Rank-Order of Potencies for Inhibition of the Secretion of Aβ40 and Aβ42 Suggests That Both Are Generated by a Single γ-Secretase*

The Alzheimer’s disease amyloid peptide Aβ has a heterogeneous COOH terminus, as variants 40 and 42 residues long are found in neuritic plaques and are secreted constitutively by cultured cells. The proteolytic activity that liberates the Aβ COOH terminus from the β-amyloid precursor protein is called γ-secretase. It could be one protease with dual specificity or two distinct enzymes. By using enzyme-linked immunosorbent assays selective for Aβ40 or Aβ42, we have measured Aβ secretion by a HeLa cell line, and we have examined the dose responses for a panel of five structurally diverse γ-secretase inhibitors. The inhibitors lowered Aβ and p3 secretion and increased levels of the COOH-terminal 99-residue β-amyloid precursor protein derivative that is the precursor for Aβ but did not alter secretion of β-amyloid precursor protein derivatives generated by other secretases, indicating that the inhibitors blocked the γ-secretase processing step. The dose-dependent inhibition of Aβ42 was unusual, as the compounds elevated Aβ42 secretion at sub-inhibitory doses and then inhibited secretion at higher doses. A compound was identified that elevated Aβ42 secretion at a low concentration without inhibiting Aβ42 or Aβ40 at high concentrations, demonstrating that these phenomena are separable pharmacologically. Using either of two methods, IC50 values for inhibition of Aβ42 and Aβ40 were found to have the same rank-order and fall on a trend line with near-unit slope. These results favor the hypothesis that Aβ variants ending at residue 40 or 42 are generated by a single γ-secretase.

The Alzheimer's disease amyloid peptide A␤ has a heterogeneous COOH terminus, as variants 40 and 42 residues long are found in neuritic plaques and are secreted constitutively by cultured cells. The proteolytic activity that liberates the A␤ COOH terminus from the ␤-amyloid precursor protein is called ␥-secretase. It could be one protease with dual specificity or two distinct enzymes. By using enzyme-linked immunosorbent assays selective for A␤40 or A␤42, we have measured A␤ secretion by a HeLa cell line, and we have examined the dose responses for a panel of five structurally diverse ␥-secretase inhibitors. The inhibitors lowered A␤ and p3 secretion and increased levels of the COOH-terminal 99-residue ␤-amyloid precursor protein derivative that is the precursor for A␤ but did not alter secretion of ␤-amyloid precursor protein derivatives generated by other secretases, indicating that the inhibitors blocked the ␥-secretase processing step. The dose-dependent inhibition of A␤42 was unusual, as the compounds elevated A␤42 secretion at sub-inhibitory doses and then inhibited secretion at higher doses. A compound was identified that elevated A␤42 secretion at a low concentration without inhibiting A␤42 or A␤40 at high concentrations, demonstrating that these phenomena are separable pharmacologically. Using either of two methods, IC 50 values for inhibition of A␤42 and A␤40 were found to have the same rank-order and fall on a trend line with near-unit slope. These results favor the hypothesis that A␤ variants ending at residue 40 or 42 are generated by a single ␥-secretase.
