitored as in Fig. 1B. Reduced levels of dimers and concomitant increase of cleavage product are detected in cells expressing ADAM10 (control C and A17 lanes). Cleavage is nevertheless increased by treatment with TPA. The experiment was repeated three times. WB, Western blot; V, empty vector; C, control. (Fig. 2, B and C) and vice versa (Fig. 2D). Consistently with the BiFC data, TPA did not enhance the association of overexpressed or endogenous CD44 and ADAM10 (Fig. 2, B and D, respectively).

KR-Mt
Association of both partners prior to cleavage would, in principle, fit with our hypothesis that their relative positional change induced by ICD modification might allow proteolysis. ADAMs, like CD44 and NRG1, indeed possess cytoplasmic tails. However, the necessity of the ADAM17 and ADAM10 ICDs for regulated cleavage has been put into question (47), and the ADAM17 ICD can be removed without compromising regulated cleavage (48). 6 We therefore tested the effect of ADAM10 ICD deletion mutants on CD44 cleavage. None of the deletion mutants tested inhibited induced release of solCD44E (Fig. 3). This argues that no structural determinants in the ADAM10 ICD regulate cleavage. It is hence unlikely that ICD modifications of CD44 in a substrate/ADAM10 heterodimer could be converted into a rotation or other positioning of the substrate relative to the protease. In addition, this observation further highlights the importance of intracellular signaling input into the substrate's ICD or, possibly, into the ICDs of other associated membrane proteins for cleavage regulation.
As the next possible option, we concentrated on whether our studied substrates could act as homodimers or oligomers in the plasma membrane, allowing relative positioning of the monomers to each other across the membrane. BiFC and co-IP of differently tagged monomers indeed showed that both substrates exist, at least in part, as homodimers in the plasma membrane. In the BiFC assay (see schematic in Fig. 2), CD44 WT or NRG1 WT monomers complemented fluorescence with or without TPA (Fig. 4, A and B). By co-precipitation of differently tagged monomers, C-terminal Myc-tagged NRG1 was associated with C-terminal GFP-tagged NRG1 (Fig. 4C). Interestingly, although NRG1 monomers were not covalently linked to each other in the dimer, CD44 monomers formed cysteine disulfide bonds and migrated in nonreducing SDS-PAGE as dimers. These dimers existed prior to and after TPA induction, and they could be dissociated by dithiothreitol (reducing gel; Fig. 6A, see also Fig. 8B).
Consequently, we asked whether the induced cleavage reaction depended on homodimerization of the substrates and whether only dimers were subject to ectodomain cleavage. To this end, we first mutated cysteines putatively responsible for the stabilization of CD44 dimers to alanines. Dimerization was prevented by double mutation of Cys-286 (C286A), which lies in the transmembrane domain, and of Cys-295 (C295A), which lies in the above described cleavage regulatory KR motif (ERM/ merlin interaction motif) (see box Fig. 5); importantly, the mutations also reduced ectodomain cleavage of CD44 (Fig. 5,  compare lanes 2 and 5 with 3 and 6). Although less efficient, mutation of only one cysteine also reduced dimer formation and cleavage (Fig. 5, lanes 9 and 12).
Monomer-monomer interaction leading to dimerization is likely mediated by sequences in the ectodomain of CD44. We thus hypothesized that we should be able to prevent dimerization by adding an excess of soluble ectodomain, and if the dimers were the targets of cleavage, preventing dimerization should inhibit cleavage. This was in fact the case. Overexpression of the soluble CD44 ectodomain (co-transfection of construct with truncation proximal to the transmembrane domain, detectable as a major band around 70 kDa) prevented CD44fl dimerization and TPA-induced solCD44E release (Fig. 6A). We note that dimers did not by themselves trigger cleavage but that TPA stimulation was also required (Figs. 5 and 6A). Another hint for dimer-dependent cleavage was obtained by the use of antibodies. By treating cells with a bivalent CD44 antibody, we could detect a moderate enhancement of basal or TPA-stimulated dimer formation (nonreducing conditions) as well as the release of solCD44E (Fig. 6B, upper and middle panels). Dimer formation and cleavage were prevented by pre-incubation of the CD44 antibody with an isotype-specific secondary antibody, presumably because the larger antibody complex sterically hindered dimerization (Fig. 6B). Further evidence for preferential cleavage of dimers is provided by our observation that down-regulation of ADAM10 (but not of ADAM17) by siRNA leads to the accumulation of CD44 dimers; cleavage, however, required TPA (Fig. 5B).
