Research ArticleAlzheimer’s Disease

A Therapeutic Antibody Targeting BACE1 Inhibits Amyloid-β Production in Vivo

See allHide authors and affiliations

Science Translational Medicine  25 May 2011:
Vol. 3, Issue 84, pp. 84ra43
DOI: 10.1126/scitranslmed.3002254

Abstract

Reducing production of amyloid-β (Aβ) peptide by direct inhibition of the enzymes that process amyloid precursor protein (APP) is a central therapeutic strategy for treating Alzheimer’s disease. However, small-molecule inhibitors of the β-secretase (BACE1) and γ-secretase APP processing enzymes have shown a lack of target selectivity and poor penetrance of the blood-brain barrier (BBB). Here, we have developed a high-affinity, phage-derived human antibody that targets BACE1 (anti-BACE1) and is anti-amyloidogenic. Anti-BACE1 reduces endogenous BACE1 activity and Aβ production in human cell lines expressing APP and in cultured primary neurons. Anti-BACE1 is highly selective and does not inhibit the related enzymes BACE2 or cathepsin D. Competitive binding assays and x-ray crystallography indicate that anti-BACE1 binds noncompetitively to an exosite on BACE1 and not to the catalytic site. Systemic dosing of mice and nonhuman primates with anti-BACE1 resulted in sustained reductions in peripheral Aβ peptide concentrations. Anti-BACE1 also reduces central nervous system Aβ concentrations in mouse and monkey, consistent with a measurable uptake of antibody across the BBB. Thus, BACE1 can be targeted in a highly selective manner through passive immunization with anti-BACE1, providing a potential approach for treating Alzheimer’s disease. Nevertheless, therapeutic success with anti-BACE1 will depend on improving antibody uptake into the brain.

Introduction

Alzheimer’s disease (AD) is the leading cause of dementia worldwide, with the primary risk factor being age. With the increase in human life span, AD will continue to afflict ever-increasing numbers of the elderly, augmenting an already serious health problem. Current AD drugs target symptomatic mechanisms with only limited benefit. Thus, the discovery of disease-modifying therapeutics that can slow or ultimately halt disease progression is paramount. One of the hallmark pathologies of AD is the presence of amyloid plaques in the brain, formed by fibrillar deposits of amyloid-β (Aβ) peptides. The amyloid hypothesis posits that pathology initiates because of an imbalance in Aβ production and/or clearance, which may result from altered expression or processing of amyloid precursor protein (APP) or changes in Aβ metabolism (1). This imbalance triggers a cascade of neurotoxicity, possibly through toxic oligomeric intermediates, beginning with defects in synaptic physiology, leading to synaptic loss, astrogliosis, and eventually cell death. Development of therapies that reduce amyloidogenic processing holds great promise but has not yet proven successful clinically. Clinical evidence for ameliorating AD by targeting the amyloid cascade is limited because of the paucity of effective therapies that can reduce Aβ concentrations in the brain (2).

Amyloidogenic processing of APP involves sequential cleavage by β-secretase followed by γ-secretase (3). The aspartyl protease β-site APP cleaving enzyme 1 (BACE1) is the primary β-secretase in the brain, making it a prime candidate for AD therapeutics. Since its discovery in 1999, BACE1 has been heavily pursued as a small-molecule drug target (4). Many potent BACE1 inhibitors have been described, but only recently have some been reported that are able to cross the blood-brain barrier (BBB) in sufficient quantities to reduce brain Aβ concentrations (5) including in humans (6). Despite these advances, small-molecule inhibitors often have the liability of imperfect selectivity, posing a substantial risk for off-target toxicity that could limit dosing, or even halt clinical programs owing to adverse effects.

Antibody-based therapeutics provide an attractive alternative for new AD treatments. In general, therapeutic antibodies have excellent selectivity, thus reducing the probability of off-target effects. Already successful in oncology and immunology indications, many antibody therapeutics are currently being tested in clinical trials for AD. These drugs target Aβ directly, aiming to neutralize its toxic activity or dissolve existing plaques (7, 8). Nevertheless, the success of antibody therapies for central nervous system (CNS) disorders is likely to be limited technically by the BBB, which impedes antibody entry into the brain from the systemic circulation. Although the amount of antibody entering the CNS may be limited, a pharmacodynamic (PD) readout of antibody efficacy would provide a better measure of the true potential of therapeutic antibodies in the CNS.

BACE1 provides an ideal target for exploring the potential of antibody therapeutics in the CNS. BACE1 has been shown genetically to account for most Aβ production in vivo (912). Thus, measuring Aβ concentrations in the CNS after BACE1 antibody administration provides a direct measure of antibody-mediated target neutralization. We describe here the generation and characterization of a function-blocking BACE1 antibody through a phage-panning approach. Anti-BACE1 delivered systemically in various in vivo models, including in nonhuman primates, reduces Aβ concentrations not only in the periphery but also in the brain. Furthermore, brain or cerebrospinal fluid (CSF) concentrations of anti-BACE1 correlate with the ability to reduce amyloidogenic processing in vivo. These findings provide promising proof of concept that passive immunization against BACE1 may be a viable approach to treat AD. However, improvements in brain uptake and cellular potency of anti-BACE1 may ultimately be necessary for success in the clinic.

Results

Identification of a phage-derived human antibody targeting BACE1

To generate and optimize high-affinity antibodies to BACE1, we used human synthetic antibody libraries that mimic natural diversity (13, 14) to pan against the enzymatically active extracellular domain (ECD) of recombinant human BACE1 (rhBACE1). Clones that bound with high affinity to rhBACE1 were formatted onto a human IgG backbone and further selected on the basis of their ability to inhibit the activity of BACE1 in a biochemical assay. Here, we describe the activity of one clone that we refer to as anti-BACE1. Anti-BACE1 binds with similarly high affinity to both human and mouse recombinant BACE1 (KD = 1.3 to 1.4 nM), as determined by binding constants from surface plasmon resonance (SPR) analysis. Binding of anti-BACE1 was confirmed not just at pH 7.0 but also at pH 5.0. This is important because BACE1 is optimally active at acidic pH, presumably in endocytic vesicles and the trans-Golgi network, so an effective antibody might need to be functional in acidic compartments.