A hallmark feature of the neuropathology of Alzheimer's disease is the abundant deposition of amyloid into neuritic and diffuse plaques in the brain parenchyma. The predominant core constituent of amyloid plaques is a 40-to 42-residue amyloid peptide, A␤. 1 A␤ is generated from the ␤-amyloid precursor protein (APP) by two sequential proteolytic cleavages. First, ␤-secretase activity liberates the NH 2 terminus of A␤, gener-ating a 99-residue COOH-terminal APP fragment (C99). Second, ␥-secretase activity cleaves C99 to liberate the A␤ COOH terminus. APP is also processed in a nonamyloidogenic manner, being cleaved within the A␤ domain by an activity termed ␣-secretase. The terms ␣-, ␤-, and ␥-secretase are conceptual terms for proteases that are not yet definitively identified (reviewed in Refs. [1][2][3]. The A␤ peptide is a constitutive secretory product of a variety of neuronal and non-neuronal cells, in which multiple NH 2and COOH-terminal cleavages generate several A␤ species. The most prevalent secreted forms of A␤ appear identical to neuritic plaque core amyloid A␤ and terminate either at residue 40 or 42 (4,5). Although A␤40 variants are the predominant component of soluble A␤ extracted from brain and secreted by cultured cells, A␤42 forms are the predominant constituents of neuritic plaques (6 -9). In vitro, forms of A␤ that terminate at residue 42 form fibril faster than forms of A␤ that terminate at residue 40, by orders of magnitude (10). Mutations in the APP gene or in two unrelated genes, presenilin-1 and presenilin-2, are a leading cause of early-onset, inherited forms of Alzheimer's disease. Strikingly, all of the mutations linked to Alzheimer's disease increase the amount of A␤42 that is secreted from cultured cells or extracted from the transgenic mouse brain and human plasma, whereas only some of the mutations are characterized by increased A␤40 (reviewed in Refs. 3 and 11-13). These observations suggest A␤42 has a central role in amyloid plaque formation and, hence, in the pathogenesis of disease.
The two major COOH-terminal variants of A␤ could be generated in several different ways. First, there may be a single ␥-secretase capable of cleaving at both A␤40 and A␤42. Alternatively, two distinct endoproteinases, with different substrate preferences and inhibitor sensitivities, could generate the two distinct species of A␤. A third possibility is that a single ␥-secretase generates A␤42, which subsequently is processed by a carboxypeptidase to form A␤40. The resolution of these alternatives will direct efforts to identify the protease, or proteases, responsible for formation of the A␤ variants and will pose different questions about molecular and cellular mechanisms of ␥-secretase processing. Resolution of these alternatives may also have important implications for therapeutic strategies aimed at inhibiting A␤ production and interfering with amyloid deposition. If there are two distinct ␥-secretases, the central role of A␤42 in plaque formation argues for the importance of inhibiting the ␥-secretase responsible for formation of A␤42 and suggests that it may be possible to block production of A␤42 without interfering with A␤40. If, on the other hand, both species are generated by a single ␥-secretase or if A␤42 serves as a precursor for A␤40, then inhibitors of A␤42 production would be expected to block formation of A␤40 as well.
Studies of the inhibition of cellular secretion of A␤ variants * 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  have suggested that different ␥-secretases may generate A␤40 and A␤42 (14 -16). The existence of two ␥-secretases has been suggested by the observation that a given concentration of peptidyl-aldehyde or -difluoroketoamide protease inhibitors blocks secretion of A␤40 but elevates A␤42. Here we reconsider this question by a detailed examination of the dose responses for inhibition of A␤40 and A␤42 secretion by a panel of five structurally diverse ␥-secretase inhibitors.

EXPERIMENTAL PROCEDURES
Materials-MEM and methionine-free MEM were from Mediatech. Opti-MEM® was from Life Technologies. Fetal bovine serum was from JRH Biosciences. [ 35 S]Met was Tran 35 S-label from ICN. Cell culturetested Me 2 SO was from Sigma. Calpeptin was from Calbiochem.
Synthesis of Protease Inhibitors-Compounds A, D, and E are aldehydes and were synthesized by standard peptide chemistry methods. Compound B, a difluoroketoamide, is structurally similar to aspartyl protease inhibitors described in Refs. 17 and 18 and was synthesized as described therein. Compound N, a dihydroxyethylene, was synthesized as described in Ref. 19 (Compound 20a in that paper).
