Antagonistic Effects of β-Site Amyloid Precursor Protein-cleaving Enzymes 1 and 2 on β-Amyloid Peptide Production in Cells

The deposition of extracellular β-amyloid peptide (Aβ) in the brain is a pathologic feature of Alzheimer's disease. The β-site amyloid precursor protein cleaving enzyme 1 (BACE1), an integral membrane aspartyl protease responsible for cleavage of amyloid precursor protein (APP) at the β-site, promotes Aβ production. A second integral membrane aspartyl protease related to BACE1, referred to as β-site amyloid precursor protein cleaving enzyme 2 (BACE2) has also been demonstrated to cleave APP at the β-cleavage site in transfected cells. The role of endogenous BACE2 in Aβ production remains unresolved. We investigated the role of endogenous BACE2 in Aβ production in cells by selective inactivation of its transcripts using RNA interference. We are able to reduce steady state levels for mRNA for each enzyme by >85%, and protein amounts by 88–94% in cells. Selective inactivation of BACE1 by RNA interference results in decreased β-cleaved secreted APP and Aβ peptide secretion from cells, as expected. Selective inactivation of BACE2 by RNAi results in increased β-cleaved secreted APP and Aβ peptide secretion from cells. Simultaneous targeting of both enzymes by RNA interference does not have any net effect on Aβ released from cells. Our observations of changes in APP metabolism and Aβ are consistent with a role of BACE2 in suppressing Aβ production in cells that co-express both enzymes.

The production and deposition of insoluble A␤ 1 peptide in the brain results in the hallmark pathological feature of Alzheimer's disease (1). The cellular enzymes responsible for production of A␤ peptide are molecular targets for therapeutic intervention in Alzheimer's disease (2,3). BACE1 (␤-site APP cleaving enzyme, ASP2, Memapsin 2, ␤-secretase), the enzyme responsible for cleavage of the amyloid precursor protein (APP) resulting in the amino terminus of A␤ peptide, is a novel integral membrane aspartyl protease (4 -8). Cellular antisense (6,7) and knock-out mouse models (9 -11) have unambiguously confirmed the role of BACE1 in promoting A␤ production. A second integral membrane aspartyl protease related to BACE1, referred to as BACE2 (ASP1, Memapsin 1) has also been demonstrated to cleave APP at the ␤-cleavage site (7,8,(12)(13)(14)(15)(16). Both enzymes also cleave APP at a second site within the A␤ region. This second cleavage site for BACE1 is between A␤ residues 10 and 11 and between A␤ residues 19 and 20 for BACE2. Cleavage of APP by BACE1 at either site is amyloidogenic, whereas the second cleavage site for BACE2 on APP precludes formation of A␤.
Overexpression of BACE2 in transfected cells produces intracellular carboxyl-terminal fragments (CTFs), as well as release of ␤-cleaved secreted APP (sAPP␤) from cells, consistent with cleavage of APP at the ␤-site by this enzyme (14, 16 -19). We have also observed decreased A␤ production from cells transfected with BACE2 (not shown), consistent with published observations (14, 16 -19). Hence, the reduction in secreted A␤ occurs in spite of the concomitant increased secretion of sAPP␤ from cells overexpressing BACE2. The decrease in A␤ production from BACE2 transfected cells has been attributed to the second cleavage site for BACE2 on APP (14, 16 -19). The observations on the cleavage specificity of BACE2 on APP substrate are derived from transfected cells that overexpress the enzyme. The role of BACE2 in A␤ production in cells expressing endogenous levels of the enzyme remains an unanswered question. Many tissues (including brain) and cell types co-express BACE1 and BACE2 mRNA (8,13,14,16). The therapeutic relevance of inhibiting one, the other, or both enzymes for A␤ production in cells/tissues that co-express the two enzymes also remains an unanswered question.
RNA interference (RNAi) is the process whereby doublestranded RNA mediates the sequence specific destruction of its cognate mRNA (20 -22). RNAi is triggered by 21-23-nucleotide synthetic double-stranded RNAs termed small interfering RNA (siRNAs, we use "oligonucleotide" synonymously with siRNA in this report for convenience) in mammalian cells (23)(24)(25). We report here the selective knock-down of endogenous BACE1 or BACE2 message as well as protein by RNAi in HEK293 cell lines stably overexpressing APP695wt (wild type allele) or APPsw, the familial Alzheimer's disease allele of APP (the K595M,N596L Swedish double mutant). The selective knockdown of either enzyme leads to complementary alterations in APP metabolites and A␤ peptide secreted from cells. Our studies suggest that BACE2 inhibition elevates secretion of A␤ in cells co-expressing BACE1 and BACE2 and are of significance for amyloid-based therapeutics targeting these enzymes.

