Presenilin-dependent Intramembrane Proteolysis of CD44 Leads to the Liberation of Its Intracellular Domain and the Secretion of an A (cid:1) -like Peptide*

Alzheimer’s disease (AD)-associated (cid:2) -secretase is a presenilin (PS)- dependent proteolytic activity involved in the intramembraneous cleavage of the (cid:1) -amyloid precursor protein, Notch, LDL receptor-related protein, E-cadherin, and ErbB-4. This cut produces the corresponding intracellular domains (ICD), which are required for nuclear signaling of Notch and probably ErbB-4, the (cid:1) -amyloid precursor protein, E-cadherin, and the LDL receptor-related protein as well. We have now investigated CD44, a cell surface adhesion mole-cule, which also undergoes an intramembraneous cleavage to liberate its ICD. We demonstrate that this cleavage requires a PS-dependent (cid:2) -secretase activity. A loss-of-function PS1 mutation, a PS1/PS2 knockout, as well as two independent and highly specific (cid:2) -secretase inhibitors, abolish this cleavage.

Intramembraneous proteolysis has been long thought to be an exceptional biochemical pathway involved in the pathological generation of A␤ 1 (1). However, intramembraneous proc-essing of membrane-bound proteins has now been demonstrated to play an important physiological role in regulated nuclear signaling (2). Currently three such proteolytic pathways are known, PS-dependent intramembraneous proteolysis (3), site 2 protease-mediated cleavage (4), and rhomboid-mediated proteolysis (5,6). The PS-dependent ␥-secretase cleavage is of pivotal importance for A␤ generation (3). A␤ is physiologically produced from the ␤-amyloid precursor protein (␤APP) through an initial ␤-secretase cleavage followed by the intramembraneous ␥-secretase cut (7). The resulting peptide is secreted and deposited in the AD-defining amyloid plaques. A knockout of the two homologous PS1 and PS2 genes fully abolishes A␤ generation (8,9). Furthermore, highly specific ␥-secretase inhibitors bind to PSs (10,11). Mutagenesis of two conserved critical aspartate residues within transmembrane domains (TMD) 6 and 7 of PS1 and PS2 also blocks ␥-secretase function (12,13), which indicates that PSs may contain an intrinsic aspartyl protease activity responsible for the ␥-secretase cut (14). Indeed, we have identified a novel active site motif in all PSs (15), which is also present in polytopic bacterial aspartyl proteases called type 4 prepilin peptidases (16). Very recently this finding was confirmed by the identification of the same motif in the signal peptide peptidase SPP, another polytopic aspartyl protease (17). However, increasing evidence indicates that ␥-secretase activity resides in a high molecular weight complex (18) composed of PS (19 -21), nicastrin (22)(23)(24), PEN-2 (25,26), and probably Aph-1 as well (25,27).
PSs also play a major role in Notch signaling (3). Deletion of the PS1 gene in mice causes a phenocopy of a Notch knockout (28). This is due to a lack of nuclear signaling via the Notch intracellular domain (NICD) (29). NICD is liberated by the PS-dependent ␥-secretase and targeted to the nucleus where it regulates gene transcription. ␤APP itself may also provide a precursor for the production of a cytoplasmic fragment involved in nuclear signaling (30). Indeed it has recently been demonstrated that the ␤APP intracellular domain (AICD) is generated by a very similar proteolytic mechanism like NICD (31)(32)(33)(34). In addition, LDL receptor-related protein (LRP) (35), E-cadherin (36), and ErbB-4 (37,38) have also been shown to undergo PS-dependent endpoproteolysis resulting in the liberation of their ICDs, which may be involved in nuclear signaling as well. However, the cleavage, which liberates such ICDs takes place at a topologically different site as the cut, which generates A␤ (see Fig. 1A). Moreover, secreted peptides similar to A␤ have so far not been observed to be generated from any of the PS substrates other than ␤APP. This may suggest that biochemically different mechanisms are involved in intramem-braneous proteolysis leading to ICD liberation and the production of A␤-like peptides (32), a hypothesis that may exclude a direct function of PSs in ␥-secretase processing.