Measurement of APP Processing and A␤ Secretion in Cell Culture-M17 human neuroblastoma cells were stably transfected with an expression construct coding for human APP695 carrying the "Swedish" mutation K595N/M596L (20). The cells were maintained in Opti-MEM plus 5% heat-inactivated fetal bovine serum. Metabolic labeling and immunoprecipitation were performed essentially as described (21). Briefly, subconfluent cells in 6-well dishes were incubated for 4 h in 1 ml of methionine-free MEM containing 200 Ci of [ 35 S]Met. A␤ and p3 were immunoprecipitated from the medium with Ab1153, raised against A␤1-28 (22), or with Ab58, raised against A␤17-40 (1). COOHterminal derivatives of APP were immunoprecipitated from cell lysates with Ab11, raised against APP666-695 (21). Immunoprecipitated proteins were separated by electrophoresis through Tris-Tricine gels and visualized by phosphorimaging.
HeLa-pNAN8 cells are a clone of HeLa cells transfected with the pNAN expression construct. This vector expresses a fusion protein consisting of the bovine growth hormone signal sequence, the COOHterminal 103 residues of APP (beginning four residues upstream of the ␤-secretase cleavage site), and a tetrapeptide extension Tyr-Cys-Phe-Ala (23). The cells were maintained in MEM plus 10% fetal bovine serum. For each experiment, HeLa-pNAN8 cells were grown to ϳ90% confluence in 24-well plates. Cells were pretreated for 1 h and then treated for 4 h, with test compound or Me 2 SO vehicle in 0.3 ml of medium per well. Each 24-well plate held three vehicle control wells and three wells each at seven doses of drug spaced at six doses per log unit. The close spacing of the doses and the triplicate cultures were chosen to allow reliable determination of IC 50 values despite the steepness of the dose-response curves (see Fig. 3). At the end of the conditioning period, the plates were centrifuged at 200 ϫ g for 5 min, and the medium was loaded onto ELISA plates at appropriate dilutions for determination of A␤40 and A␤42. The A␤40-and A␤42-selective ELISAs have been described and characterized extensively (24). The compounds used in this study neither interfered with detection of A␤40 or A␤42 by ELISA nor enhanced detection of A␤42. 2 Dose-response curves were expressed as percent of control value and were fitted (by nonlinear least squares using the program GraphPad Prism) to sigmoidal dose-response curves with variable slope, the bottom fixed at zero and the top fixed to 100% or the observed maximum, whichever was greater.
Dose-response curves in the presence of 10 M Compound N were measured as above, except that only six doses of test compound were included on each plate, allowing measurement of A␤ secretion by both untreated cells (no Compound N, no test compound) as well as Compound N-treated control (10 M Compound N, no test compound).

RESULTS
The five inhibitors of A␤ secretion chosen for this work incorporate either aldehyde or difluoroketoamide isosteres as enzyme-reactive groups (Fig. 1). Compounds incorporating these groups, including Compound C, have been demonstrated previously to be inhibitors of ␥-secretase processing (14 -16, 25). Diagnostic of ␥-secretase inhibition is blockade of both A␤ and p3, secreted species that have the ␥-secretase cleavage in common, without concomitant effects on the secretion of APP fragments derived from the activity of other secretases (1). Both Compound A and Compound B inhibited secretion not only of A␤ but also of p3, as shown by metabolic labeling and immunoprecipitation following treatment with the inhibitors (Fig. 2, top panel). Comparable doses of the inhibitors did not alter APP synthesis or the secretion of APP␣ and APP␤, the NH 2 -terminal products of ␣and ␤-secretases. 2 Moreover, the compounds blocked ␥-secretase processing of APP in a cell-free system, suggesting that they inhibit ␥-secretase directly. 3 To characterize further the effects of the five protease inhibitors on APP processing, levels of COOH-terminal derivatives of APP were measured. As shown in Fig. 2 (bottom panel), at a dose that blocked A␤40 and A␤42 secretion by more than 90%, compounds A and B elevated two [ 35 S]Met-labeled COOH-terminal APP derivatives that serve as ␥-secretase substrates, the ϳ12-kDa ␤-secretase-derived C99 fragment, and the ϳ9-kDa ␣-secretase-derived C83 fragment (5,26,27). (The 6 kDa ␥-secretase-derived C57 fragment was not detected in these experiments.) Collectively, these data demonstrate that compounds used in the present analysis do not markedly reduce APP synthesis or block ␣or ␤-secretase processing but instead inhibit the ␥-secretase processing step.