Anti-BACE1 selectively inhibits BACE1 in biochemical assays

We examined the inhibitory activity of anti-BACE1 in vitro using several independent assays. First, we used a 27–amino acid synthetic APP substrate with rhBACE1 in a homogeneous time-resolved fluorescence (HTRF) assay. Anti-BACE1 potently inhibited rhBACE1 proteolysis, showing an IC50 (median inhibitory concentration) of 1.7 nM, with a maximal inhibition reaching 77% (Fig. 1A). This inhibition was comparable to that achieved with a BACE1 small-molecule inhibitor (Inhibitor IV, Calbiochem), although less potent and less effective than the peptidic inhibitor OM99-2 (15). A control IgG that does not bind to BACE1 did not inhibit activity. The 27–amino acid substrate we use in the HTRF assay is longer than conventional APP substrates used in inhibitor screens and was designed to more closely mimic the full-length APP substrate through an extended binding interface. To test the activity of anti-BACE1 in a more conventional assay, we used a standard short APP peptide substrate somewhat more restricted to the active site in a fluorescence resonance energy transfer (FRET) assay. Under these conditions, anti-BACE1 inhibited BACE1 activity with an IC50 of 17 nM (fig. S1A). The reduced IC50 observed when using the shorter BACE1 substrate may relate to the mechanism by which anti-BACE1 inhibits BACE1 activity.

Fig. 1

Anti-BACE1 is a potent and selective inhibitor of BACE1. (A) Enzymatic activity of rhBACE1 ECD was tested with the 27–amino acid peptide substrate in an HTRF assay. Activity was measured in the presence of control IgG, anti-BACE1, OM99-2, or a small-molecule inhibitor (SMI) of BACE1 (Inhibitor IV, Calbiochem). Anti-BACE1 inhibits BACE1 activity with an IC50 of 1.7 nM. Values reflect BACE1 enzymatic activity relative to control (no antibody present) ± SEM. (B) In vitro enzymatic activity of rhBACE1 ectodomain, rhBACE2 ectodomain, or cathepsin D ectodomain was assayed by a microfluidic capillary electrophoresis assay with an APP peptide substrate. Anti-BACE1 inhibited rBACE1 activity with an IC50 of 80 pM but did not inhibit rBACE2 or cathepsin D. Values reflect percent substrate conversion ± SEM.

Most BACE1 small-molecule inhibitors not only block BACE1 activity but also are potent inhibitors of related members of the pepsin family of aspartyl proteases, particularly BACE2. To evaluate the selectivity of anti-BACE1, we used microfluidic capillary electrophoresis assays to examine inhibition of BACE2 and cathepsin D. As shown in Fig. 1B, anti-BACE1 robustly inhibited BACE1 activity in the microfluidic capillary electrophoresis assay. In contrast, anti-BACE1 had no effect on BACE2 or cathepsin D proteolytic activity. Enzyme-linked immunosorbent assay (ELISA) binding analysis also confirmed that anti-BACE1 does not bind to BACE2 (fig. S1B), the most highly related protease to BACE1. Together, these data indicate that anti-BACE1 is a potent and exquisitely selective BACE1 antagonist.

A co-crystal structure reveals mode of binding and noncompetitive inhibition

Given that anti-BACE1 inhibits enzymatic activity, it seemed possible that it might bind to the BACE1 active-site region. Thus, we carried out competition binding ELISA experiments with the previously described active-site binding peptidic inhibitor OM99-2 (15). As seen in Fig. 2A, whereas anti-BACE1 competes with itself for binding to BACE1, OM99-2 does not, implying that anti-BACE1 binds to an exosite.

Fig. 2

Anti-BACE1 inhibits BACE1 by binding to an exosite. (A) Competitive ELISA assay was used to study the anti-BACE1/BACE1 interaction compared to a known active-site BACE1 inhibitor. Active-site inhibitor OM99-2 does not compete with anti-BACE1 for BACE1 binding. (B) The crystal structure of the anti-BACE1 Fab and rhBACE1 catalytic domain showing anti-BACE1 binding to a BACE1 exosite. (C) Close-up view of the antibody complementarity-determining regions (CDRs) and the BACE1 epitope. Both heavy chain (HC) and light chain (LC) are involved in the interaction. (D) Amino acid sequence at and around the binding epitope. Residues in direct contact are boxed. For comparison, sequences of BACE2 and cathepsin D (CatD) are aligned to BACE1 based on crystal structures.

To further characterize this binding site on BACE1, we carried out crystallographic analysis of the anti-BACE1 Fab–BACE1 complex. The two proteins were independently expressed in Escherichia coli, purified to homogeneity, and then mixed together to form a complex that was purified by size exclusion chromatography. A crystal structure was determined at 2.8-Å resolution (Fig. 2B). Here, we use the amino acid numbering system that starts at the mature protease domain of BACE1 (16). The overall structure of BACE1 in the complex largely resembles its free form published previously (15), which can be aligned with 0.63-Å root mean square deviation (RMSD) at the Cα atom positions of 96% (373 of 385) of the residues. The anti-BACE1 Fab does not bind in the vicinity of the active site, but to an exosite on BACE1. The epitope comprises structural elements denoted as loop C (254 to 257), D (270 to 274), and F (309 to 320), which are closely located in three-dimensional space. Both the heavy and light chains of the antibody are involved in molecular recognition (Fig. 2C), and bury 840 Å2 of surface area on BACE1 with a complementarity score (17) of 0.71, consistent with high affinity. Unlike the free form, where the BACE1 epitope region is more dynamic as indicated by high-temperature factors, the antibody-bound structure is stabilized in a unique conformation, which deforms the P6 and P7 sites of the secretase (18). Adjacent to those sites, residues 157 to 170, which adopt an α-helical structure in the substrate-bound complex, become a random loop in the antibody complex, which adversely affects APP proteolytic cleavage, perhaps preventing APP from reaching into the BACE1 catalytic cleft in a catalytically competent manner. It is noteworthy that a previously described set of phage-derived peptides that bind outside the active site are also capable of modulating BACE1 activity (19). Further studies are needed to assess how these peptides relate to anti-BACE1 described herein. Overall, the structure suggests that targeting this exosite hinders enzymatically important interactions between APP and BACE1 and translates into allosteric modulation of enzyme activity.