Identification of PS substrates is of pivotal importance for the development of safe ␥-secretase inhibitors. In that regard it has been shown that ␥-secretase inhibitors not only block A␤ generation but also severely interfere with T-lymphocyte differentiation in cultured cells (39,40) and lateral inhibition in zebrafish (Danio rerio) (41). Very recently CD44, a cell surface adhesion protein, has been shown to undergo an intramembraneous cleavage, which results in the generation of CD44-ICD (42). CD44-ICD is targeted to the nucleus where it regulates genes containing TPA-responsive elements (42). Based on the inhibition of this cleavage by MG132, an unspecific proteasome inhibitor (43), which also blocks ␤and ␥-secretase (44), it has been hypothesized that CD44-ICD generation could be mediated by ␥-secretase (42). However, as described above at least three different intramembraneous cleavage mechanisms are currently known and evidence for a PS-dependent CD44-ICD generation is missing. We therefore investigated if inactivation of PS function through highly selective inhibitors, a complete knockout of PS genes, or a dominant negative PS1 mutation affects CD44-ICD production. Upon demonstration of a PS-dependent CD44-ICD generation, we further found that a second PS-dependent ␥-secretase cut located in the middle of the TMD liberates a small and hydrophobic peptide similar to A␤. Therefore our data suggest a novel function of PSs in TMD removal. Moreover, our findings demonstrate very similar mechanisms of ␤APP and CD44 ␥-secretase processing and therefore further support a direct role of PSs in intramembraneous proteolysis.
cDNA Constructs-A full-length CD44 construct with a C-terminal Myc epitope tag was generated by PCR using a human brain cDNA library (Clontech) and the following primers: CD44FL forward, cgcaagcttccggacaccatggacaagttttgg and CD44FL reverse, cgcctcgagtcaCA-GATCCTCTTCTGAGATGAGTTTTTGTTCcaccccaatcttcatgtcc. Capital letters denote the C-terminal Myc epitope tag. The PCR product was subcloned into the HindIII/XhoI sites of pcDNA3.1/Hygro(ϩ) (Invitrogen). The CD44⌬E and CD44⌬E-FLAG constructs were generated by PCR using the following primers: CD44⌬E forward, cgcaagcttcaagaaggtggagcaaac or CD44⌬E-FLAG forward, cgcaagcttGACTACAAAGAC-GATGACGACAAGcaagaaggtggagcaaac, respectively and CD44⌬E reverse, ccgctcgagccaccccaatcttcatgtcc and subcloned into the HindIII/ XhoI sites of pSecTag2/HygroB (Invitrogen). Capital letters denote the N-terminal FLAG-epitope-tag. The resulting cDNA constructs were sequenced for verification.
Antibodies-The monoclonal antibody 9E10 to the Myc epitope was obtained from the hybridoma bank. Anti-FLAG-M2-agarose, anti-Mycagarose and the monoclonal FLAG-M2 antibody were from Sigma.
Analysis of CD44 Expression-HEK293 cells or HEK293 cells stably expressing wild type PS1 or PS1 D385N were transiently transfected with cDNA encoding CD44FL. 24 h after transfection, cell lysates were prepared and analyzed for CD44FL by immunoblotting with antibody 9E10. Detection was performed using enhanced chemiluminescence (ECL, Amersham Biosciences).