EXPERIMENTAL PROCEDURES
Quantitative PCR-For RNA analysis, total RNA was prepared from cells at various time points post-transfection using Qiagen RNAeasy * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. total RNA extraction kits. Total RNA was treated with DNase and quantitated by absorbance before quantitative PCR assay. Absence of contaminating DNA in the final RNA preparations was confirmed by a PCR assay with input RNA as template, omitting the reverse transcriptase. Quantitative RT-PCR was performed using an ABI 7700 Sequence Detector. All of the samples were assayed in triplicate. BACE1 primers, 5Ј-CCCGAAAACGAATTGGCTTT and 3Ј-GCTGCCG-TCCTGAACTCATC, and probe, CTGTCAGCGCTTGCCATGTGCA; and BACE2 primers, 5Ј-CGTTTTCTCCATGCAGATGTGT and 3Ј-CCT-CCGTTGGTCCCAGATC, and probe, AGCCGGCTTGCCCGTTGCT. Standard curves for BACE1 and 2 in HEK293 cells treated with siR-NAs, as well as in primary brain cultures, were obtained using serial dilution of total RNA from untreated HEK293 cells. BACE message values per ng of input RNA were normalized to GAPDH message per ng of input RNA for each sample. BACE1 and 2 mRNA levels in cell lines were determined with standard curves using an in vitro synthesized RNA transcript as the standard.
RNAi-mediated Knock-down of BACE1 and BACE2 in HEK293 Cells-The synthetic siRNAs to BACE1 used in this study are: B1 sense, CAGGAUCUGAAAAUGGACUGtt; B1 anti-sense, CAGUCCAU-UUUCAGAUCCUGtt; B2 sense, UCUACGUUGUCUUUGAUCGGtt; B2 antisense, CCGAUCAAAGACAACGUAGAtt; B3 sense, CAGACGC-UCAACAUCCUGGUtt; and B3 antisense, ACCAGGAUGUUGAGCG-UCUGtt. The siRNAs to BACE2 used in this study are: C1 sense, GUGGGCAUGGGCGCACUGGtt; C1 antisense, CCAGUGCGCCCAU-GCCCACtt; C3 sense, ACAGAGAGGUCUAGCACAUtt; C3 antisense, AUGUGCUAGACCUCUCUGUtt; C4 sense, UUGAAUCAGAGAAUU-UCUUtt; and C4 antisense, AAGAAAUUCUCUGAUUCAAtt. These siRNA were synthesized and used essentially as described (23). HEK293 cells stably transfected with wild type APP695 (Amy5) (26) or APP695sw (293sw) were plated at a density of 200,000 cells/well in 24-well plates the morning of transfection and allowed to settle for 5 h prior to transfection. The Cells were transfected overnight with 200 ng of double-stranded RNA/well using Lipofectamine 2000® (Invitrogen) per the manufacturer's instructions. For quantitative analysis of APP metabolites and A␤ production in cells, all of the transfections were performed in triplicate, and the samples were assayed independently. For the experiment shown in Fig. 5, transfections with a combination of BACE1 and BACE2 oligonucleotides were conducted with 100 ng of each oligonucleotide, for a total of 200 ng of oligonucleotide/ transfection (standard conditions) or 20 ng of total oligonucleotide (0.1ϫ condition). No additional carrier nucleic acid was added for the 0.1ϫ transfections. BACE1 and 2 protein knock-down was confirmed by co-transfecting siRNAs with 0.8-g plasmid expression vectors encoding BACE1 or BACE2-Fc.
Western Blot Analysis and A␤ Measurement in Transfected Cells-For protein knock-down studies, the cell lysates and conditioned media were collected at 48 h post-transfection. The cells were lysed in phosphate-buffered saline with 0.5% Nonidet P-40 supplemented with complete protease inhibitor mixture (Roche Applied Science) and analyzed by Western blot with the antibodies as noted below. Protein concentration in lysates was determined by BCA assay (Pierce). Fresh conditioned medium was collected from cells for 4 h at 2 days post-transfection by replacing transfection medium, and 10 -20 l/lane (normalized for protein concentration of the cognate lysate) were loaded onto gels for Western blot analysis.