Compared to BACE2 and cathepsin D, two main counter-targets of small-molecule BACE1 inhibitor design, the amino acid composition in the epitope is poorly conserved, as shown in Fig. 2D, consistent with our observation that the antibody is highly selective toward BACE1. Together, the crystallographic data and competition binding studies show that anti-BACE1 binds and inhibits BACE1 in a unique manner, distinct from described small-molecule inhibitors, which bind at the active site.

Anti-BACE1 inhibits cellular BACE1 activity

We next asked whether anti-BACE1 could inhibit BACE1 in a cellular context. Despite being very potent in biochemical assays, peptidic inhibitors of BACE1 often show reduced activity in cell-based assays because of poor cellular permeability (20). Macromolecules such as antibodies do not readily cross cell membranes; however, they may be taken into cells via receptor-mediated endocytosis. BACE1 is generally believed to act in endosomes and other intracellular vesicles; thus, targeting cellular BACE1 with antibodies may not be feasible. Nevertheless, we explored BACE1 inhibition in cells by meauring the levels of Aβ peptide being released by APP-expressing cells in the presence or absence of anti-BACE1. We first looked at Aβ1–40 peptide release from 293-HEK (human embryonic kidney) cultured cells stably expressing wild-type human APP (293-hAPP). Aβ1–40 release from 293-hAPP cells is inhibited by anti-BACE1, with an IC50 of 17 nM and a maximal extent of inhibition of ~90% (Fig. 3A). As a positive control, we used a potent small-molecule inhibitor of BACE1 (inhibitor 8e) (21).

Fig. 3

Anti-BACE1 blocks Aβ production in heterologous cells and primary neurons. Cells were incubated with medium containing control IgG, anti-BACE1, or a BACE1 small-molecule inhibitor (SMI) for 24 hours. (A) 293-hAPP cells show a dose-dependent reduction in Aβ production in the presence of anti-BACE1, with an IC50 of 17 nM. (B to D) Primary mouse neurons cultured from (B) E13.5 dorsal root ganglia (DRG), (C) E16.5 cortex, and (D) E16.5 hippocampus similarly show dose-dependent reductions in Aβ, with IC50 of 15 nM (DRG neurons), 3 nM (cortical neurons), or 4 nM (hippocampal neurons). (E and F) Western blot of sAPPβ and sAPPα with associated quantification from conditioned media of 293-hAPP cells (E) and mouse cortical neuron cultures (F) treated with anti-BACE1 at increasing concentrations shows a selective decrease in sAPPβ with no reduction observed for sAPPα.

We next looked at Aβ production in a variety of different primary neuronal cell types isolated from mouse. As shown in Fig. 3, B to D, anti-BACE1 inhibited Aβ1–40 production in primary neurons derived from mouse embryonic dorsal root ganglia (DRG) (IC50 = 15 nM), cortex (IC50 = 3 nM), and hippocampus (IC50 = 4 nM). In neurons, the maximal Aβ1–40 reduction achieved with anti-BACE1 was ~70% compared to ~85% with a small-molecule inhibitor control. These findings indicate that the anti-BACE1 antibody potently inhibits BACE1 activity. Furthermore, the anti-BACE1 antibody appears to show the highest potency in neurons of the CNS.

When APP is cleaved by BACE1, other APP fragments are generated. Specifically, BACE1 cleavage of APP gives rise to sAPPβ, whereas α-secretase cleavage of APP gives rise to sAPPα. To further prove the specificity of anti-BACE1, we evaluated the reduction of sAPPβ with anti-BACE1 treatment on 293-hAPP cells (Fig. 3E) and mouse cortical neurons (Fig. 3F). As expected, sAPPβ is reduced in a dose-dependent fashion, whereas no changes were observed for sAPPα.

For anti-BACE1 to function in cells, we hypothesized that it must bind to BACE1 expressed on the surface of membranes and subsequently get internalized into vesicular compartments where BACE1 is presumably active. Using primary neuron cultures, we tested anti-BACE1 binding to neurons, the internalization of anti-BACE1 into neurons, and the colocalization of anti-BACE1 relative to markers of various cellular compartments. Staining of fixed cortical neurons in either nonpermeabilized or permeabilized conditions showed a significant amount of BACE1 staining, with a higher signal observed after permeabilization (Fig. 4, A and B). These data are consistent with other reported findings showing that most BACE1 is found intracellularly (22); however, significant levels of BACE1 are also localized to the cell surface of these cortical neurons.

Fig. 4

Anti-BACE1 binds to surface BACE1 and is internalized into intracellular vesicles in neurons. (A and B) Anti-BACE1 detects BACE1 both on the cell surface in nonpermeabilized cells and in internal compartments in permeabilized cells by direct immunocytochemistry. (C and D) Control IgG or anti-BACE1 (1 μM) was incubated with embryonic cortical neurons at 37°C for the times indicated. Cells were washed thoroughly to remove free antibody. Anti-BACE1 signal was detected both on the surface of cells as well as intracellularly, but most antibody signal was found to be internalized. (E) Anti-BACE1 was colocalized with a marker of early endosomes (EEA1) but not with lysosomes (LAMP1) in cells that were incubated with anti-BACE1 for 3 hours. Scale bars, 50 μm [(A) and (C)] and 10 μm [(B), (D), and (E)].

To assess whether anti-BACE1 is internalized, we cultured mouse cortical neurons in the presence of anti-BACE1 or a control IgG for 10 min or 3 hours, after which we detected anti-BACE1 by immunostaining. We analyzed cell surface anti-BACE1 localization in nonpermeabilized cells, as well as internalized anti-BACE1 in permeabilized cells. Staining was observed in both conditions, but with a higher overall signal observed when cultures were permeabilized, suggesting that anti-BACE1 is internalized into intracellular compartments (Fig. 4, C and D). Internalization was evident after only 10 min of anti-BACE1 treatment and could be prevented by cold incubation (fig. S2A), suggesting that the antibody is actively taken up by early endosomes. Much of the anti-BACE1 signal was punctate (Fig. 4D), presumably reflecting uptake into vesicles. Furthermore, most of the signal detected is target-mediated, because neurons cultured from BACE1−/− mice failed to show anti-BACE1 internalization (fig. S2B). Nevertheless, cell body staining is observed in all conditions, including control immunoglobulin G (IgG) and BACE1−/− cultures. This background staining may be either nonspecific binding of secondary antibodies or binding of IgG to Fc receptors, possibly expressed on microglia in these mixed cortical neuron cultures.

