NFκB-dependent Control of BACE1 Promoter Transactivation by Aβ42*

β-Amyloid (Aβ) peptides that accumulate in Alzheimer disease are generated from the β-amyloid precursor protein (βAPP) by cleavages by β-secretase BACE1 and by presenilin-dependent γ-secretase activities. Very few data document a putative cross-talk between these proteases and the regulatory mechanisms underlying such interaction. We show that presenilin deficiency lowers BACE1 maturation and affects both BACE1 activity and promoter transactivation. The specific γ-secretase inhibitor DFK167 triggers the decrease of BACE1 activity in wild-type but not in presenilin-deficient fibroblasts. This decrease is also elicited by catalytically inactive γ-secretase. The overexpression of APP intracellular domain (AICD), the γ/ϵ-secretase-derived C-terminal product of β-amyloid precursor protein, does not modulate BACE1 activity or promoter transactivation in fibroblasts and does not alter BACE1 expression in AICD transgenic brains of mice. A DFK167-sensitive increase of BACE1 activity is observed in cells overexpressing APPϵ (the N-terminal product of βAPP generated by ϵ-secretase cleavage harboring the Aβ domain but lacking the AICD sequence), suggesting that the production of Aβ could account for the modulation of BACE1. Accordingly, we show that HEK293 cells overexpressing wild-type βAPP exhibit a DFK167-sensitive increase in BACE1 promoter transactivation that is increased by the Aβ-potentiating Swedish mutation. This effect was mimicked by exogenous application of Aβ42 but not Aβ40 or by transient transfection of cDNA encoding Aβ42 sequence. The IκB kinase inhibitor BMS345541 prevents Aβ-induced BACE1 promoter transactivation suggesting that NFκB could mediate this Aβ-associated phenotype. Accordingly, the overexpression of wild-type or Swedish mutated βAPP does not modify the transactivation of BACE1 promoter constructs lacking NFκB-responsive element. Furthermore, APP/β-amyloid precursor protein-like protein deficiency does not affect BACE1 activity and expression. Overall, these data suggest that physiological levels of endogenous Aβ are not sufficient per se to modulate BACE1 promoter transactivation but that exacerbated Aβ production linked to wild-type or Swedish mutated βAPP overexpression modulates BACE1 promoter transactivation and activity via an NFκB-dependent pathway.

␤-Amyloid (A␤) peptides that accumulate in Alzheimer disease are generated from the ␤-amyloid precursor protein (␤APP) by cleavages by ␤-secretase BACE1 and by presenilindependent ␥-secretase activities. Very few data document a putative cross-talk between these proteases and the regulatory mechanisms underlying such interaction. We show that presenilin deficiency lowers BACE1 maturation and affects both BACE1 activity and promoter transactivation. The specific ␥-secretase inhibitor DFK167 triggers the decrease of BACE1 activity in wild-type but not in presenilin-deficient fibroblasts. This decrease is also elicited by catalytically inactive ␥-secretase. The overexpression of APP intracellular domain (AICD), the ␥/⑀-secretase-derived C-terminal product of ␤-amyloid precursor protein, does not modulate BACE1 activity or promoter transactivation in fibroblasts and does not alter BACE1 expression in AICD transgenic brains of mice. A DFK167-sensitive increase of BACE1 activity is observed in cells overexpressing APP⑀ (the N-terminal product of ␤APP generated by ⑀-secretase cleavage harboring the A␤ domain but lacking the AICD sequence), suggesting that the production of A␤ could account for the modulation of BACE1. Accordingly, we show that HEK293 cells overexpressing wild-type ␤APP exhibit a DFK167-sensitive increase in BACE1 promoter transactivation that is increased by the A␤-potentiating Swedish mutation. This effect was mimicked by exogenous application of A␤42 but not A␤40 or by transient transfection of cDNA encoding A␤42 sequence. The IB kinase inhibitor BMS345541 prevents A␤-induced BACE1 promoter transactivation suggesting that NFB could mediate this A␤-associated phenotype. Accordingly, the overexpression of wild-type or Swedish mutated ␤APP does not modify the transactivation of BACE1 promoter constructs lacking NFB-responsive element. Furthermore, APP/␤-amyloid precursor protein-like protein deficiency does not affect BACE1 activity and expression. Overall, these data suggest that physiological levels of endogenous A␤ are not sufficient per se to modulate BACE1 promoter transactivation but that exacerbated A␤ production linked to wild-type or Swedish mutated ␤APP overexpression modulates BACE1 promoter transactivation and activity via an NFB-dependent pathway.