APP metabolites were detected with the antibodies previously described. Briefly, we used 8E5 for total sAPP (27), anti-6 for intracellular mature APP, as well as intracellular CTFs (28), and 192wt and 192sw for sAPP␤ species produced from APPwt and APPsw, respectively (29). The mouse monoclonal 7A6, preferentially recognizing the sAPP␣ neoepitope, was produced using peptide CEVHHQK (residues 607-612 from APP695/A␤ residues 11-16), coupled via the amino-terminal cysteine residue to sheep anti-mouse IgG as immunogen. Hybridomas from mice immunized with this peptide were screened for reactivity to peptide sequence CGGYEVHHQK (A␤ residues 10 -16). Hybridoma clone 7A6 showed stronger reactivity toward the A␤10 -16 peptide than to a peptide spanning the ␣-secretase cleavage site (A␤ residues 1-28). A quantitation of the difference in reactivity of 7A6 toward the two peptides, presented in the supplemental figure, shows that 7A6 reactivity toward sAPP in conditioned medium is competed 100% by the immunizing peptide at 100 g/ml but only partially by the overlapping peptide at 100 g/ml. Competition by 10 g/ml A␤ 11-16 is approximately equal to the level of competition seen with 100 g/ml A␤ 1-28. The result presented in the supplemental figure indicates that 7A6 binds with ϳ10 times greater preference to the sAPP␣ neo-epitope (ending at A␤ residue 16) than it does to a peptide which spans this site. Hence, although 7A6 is not expected to detect the sAPP species from BACE2 cleavage at A␤ residues 19 and 20 by virtue of the immunogen against which it was raised, it can potentially detect sAPP species extending beyond A␤ residue 16 because of a 10-fold weaker reactivity for residues between A␤11 and A␤16. Further characterization of monoclonal 7A6 included the ability to capture iodinated A␤10 -16 peptide.
Total A␤ was quantitated by sandwich ELISA using monoclonal antibody 266-coated microtiter plates followed by biotinylated 3D6 as secondary reporter (30). This assay measures all A␤ species starting at position 1 and extending beyond position 28 (i.e. A␤ 1-x). Statistical analysis of A␤ values was performed using StatView software package (SAS, Cary, NC). All of the samples were analyzed by analysis of variance, followed by post-hoc analysis using Fischer's protected least significant difference test to determine p values, comparing the means of the normalized A␤ values of each treatment group to the mean of the normalized A␤ value from the no treatment control group and also to the mean of the normalized A␤ value from the lamin and GFP control groups.
BACE1 protein expression was detected by rabbit polyclonal antisera 264 (immunized with a peptide corresponding to BACE1 residues 48 -66 of the nascent polypeptide). BACE2 protein was detected as a chimeric Fc construct (residues 1-465 of BACE2 extracellular domain fused to H-CH2-CH3 domains of human C␥1) followed by an horseradish peroxidase-conjugated goat anti-human IgG (Jackson Immunoresearch) to detect and quantify suppression of transfected BACE2. Parallel blots from equivalently loaded gels were probed with antibody to ␤-tubulin (Sigma) to confirm specificity of siRNAs for the targeted protein, as well as equal sample loading.

BACE1 and BACE2 mRNA Are Co-expressed in Cells-A
quantitative RT-PCR survey of different cell types reveals the co-expression of mRNA for BACE1 and BACE2 enzymes in a variety of cell lines (Fig. 1A). Of the brain cell types surveyed, we observed co-expression of the mRNA for both enzymes in astrocytes, whereas BACE1 was exclusively expressed in neurons and microglia (Fig. 1B). Oligodendrocyte precursor cells (purified from rat optic nerve) do not express detectable levels of either message (not shown). Although a discrepancy has been noted with regard to message and activity distribution of BACE1 (5, 6), we note that mRNA degradation is directly linked with protein expression, as well as with APP metabolite production, in our experiments using RNAi (see below) in HEK293 cells.
siRNAs to BACE1 and BACE2 Reduce Expression of the Cognate Protein and mRNA in a Sequence-specific Manner-A detailed comparison of residue specificity between 8 substrate subsites spanning P 4 -P 4 Ј revealed a very high degree of similarity between BACE1 and 2 (31,32), underscoring the challenge in obtaining selective small molecule inhibitors. Hence, we addressed the role of BACE1 and BACE2 in A␤ production in HEK293 cells (a cell type that expresses message for both enzymes, labeled as A293 in Fig. 1A) by selective degradation of mRNA for each enzyme using RNAi.