To better identify the subcellular compartments to which anti-BACE1 was localized, we costained with markers of different vesicular compartments: early endosomes (EEA1) and lysosomes (LAMP1). Anti-BACE1 immunoreactivity colocalized with markers for early endosomes, but not lysosomes (Fig. 4E). This pattern is consistent with antibody localizing to compartments where BACE1 is active.

Anti-BACE1 pharmacokinetics are dependent on BACE1 expression in vivo

On the basis of positive cellular inhibition data, we set out to test anti-BACE1 activity in vivo, beginning with examination of pharmacokinetics (PK) in mice. Systemic administration of anti-BACE1 revealed interesting insights into BACE1 biology. The clearance of anti-BACE1 appeared dose-dependent, suggesting the presence of a target-mediated clearance process (Fig. 5A). When dosed at 1 mg/kg, anti-BACE1 was cleared at a rate of 33 ml/kg per day compared to 12 ml/kg per day when dosed at 10 mg/kg. Furthermore, serum anti-BACE1 concentrations observed in two separate ELISA assays that differentiate between total anti-BACE1 (unbound drug plus drug complexed with BACE1) and free anti-BACE1 (unbound drug only) indicated that some of the anti-BACE1 in serum was likely to be bound to soluble BACE1 (Fig. 5A). To confirm that the kinetics of anti-BACE1 was target-dependent, we further evaluated the PK in mice with varying degrees of BACE1 expression (Fig. 5B). Here, we again observed similar kinetics of antibody clearance (32 ml/kg per day) after a single administration to mice with normal endogenous expression of BACE1 (BACE1+/+). Comparatively, in animals with no BACE1 expression (BACE1−/−), slower clearance was observed (6.4 ml/kg per day). This value is similar to that expected for a typical IgG1 with linear PK in mice (23). The clearance of anti-BACE1 in mice with reduced levels of BACE1 expression (BACE1+/−) was reduced by a factor of ~2 (16 ml/kg per day) compared to BACE+/+ mice.

Fig. 5

BACE1−/− mice reveal target-mediated clearance of anti-BACE1 and a baseline for maximal Aβ reduction in vivo. (A) A single dose of anti-BACE1 (1 or 10 mg/kg) was delivered via intravenous injection to C57BL/6J mice. Serum pharmacokinetics (PKs) were analyzed out to 21 days after dose. Observed kinetics were nonlinear, and the difference in total versus free antibody concentrations for samples where anti-BACE1 concentration is <10 μg/ml was suggestive of a target-mediated clearance process. mAb, monoclonal antibody. (B) Serum concentration-time profiles in BACE1+/+, BACE1+/−, and BACE1−/− mice confirm that the clearance of anti-BACE1 is indeed target-mediated and dependent on the levels of endogenous expression of BACE1. (C) To examine the specific contribution of BACE1 to Aβ production in mice, we looked at Aβ1–40 in BACE1+/+ versus BACE1−/− mice. Plasma Aβ1–40 is reduced by 55% (117 ± 19 versus 53 ± 6 pg/ml, P < 0.0005) and forebrain Aβ1–40 by 80% (933 ± 50 versus 185 ± 15 pg/ml, P < 0.0005) in BACE1−/− mice. (D) Single doses of anti-BACE1 (0.03, 0.3, 3, and 30 mg/kg) were delivered by intraperitoneal injection to wild-type C57BL/6J mice. Twenty-four hours later, plasma samples were harvested to analyze Aβ1–40. Plasma Aβ1–40 is significantly reduced at all four doses. Baseline plasma Aβ1–40 [123 ± 34 pg/ml in animals dosed with control IgG (30 mg/kg)] was reduced to 80% (98 ± 22 pg/ml) with anti-BACE1 (0.03 mg/kg) (P = 0.05), 79% (97 ± 10 pg/ml) with anti-BACE1 (0.3 mg/kg) (P = 0.04), 61% (75 ± 22 pg/ml) with anti-BACE1 (3 mg/kg) (P = 0.002), and 70% (86 ± 22 pg/ml) with anti-BACE1 (30 mg/kg) (P = 0.01). Dashed line represents Aβ concentrations in BACE1−/− mice. (E) Associated plasma concentrations of anti-BACE1 at the respective doses showing an expected dose-dependent rise in peripheral drug concentrations with increasing dose. All doses, with the exception of 0.03 mg/kg, are in excess of the cellular IC50 as reported in Fig. 3. *P < 0.05; **P < 0.005; ***P < 0.0005 (t test, compared to control IgG–dosed animals).

Overall, these data provided two important insights. First, the kinetics observed for anti-BACE1 were dependent on the expression of BACE1 in vivo, because simply removing BACE1 genetically slows antibody clearance. Second, BACE1 expression in mice was sufficient to alter antibody kinetics at low anti-BACE1 concentrations. However, modestly higher doses improved exposure, suggesting that, at least peripherally, as long as the level of circulating anti-BACE1 is adequate to overcome the level of BACE1 expression, a therapeutically reasonable dose or dosing regimen can be achieved.