Alzheimer disease (AD) 3 is characterized by abnormal deposition of a set of hydrophobic peptides called amyloid ␤ (A␤) peptides. The increase of cerebral A␤ levels is one of the common denominators characterizing both sporadic and familial forms of AD and therefore, if not demonstrated as the etiological cause of AD pathology, is often considered as a key factor contributing to the degenerative process (1).
The mechanisms by which A␤ peptides are generated are a matter of intense research in the AD field. A␤ peptides are released from a transmembrane protein, ␤-amyloid precursor protein (␤APP), by the sequential attacks by ␤and ␥-secretase that liberate the N-and C-terminal moieties of A␤ peptides, respectively (2). All ␤-secretase-like activity appears to be borne by an aspartyl protease referred to as BACE1, ASP2, or memapsin 2 (3-7), whereas ␥-secretase seems to be due to both presenilin (PS)-dependent and PS-independent activities (8 -13).
A lot has been learned recently about the biology of BACE1. BACE1 is a type I transmembrane protein that undergoes several post-transductional modifications. BACE1 is N-glycosylated in its ectodomain where six cysteine residues form intramolecular disulfide bridges (14). Immature glycosylated BACE1 rapidly undergoes a furin-mediated removal of its prodomain (15) to generate the 70-kDa mature form of the enzyme. Mature BACE1 also undergoes palmitoylation at three cysteine residues and sulfation of N-glycosylated moieties (16). Of most importance, BACE1 depletion leads to full impairment of A␤ production and is apparently safe (17,18). Thus, invalidated embryos are viable whereas adult BACE1 Ϫ/Ϫ mice are fertile. Although recent data indicate that BACE1 could have other substrates besides ␤APP, which could be involved in important myelination processes (19,20), the above data identify BACE1 as a suitable therapeutic target.
PS belong to the high molecular weight ␥-secretase complex, which also includes Aph-1, Pen-2, and nicastrin (21,22). Besides their contribution to the intramembranous cleavage of ␤APP, PS participate in the proteolytic activation/degradation of a series of other substrates involved in many vital functions (23), thereby explaining the apparent remarkable pleiotropy of these proteins (24). Among other functions, PS contribute to cellular adhesion, cell death control, and cell signaling (23,25,26). Another remarkable property of PS concerns their ability to act as a molecular chaperone as was demonstrated for TrkB (27), telencephalin (28), and N-cadherin (29).
Of most interest, Hébert et al. (30) demonstrated by yeast two-hybrid assay, co-immunoprecipitation procedure, and pulldown experiments that BACE1 and PS1 physically interacted. More recently, Kuzuya et al. (31) established that PS1 could modulate BACE1 maturation. The above data agreed well with the demonstration that ␤APP, PS1, and BACE1 co-localized in membrane vesicles that undergo axonal transport by a kinesin-dependent mechanism (32). However, the functional cross-talk between the two partners and the mechanisms by which PS could control BACE1 activity remained elusive. Here we show that PS control the promoter transactivation of BACE1, and we establish that this PS-dependent effect is related to its associated ␥-secretase activity. Furthermore, we demonstrate that BACE1 activity and promoter transactivation were directly linked to exacerbated production of A␤ and unrelated to AICD generation. Finally, we establish that A␤-mediated transactivation of BACE1 promoter involves the NFB pathway.
Transactivation of BACE1 Promoter-Empty vector or cDNA encoding rat BACE1 promoter or deletion constructs in-frame with luciferase (44) were co-transfected with ␤-galactosidase (to normalize transfection efficiencies). HEK293 cells and fibroblasts devoid of presenilins were cultured in 12-well dishes or 60-mm-diameter dishes and transfected with 2 or 8 g of the total amount of cDNA, respectively. For A␤ transfection experiments, cells were transfected with a total amount of 3 g of cDNA encoding rat BACE1 promoter, ␤-galactosidase, and either empty vector or constructs harboring A␤-(1-42), A␤-(42-1), or A␤-(1-40) (45). Thirty hours after transfection, cells were harvested with phosphate-buffered saline/EDTA (5 mM), pelleted by centrifugation (1000 ϫ g, 5min), lysed with 50 l of lysis buffer (luciferase kit Promega), centrifuged for 5 min at 2000 rpm, and then luciferase activity was measured with 10 l of supernatant and 50 l of luciferase assay (Promega).
Effect of ␥-Secretase Inhibitor Treatment on BACE1 Promoter Transactivation and Activity-Cells were treated for 16 h with various concentrations of the ␥-secretase inhibitor DFK167 (46) or corresponding amounts of Me 2 SO (all DFK167 stocks were made at 10 mM in 100% Me 2 SO). Cells were harvested with phosphate-buffered saline/EDTA (5 mM), pelleted by centrifugation (1000 ϫ g, 5 min), and lysed with Tris-HCl (10 mM, pH 7.5) or with 50 l of lysis buffer (luciferase kit Promega) to measure BACE1 activity and rat BACE1 promoter activity, respectively.