CD44-ICD and CD44-␤ Generation-HEK293 cells stably expressing wild type PS1 or PS1 D385N were grown on poly-L-lysine coated 6-cm dishes. 6 h after transfection of the cells with CD44⌬E or CD44⌬E-FLAG, cells were incubated for 15 h in DMEM containing 1 M DAPT, 5 M L-685,458, or Me 2 SO as vehicle. In order to detect CD44 ␤-peptides, media were collected, cleared from debris by centrifugation at 16,000 ϫ g at 4°C for 10 min and subjected to immunoprecipitation with anti-FLAG-M2-agarose for 4 h at 4°C. Immunoprecipitates were washed three times with phosphate-buffered saline, subjected to SDS-PAGE on 10 -20% Tris-Tricine gels (Invitrogen), and blotted onto nitrocellulose membranes. The membranes were boiled for 5 min in phosphate-buffered saline, probed with antibody FLAG-M2, and developed using Western Star (Tropix). For detection of CD44-ICD, cell lysates were immunoprecipitated with anti-Myc-agarose for 2 h at 4°C. Immunoprecipitates were washed three times with phosphate-buffered saline subjected to SDS-PAGE, immunoblotted with antibody 9E10, and analyzed for CD44-ICD as described above. Mouse embryonic fibroblasts were cultured in 10-cm dishes, and CD44-ICD and CD44-␤ were detected as described for HEK293 cells.
In Vitro Generation of CD44-ICD-CD44-ICD was generated in vitro as previously described for the generation of AICD (31) except that both homogenization buffer and assay buffer were supplemented with 5 mM EDTA, 1 mM 1,10-phenanthroline, and 10 M lactacystin. Following SDS-PAGE, CD44-ICD was analyzed by immunoblotting with 9E10 as described above.
Combined Immunoprecipitation/MALDI-TOF MS Analysis of CD44 ␤-peptides-Cell lines transiently expressing CD44⌬E-FLAG were grown to confluency in 10-cm dishes. Media were replaced with 5 ml of Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and were incubated for 3 h. From four dishes, 20 ml of conditioned media were collected, immediately put on ice, and subjected to a clarifying spin. Following addition of a protease inhibitor mix (1:1000, Sigma) and 0.025% sodium azide, conditioned media were immunoprecipitated with anti-FLAG-M2-agarose for 4 h at 4°C. Immunoprecipitates were washed three times for 10 min at 4°C with wash buffer 1 (0.1% n-octylglucoside, 140 mM NaCl, 10 mM Tris (pH 8.0), 0.025% sodium azide) and once with wash buffer 2 (10 mM Tris, pH 8.0, 0.025% sodium azide). Immunoprecipitated peptides were eluted with trifluoroacetic acid/acetonitrile/water (1:20:20) saturated with ␣-cyano-4-hydroxycinnamic acid. The dissolved samples were dried on a stainless plate and subjected to MALDI-TOF MS analysis. The MS peak heights and molecular masses were calibrated with angiotensin (Sigma) and bovine insulin ␤-chain (Sigma).

RESULTS
The cell surface protein CD44 was previously shown to undergo ectodomain shedding in various cancer cell lines to produce a membrane-anchored C-terminal fragment (CTF) (45). Very recently it was demonstrated that this CTF could further be proteolytically cleaved within the transmembrane region very close to the putative cytoplasmic border to liberate CD44-ICD from the membrane (42). Thus this cut is reminiscent to the PS/␥-secretase-dependent generation of AICD (Fig. 1A). We therefore investigated whether intramembraneous proteolytic processing of CD44 is dependent on PS activity.
PS1-dependent CD44-ICD Generation-HEK293 cells express endogenous PS-dependent ␥-secretase activity and have been widely used to investigate PS-dependent processing of ␤APP (15,46), LRP (35), ErbB-4 (38), E-cadherin (36), and Notch (13). Therefore we transiently transfected HEK293 cells or HEK293 cells stably expressing either wild type PS1 or the biologically inactive mutant PS1 D385N (12) with a cDNA construct encoding C-terminal Myc-tagged full-length CD44 (CD44FL) (Fig. 1B). CD44FL is expressed as a 90 -100 kDa protein in transfected cells (Fig. 2). Additionally, we observed a robust accumulation of a protein species of ϳ25 kDa in cells expressing the non-functional dominant negative mutant PS1 D385N (Fig. 2). This protein most likely corresponds to the C-terminal fragment of CD44 (CD44-CTF) derived by ectodomain cleavage (42). In contrast almost no CD44-CTF was observed in HEK293 cells or cells expressing wild type PS1 (Fig.  2). This result is reminiscent to the effect of biologically inactive PS1 on the accumulation of ␤APP-CTFs (12) and Notch⌬E (41). The accumulation of CD44-CTF in the presence of a nonfunctional PS1 variant thus suggests that CD44 is a physiological substrate for PS-mediated ␥-secretase processing.