Synthetic siRNAs to BACE1 and BACE2 ("Experimental Procedures") were tested for activity in an overexpression paradigm by co-transfecting HEK293 cells with the doublestranded RNAs along with BACE1 or BACE2-Fc expression vectors. Expression of BACE1 and BACE2-Fc in lysates from transfected cells was assessed by Western blot analysis. The results, shown in Fig. 2, indicate that BACE1 expression is reduced specifically by oligonucleotide B3, and BACE2-Fc expression is reduced specifically by oligonucleotide C3. Expression of BACE1 is not affected by an irrelevant lamin A/C oligonucleotide (23) nor any of the BACE2 oligonucleotides tested. Likewise, BACE2 expression is not affected by any of the BACE1 oligonucleotides tested nor by irrelevant oligonucleotides targeting lamin A/C or GFP.
We further characterized the specificity as well as the kinetics of the active siRNAs B3 (anti-BACE1) and C3 (anti-BACE2) for degradation of the cognate mRNA in 293sw cells. Total RNA was isolated from 293sw cells (expressing endogenous levels of BACE1 and BACE2) at varying time points following transfection with oligonucleotides B3, C3, or B3ϩC3, and mRNA levels were measured by quantitative RT-PCR. Lamin siRNA was used as a specificity control. The results from this experiment, shown in Fig. 3, confirm the specificity and reveal the kinetics In cells transfected simultaneously with both oligonucleotides, suppression of BACE1 mRNA at time points earlier than 48 h is attenuated when compared with BACE2 mRNA suppression (Fig. 3, C and D). Furthermore, the overall potency of the RNAi response to each target appeared to be attenuated when cells were treated with both oligonucleotides (see legend to Fig. 3).
Extracts from cells transiently transfected with B3 siRNA plus BACE1 expression plasmid or C3 siRNA plus BACE2-Fc expression plasmid were analyzed by quantitative Western blots to determine the magnitude of protein suppression in an overexpression paradigm. Our results, shown in Fig. 4, indicate that BACE1 expression is reduced at least 8 -16-fold (Fig. 4, A  and B), whereas BACE2 is suppressed at least 16-fold (Fig. 4C). In summary, RNAi mediated suppression of BACE1 by oligonucleotide B3 and that of BACE2 by oligonucleotide C3 are highly sequence-specific. In addition, mRNA degradation correlates with a suppression of the cognate overexpressed protein.
BACE1 and BACE2 Knock-down Alters APP Metabolites Released from Cells-The consequence of knocking down endogenous BACE2 or BACE1 on APP metabolism and A␤ production was studied in stably transfected cells overexpressing APPsw (293sw cells; Fig. 5) or APP695wt (Amy5 cells; Fig. 6). The high level of APP expression in these cells facilitates detection of metabolites produced by the two enzymes.
Treatment with siRNAs targeting BACE1 or BACE2 does not affect the level of intracellular holo-APP or that of a control protein, ␤-tubulin (Fig. 5A). Intracellular CTFs of APP, signature ␣and ␤-secretase cleavage products are modulated in cells treated with B3 or C3 siRNAs (Fig. 5A, CTFs). Oligonucleotide B3 and C3 individually stimulate total sAPP (sAPPtot) secretion relative to untransfected (labeled 0) or lamin transfected cells (Fig. 5B). The stimulation of sAPPtot, and more notably sAPP␤, observed with the control lamin oligonucleotide is variable from experiment to experiment. Nonetheless, sAPP␤ stimulation by the lamin oligonucleotide is consistent with the "nonspecific" stimulation of A␤ secretion effected by this oligonucleotide (Fig. 7). We cannot explain the basis for this effect of the lamin oligonucleotide. BACE1 oligonucleotide B3 suppresses sAPP␤ released from cells (Fig. 5B), consistent with its activity in suppressing BACE1 message and protein as described above. The suppression of sAPP␤ from cells treated with B3 correlates with an increase in sAPP␣, suggesting that the increased sAPPtot secreted from these cells results principally from the ␣-secretase pathway.
In contrast with the effects of BACE1 suppression, BACE2 suppression effected by oligonucleotide C3 results in a significant increase in sAPP␤ secretion from 293sw cells (Fig. 5B). The increase in sAPP␤ levels effected by C3 is consistent with a shunting of intracellular APP into the BACE1 secretase pathway. In cells treated with BACE2 oligonucleotide C3, a greater proportion of the increase in sAPPtot is comprised of sAPP␤ than sAPP␣ (Fig. 5B). This observation suggests that reduction of the ␣-like cleavage after A␤ residues 19 and 20 attributed to BACE2 modulates sAPP␤ versus sAPP␣ production. The 7A6 antibody used to detect sAPP␣ in the blots shown in Figs. 5 and 6 shows ϳ10-fold higher specificity for A␤16 neo-epitope (resulting from an ␣-secretase cleavage of APP) than for an overlapping peptide (the supplemental figure and "Experimental Procedures"). Although 7A6 would not recognize the A␤19 and A␤20 neo-epitope produced by BACE2 cleavage of APP (see "Experimental Procedures"), it could potentially recognize the BACE2 cleavage product as a consequence of the 10-fold lower reactivity to the overlapping peptide. The effect on APP metabolites in 293sw cells co-transfected with the active oligonucleotides (B3ϩC3) is discussed separately below.