Effect of Exogenous 〈␤ on BACE1 Promoter Transactivation-Twenty four hours after transfection of rat BACE1 promoter, medium was replaced with Opti-MEM medium (Sigma) containing fetal bovine serum (2%) and complemented with phosphoramidon (10 M) to prevent A␤ degradation. The cells were then treated for 48 h with various concentrations of synthetic A␤42 or A␤40 (Bachem).
Analysis of A␤40 Production-Cells were allowed to secrete A␤ in Opti-MEM medium (Sigma) for 8 or 16 h in the presence of phosphoramidon (10 M) (to prevent the degradation of secreted A␤), in the absence or in the presence of DFK167 (50 M), collected, supplemented with RIPA buffer (10 mM Tris-HCl, pH 8, EDTA 5 mM, NaCl 150 mM), and then incubated overnight with a 100-fold dilution of FCA18 or FCA3340 (47). A␤40 was immunoprecipitated with protein A-Sepharose, analyzed onto a 16.5% Tris/Tricine gel, Western blotted, and revealed with 6E10 as described previously (36).
Stastistical Analysis-Statistical analysis was performed with Prism software (Graphpad, San Diego) using the Student-Newman-Keul's multiple comparison test for one-way analysis of variance or unpaired t test for pairwise comparison.

Presenilin 1 and Presenilin 2 Deficiency
Affects BACE1 Maturation, Activity, and Promoter Transactivation-We examined the influence of presenilins on the expression of BACE1. In wild-type fibroblasts, pro-BACE1 and mature BACE1 were detected at their previously reported molecular weights ( Fig.  1A) (48). Interestingly, the depletion of both PS1 and PS2 lowers the level of mature BACE1 (44.4 Ϯ 10.1% compared with control, n ϭ 6, p Ͻ 0.01) without affecting pro-BACE1 (Fig. 1, B and C). In PS-deficient fibroblasts, an additional band harboring BACE-1-like immunoreactivity was detected above the mature form of the enzyme (Fig. 1A). The identity of this protein is still unknown, but we can rule out the possibility that this protein derives from abnormal prohormone convertase proc-essing that would have occurred in the absence of PS because this protein remains totally insensitive to the convertase inhibitor ␣ 1 -antitrypsin Portland variant (49, 50 and data not shown).
We examined whether PS deficiency-associated reduction of mature BACE1 expression was associated with a decrease of BACE1 enzymatic activity. Thus, the depletion of endogenous PS triggers a significant reduction of BACE1 activity (27.4 Ϯ 4.8% compared with control, n ϭ 6, p Ͻ 0.0001; see Fig. 1D). The similar decrease in both mature BACE1 expression and activity suggests that the above-described unidentified protein apparently does not harbor catalytic properties. Of most interest, we establish that fibroblasts devoid of PS display drastically lower BACE1 promoter transactivation (77.4 Ϯ 7.4% compared with control, n ϭ 9, p Ͻ 0.0001; see Fig. 1E). Overall, the above data suggested that PS modulate BACE1 activity by apparently controlling this protease at both transcriptional and post-transcriptional levels.
Presenilin-dependent ␥-Secretase Activity Is Involved in the Control of BACE1-We examined whether the modulation of BACE1 activity could be linked to the enzymatic activity displayed by the PS-dependent ␥-secretase complex. Two distinct sets of experiments suggest that it is indeed the case. First, we assessed the influence of a double mutation thought to yield a catalytically inactive PS1 (D257A/D385A-PS1, AA-PS1 (51)). Unlike wild-type PS1, which slightly increases BACE1 activity (Fig. 2B), AA-PS1 overexpression ( Fig. 2A) reduces BACE1 activity (53.4 Ϯ 8.2% compared with WT-PS1, n ϭ 3, p Ͻ 0.01; Fig. 2B). Second, we assessed the influence of DFK167, a ␥-secretase inhibitor that physically interacts with PS (52), on BACE1 activity. As a matter of controls, we first show that DFK167 does not directly interfere with BACE1 activity (Fig.  3A) but indeed abolishes A␤ production in HEK293 cells expressing Swedish mutated ␤APP (Fig. 3B). Clearly, DFK167 reduces cellular BACE1 activity in wild-type fibroblasts (40.9 Ϯ 3.6% for 10 M DFK167 and 46.9 Ϯ 4% for 50 M DFK167 compared with control untreated cells, n ϭ 16, p Ͻ 0.001; Fig.  3C) as well as BACE1 expression (Fig. 3C). However, the ␥-secretase inhibitor neither alters BACE1 expression nor activity in PS-deficient cells (Fig. 3D). Altogether, both mutational and pharmacological data indicate that PS-induced modulation of BACE1 activity was apparently linked to PS-associated ␥-secretase activity.