Next we investigated if the massive accumulation of CD44-CTF in the presence of the non-functional PS1 D385N mutation also results in a concomitant decrease of CD44-ICD. Therefore HEK293 cells stably expressing wild type PS1 or PS1 D385N were transiently transfected with CD44⌬E (Fig. 1B), which mimics the product of the ectodomain-shedded CD44 and is reminiscent to the constitutively active Notch⌬E frequently used to monitor PS-dependent Notch endoproteolysis (47). Consistent with Okamoto et al. (42) we obtained proteins with a molecular mass of 25-30 kDa upon expression of CD44⌬E.
These fragments accumulated to high levels either in the presence of the non-functional PS1 D385N mutant or in wild type PS1-expressing cells treated with specific ␥-secretase inhibitors (48,49) (Fig. 3A, upper panels). This observation is reminiscent of the accumulation of ␤APP-CTFs upon ␥-secretase inhibitor treatment (31). Consistent with previous results (42), immunoprecipitation of cell lysates with anti-Myc-agarose resulted in the detection of CD44-ICD (Fig. 3A, lower panels). CD44-ICD generation is PS-dependent, since this fragment was not detectable in cells expressing the non-functional mutant PS1 D385N (Fig. 3A, lower panels). Moreover the generation of CD44-ICD was also inhibited in the presence of two highly specific ␥-secretase inhibitors, DAPT (48) and L-685,458 (49) (Fig. 3A, lower panels).
These results were confirmed by a previously established in vitro assay for PS-dependent AICD generation (31). Membranes were prepared from HEK293 cells transiently expressing CD44⌬E and incubated either at 4 or 37°C. As shown in Fig. 3B incubation of membranes at 37°C led to the production of CD44-ICD. When L-685,458 was added to inhibit PS-dependent ␥-secretase activity no CD44-ICD was observed (Fig. 3B). Moreover, CD44-ICD was also not produced when membranes isolated from HEK293 cells expressing PS1 D385N were used in the in vitro assay (Fig. 3B). The lack of CD44-ICD production upon interference with PS activity thus unequivocally proves FIG. 1. Generation of AICD and CD44-ICD. A, AICD and CD44-ICD are generated through cleavage in the hydrophobic TMD close to the cytoplasmic border. In addition ␤APP is cleaved by ␥-secretase in the middle of the TMD after amino acids 40 and 42 (numbering according to the A␤-domain), which results in the secretion of A␤. Whereas AICD and A␤ are generated in a PS/␥-secretase-dependent manner, the proteolytic mechanism, which is responsible for CD44-ICD generation is unknown. B, schematic representation of the CD44 constructs used in this study. The signal sequence of CD44FL is shown by a white box and the spacer (from pSecTag2/HygroB) between the signal peptide and amino acid Q250 (CD44⌬E) or the FLAG tag (CD44⌬E-FLAG) is shown by a striped box. The epitopes of antibodies 9E10 and M2 are indicated. The reaction mixtures were separated by ultracentrifugation into pellet (P100) and soluble fraction (S100). The S100 fraction was further analyzed for CD44-ICD by immunoblotting with antibody 9E10. that CD44 is a novel substrate for PS-mediated ␥-secretase activity.