To corroborate the observation that dimer formation is not sufficient to induce cleavage, we forced dimerization of CD44 by C-terminal fusion of FKBP dimerization domains and addition of the FKBP ligand AP20187 (49). AP20187 indeed led to increased CD44-FKBP dimerization. However, TPA was still needed to induce release of CD44-FKBP C-terminal cleavage products (CD44-FKBP⌬E) (Fig. 7A). The same result was obtained when stringency of dimerization was enhanced by the addition of two FKBP domains (Fig. 7B). However, addition of AP20187 strongly enhanced TPA-stimulated cleavage (compare 3rd and 4th lanes in Fig. 7, A and B). Taken together, these data are suggestive of a role of dimerization/oligomerization as a pre-condition of induced ectodomain cleavage of CD44 and NRG1 and for a regulatory role of the ICD in dimer formation and dependent cleavage.
These results predicted that cleavage regulatory ICD modifications would affect the ability of CD44 to dimerize. Indeed, the uncleavable CD44 mutants KR-Mt and S291D did not form dimers at all, whereas the constitutively cleaved S291A and the ICD deletion mutant CD44⌬ICD formed dimers (Figs. 5A and 8A) or even oligomers (Fig. 8, B and C) irrespective of TPA stimulation. The presence of the reducing agent DTT in cell lysates could partially reduce dimers and oligomers (Fig. 8B,  lower panel). Interestingly, CD44⌬ICD dimerization did partially depend on stabilization by the putative Cys-286 disulfide bridge (Fig. 8B, lower panel, 3rd lane; for comparison see C286A in CD44 WT Fig. 5A), but dimerization was not completely reduced by mutation of Cys-286 because part of it is likely mediated by interactions between the ectodomains. This interaction of ectodomains was also the basis of

Substrate Dimerization Regulates Ectodomain Cleavage
the experiment in Fig. 6A. CD44⌬ICD, in contrast to the full-length CD44 with S291A mutation (see CD44S291A in Fig. 5A), was constitutively cleaved irrespective of the degree of dimerization (Fig. 8B, upper panel shows the N-terminal ectodomain cleavage product, solCD44E; compare with lower dimerization panel). As predicted by our results so far, the ICD deletion protein apparently does not need a signal transfer through the membrane, and the ectodomain is apparently open for protease access. This is not the case when the "repressive" ICD is present. The signal transfer is required, and accordingly, both dimerization and cleavage are regulated. This is further highlighted in Fig. 8D. Binding of a tumor suppressor merlin (neurofibromatosis 2, Nf2) to the CD44 ICD blocks its ectodomain cleavage (50). A constitutively active mutant of merlin (S518A mutant) interfered with CD44 dimerization, consistent with our prediction of dimerization as a prerequisite for cleavage (Fig. 8D).