Reduction in peripheral and brain Aβ levels with anti-BACE1 dosing in mice

We next looked to see whether anti-BACE1 could block BACE1 processing of APP in vivo. To establish the maximal Aβ reduction that a BACE1-specific inhibitor could achieve, and to validate our sensitive ELISA assay designed to detect rodent Aβ1–40, we looked at the contribution of BACE1 to Aβ production in plasma and forebrain of BACE1−/− mice compared to BACE1+/+ controls (Fig. 5C). We found that the plasma Aβ1–40 signal is reduced by 55% (117 ± 19 versus 53 ± 6 pg/ml, P < 0.0005), and the brain Aβ1–40 signal by 80% (933 ± 50 versus 185 ± 15 pg/ml, P < 0.0005) in BACE1−/− mice. These results imply that BACE1 is indeed the major β-secretase in the forebrain, but that in the periphery, BACE1 accounts for only partial Aβ1–40 production, with the remainder coming from another β-secretase. These data also set the baseline for maximal Aβ reduction in both the plasma (~50%) and the brain (~80%) with a BACE1-selective antibody (represented by a dashed line as a reference in all wild-type mouse dosing studies).

To determine how effective anti-BACE1 antibodies are at reducing Aβ production in vivo, we first conducted a single-dose, dose-escalating study in wild-type C57BL/6 mice evaluating plasma Aβ reduction (Fig. 5D). On the basis of predicted exposures from our PK studies (Fig. 5A), we selected four dose levels that ranged above and below the cellular IC50 in the periphery 24 hours after dosing. All four doses reduced Aβ concentrations significantly (Fig. 5D). Baseline plasma Aβ concentrations of 123 ± 34 pg/ml were reduced to 98 ± 22 pg/ml with anti-BACE1 (0.03 mg/kg) (P = 0.05), 97 ± 10 pg/ml with anti-BACE1 (0.3 mg/kg) (P = 0.04), 75 ± 22 pg/ml with anti-BACE1 (3 mg/kg) (P = 0.002), and 86 ± 22 pg/ml with anti-BACE1 (30 mg/kg) (P = 0.01). Doses of 3 and 30 mg/kg neared maximal Aβ reduction based on the BACE1 knockout predictions (dashed line). These data also suggest that anti-BACE1 amounts modestly above (0.3 mg/kg) or below (0.03 mg/kg) the predicted cellular IC50 (Fig. 5E) are sufficient to significantly reduce BACE1 activity in vivo.

It is generally believed that antibodies do not efficiently cross the BBB, with estimates on the order of 1:1000 steady-state [brain]/[blood] ratio of IgG (24). On the basis of these initial assumptions, we reasoned that significantly higher doses of anti-BACE1 would be needed to reduce brain Aβ compared to reducing peripheral Aβ. Furthermore, we wanted to establish a relationship between brain and peripheral amounts of anti-BACE1 to Aβ reduction in both compartments (PK/PD relationship). We next dosed wild-type mice with anti-BACE1 or a control IgG systemically, and then examined Aβ1–40 concentrations in both the plasma and the forebrain of treated animals 4 hours after a single intraperitoneal injection of 100 mg/kg. As shown in Fig. 6A, this dose was able to reduce plasma Aβ1–40 by ~50% of control levels, from 108 ± 24 to 56 ± 24 pg/ml (P < 0.0005), similar to what was observed in the BACE1−/− mice (dashed line). However, we detected no change in forebrain Aβ1–40 at this early time point. Anti-BACE1 concentrations in serum were very high (1040 μg/ml) by 4 hours after administration, and brain concentrations (0.75 μg/g) were 0.07% of the concentration in serum (Table 1), closely approximating the predicted 0.1% steady-state penetration of antibodies into the CNS (24). The anti-BACE1 concentrations achieved in the brain, 5 ± 3 nM, are near the cellular IC50 we observed in neurons (Fig. 3). Therefore, we speculated that later time points would result in Aβ reduction at this dose level, because the accumulation of antibody in the brain may be slower than the accumulation of freely diffusible small molecules, where observed Aβ reductions are rapid and transient (5). Furthermore, measurements of Aβ are a balance between production and clearance; thus, 4 hours after dose may be too early to see a reduction in brain Aβ with antibody therapy.

Fig. 6

Systemic dosing of anti-BACE1 inhibits Aβ production in both the periphery and CNS of wild-type mice. (A and B) A single dose of control IgG or anti-BACE1 (100 mg/kg) was delivered by intraperitoneal injection to C57BL/6J mice. (A) Four hours after dose, plasma Aβ1–40 is reduced by 48% (108 ± 24 to 56 ± 24 pg/ml; P < 0.0005), but forebrain Aβ1–40 is not reduced at this time point. (B) One, 2, or 5 days after dose, both plasma and forebrain Aβ1–40 concentrations are significantly reduced. Plasma Aβ1–40 was reduced from a baseline of 144 ± 22 to 104 ± 26 pg/ml (P < 0.005) at 1 day, from 142 ± 15 to 98 ± 10 pg/ml (P < 0.0005) at 2 days, and from 167 ± 20 to 102 ± 44 pg/ml (P < 0.0005) at 5 days. Brain Aβ1–40 was reduced from a baseline of 3150 ± 330 to 2530 ± 240 pg/g (P < 0.0005) at 1 day, from 3050 ± 370 to 2400 ± 220 pg/g (P < 0.0005) at 2 days, and from 3070 ± 320 to 2830 ± 200 pg/g (P = 0.05) at 5 days. (C) Control IgG or anti-BACE1 (30 or 100 mg/kg) was delivered by three intraperitoneal injections, each 4 days apart. Four hours after the last dose, plasma Aβ1–40 is reduced by 50 to 53% (from 144 ± 27 to 68 ± 8 pg/ml at 30 mg/kg, P < 0.0005; from 135 ± 35 to 69 ± 12 pg/ml at 100 mg/kg, P < 0.0005), whereas forebrain Aβ1–40 is not reduced by dosing at 30 mg/kg, but is reduced by 42% (from 2780 ± 210 to 1610 ± 150 pg/g, P < 0.0005) when dosed at 100 mg/kg. Values plotted are means ± SEM. Dashed line represents Aβ concentrations observed in BACE1−/− mice. *P = 0.05; **P < 0.005; ***P < 0.0005 (t test, compared to control IgG–dosed animals).

Table 1

Antibody concentrations from mouse efficacy studies.