Secretion of CD44 ␤-Peptides Similar to A␤-Intramembraneous proteolysis of ␤APP not only results in the generation of AICD via a cleavage close to the cytoplasmic site of the TMD, but also in the generation of A␤ by an additional cut in the middle of the TMD (Fig. 1A). We therefore investigated if we could detect the N-terminal counterpart of CD44-ICD in conditioned media similar to A␤. To do so we transiently transfected cells either expressing wild type PS1 or PS1 D385N with an N-terminal FLAG-tagged and C-terminal Myc-tagged CD44⌬E construct (CD44⌬E-FLAG) (Fig. 1B). As expected expression of CD44⌬E-FLAG led to CD44-CTFs of the expected slightly higher molecular mass compared with the expression of CD44⌬E (Fig. 4, upper panels). Conditioned media of transfected cells treated with or without ␥-secretase inhibitors were precipitated with anti-FLAG-M2-agarose. As shown in Fig. 4, we observed a triplet of peptides with a molecular weight around 5 kDa (hereafter called CD44-␤) in cells expressing endogenous or ectopic wild type PS1, suggesting heterogeneous processing. In conditioned media from cells expressing CD44⌬E no such peptides were detected demonstrating the specific isolation of CD44-␤ (Fig. 4, lower panels). The generation of CD44-␤ is PS-dependent, because cells expressing the non-functional PS1 D385N mutant secreted no CD44-␤ (Fig. 4, lower panels). In addition, treatment with the ␥-secretase inhibitors DAPT (48) or L-685,458 (49) also significantly reduced the production of CD44-␤ (Fig. 4, lower panels).
In order to determine the cleavage site, which results in the release of CD44-␤ we performed MALDI-TOF mass spectrometry using immunoprecipitated CD44-␤. This revealed a major peptide species with a molecular mass of 4385 Da (Fig. 6A).
Computer-based analysis revealed that the proteolytic cleavage, which results in the secretion of this peptide occurs between Ala 278 and Leu 279 (Fig. 6B) in cells expressing endogenous or ectopic wild type PS1. Similar to the heterogeneous ␥-secretase generated A␤ additional peptides were observed (Fig. 6, A and B). All peptides were generated by a PS-dependent activity, since the corresponding peaks were not observed upon expression of PS1 D385N (Fig. 6A, inset). Thus, like ␤APP (31-34), CD44 is cleaved at two topologically different sites within the TMD (Fig. 7). Both cleavages are dependent on PS function, since the ␥-secretase inhibitors DAPT and L-685,458, and the biologically inactive PS1 D385N mutant inhibit formation of CD44-␤ and of CD44-ICD. This suggests a dual cleavage mechanism directly mediated by a PS1-dependent ␥-secretase activity. DISCUSSION Overall, our data demonstrate that intramembraneous processing of CD44 occurs by a PS-dependent ␥-secretase activity at two topologically different sites (summarized in Fig. 7). One cleavage (close to the cytoplasmic border) liberates CD44-ICD for putative nuclear signaling, the other cut (in the middle of the TMD predominantly between Ala 278 and Leu 279 ) splits the TMD and results in the secretion of a peptide similar to A␤.
Cleavage close to the cytosolic border of the TMD results in the generation of CD44-ICD. This cut is remarkably similar to AICD and NICD generation, since it not only occurs at a topologically similar site but it is also fully PS-dependent. This is demonstrated by the inhibition of CD44-ICD generation through DAPT, L-685,458, and by a dominant negative mutation of PS1. Moreover no CD44-ICD could be detected in PS 1/2 double knockout cells.