Discussion
In summary, our results suggest that CD44 and NRG1 cleavage, and possibly the cleavage of other ADAM substrates, is regulated by an "inside-out" signaling mechanism that requires prior substrate homodimerization and specific modification of the substrate's ICD. Dimerization is a precondition that per-mits signal transfer from the modified ICD to the ectodomain. According to our BiFC data, substantial amounts of CD44 exist as homodimers on the cell surface (Fig. 4, AЈ/BЈ). Dimerization of CD44 required stabilization by putative cysteine disulfide bridges in the transmembrane domain (Cys-286) and the juxtamembrane ICD (Cys-295) (Fig. 5A). However, we cannot rule out that the cysteine mutations affect ICD modifications or interactions with partner proteins and that these putative interferences are in fact responsible for the reduced cleavage. Furthermore, we showed that dimers are also mediated by ectodomain interactions. This was corroborated by the fact that excess soluble CD44E or the combined treatment with primary (anti-CD44) and secondary (isotype-specific) antibody inhibited dimerization and the induced cleavage of CD44 (Fig. 6). Dimerization was however not sufficient for cleavage, even when forced by FKBP domains (Fig. 7), as cleavage generally required the presence of a cleavage stimulus (Figs. [5][6][7] or at least the presence of a cleavage promoting modification (e.g. CD44 S291A in Fig. 8A). Cleavage inhibitory modifications consistently prevented dimerization (KR-Mt, S291D, and merlin S518A overexpression; Fig. 8, A and D). An early report on antibody-induced shedding of CD44 (51) probably already indicated dimer-dependent cleavage. In a previous report, phorbol ester stimulated CD44 dimer formation (52), an effect that is barely visible in our experiments. However, consistent with our results on the importance of cysteines Cys-286 and Cys-295, CD44 dimers were also found to be stabilized by cysteine links (52). Nevertheless, our results suggest that the ectodomains also have a significant role in dimer formation, as shown by the high level of dimerization of the CD44 ICD deletion (no C295) (Fig. 8B) and also by the effective inhibition of CD44 WT cleavage by addition of soluble CD44 ectodomain (Fig. 6A).
BiFC assays (Fig. 4B) indicate that NRG1 is also associated as dimers on the cell surface. In the case of NRG1, dimerization appears to involve the intracellular domain. Cross-linking studies of the NRG1␣2c ICD expressed alone showed that it spontaneously dimerizes and that force dimerization of a previously cleavage-resistant NRG1␣2c ICD deletion mutant rescued its cleavage (an Fc region was added instead of its ICD; Fc dimerizes by disulfide bonds) (53). Based on our observations, we postulate that other ADAM10 or ADAM17 cleaved transmembrane proteins might also require dimerization for their cleavage. In fact, many membrane proteins are known to form homo-or heterodimers, e.g. ICAM1, angiotensin I-converting enzyme, and Alzheimer precursor protein (54 -57), that would allow regulation similar to the one described here.
Although we have not strictly ruled out that the antibodies or excess soluble ectodomain used in our studies also interfered with the ADAM-substrate interaction, the fact that ADAMs were found associated with their substrates even under conditions of absent proteolysis (e.g. due to mutation of the substrate) (Fig. 2) strengthens the argument that the substrates need to be made cleavage-competent to be in a productive interaction with the ADAMs. This is accomplished by the above-mentioned inside-out signaling process that makes the substrate accessible for cleavage. 5 Our data suggest an additional mechanism of cleavage regulation on the level of the substrate. We do not conclude from our data that regulation of cleavage does not also occur on the protease level. Protease activity regulation has been well documented. Numerous reports have shown that the expression level, surface localization, and catalytic activity of ADAMs are regulated (58 -60).
Would cells carry enough ADAM molecules to permit silent associations with numerous substrates? Probably yes. The cleavage reactions trigger the release of highly active regulatory FIGURE 9. Schematic representation of cleavage regulation through dimerization. CD44 monomers and dimers co-exist on the cell surface. CD44 dimers are stabilized by putative cysteine bridges in the ICD and by putative ectodomain interactions. Our data suggest that ectodomain cleavage regulation depends on ICD modification and interaction with either ERM proteins or merlin on CD44 dimers. At high cell density merlin is dephosphorylated and active and bound to the phosphorylated CD44 ICD. Under low cell density or after TPA stimulation (PKC activator), Ser-291 is dephosphorylated, and phosphorylated/activated ERM proteins displace merlin. This releases a restrictive ICD conformation in the dimer and leads to a positional structural change of the dimerization partners that enable ectodomain accessibility to the ADAM protease. CD44 without its ICD is missing the restrictive ICD conformation and is spontaneously cleaved. A link of ERM proteins to the actin cytoskeleton is possibly important in this regulation. How can CD44S291 be dephosphorylated in response to TPA-dependent activation of PKC? PP1/2 serine phosphatase can indeed be activated by PKC and is regulated by endogenous PKC-activated inhibitors (8,41,46). molecules. Therefore, it would suffice if only a minority of substrate molecules were in fact subjected to proteolysis. Indeed, only a small fraction of CD44 on the cell surface (some 15%) is subjected to cleavage (50). Correspondingly, according to the BiFC data of Fig. 2A, only a similar percentage of CD44fl (about 10%) is associated with ADAM10 in the absence of cleavage. It is not clear from the BiFC assay whether CD44 is associated with ADAM10 as dimer. Our other experiments support that dimers are preferentially cleaved by ADAM10 (Fig. 5B). NRG1 cleavage however, is particularly effective suggesting that for some substrates the association with ADAMs is close to 100%. Ectodomain release may be required rapidly. Thus, close proximity of enzyme and substrates would permit fast responses that are needed for factors involved in signaling. Induced cleavage has been detected in under 5 min in many studies (4,61).