A␤40 and A␤42 are generated physiologically by ␥-secretase cleavage of C99, the 99 COOH-terminal residues of APP (5,25,28,29). We measured dose responses for inhibition of A␤40 and A␤42 secretion by the HeLa-pNAN8 cell line. This cell line stably overexpresses C99 with tetrapeptide extensions, EVKM-(C99)-YCFA (23). A similar construct, with most of the NH 2terminal APP sequence deleted, has been used previously to study sequence specificity of ␥-secretase cleavage (30). Under the protocol described here, medium conditioned by HeLa-pNAN8 cells for 4 h accumulates 7.0 to 8.5 ng of A␤40/ml and 0.9 to 1.5 ng of A␤42/ml. A␤42 constitutes ϳ14% of the total, as it does in medium conditioned by cell lines expressing fulllength APP (e.g. Ref. 31). A␤ accumulation is linear with respect to time over 4 h. The high level of A␤ secretion by HeLa-pNAN8 cells facilitates measurement of dose-dependent alterations in secretion of both A␤40 and A␤42 by protease inhibitors. Discriminative electrophoretic and ELISA analyses indicate that A␤40 and A␤42 are the predominant A␤ forms being measured, although a minor contribution of A␤ forms whose NH 2 termini are within 2-3 residues of Asp-1 cannot be ruled out. 2 Fig . 3A illustrates the dose-response curves for inhibition of A␤40 and A␤42 secretion by the difluoroketoamide Compound B. The qualitative features of the dose-response curve are the same for all ␥-secretase inhibitors tested. The most striking feature is the approximately 7-fold elevation of A␤42 secretion at sub-inhibitory doses. Among the five ␥-secretase inhibitors tested, elevations of A␤42 ranged from 3-to 8-fold. At the peak of the elevation, A␤42 constituted as much as 50% of the total secreted A␤. Another feature of the dose-response curves is their steepness. Hill slopes ranged from 1.8 to 4.9 for A␤40 and from 3.0 to 5.3 for A␤42. The steepness of the dose-response curves dictated the close spacing of compound doses chosen in these experiments (six doses per log unit).
A compound was identified that elevated A␤42 secretion at relatively low doses, without causing inhibition even at much higher doses. Fig. 4 shows dose-dependent effects on A␤ secretion of compound N, described previously as a potent inhibitor of a variety of aspartyl proteases (IC 50 Ͻ60 nM for cathepsin D, endothelin-converting enzyme, and renin; see Ref. 19). Compound N elevated A␤42 secretion nearly 6-fold, with an estimated EC 50 about 500 nM. At doses 100-fold higher, compound N did not inhibit secretion of either A␤40 or A␤42. In the presence of 10 M compound N, a saturating concentration for elevation of A␤42, dose-response curves for the five ␥-secretase inhibitors showed little or no further elevation of A␤42 (Fig.  3B), indicating that the ␥-secretase inhibitors and compound N share a common mechanism for elevation of A␤42 secretion. The presence of compound N did not interfere with inhibition of A␤ secretion by the ␥-secretase inhibitors tested; indeed, the were immunoprecipitated from the medium with Ab1153. Qualitatively similar results were obtained when the medium was immunoprecipitated with Ab58 (not shown). Bottom panel, COOH-terminal derivatives of APP were immunoprecipitated from the corresponding cell lysates with Ab11. The C99 fragment was identified from the literature and by its co-migration with recombinant Met-C99. The C83 fragment was identified from the literature (5,26,27). The ␥-secretase-derived C57 fragment was not detected in these experiments; the migration of recombinant Met-C57 indicated that it would run below the 6.5-kDa standard. 2 These experiments were performed in M17 cells because HeLa-pNAN8 cells do not secrete appreciable amounts of p3. IC 50 values in the presence of compound N are somewhat lower than in its absence. Consequently, the elevation of A␤42 can be separated pharmacologically from inhibition of ␥-secretase and blockade of A␤42 and A␤40 secretion.