The effects of BACE1 oligonucleotide B3 and BACE2 oligonucleotide C3 on APP metabolites from Amy5 cells (expressing APPwt substrate) are shown in Fig. 6. Intracellular expression of holo-APP and ␤-tubulin are not effected by the oligonucleotides (Fig. 6A). Oligonucleotides B3 and C3 increased sAPPtot, as well as sAPP␣, compared with control lamin oligonucleotide (Fig. 6B). The increase in sAPPtot species effected by the knock-down of BACE2 with oligonucleotide C3 appears to be qualitatively greater in Amy5 cells than the increase in sAPPtot observed upon knock-down of BACE2 in 293sw cells (compare lanes C3 in Figs. 5B and 6B). In the absence of a quantitative assessment of secreted APP forms between the two cell types, it is difficult to ascertain the significance of this observation. The reduced sensitivity of the 192wt antibody precludes any conclusion regarding the effect of oligonucleotide B3 on sAPP␤ production from APPwt substrate compared with APPsw substrate (Fig. 6B, compare lamin versus B3). Consistent with the observations from 293sw cells, anti-BACE2 oligonucleotide C3 leads to an increase in all species of sAPP, particularly sAPP␤ (Fig. 6B). Fig.  7 (A and B, respectively). In 293sw cells (Fig. 7A), oligonucleotide B3 results in an ϳ60% decrease in A␤ secretion, whereas oligonucleotide C3 results in an ϳ125% increase in A␤ secre- tion (compared with the average A␤ level from all three control samples). The observed changes in A␤ production are consistent with the effect on mRNA, protein, and APP metabolites in 293sw cells produced by knock-down of BACE1 and BACE2. ELISA quantitation of total A␤ secreted in conditioned medium from APPwt expressing cells transfected with control oligonucleotides or B3 and C3 oligonucleotides is shown in Fig. 7B. Consistent with the observations in APPsw cells, A␤ secretion is significantly reduced in cells transfected with oligonucleotide B3 and stimulated in oligonucleotide C3 transfected cells. For both experiments shown in Fig. 7 (A and B), a "nonspecific" stimulation of A␤ secretion is observed from cells treated with either of the control siRNAs to lamin A/C or GFP. The elevation in A␤ produced by either control oligonucleotide reaches statistical significance. However, the changes in A␤ effected by knock-down of BACE1 or BACE2 are associated with much lower p values (see legend to Fig. 7).

BACE1 and BACE2 Knock-down Effect A␤ Secretion in an Opposite Manner-Quantitation of A␤ secreted into the conditioned medium from 293sw and Amy5 cells is presented in
In summary, RNAi-mediated knock-down of BACE1 is reflected in concomitant decreases in total A␤ peptide and sAPP␤ and increased levels of sAPPtot, as well as sAPP␣ secreted from cells. Conversely, RNAi-mediated knock-down of BACE2 is reflected in a concomitant increase in total A␤ peptide, sAPP␤, and sAPP␣ secretion from cells.
Double Knock-down of BACE1 and 2 in Cells following Co-transfection with B3 and C3-Co-transfection of cells with oligonucleotides B3ϩC3 results in degradation of the endogenous BACE1 as well as BACE2 mRNAs (Fig. 3, C and D). The kinetics of mRNA degradation with both oligonucleotides is similar to that with the individual oligonucleotides, whereas the magnitude of the effect appears attenuated (most notably with BACE1 at time points earlier than 48 h). Based on our results presented above, we assume that the decrease in mRNA in cells co-transfected with both oligonucleotides reflects a decrease in the steady state levels of endogenous BACE1 and BACE2 protein.