View this table:

To test this prediction, we dosed wild-type mice with anti-BACE1 or a control IgG systemically, and then examined Aβ1–40 concentrations in both the plasma and the forebrain of treated animals 1, 2, or 5 days after a single intraperitoneal injection of 100 mg/kg. As shown in Fig. 6B, this dose was able to reduce plasma Aβ1–40 concentrations at all time points. Plasma Aβ1–40 was reduced from a baseline of 144 ± 22 to 104 ± 26 pg/ml (P = 0.001) at 1 day, from 142 ± 15 to 98 ± 10 pg/ml (P < 0.0005) at 2 days, and from 167 ± 20 to 102 ± 44 pg/ml (P < 0.0005) at 5 days. We also see a significant reduction in brain Aβ1–40 concentrations at all time points, with reductions correlating with concentrations of anti-BACE1 in the brain (Table 1). Brain Aβ1–40 was reduced from a baseline of 3150 ± 330 to 2530 ± 240 pg/g (P < 0.0005) at 1 day, from 3050 ± 370 to 2400 ± 220 pg/g (P < 0.0005) at 2 days, and from 3070 ± 320 to 2830 ± 200 pg/g (P = 0.05) at 5 days. Nevertheless, reductions in brain Aβ with a single dose of anti-BACE1 were modest, ~20% of control 2 days after dose.

To see whether we could improve efficacy by elevating brain antibody concentrations through repeated dosing, we performed another experiment administering anti-BACE1 or control IgG at 30 or 100 mg/kg intraperitoneally given every 4 days for a total of three doses. Here, we examined concentrations of Aβ1–40 in both the plasma and the forebrain of treated animals 4 hours after the last dose. Again, we saw ~50% reduction in plasma Aβ1–40 concentrations after multidosing at both 30 mg/kg (from 144 ± 27 to 68 ± 8 pg/ml, P < 0.0005) and 100 mg/kg (from 135 ± 35 to 69 ± 12 pg/ml, P < 0.0005) (Fig. 6C), which represents ~100% reduction relative to the contribution of BACE1 to peripheral Aβ concentrations based on the BACE1−/− mice (dashed line). We also saw a 42% reduction in Aβ1–40 in mouse forebrain at the high dose of anti-BACE1 (from 2780 ± 210 to 1610 ± 150 pg/g, P < 0.0005), which is greater than 50% of BACE1’s contribution to brain Aβ production (dashed line), although no reduction was observed at the low dose. These reductions in brain Aβ correlate with brain antibody concentrations (Table 1). Thus, as predicted, higher antibody concentrations in the brain resulted in robust reductions in Aβ. Notably, there was no difference in peripheral Aβ concentrations at the 30 mg/kg dose compared to the 100 mg/kg dose; thus, simply reducing peripheral Aβ is not sufficient to reduce the concentration of Aβ in the brain.

We wished to evaluate the effect of anti-BACE1 treatment on Aβ production in a mouse model of AD, such as the Tg2576 line, which overexpresses a form of APP carrying the Swedish (Swe) mutation found in a family with the rare familial form of AD (25). We recognized, however, that a potential complication arises from the fact that the Swe mutation appears to render APP a better substrate for BACE1 and may also lead to processing of APP in the secretory pathway (26, 27). Because of these complications, anti-BACE1 may be much less effective in blocking Aβ production from transgenically overexpressed APPSwe than from endogenous wild-type APP. Our results, unfortunately, support this possibility.

Five-month-old Tg2576 mice were treated with anti-BACE1 antibody (30 or 100 mg/kg) or vehicle by intraperitoneal injection every 4 days for a total of three doses, similar to the dosing regimen used in wild-type mice just described. Plasma, cortex, and hippocampus were analyzed for concentrations of soluble human Aβ1–40 and Aβ1–42 (derived purely from the transgene). Plasma Aβ1–40 and Aβ1–42 concentrations showed a 70% reduction compared to control at anti-BACE1 antibody doses of 30 and 100 mg/kg (fig. S3A). Plasma Aβ1–40 was reduced from a baseline of 14,900 ± 880 to 4830 ± 290 pg/ml (P < 0.0005) with anti-BACE1 (30 mg/kg), and to 4130 ± 260 pg/ml (P < 0.0005) with anti-BACE1 (100 mg/kg). Plasma Aβ1–42 was reduced from a baseline of 1360 ± 98 to 360 ± 48 pg/ml (P < 0.0005) with anti-BACE1 (30 mg/kg), and to 340 ± 53 pg/ml (P < 0.0005) with anti-BACE1 (100 mg/kg). However, in contrast to what we saw in wild-type mice, we noted only modest reductions (15 to 22%) in brain Aβ1–40 and Aβ1–42 at the 100 mg/kg dose of the anti-BACE1 antibody (fig. S3A). Mouse hippocampal Aβ1–40 was reduced from a baseline of 4400 ± 130 to 3830 ± 145 pg/g (P = 0.01), and hippocampal Aβ1–42 was reduced from a baseline of 2000 ± 60 to 1560 ± 68 pg/g (P < 0.0005) with anti-BACE1 (100 mg/kg). The concentration of anti-BACE1 antibody in the brain of treated animals increased in a dose-dependent manner (5 ± 4 nM at 30 mg/kg and 14 ± 9 nM at 100 mg/kg) (table S2), and was similar to what we saw in wild-type mice (6 ± 4 nM at 30 mg/kg and 20 ± 11 nM at 100 mg/kg) (Table 1). We believe that the reduced efficacy in Tg2576 mice is a consequence of the animal model, because when we delivered anti-BACE1 via intracerebroventricular delivery to maximize brain concentrations of drug, Aβ reduction remained modest (fig. S3B), despite extremely high concentrations of antibody in the brain after infusion (~20 to 400 times the cellular IC50; table S2). These studies together suggest that high-dose systemic administration of anti-BACE1 may be able to maximally reduce Aβ concentrations in the brain. However, this reduction is modest in Tg2576 mice compared to what was observed in wild-type mice. Thus, we conclude that the Tg2576 mouse is not a good model to evaluate the potential of anti-BACE1 to reduce Aβ production and ameliorate disease pathology.