Two methods were used to compare the rank-order of inhibitor potencies for A␤40 and A␤42. In the first, the IC 50 for A␤42 was empirically defined as the dose at which A␤42 secretion declines to half of its peak (elevated) value and was estimated by fitting the data to a sigmoidal dose-response with variable slope and the top fixed to the observed peak. Fig. 5A summarizes the relationship between the IC 50 values measured for A␤40 and for A␤42. For the five ␥-secretase inhibitors tested, the IC 50 values fall near a trend line with unit slope. The rank-order of these compounds for inhibition of A␤42 is identical to their rank-order for inhibition of A␤40. In the second method, IC 50 values for A␤40 and A␤42 were compared in the presence of 10 M Compound N to saturate the elevation of A␤42 at low ␥-secretase inhibitor concentrations. Because the A␤42 dose responses look like pure inhibition under this condition, determination of the A␤42 IC 50 by this method is straightforward. IC 50 values measured by this method fall near a trend line with unit slope, exactly like the empirical IC 50 values (Fig. 5B). Table I summarizes IC 50 values determined by both methods. By both methods, the rank-order of ␥-secretase inhibitors for inhibition of A␤42 is identical to their rank-order for inhibition of A␤40. DISCUSSION If both A␤40 and A␤42 are generated by a single ␥-secretase, the rank-order of inhibitor potencies against these two cleavages should be the same. The potencies need not be identical, because in each case the inhibitor is competing with a distinct substrate sequence for which the protease may have different affinities. If, on the other hand, two different ␥-secretases cleave the precursor at these two sites, there ought to be good inhibitors of one that are poor inhibitors of the other, and the rank-order of inhibitor potencies for the two A␤ species should differ. To distinguish between these possibilities, we have examined the dose-dependent inhibition of A␤40 and A␤42 secretion by a panel of five structurally diverse ␥-secretase inhibitors. Previous studies using only one or two doses of protease inhibitor have demonstrated differential effects on A␤40 and A␤42 secretion, and these data have been used to argue for the existence of two distinct ␥-secretases (14 -16). We demonstrate here that comparison of inhibitor effects at only one or two concentrations is not straightforward, because the dose response for A␤42 is the superposition of two effects as follows: elevation of A␤42 production at relatively low doses and inhi-bition at higher doses.
Elevations in A␤42 secretion have been observed previously, as synthetic intermediates of a difluoroketoamide ␥-secretase inhibitor are inactive against A␤40 secretion at 200 M but elevate A␤42 (Compounds 5-7 in Ref. 16). Peptide aldehydes, described as calpain inhibitors, have also been reported to elevate secretion of both A␤40 and A␤42 and to increase the ratio of A␤42 to total A␤ (32). The elevation of A␤40 reported there was not observed in the present work. We have found that the two effects, elevation of A␤42 production at relatively low doses and inhibition at higher doses, are separable pharmacologically. Compound N potently and markedly elevates A␤42 production with no inhibition even at a dose 100-fold higher (Fig. 4). Furthermore, a saturating concentration of Compound N for elevation of A␤42 does not interfere with inhibition of A␤42 by ␥-secretase inhibitors.