APP metabolism in cells co-transfected with both oligonucleotides under standard conditions (30 nM total concentration of oligonucleotide) leads to elevated secretion of all sAPP species, including sAPP␤ (Fig. 5B, compare B3ϩC3 triplicate lanes with B3 lanes). We considered the possibility that co-transfection led to "saturation" of the RNAi pathway. Hence, we repeated the experiment by reducing the dose of oligonucleotide in the transfection. A 10-fold lower dose of oligonucleotides (3 nM) transfected separately produced qualitatively similar effects on APP metabolism as the standard 30 nM transfection (Fig. 5B, compare B3 with 0.1ϫ B3 triplicate lanes and C3 with  0.1ϫ C3 triplicate lanes). Again, co-treatment with both oligonucleotides at a 0.1ϫ concentration produced similar results as at standard concentration: elevated production of all species of  20), to quantitate the magnitude of protein suppression in an overexpression paradigm. The serial dilutions of crude extract from cells co-transfected with siRNA to lamin (odd numbered lanes) or BACEspecific siRNA (even numbered lanes) were probed on Western blots with antibody to BACE1 or BACE2. Parallel blots from equivalently loaded gels were probed with antibody to tubulin to confirm equal loading between pairs of lanes. A, BACE1 suppression by oligonucleotide B3 (even numbered lanes) but not lamin oligonucleotide (odd numbered lanes) in 293sw cells overexpressing BACE1, as revealed by the anti-BACE1 antisera 264. The tubulin blot indicates that the B3treated samples (even numbered lanes) are slightly overloaded with respect to the lamin oligonucleotide-treated samples (odd numbered lanes). B, suppression of BACE1 in Amy5 (stable wt APP expressing) cells transiently transfected to overexpress BACE1. The lysates were probed as described for A. C, suppression of BACE2-Fc expression in Amy5 cells, as revealed by an anti-human Fc antibody. sAPP, including sAPP␤ (compare 0.1ϫ B3ϩC3 triplicate transfection lanes with 0.1ϫ B3 lanes in Fig. 5B).
ELISA quantitation of total A␤ secreted from cells in this experiment (measured in the same conditioned medium samples as shown in Western blots of Fig. 5B) is presented in Fig.  7C. As above, BACE2 knock-down by oligonucleotide C3 results in elevated A␤ secretion from cells. The samples treated with the anti-BACE2 oligonucleotide C3 consistently reached statistical significance at both doses of oligonucleotide tested. Lowering of A␤ secretion effected by the anti-BACE1 oligonucleotide B3 only produced a significant effect at the 0.1ϫ dose. None of the other treatments produced a significant effect on A␤ secretion from cells, including (in this experiment) cells transfected with the lamin oligonucleotide, as well as cells co-transfected with both oligonucleotides simultaneously. The lack of an effect on A␤ secretion from cells co-transfected with B3 and C3 oligonucleotides (B3ϩC3) stands in contrast to the effects on A␤ when each enzyme is targeted individually (see above). This observation suggests that simultaneous partial inhibition of both enzymes results in a negligible effect on A␤ production in cells that co-express the two enzymes. DISCUSSION We employed RNAi to selectively inactivate endogenous BACE1 and BACE2 to address the role of endogenous BACE2 in APP metabolism and A␤ formation. We demonstrate that siRNAs targeting these two enzymes work via degradation of the cognate endogenous mRNAs, consistent with the mechanism of action for RNAi in invertebrate systems (20 -22) as well as in mammalian cells (33)(34)(35)(36). The apparently less robust suppression of BACE1 mRNA (50% reduction at 48 h) versus BACE2 mRNA (90% reduction at 48 h; see Fig. 3) may reflect a difference in potency between the two active oligonucleotides for their respective targets.
The suppression of message correlated with an 88 -94% suppression of BACE1 protein and Ͼ94% suppression of BACE2 protein in an overexpression paradigm (Fig. 4). Pulse-chase studies indicate BACE2 is turned over more rapidly than BACE1 (18,(37)(38)(39), and this may explain the apparently greater suppression observed for BACE2. Direct confirmation of endogenous BACE1 and BACE2 protein knock-down in our experiments is technically limited by the low expression of these proteins in our system, combined with the sensitivity of available antibodies (see also Fig. 3B in Ref. 40). 2 Selective activity assays for each enzyme in cells that express both activities are encumbered by the very similar substrate preferences and the overlapping activities of BACE1 and BACE2 (31,32,41). Thus, measurement of a reduction in enzyme activities mediated by RNAi in our system is limited by the lack of a specific assay. However, we note that RNAi targeting BACE1 leads to a reduction of sAPP␤, a direct (and well acknowledged) product of BACE1 activity. Hence, although we cannot directly demonstrate reduction of endogenous protein by RNAi in our experiments, it is reasonable to assume that endogenous BACE enzymes are lowered by RNAi on the basis of the following observations: 1) reduction of endogenous message, 2) reduction of overexpressed protein; 3) reduction of endogenous product of BACE1 activity; and 4) alterations in APP metabolites effected by knock-down of BACE enzymes.