In the examples reported here, the interaction of ADAMs and substrates is made productive by activation of intracellular signaling pathways that induce specific substrate ICD modifications on CD44 and NRG1 5 (4,50). We showed that these ICD modifications regulate accessibility of the substrate's ectodomain to the ADAM protease. 5 Thus, our finding of substrate homodimerization as a prerequisite of cleavage of CD44 and NRG1 (Figs. 4 and 5) offers an explanation of how the intracellular ICD modification could affect the accessibility of the ectodomain, namely by ICD modification-induced relative positional change of the substrate dimerization partners to each other, e.g. by rotation or movement within the plane of the membrane, allowing access of the extracellular ADAM catalytic domain to the substrate's ectodomain. This conclusion is further supported by the fact that the observed effects of our CD44 ICD mutants on dimerization are consistent with their observed effects on cleavage; cleavage inhibitory ICD mutations inhibit dimerization, and cleavage activating ICD mutations enhance dimerization (Figs. 5 and 8).
Besides the examples reported here, ectodomain accessibility regulation has also been described for another ADAM substrate pair. Notch and ADAM10 are also associated in a nonproductive manner prior to ligand binding to Notch (62)(63)(64). Interestingly, ligand binding followed by its endocytosis exerts a "pulling" force on the Notch ectodomain, inducing structural changes that facilitate ADAM10 cleavage (62,65) and providing evidence for the existence of other modes of ectodomain cleavage accessibility regulation than those requiring substrate homodimerization reported here. Along these lines, certain ADAM17 substrates (and ADAM17 itself) are known to interact/heterodimerize with "third" proteins that could relay such a function, e.g. annexins (66 -68). In addition, other transmembrane proteins that interact with ADAM17 could potentially participate in such cleavage regulation, e.g. iRhoms (69).
An interesting comparable case of inside-out signaling concerns the ectodomain ligand binding affinity of integrins (25,70). Integrins are heterodimers of ␣ and ␤ chains. Similar to our proposed model for CD44 and NRG1 cleavage, interaction partners of the ␤ chain ICD presumably trigger a conformational change of the ectodomains, requiring a change in the relative positioning of the heterodimerization partners. Interestingly, activation of integrin requires linkage to the actin cytoskeleton via talin (71)(72)(73). Similar to CD44 and ERM/merlin, binding and release of talin is regulated by phosphorylation/ dephosphorylation of the integrin ␤ chain. ERM proteins are also known to interact with actin, and we observed that this actin link function is required for cleavage of CD44, and we speculate that this may anchor and stabilize a CD44 dimer (50). However, we did not detect a clear effect of an ezrin actin link mutant on dimerization of CD44 (data not shown). Thus, the exact role of actin interaction for substrate cleavage has not yet been revealed.
A schematic representation of our data is summarized in Fig.  9. Our results and the example of integrin heterodimer activation by an inside-out signaling mechanism in principle support the notion that dimerization of substrates may represent a more generally used regulatory mechanism of ectodomain cleavage; this mechanism allows intracellular signaling to induce inside-out signaling via modification of the substrate's ICD.