We therefore adopt two approaches to separate elevation of A␤42 secretion from its inhibition. In the first, we consider only the descending portion of the dose-response curve and fix the top of the fitted curve to the observed peak A␤42. This approach supposes the elevation of A␤42 to have gone to completion before inhibition begins, as is apparent from the doseresponse curves (Fig. 3A). In the second approach, the elevation of A␤42 is saturated by treatment of the cells with 10 M Compound N. Because the ␥-secretase inhibitors cause no further elevation of A␤42 under this condition, their dose-response curves reflect pure inhibition. IC 50 values determined by both approaches rank the ␥-secretase inhibitors in the same order for A␤40 and for A␤42 (Fig. 5).
The dose-response data presented here are consistent with previously published data on inhibition of A␤ secretion (14 -16), in which inhibitors of ␥-secretase processing, including compound C of the present report, were shown to have no effect or to elevate A␤42 secretion at a dose that inhibited A␤40 secretion. Inspection of Fig. 3A reveals how such a dose may be chosen. At higher doses, however, every ␥-secretase inhibitor that inhibits A␤40 secretion also inhibits A␤42 secretion, and with the same rank-order potency.
The observation that certain protease inhibitors cause a marked and relatively selective increase in secretion of A␤42 reveals a potentially important step in the regulation of A␤42 levels, but at present we can only speculate as to the mechanism. One explanation would be that the protease inhibitors block, in addition to ␥-secretase, a protease that degrades A␤42. To test this hypothesis, cell-free conditioned medium was incubated for an additional 4 h (the usual conditioning period) in the presence or absence of the ␥-secretase inhibitors at doses saturating for the elevation of A␤42. No degradation of A␤40 or A␤42 was observed during the cell-free incubation, and no elevation of A␤42 occurred on inclusion of the ␥-secretase inhibitors. 2 Note that this experiment does not address the possible intracellular degradation of A␤42. But intracellular A␤42 has been shown to be degraded more slowly than intracellular A␤40, too slowly to contribute to the experiments here (33). Furthermore, A␤ is found in cultured cells at much lower levels than secreted A␤ (29, 34 -36), suggesting that A␤ is rapidly secreted upon generation and, hence, that its intracellular degradation is unlikely to contribute to its turnover. Inhibition of A␤42 degradation is therefore unlikely to account for the elevation of A␤42 secretion reported here.
Another possible explanation for the elevation of A␤42 would be the generation of A␤40 by carboxypeptidase-mediated processing of A␤42, with the protease inhibitors actually inhibiting the carboxypeptidase, resulting in an accumulation of A␤42. Fig. 3 shows, however, that the elevation of A␤42 begins at lower doses than the inhibition of A␤40 and that modestly higher doses inhibit A␤42 as well as A␤40. Fig. 4 demonstrates that Compound N has no effect on secretion of A␤40 at doses at which its elevation of A␤42 has saturated. Neither of these observations is consistent with inhibition of carboxypeptidasemediated processing of A␤42 resulting in the elevation of A␤42 secretion reported here.
Several lines of evidence suggest that C99, the immediate precursor of A␤40 and A␤42, is available to ␥-secretase in kinetically limiting amounts. That inhibition of ␥-secretase increases the level of C99 ( Fig. 2; see also Ref. 25) is inconsistent with the level of available C99 being saturating for ␥-secretase processing and, instead, indicates that the availability of C99 is rate-limiting. A limiting role for C99 availability is further supported by the many reports that a double missense mutation in APP linked to familial Alzheimer's disease (the so-called Swedish mutation, see Ref. 20) produces parallel increases in A␤ secretion and C99 levels through enhanced ␤-secretase processing (5,27,37). The steepness of the doseresponse curves for ␥-secretase inhibition (Fig. 3) is also consistent with limiting amounts of available C99 substrate. Therefore, another way by which sub-inhibitory doses of ␥-secretase inhibitors might increase secretion of A␤42 is by relieving the limited availability of C99.