AICDs Do Not Modulate BACE1 Activity, in Vitro and in Vivo-Because the modulation of BACE1 was apparently linked to PS-dependent catalytic events, we postulated that the control of BACE1 should involve catabolites generated by PSdependent ␥-secretase-associated processes and harboring transcription factor properties. In this context, we examined whether the C-terminal products of ␤APP hydrolysis could contribute to the modulation of BACE1 activity. Thus, these fragments called AICD (APP intracellular domain) have been shown to be derived from DFK167-sensitive ␥and ⑀-secretase proteolytic cleavages (53,54) and to translocate into the nucleus (55,56) where they participate in the transactivation of several promoters (41,(57)(58)(59)(60). AICDC59 (␥-secretase-derived) and AICDC50 (⑀-secretase-derived) cDNAs were co-transfected with Fe65 and Tip60, two proteins reported to stabilize FIGURE 1. Presenilin deficiency affects BACE1 expression, activity, and promoter transactivation. A and B, endogenous BACE1 expression was established by Western blot in wild-type (PS ϩ/ϩ ) and presenilin-deficient (PS Ϫ/Ϫ ) fibroblasts as described under "Materials and Methods." Note that an additional band above mature BACE1 was detectable only in PS Ϫ/Ϫ fibroblasts. Bars in B and C correspond to the densitometric analysis of mature and immature BACE1 immunoreactivities, respectively, correspond to the densitometric analysis of mature BACE1 immunoreactivity expressed as the control expression observed in PS ϩ/ϩ (taken as 100), and are the means Ϯ S.E. of six experiments. D, BACE1 activity was fluorimetrically monitored as described under "Materials and Methods" in homogenates of the indicated cell lines. Bars correspond to BACE1 activity expressed as percent of that observed in PS ϩ/ϩ (control taken as 100) and are the means Ϯ S.E. of six independent determinations. E, PS ϩ/ϩ and PS Ϫ/Ϫ fibroblasts were transiently transfected with rat BACE1 promoter-luciferase cDNA together with ␤-galactosidase cDNA to normalize data for transfection efficiency as described under "Materials and Methods." Thirty hours after transfection, luciferase reporter activity was measured as described under "Materials and Methods." Bars correspond to luciferase activity expressed as percent of that observed in PS ϩ/ϩ and are the means Ϯ S.E. of three independent determinations. *, p Ͻ 0.01; **, p Ͻ 0.0001. ns, not statistically significant.
AICD and favor their nuclear translocation (57). In these conditions, AICDs are readily detectable in both PS ϩ/ϩ and PS Ϫ/Ϫ fibroblasts (Fig. 4A). As shown in Fig. 1, in this independent set of experiments, both BACE1 expression (31% reduction when compared with control) and activity (compare empty bars) are reduced in PS Ϫ/Ϫ fibroblasts (Fig. 4B). However, AICDC59 and AICDC50 are unable to affect BACE1 activity in both wild-type and PS-deficient fibroblasts (Fig. 4B) and do not modify BACE1 promoter transactivation (Fig. 5). It should be noted that the transfection of cDNA coding for E-CTF2, ALID1, and ALID2, the ␥-secretase-derived intracellular fragments of E-cadherin, APLP1, and APLP2, respectively, did not increase BACE1 expression, activity, and promoter transactivation (data not shown). These data were confirmed in vivo. Thus, we examined the expression and activity of BACE1 in single Fe65 and double Fe65/AICD transgenic mice brains. As described previously (61), AICD-like immunoreactivity is clearly observed in double transgenic mice brain (Fig. 6A), whereas endogenous levels of cerebral AICDs are poorly detectable, in agreement with the low catabolic stability reported for these fragments (62). However, the overexpression of AICD does not modify BACE1 expression (Fig. 6, B and C) and activity (Fig. 6D) in these transgenic mice brains.