The suppression of BACE1 was reflected in a suppression of A␤ secretion, whereas the suppression of BACE2 was reflected by increased A␤ secretion from both wild type and familial Alzheimer's disease mutant forms of APP. Because we cannot relate the protein knock-down by RNAi in cells, measured by Western blots, to relative decrease in enzyme activity (for reasons cited above), it is possible that the low level of A␤ observed upon BACE1 knock-down is due to the residual enzyme activity remaining in our system. Yan et al. (19), using antisense oligonucleotides, did not observe increased A␤ production upon BACE2 knock-down. Although Yan et al. demonstrated mRNA suppression, they did not quantify the magnitude of BACE2 protein suppression effected by antisense. Hence, in addition to different time points at which medium was harvested for A␤ analysis between the two studies, our discordant observations may be attributable to differences in the magnitude BACE2 protein suppression effected by antisense compared with RNAi.
RNAi targeting BACE1 led to a decrease in secreted sAPP␤, as well as A␤ peptide, consistent with earlier studies (6, 7, 9 -11). In contrast, targeting BACE2 by RNAi led to increased sAPP␤ (Figs. 5 and 6) and A␤ peptide secretion from cells (Fig.  7). The elevation of A␤ upon BACE2 knock-down is a novel finding. The opposing effects on A␤ production suggest that both enzymes act on the substrate, and the shut-down of 2 G. Basi, N. Frigon, and S. Sinha, unpublished observations. BACE1 activity shunts intracellular APP into the ␣-secretase pathway, which now includes the ␣-like cleavage attributable to BACE2. Conversely, shut-down of BACE2 activity shunts intracellular APP into the ␤-secretase pathway, illustrated schematically in Fig. 8. These results suggest that BACE1 and BACE2 activities regulate A␤ secretion in cells co-expressing the two enzymes. Thus, our observations from the selective inactivation of BACE1 and 2 transcripts by RNAi provides a possible insight into the normal physiological activity of these enzymes on APP metabolism and A␤ production, particularly in cells that co-express both enzymes.
An alternative explanation for our observations from BACE2 knock-down can be envisioned. In this scenario, BACE2 does not exert its effects directly as an APP processing enzyme but negatively regulates APP secretion (either directly or indirectly via negatively regulating transport of APP to cellular compartments where secretory cleavage takes place). Thus, BACE2 knock-down would lead to the elevations in sAPPtot, sAPP␤, and A␤, as we observe. A corollary of this hypothesis is that an overexpression of BACE2 would be predicted to lower all forms of sAPP and A␤ production in cells. The former prediction is not consistent with published data showing sAPP is not lowered upon BACE2 overexpression (14,16,18,19). The latter prediction (decrease in A␤ from BACE2 overexpression) is well established as being due to cleavage of APP at A␤ residues 19 and 20. Hence, the elevation of A␤ we observe upon knock-down of BACE2 is most consistent with a direct role for this enzyme as an APP processing enzyme.
BACE1 cleavage between A␤ residues 10 and 11 correlates with high expression levels of BACE1 (42,43), suggesting that in cells expressing relatively low levels of BACE1, cleavage at A␤ position 1 predominates. Cleavage at A␤ position 1 by BACE2 has only been observed in transiently transfected cells overexpressing the enzyme (14, 16 -19). If the endogenous BACE2 in our cell lines cleaved primarily at position 1, we would expect to have observed a decrease in sAPP␤ from RNAi against BACE2, additive with the BACE1 RNAi observation. In contrast, we observe an increase in sAPP␤ as well as A␤ release following the selective inactivation of endogenous BACE2. Our results targeting BACE2 suggest that it is primarily the cleavage at position 19/20 that is abrogated. Although our studies do not directly address position 1 cleavage by BACE2, by analogy with earlier findings with BACE1 (42,43), our data are conlamin transfected cells. #, sample 0.1B3 was statistically significant (p ϭ 0.0321) relative to lamin transfected cells. Sample B3 did not reach statistical significance in this experiment (p ϭ 0.0875 relative to lamin transfected cells).

FIG. 7.