One mechanism by which ␥-secretase inhibitors may relieve a limiting amount of C99 is through the accessibility of the ␥-secretase cleavage site. The cleavage site for ␥-secretase is in the transmembrane segment of APP. The ␥-secretase inhibitors are relatively hydrophobic. At the near-micromolar concentrations required for ␥-secretase inhibition, they may perturb the membrane structure in such a way as to make the cleavage site more available to the enzyme. The A␤42 site is not buried as deeply as the A␤40 site, so membrane perturbation could have a greater effect on the availability of the A␤42 site than the A␤40 site, as observed. A similar physical mechanism has been proposed to explain preferential generation of A␤42 in the endoplasmic reticulum, where the membrane is thinner than in the trans-Golgi network (38). The mechanism proposed here predicts that more potent inhibitors will inhibit ␥-secretase cleavage at concentrations below those necessary to perturb the membrane; in consequence, the elevation of A␤42 will not be observed. Testing of this prediction awaits the discovery of more potent ␥-secretase inhibitors.
Another mechanism is to hypothesize that the ␥-secretase inhibitors also interfere with C99 turnover. The marked increase in A␤42 secretion by relatively low doses of the peptidyl aldehydes, which act as inhibitors of cysteine, serine, and certain aspartyl proteases, and compound N, a potent broad-spectrum aspartyl protease inhibitor, may be due to their potent inhibition of a nonamyloidogenic degradation of C99, thereby increasing C99 availability and, consequently, A␤42 formation. Preferential elevation of A␤42 levels raises the possibility that C99 availability may be particularly rate-limiting for ␥-secretase processing in the cellular compartment that generates A␤42 (38,39). According to this model, C99 elimination by an aspartyl protease plays a role in regulating A␤42 production. An inhibitor of C99 elimination such as compound N may be a useful tool for identifying and characterizing a protease responsible for down-regulation of A␤42.
The ␥-secretase has received only limited prior characterization, although mutagenesis studies of APP-based substrates have suggested that the protease has a loose preference for hydrophobic residues in the vicinity of the cleavage site (30,40). The inhibitor data presented here with four dipeptidyl aldehydes and a dipeptidyl difluoroketoamide are consistent with this model. Assuming that the aldehyde and difluoroketoamide moieties interact with critical active site residues, a variety of neighboring aliphatic and aromatic substituents appear to be well tolerated, leading to relatively minor alterations in affinity. Even introducing an additional spacer between the two amide functionalities, through the ␤-alanyl residue in Compound E, causes only a small decrease in potency, indicating that the enzyme has broad steric tolerance in the active site. Broad steric tolerance is consistent with a single enzyme capable of cleaving APP at two different sites to generate either A␤40 or A␤42.
We point out some qualifications to our suggestion that both A␤40 and A␤42 are generated by a single ␥-secretase. The suggestion applies to secreted A␤, as is the case for previous work on this question (14,15), but does not address the nature of the protease or proteases responsible for intracellular formation of A␤ variants (38,39). The possibility that secreted A␤40 and A␤42 are generated by two distinct but closely related proteases with very similar inhibitor sensitivities cannot be ruled out but may be addressed by applying the approach described here to a larger and more structurally diverse set of more potent protease inhibitors.
In conclusion, we have determined the dose response for A␤40 and A␤42 secretion by HeLa-pNAN8 cells for a series of ␥-secretase inhibitors. The inhibitors elevate A␤42 secretion at relatively low doses but inhibit it at higher concentrations. The elevation of A␤42 is likely due to a different mechanism than inhibition, as compounds exist that cause one without the other. Inhibition of A␤ secretion is accompanied by a reciprocal increase in levels of C99, its immediate precursor. Two different ways of extracting an IC 50 from the complex dose response for A␤42 secretion result in the same rank-order for inhibition of A␤42 as for inhibition of A␤40. Taken together, these data suggest that A␤40 and A␤42 are generated by a single ␥-secretase which uses a limiting amount of C99 as substrate.