APP⑀-associated and DFK167-sensitive Modulation of BACE1 Activity-In search of another plausible ␥-secretasederived product able to modulate BACE1 but distinct of AICDs, we examined the influence of APP⑀ on the activity of BACE1. APP⑀ corresponds to the N-terminal ␤APP fragment generated by ⑀-secretase and therefore lacks the C-terminal AICD domain (36). We established previously that APP⑀ undergoes both ␤-secretase and DFK167-sensitive ␥-secretase cleavages (36), yielding the A␤. Therefore, cells overexpressing APP⑀ (Fig. 7A) could be seen as a theoretically useful model to examine whether the PS-dependent ␥-secretase-mediated control of BACE1 was linked to A␤ production. First, we show that the transient transfection of either wild-type ␤APP or APP⑀ cDNA in HEK293 cells, which enhances A␤ recovery (Fig. 7A), triggers a similar increase in BACE1 promoter transactivation (Fig. 7B). Second, in stably transfected APP⑀-expressing HEK293 cells (Fig. 7C), we observed a DFK167-sensitive reduction of BACE1 activity (29.8 Ϯ 7.9% compared with untreated APP⑀ cells, n ϭ 12, p Ͻ 0.001; Fig. 7D). Overall, these data indicate that A␤ but not AICD contributes to the modulation of BACE1 activity.
A␤ Controls BACE1 Promoter Transactivation but Does Not Affect ␣-Secretase Activity-We examined whether A␤-associated control of BACE1 activity could be directly linked to the level of A␤. In this context, we took advantage of cells lines overexpressing wild-type (WT-APP) or Swedish-mutated  (Swe-APP) ␤APP (Fig. 8A). It has been consistently established that the Swedish double mutation increases the production of A␤ peptides (63)(64)(65). This was confirmed here (Fig. 8A). Thus, by means of FCA18 antibody that recognizes genuine A␤ begin-  . AICD expression does not affect rat BACE1 promoter-luciferase activity. HEK293 cells were transiently transfected with rat BACE1 promoterluciferase cDNA together with either empty pcDNA3, AICDC50 (C50), or AICDC59 (C59) cDNA. Thirty hours after transfection, luciferase reporter activity was measured as described under "Materials and Methods." Bars correspond to luciferase activity expressed as percent of that observed in mocktransfected HEK293 cells (taken as 100) and are the means Ϯ S.E. of 11 independent determinations. ns, not statistically significant. FIGURE 6. Endogenous BACE1 expression and activity in Fe65-AICD transgenic mice brains. A, Western blot analyses of AICD and Fe65 expressions in Fe65 (three independent animals) or AICD-Fe65 doubly transgenic (two independent animals) mice brain homogenates were carried out as described under "Materials and Methods." BACE1 expression (B and C) and activity (D) of indicated brains homogenates were monitored as described under "Materials and Methods." Bars correspond to BACE1 expression (C) or activity (D) expressed as percent of that observed in Fe65 transgenic mice brain (taken as 100) and are the means Ϯ S.E. of three independent determinations. ns, not statistically significant. FIGURE 7. Influence of APP⑀ on BACE1 activity and promoter transactivation. A, mock-transfected HEK293 cells were transiently transfected with empty vector (pcDNA3), wild-type ␤APP (Wt-APP), or APP⑀ cDNAs. Secreted A␤ was monitored after immunoprecipitation with FC3340 antibody and analyzed after 16.5% Tris/Tricine electrophoresis and Western blot with WO2 antibody as described under "Materials and Methods." Std indicates synthetic A␤40 run in the same conditions. ␤APP and APP⑀ were detected with 2H3 antibody as described under "Materials and Methods." B, HEK293 cells were transiently transfected with rat BACE1 promoter-luciferase cDNA with empty vector (pcDNA3), WT-␤APP (Wt-APP), or APP⑀ cDNAs. Forty eight hours after transfection, luciferase reporter activity was measured as described under "Materials and Methods." Bars correspond to luciferase activity expressed as percent of that observed in empty vector-transfected HEK293 cells (taken as 100) and are the means Ϯ S.E. of eight independent determinations. C, Western blot analyses of APP-like immunoreactivity and BACE1 activity in mock-transfected (Ct) HEK293 cells or in HEK293 cells stably overexpressing WT-APP or APP⑀ were performed by Western blot using either BR188 or 22C11 antibodies as described under "Materials and Methods." D, APP⑀-expressing cells were incubated for 16 h with DFK167 (50 M) or corresponding amount of Me 2 SO (DMSO), and then BACE1 activity was fluorimetrically assayed in cell homogenates as described under "Materials and Methods." Bars in D correspond to BACE1 activity expressed as percent of that observed in control (Me 2 SO-treated) APP⑀-expressing cells (taken as 100) and are the means Ϯ S.E. of 12 independent determinations. *, p Ͻ 0.01; **, p Ͻ 0.001; ns, not statistically significant.
ning at Asp-1 residue (47), we show that Swe-APP cells produce more A␤ than WT-APP cells (Fig. 8A). As expected, this production was drastically reduced by DFK167 (Fig. 8A, dark). It should be noted here that prolonged exposure of the gels allows revealing DFK167-sensitive production of A␤ in mock-transfected cells (Fig. 8A). Interestingly, WT-APP expression increases BACE1 promoter transactivation (compare Ct and WT-APP in Fig. 8B), a phenotype that was further accentuated in cells overexpressing Swe-APP (compare WT-APP and Swe-APP in Fig. 8B). DFK167 significantly reduced but did not abolish APP-associated increase of BACE1 promoter transactivation (Fig. 8B).