DAPT and L-685,458 specifically inhibit the ␥-secretase in cultured cells and in vivo (48,49). In transgenic mice, which develop amyloid plaques due to the overexpression of mutant ␤APP, DAPT significantly reduces the A␤ burden (48). In addition PS-dependent Notch endoproteolysis is also blocked by ␥-secretase inhibitors in cultured cells (29,50,51). Moreover, DAPT causes a severe Notch phenotype in zebrafish FIG. 5. Lack of CD44-ICD or CD44-␤ generation in the absence of PS. PS1 ϩ/ϩ /PS2 ϩ/ϩ and PS1 Ϫ/Ϫ /PS2 Ϫ/Ϫ mouse embryonic fibroblasts were transiently transfected with CD44⌬E-FLAG. In addition PS1 ϩ/ϩ / PS2 ϩ/ϩ mouse embryonic fibroblasts transfected with CD44⌬E-FLAG were also treated with the ␥-secretase inhibitor L-685,458. CD44-ICD or CD44-␤ was precipitated as described above. Note that the ␥-secretase inhibitor L-685,458 as well as the PS1/PS2 double knockout fully abolished CD44-ICD or CD44-␤ production. embryos due the inhibition of NICD generation (41). Therefore several substrates for the PS-dependent ␥-secretase have now been identified: ␤APP (and its homologues) (1), Notch1-4 (29,52,53), LRP (35), E-cadherin (36) ErbB-4 (37,38), and CD44. Interestingly, all of these are type 1 transmembrane proteins. Moreover, they all undergo ectodomain shedding, which apparently is a prerequisite for the subsequent ␥-secretase cleavage. The resulting cytoplasmic domains may then be targeted to the nucleus where they could regulate transcription of selective target genes (2, 3). Nuclear targeting may require binding of additional cofactors and the efficiency of nuclear targeting appears to be very different for all known ICDs. Whereas NICD can be detected in large amounts within the nucleus (54,55), in vivo produced authentic AICD has so far not been detected within nuclei. Moreover, natural target genes have so far been identified only for CD44 (42) and Notch1 (47). In all cases the intramembraneous cleavage leading to ICD generation seems to take place close to the cytoplasmic domain. PS-dependent ICD generation occurs constitutively immediately after ectodomain shedding. This raises the question how nuclear signaling of the different ICDs is regulated. Up-or down-regulation of PS genes has so far not been reported. Moreover, artificial overexpression of PSs by transfection does not lead to substantially higher PS levels since only endogenous PSs are replaced by the overexpressed variant (56). This may also be the case for other components of the PS/␥-secretase complex such as nicastrin (22)(23)(24). It appears therefore that the activity of ICDs like that of conventional transcription factors may be regulated by proteolytic degradation. In fact, NICD (57,58) and LRP-ICD (35) have been shown to be degraded by the proteasome whereas AICD is predominantly removed by a metalloprotease activity, most likely IDE (59).
In addition to ICD generation we also found a PS-dependent production of A␤-like CD44 peptides, which we called CD44-␤. MALDI-TOF MS revealed that these peptides are generated by cleavage within the middle of the TMD. This cut is apparently homologous to the cleavages, which lead to the secretion of the two major A␤ species (Fig. 7). Our findings therefore reveal a remarkably similar intramembraneous cleavage mechanism of ␤APP and CD44. Moreover, we have recently identified a secreted Notch peptide, which also resembles A␤ and CD44-␤ (60). Therefore we identified a dual intramembraneous cut mediated by a PS-dependent ␥-secretase activity. Apparently one cut allows the generation of ICDs, which may be involved in nuclear signaling. The other cut, which occurs within the middle of the TMD allows the efficient removal of the remaining membrane associated stub. Thus we propose that PS-mediated endoproteolysis is required at least in some cases for nuclear signaling but also for the general clearance of TMDs. This is supported by the finding that even artificial TMDs may allow ␥-secretase processing in cultured cells (61) and in Drososphila (62). FIG. 6. Identification of the cleavage site, which liberates CD44-␤, by MALDI-TOF mass spectrometry. A, conditioned media from cells expressing endogenous PS1 were analyzed. Molecular masses in Da of the individual peaks are indicated. Inset, MALDI-TOF spectra of CD44 peptides isolated from conditioned media of cells expressing, ectopic PS1 wild type and the non-functional PS1 D385N. Note that all peptides (including the major peptide with a molecular mass of 4385 Da) observed are generated via a PS1-dependent pathway, since no signals were observed in conditioned media of cells expressing the dominant negative mutant PS1 D385N. B, schematic representation of identified peptide fragments according to the observed mass. The molecular mass of the major peak suggests that after cleavage of the signal peptide an additional N-terminal cleavage occurs, which could be mediated by furin at a pseudo-cleavage site introduced by the pSecTag2 vector between signal sequence and the N-terminal FLAG tag. The peak with the molecular mass of 4757 Da could not be assigned.