Total A␤ secreted into medium from cells transfected with siRNAs to BACE1 and BACE2 as compared with untransfected cells or control oligonucleotides to lamin and GFP. A␤ secretion was determined by ELISA as described under "Experimental Procedures." The values were normalized for total protein concentration in lysates from the respective wells. The ELISA values shown are average values from triplicate transfection with the indicated siRNA oligonucleotides. Utfx denotes medium harvested from untransfected cells. A, total A␤ from 293sw cells treated with siRNA oligonucleotides indicated below the bars. *, samples B3 and C3, p Ͻ 0.0001 relative to untransfected cells, lamin, and GFP samples; #, p ϭ 0.028 relative to untransfected cells; ϩϩ, p ϭ 0.017 relative to Lamin transfected cells. A␤ secreted from GFP oligonucleotide transfected sample is not statistically significant from untransfected cells (p ϭ 0.784). B, A␤ levels from Amy5 cells transfected with siRNA oligonucleotides indicated below the bars. *, for sample B3 p ϭ 0.013 relative to untransfected cells, and Ͻ0.0001 relative to lamin or GFP transfected cells; for sample C3 p Ͻ 0.0001 relative to untransfected or lamin transfected cells, and p ϭ 00013 relative to GFP transfected cells. #, for lamin transfected cells, p ϭ 0.0123 relative to untransfected cells; ϩϩ, for GFP transfected cells, p ϭ 0.0029 relative to untransfected cells, GFP transfected sample is not statistically significant from lamin transfected cells (p ϭ 0.405). C, the same conditioned medium samples as shown in Fig. 5 (i.e. from 293sw cells) were assayed for A␤ levels following treatment with siRNAs (noted below the bars) to determine the effect of combined knock-down of BACE1 and 2 on A␤ production. *, for samples treated with siRNA C3, p ϭ 0.0006 relative to untransfected cells, and p ϭ 0.015 relative to lamin transfected cells; for samples treated with 0.1C3, p ϭ 0.0013 relative to untransfected cells, and p ϭ 0.0321 relative to sistent with BACE2 cleavage at A␤ position 19/20 as the predominant activity in cells expressing low levels of BACE2 activity. By extension, the ␤-cleavage at position 1 that is observed in cells overexpressing BACE2 is likely a minor activity, undetectable in cells expressing low levels of BACE2. Signature A␤ fragments ending at the BACE2 cleavage site 19/20 have been detected using matrix-assisted laser desorption ionization time-of-flight mass spectrometry analysis of immunoprecipitated conditioned medium from N2a cells (44), as well as in formic acid solubilized Alzheimer's brain extracts (45). Taken together, the observations of Wang et al. (44) and Kalback et al. (45) are consistent with our observations that BACE2 is indeed an APP processing secretase under physiological conditions.
The opposing changes in A␤ from cells treated with B3 or C3 siRNAs in isolation stands in contrast with the relative lack of effect on A␤ from cells treated with both siRNAs in combination, as well as with observations from antisense studies (19). The protein suppression observed with RNAi, even in cells transfected with individual siRNAs, falls short of a genetic knock-out of enzyme activity effected by gene targeting in vivo, in as much as we are able to detect residual levels of gene product in siRNA treated cells. Thus, our observations from cells co-treated with both oligonucleotides more closely mimic partial inhibition of enzyme activity by nonselective small molecule compounds. In this scenario, a partial inhibition of both BACE1 and BACE2 has minimal net effect on APP metabolites produced by the two enzymes (Fig. 5), particularly A␤ (Fig. 7).
Our quantitative RT-PCR assay of the major brain cell types revealed that astrocytes co-express BACE1 and BACE2 mRNA in almost equal proportion (Fig. 1B). Farzan et al. (16) reported cloning of cDNAs encoding both enzymes from U87 astroglioma cells, consistent with our observation of BACE2 expression in astrocytes. With regard to A␤ production in the brain, the glial expression of BACE2 is consistent with both our observations using RNAi to target BACE2 in cells, as well as with BACE1 knock-out phenotype in mice (i.e. no detectable A␤ in the brain from knock-out mice) (9 -11). Specifically, our conclusion that low expression levels of BACE2 in cells in the absence of BACE1 would not elevate A␤ production is consistent with the conclusions from the knock-out studies in mice. However, with regard to compounds targeting BACE1 to lower brain A␤, glial BACE2 should be taken into consideration. Glial cells outnumber neurons in the brain by 10 -50-fold (46), and production of A␤ peptide in primary astrocyte cultures has been documented (47). Because our studies suggest that BACE2 inhibition would elevate A␤ production, the consequence of targeting BACE1 and BACE2 by RNAi for A␤ production in primary astrocyte cultures may be instructive.
In summary, we provide evidence that BACE1 and BACE2 regulate A␤ production in cells in an antagonistic manner. Our results suggest that inhibition of BACE2 by nonselective BACE inhibitors could compromise the desired lowering of A␤.