APP⑀-, WT-APP-, and Swe-APP-associated modulation of BACE1 promoter transactivation strongly suggested that A␤ was indeed the molecular determinant responsible for this phenotype. Two direct lines of evidence directly reinforced this view. First, exogenous application of A␤42 (Fig. 9A) but not A␤40 (Fig. 9B) triggers the transactivation of the BACE1 promoter. Second, the transient transfection of DNA constructs harboring the A␤-(1-42) sequence (45) also increases BACE1 promoter transactivation, whereas vectors coding for the "reverse peptide" A␤-(42-1) or for A␤-(1-40) remained biologically inert on this paradigm (Fig. 9C). Therefore, we can definitely conclude that A␤42 acts as a modulator of BACE1 transactivation.
To establish whether A␤ could also influence other activities involved in the processing of ␤APP, we examined the influence of A␤ levels on the ␣-secretase pathway. We show that ADAM10 expression (Fig. 10A) and whole ␣-secretase-like activity (Fig. 10, B and C) remained unaffected by the overexpression of either WT-or Swe-APP.
A␤-induced Regulation of BACE1 Promoter Transactivation Is NFB-dependent-Several lines of evidence indicated that BACE1 promoter activity was regulated by the transcription factor NFB (66). It was therefore tempting to speculate on the involvement of the NFB pathway in the A␤-associated control of BACE1. NFB could be indirectly targeted by BMS345541, a selective blocker of IB kinases 1 and 2 (67), thereby leading to NFB inactivation. However, we establish, in agreement with a previous study documenting the influence of a series of NFB inhibitors (68), that BMS345541 indeed blocks NFB transcriptional activity (our data not shown) and also inhibits the production of both A␤40 and A␤42 by Swe-APP-expressing cells (Fig. 11A). Therefore, the use of BMS345541 on cells would not allow discriminating between a direct effect of the inhibitor on NFB-dependent transactivation of BACE1 promoter and an indirect effect because of an upstream inhibition of A␤ production. To circumvent this problem, we examined the effect of BMS345541 on BACE1 modulation by exogenous application  of A␤42. As described in Fig. 9A, in this independent set of experiments, A␤42 triggered an increased BACE1 promoter transactivation (Fig. 11B). Interestingly, this appears to be prevented by BMS345541 (Fig. 11B). Accordingly, BMS345541 drastically reduces the increase of BACE1 promoter transactivation triggered by WT-APP overexpression (Fig. 11C). These data strongly suggest that BMS345541 triggers its effect downstream to A␤ production and that the NFB pathway mediated the A␤-induced modification of BACE1 transactivation. To reinforce this hypothesis, we examined the ability of cell lines expressing various levels of A␤ to transactivate promoter constructs of BACE1, some of which were deleted of NFB-responsive elements (Fig. 12A). As expected, the overexpression of WT-APP or Swe-APP increases the transactivation of the fulllength promoter of BACE1 (Fig. 12B). Interestingly, the 3Ј deletion of a domain, including the NFB-responsive element (BPR-Nco), drastically reduces BACE1 activation (Fig. 12B), whereas the shortest 3Ј construct (BPR-Avr) and 5Ј-deleted construct (1-2 Del) were not transactivated in these cells (Fig.  12B). It should be noted that the transcription of BPR-Nco was also reduced in mock-transfected cells. Together with the fact that BMS345541 also reduces BACE1 promoter transactivation in mock-transfected cells (Fig. 11, B and C), this could indicate that endogenous NFB participates in the control of BACE1 besides its role in A␤-mediated phenotype. Overall, our data firmly suggest that A␤-mediated control of BACE1 involves NFB-associated transactivation of the BACE1 promoter.   , 18 h, C), or equal amounts of Me 2 SO, and then luciferase reporter activity was measured as described under "Materials and Methods." Bars in B and C correspond to luciferase activity expressed as percent of that observed in untreated cells (taken as 100) and are the means Ϯ S.E. of 12 or 6 independent determinations, respectively. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. Endogenous A␤ Does Not Modify BACE1 Expression-It is noteworthy that DFK167 did not modify BACE1 promoter transactivation in Mock-transfected HEK293 cells (Fig. 8B). This suggested that the low levels of endogenous A␤ in these cells (Fig. 8A) would not be sufficient to alter BACE1 transactivation. To confirm this hypothesis, we examined the status of BACE1 expression and activity in double knock-out ␤APP/ APLP2 fibroblasts. These cells were chosen because they do not produce A␤, AICD, or AICD-like (ALIDs) fragments. We show that the lack of these proteins does not alter BACE1 expression (Fig. 13, A and B) or activity (Fig. 13C). Overall, our data indicate BACE1 is not controlled by endogenous AICD or A␤ production and that the modulation of BACE1 could be triggered by elevated levels of A␤.