To further investigate whether APPSwe animal models indeed have limited value for assessing BACE1 activity in vivo, we tested the PS2APP mouse model (28). Three-month-old PS2APP mice were treated with anti-BACE1 antibody (100 mg/kg) or control IgG by intraperitoneal injection every 4 days for a total of three doses. In this case, we once again saw reductions in Aβ in the periphery, but no reduction in brain Aβ (fig. S3C). These data are consistent with the hypothesis derived from previous reports that the APPSwe mutation renders APP a poor substrate for evaluating BACE1 inhibition, likely because APP is cleaved before reaching the cell surface in neurons (26, 27). Because most AD patients carry a wild-type allele of APP, the use of APPSwe animal models may not be representative of the AD patient population.

Anti-BACE1 reduces peripheral and CSF Aβ in nonhuman primates

To investigate the potential use of anti-BACE1 in lowering Aβ production in nonhuman primates, we dosed cynomolgus monkeys with anti-BACE1 or control IgG at 30 mg/kg by intravenous delivery (n = 5 per group). Plasma and CSF were sampled repeatedly in the week before dosing to set mean baseline Aβ1–40 concentrations for each individual animal, and then at various times after dosing. Baseline plasma Aβ1–40 concentrations were fairly uniform across animals (Fig. 7C), whereas CSF Aβ1–40 concentrations were highly variable (Fig. 7D). The high variability in CSF Aβ measurements may represent biological variability because most other parameters were controlled for, eg. diurnal variation, CSF contamination, sex, and age. Thus, all Aβ1–40 measurements were normalized to baseline for each individual monkey. As shown in Fig. 7A, anti-BACE1 treatment reduced plasma Aβ1–40 concentrations rapidly to ~50% of baseline across all individuals. The 50% maximal plasma reductions in Aβ were sustained throughout the 7-day observation period. The serum concentration-time profile for anti-BACE1 was similar to that observed for the control IgG antibody, suggesting kinetics similar to that of a typical IgG1 dosed in the linear range (Fig. 7E). Peak serum anti-BACE1 concentrations of ~800 μg/ml (5 μM) were observed at the time of first sample collection at 15 min after administration and fell to 230 μg/ml (1.5 μM) by 7 days after dose. Notably, at all time points measured after dosing, the serum concentrations of anti-BACE1 exceeded the cellular IC50 of 3 to 17 nM (Fig. 3).

Fig. 7

Anti-BACE1 inhibits peripheral and central Aβ production in nonhuman primates after systemic dosing. Cynomolgus monkeys were dosed with control IgG or anti-BACE1 (30 mg/kg) by intravenous delivery. Hatched lines show data for individual animals, and solid lines show group means. (A) Plasma Aβ1–40 concentrations were reduced to ~50% of baseline across all animals dosed with anti-BACE1. When compared to animals dosed with control IgG, plasma Aβ1–40 concentrations of animals treated with anti-BACE1 were significantly reduced from 6 hours to 7 days after dose (P ≤ 0.005). (B) Compared to control IgG–treated animals, CSF concentrations of Aβ1–40 were significantly reduced to ~50% at 1 and 3 days after dosing with anti-BACE1 (P < 0.005 at 1 day; P < 0.05 at 3 days), with values rising to baseline by 7 days after dose. (C and D) Inherent variability in baseline plasma (C) and CSF (D) Aβ1–40 concentrations. (E and F) Plasma (E) and CSF (F) concentrations of antibody after dosing. (G and H) Correlations between pharmacokinetic (PK) and pharmacodynamic (PD) effects of anti-BACE1 in the periphery (G) and CSF (H).

1–40 concentrations in the CSF of monkeys after anti-BACE1 treatment, although variable, showed a mean reduction of up to 50% from baseline at 1 and 3 days after administration, followed by a trend back toward baseline Aβ concentrations at day 7 after dose (Fig. 7B). We also observed a transient increase in CSF Aβ concentrations just after dosing of both the control IgG and anti-BACE1. This rise may be related to technical limitations of the indwelling catheter system and timing of CSF sampling, or a biological stress response to treatment. Nevertheless, anti-BACE1 was able to counteract this rise and resulted in Aβ concentrations below baseline by 24 hours after dose. These data show that a single dose of anti-BACE1 in monkey significantly reduces both plasma and CSF Aβ.

In the CSF, anti-BACE1 concentrations of 0.2 to 0.3 μg/ml were observed over this time period (Fig. 7F), which translates to ~2 nM. From this, we infer that the brain concentrations of anti-BACE1 are in a similar range. Comparing the PK and PD data (Fig. 7, G and H), these results show that drug exposure in plasma is sufficient to robustly inhibit Aβ production over a 7-day window (Fig. 7G), whereas concentrations in CSF near the cellular IC50 were associated with transiently reduced Aβ concentrations (Fig. 7H). Notably, the only time point where Aβ is not maximally reduced in plasma is at 15 min after dose, regardless of drug concentrations far in excess of the cellular IC50. These correlations suggest that monkeys may be similar to mice, in that ~50% of the Aβ found in plasma may be a result of another β-secretase (Fig. 5C), because increasing serum anti-BACE1 concentrations in monkeys do not improve Aβ reduction beyond 50% (Fig. 7G). In summary, these data suggest that systemically administered anti-BACE1 can reduce BACE1 activity in the brain, as determined by CSF Aβ measurements in a nonhuman primate. However, the doses required to achieve this transient reduction are high, and for a sustained effect, even higher doses may be needed.

Discussion

Supported by both human genetic evidence and clinical histopathological findings, inhibition of amyloidogenic processing of APP remains one of the most promising therapeutic approaches for AD. Recent clinical trial failures highlight the need for more effective therapeutic molecules with which to test the amyloid hypothesis (2). These molecules will be needed to prove whether Aβ is driving AD initiation and progression. BACE1 is a key target for therapeutic drug development because it is at the top of the amyloidogenic cascade for Aβ production. Here, we describe the discovery and characterization of a therapeutic human antibody targeting BACE1. Anti-BACE1 blocks activity of BACE1 in vitro in cells, and in vivo both in the periphery and in the CNS. Anti-BACE1 is unique when compared to previously described BACE1 inhibitors. First, it inhibits BACE1 by binding to an exosite that is distal to the active site. This contrasts with most BACE small-molecule inhibitors, which bind directly to the two active-site aspartate residues. Second, anti-BACE1 is highly selective for BACE1 and does not inhibit the activity of the related proteases, BACE2 or cathepsin D. The selectivity of anti-BACE1 for BACE1 may be a consequence of the mode of binding, because the crystal structure reveals that the antibody binding site on BACE1 is unique and does not share homology with related proteases.