DISCUSSION
One of the canonical hallmarks in Alzheimer disease-affected brain is the widespread distribution of extracellular deposits called senile plaques. A network of evidence based on biochemical, biological, anatomical, and genetic approaches indicates that amyloid ␤-peptides, the main component of senile plaques, play a central role in AD etiology (69). As a key argument to support this hypothesis is the observation that most, if not all, mutations responsible for aggressive and early onset AD cases have in common the ability to trigger modifications of the levels, the nature, or the biophysical properties of the A␤ peptides generated (for reviews see Refs. 2,24). The definitive identification of the cellular and molecular mechanisms underlying such alterations is still lacking. One of the likely hypotheses would be that ␥-secretase activity could be enhanced in AD, thereby increasing A␤ production. However, the demonstration of an up-regulation of ␥-secretase associated with sporadic AD remains to be established. Furthermore, whether the alterations triggered by pathogenic mutations also account for those responsible for sporadic AD cases remains doubtful. It is more generally admitted that mutations modify ␥-secretase-mediated cleavage specificity rather than alter its catalytic efficiency. Thus, mutations located close to the C terminus of A␤ apparently modify the ratio of A␤X-42 over A␤X-40 rather than the total A␤ load. Conversely, the Swedish mutation located downstream to the A␤ N terminus triggers an overload of A␤40. Therefore, although the overall alteration of A␤ production could be seen as a common denominator between sporadic and genetic cases of AD, close inspection reveals specific molecular dysfunctions.
Another way of modifying cellular A␤ homeostasis could occur downstream to A␤ production. Neprilysin (NEP) and insulin-degrading enzyme are the two major putative metalloproteases involved in A␤ catabolism (70 -72). NEP was shown to display higher catalytic efficiency toward A␤40 versus A␤42 (73). Therefore, early deterioration observed in those cases of familial AD where A␤42 production is more specifically altered could be due to NEP-mediated exacerbation of the A␤42 over A␤40 ratio. On the other hand, it has been documented that NEP and insulin-degrading enzyme display age-related decrease in their cerebral levels (74). This observation could account for the late increase in A␤ levels observed in sporadic cases of AD.
Here we show that PS deficiency decreases the activity and promoter transactivation of the ␤-secretase BACE1. As described previously (31), we show that part of the modulation of BACE1 activity could be accounted for by a PS-dependent defect in BACE1 maturation. However, two lines of data demonstrate that PS-dependent ␥-secretase activity could also contribute to the control of A␤ production. First, the overexpression of a catalytically inactive PS1 mutant lowers BACE1 activity. Second, the ␥-secretase inhibitor DFK167 lowers BACE1 expression and activity in wild-type but not in PS-deficient fibroblasts.
The above data prompted us to examine which putative product generated by PS-dependent ␥-secretase could harbor the potential of modulating BACE1. We first focused on AICD, the ␥-secretase-derived intracellular fragment of ␤APP because this product displays transcription factors properties that could therefore account for PS-dependent enhancement of BACE1 promoter transactivation. Even in conditions where AICD translocation and stability are favored (i.e. after co-expression with Tip60 and Fe65), we were unable to demonstrate AICD-associated increase in BACE1 activity in wild-type and PS-deficient fibroblasts. That AICD was biologically inert on BACE1 was confirmed in vivo, in double transgenic mice overexpressing both AICD and Fe65 (38), a model where neprilysin (61) and p53 4 had been shown to be up-regulated. These data were corroborated by our observation that transient transfection of wild-type ␤APP and APP⑀, the ⑀-secretase-derived N-terminal fragment of ␤APP lacking AICD, both trigger similar increase of BACE1 promoter transactivation. Several lines of data support the view that A␤ itself accounts for the observed ␥-secretase-mediated effect on BACE1. First, BACE1 activity is inhibited by DFK167 in APP⑀-expressing cells. Second, the comparison of stably transfected HEK293 cells overexpressing either wild-type or Swedish mutated ␤APP indicates that the mutation that further enhances A␤ production also potentiates the ␤APP-associated increase of BACE1 promoter transactivation. Third, exogenous application of A␤42 but not A␤40 increases both promoter transactivation of BACE1. Fourth, transient transfection of cDNA encoding A␤-(1-42) but not A␤-(42-1) or A␤-(1-40) also trigger increased BACE1 promoter activity.
The above observation is of major importance because this suggests that the control of BACE1 is intrinsically linked to the level of A␤42. Indeed, several lines of evidence indicate that normal physiological amounts of A␤ do not influence BACE1. First, although mock-transfected HEK293 produce low but measurable DFK167-sensitive A␤, this inhibitor did not modulate BACE1 promoter transactivation in these cells. Second, BACE1 expression and activity were similar in wild-type and ␤APP/APLP2-deficient fibroblasts, which do not produce endogenous levels of A␤. The above observations suggest that the control of BACE1 by A␤ only takes place when endogenous concentrations of the peptide increase significantly. This also suggests that the decrease of BACE1 activity observed in PSdeficient fibroblasts was not because of endogenous A␤ depletion but more likely attributable to an AICD-independent ␥-secretase-mediated product that remains to be identified. One can postulate that this apparent concentration-dependent A␤-mediated effect is because of the propensity of this peptide to aggregate when reaching the threshold of its solubility because several studies have clearly demonstrated that A␤ fibrils are the toxic species (75). It is tempting to speculate that this could reflect the dysfunction occurring in sporadic cases of AD.
The question arises as to whether A␤ either directly modulates BACE1 promoter transactivation, indirectly targets an intermediate cellular modulator of BACE1 transactivation, or if both direct and indirect mechanisms could take place. We have identified NFB as a molecular intermediate involved in the A␤-mediated control of BACE1. First, the inhibitor of IB kinase that blocks NFB transcriptional activity fully reverses the A␤42-induced increase of BACE1 promoter transactivation. Second, deletions that disrupt an NFB-responsive element in the promoter of BACE1 abolish the ability of WT-APP and Swe-APP cells to control BACE1 promoter transactivation. These data agree well with a previous study showing that BACE1 promoter transactivation could be regulated by NFB in neuronal cells (66).
It is noteworthy that the dose of exogenous A␤ able to trigger BACE1 promoter activation was a priori high but in the range of those used in a previous study aimed at comparing BACE1 promoter activity in neurons and glial cells (66). These high doses likely do not reflect the real effective dose because A␤ treatment requires long incubation times (see "Materials and Methods") during which several parameters, including degradation processes, bioavailability, and intracellular accessibility, can not be controlled. Furthermore, several lines of evidence indicate that the effect triggered by exogenous A␤42 cannot be seen as artifactual. First, the same dose of A␤40 remains biologically inert on BACE1 promoter activation. Second, in transfection experiments similar to those described in Ref. 45, A␤-(1-42) but not A␤-(42-1) or A␤-(1-40) mimic the phenotype triggered by exogenous A␤- , although the A␤ levels were drastically lower in transfection experiments. Third, both exogenous application of A␤42 and expression of wild-type or Swedish-mutated APP activate BACE1 promoter in a BMS345541-sensitive manner.
It should be noted that the disruption of the NFB-responsive element did not totally reverse WT-APP-and Swe-APPassociated increase of BACE1 promoter activity (see Fig. 12B), suggesting that additional mechanisms may take place. Thus, BACE1 promoter harbors a canonical CREBIO site that appears to act as a repressor of BACE1 transactivation because mutation at this site increases reporter protein expression (76). It is noteworthy that A␤ has been previously shown to down-regulate CREB-induced signaling in cultured neurons (77), apparently by blocking the nuclear translocation of phosphorylated CREB (78). Of most interest, this process appears dependent on high intracellular A␤ concentrations (78). Therefore, A␤-induced lowering of CREB-associated pathway could ultimately lead to abolish CREB-induced repression of BACE1, thereby increasing transcription of its promoter.
Besides such possible indirect modulation of BACE1 by A␤, one should also underline that several studies reported on a genuine role of A␤ as transcription factor. Thus, Ohyagi et al. (45) demonstrated that A␤ directly binds to elements of the p53 promoter. Interestingly, the A␤ binding sequence found in the p53 promoter resembles some sequence domains found on the BACE1 promoter. In agreement with this observation, Maloney et al. (79) showed that these sequences bind A␤ in vitro and influence BACE promoter constructs transactivation in cells.
Overall, our work identifies A␤42 as a regulator of BACE1 and suggests that only abnormally high levels of this pathogenic species of A␤ could influence BACE1 promoter transactivation, at least in part via the NFB pathway. This vicious circle by which pathological production of A␤42 feeds its own production could contribute to both sporadic and genetic cases of Alzheimer disease.