BACE1 cellular activity is believed to be restricted to cellular environments with acidic pH (22). The biochemical properties of aspartyl proteases, combined with recent discoveries that membrane targeting of BACE1 inhibitory peptides enhances cellular potency (20), suggest that BACE1 is acting in endosomes (29). We were therefore concerned that targeting BACE1 with an antibody would not be feasible because the cellular site of action may be inaccessible. We were surprised to see that anti-BACE1 potently inhibited APP processing in cells. These findings suggested either that APP is processed on the cell surface or that anti-BACE1 is internalized into cells where it acts in endosomes to inhibit Aβ production. Consistent with the latter idea, we observed rapid internalization of anti-BACE1 and colocalization with an endosomal marker. Combining these observations with previous studies showing that endocytosis is necessary for APP processing (30), we conclude that anti-BACE1 is likely acting in intracellular vesicles to inhibit BACE1 processing of APP.

In vivo activity of BACE1-targeting small molecules has been limited by brain penetration of many of these compounds. Considering that antibodies (150 kD) are ~375 times larger than a typical small molecule (400 daltons), it would seem unrealistic to consider antibodies as potential CNS drug candidates. Nevertheless, the stability and excellent PK of antibodies (31) allow for less frequent dosing with substantially better exposure in the periphery compared to small-molecule inhibitors. Recent clinical data have also shown that antibodies delivered systemically can alter CNS pathology (32). The question still remains as to how much antibody actually penetrates the CNS and by what mechanism. Past studies have shown that roughly 0.1% of serum antibody concentrations can be measured in the CNS of animals, a ratio that is also observed in humans (24). Consistent with these reports, we observed [brain]/[blood] ratios of a similar magnitude in both our mouse and monkey studies. In assessing the ability of such antibody concentrations in the brain to produce a therapeutic effect, anti-BACE1 is a valuable tool because it enables determination of PK/PD relationships. BACE1 is an enzyme with constitutive function because its product, Aβ, is constantly being generated, and cleared or degraded. Therefore, acute dosing with anti-BACE1 allows one to quantitatively compare peripheral and brain exposure to inhibitory activity by direct measurement of Aβ concentrations in respective compartments. Our studies show that at doses as low as 3 mg/kg or lower, we could maximally inhibit peripheral Aβ production. However, only at high doses correlating with significant brain concentrations of anti-BACE1 did we detect reductions in brain Aβ. These data were further substantiated in monkey, where we observed transient reductions in CSF Aβ when CSF anti-BACE1 approached neuronal IC50 predictions. Again, in monkey, peripheral concentrations of Aβ were robustly reduced over the entire 7-day observation time, matched by anti-BACE1 concentrations far in excess of cellular IC50 measurements at these same time points.

Our data from systemic administration of anti-BACE1 in animals therefore suggest an important way to improve potency of CNS-targeting antibodies, namely, to increase their affinity and subsequent cellular potency, thereby reducing the dose needed to show an effect in the CNS. Because micromolar concentrations in blood correlated to nanomolar concentrations in the brain, nanomolar to subnanomolar binding antibodies are likely to have an effect on CNS targets, as demonstrated here for anti-BACE1. Another approach to enhancing efficacy of CNS-targeting antibodies would be to improve the [brain]/[blood] ratio by increasing CNS exposure. Mechanisms using active transport pathways such as receptor-mediated transcytosis at the BBB combined with anti-BACE1 targeting antibodies are currently being explored to enhance antibody uptake in the brain [see companion study by Yu et al. (33), this issue].

We have shown that BACE1 can be effectively targeted with high-affinity antibodies to selectively reduce BACE1 activity in vivo. With clinical data lacking to definitively prove or disprove the amyloid hypothesis, new ways of tackling promising targets are needed. Further improvements of cellular potency and developing technologies to increase antibody uptake in the brain may help to advance anti-BACE1 antibody therapeutics toward clinically feasible dosing regimens. Nevertheless, our results illustrate that high-affinity antibodies may indeed be effective at targeting enzyme activity in the CNS, and we therefore propose that other promising CNS targets may also be effectively targeted with antibodies.

Materials and Methods

Generation of anti-BACE1 antibodies

Human synthetic antibody libraries with diversities in the H1, H2, H3, and L3 regions were used for selection. Libraries were sorted against biotinylated BACE1 ECD and were captured in neutravidin- or streptavidin-coated plates. Positive clones were identified by ELISA and DNA sequencing and reformatted to full-length IgGs. Affinity maturation was performed with combinatorial complementarity-determining region (CDR) mutagenesis. Stop codons were incorporated into various CDRs. A soft randomization strategy was used for affinity maturation, which introduced a mutation rate of ~50% at selected positions. Antibody clones were screened in vitro for their ability to block proteolytic processing by recombinant BACE1 with the BACE1 HTRF activity assay (see below). Detailed methods can be found in the Supplementary Material.

Affinity measurement with BIAcore

Binding affinities of anti-BACE1 were measured by SPR with a BIAcore 3000 instrument. Anti-BACE1 human IgG was captured by mouse anti-human Fc antibody (GE Healthcare, catalog no. BR-1008-39) coated on CM5 biosensor chips to achieve ~100 response units. For kinetic measurements, twofold serial dilutions (0.98 to 125 nM) of rhBACE1 ECD (amino acids 1 to 457, Genentech) or murine BACE1 ECD (amino acids 1 to 457, Genentech) were injected in PBT buffer [phosphate-buffered saline (PBS) with 0.05% Tween 20] at 25°C with a flow rate of 30 μl/min. Association rates (kon) and dissociation rates (koff) were calculated with a simple one-to-one Langmuir binding model (BIAcore Evaluation Software version 3.2). The equilibrium dissociation constant (KD) was calculated as the koff/kon ratio.

BACE1 HTRF IC50 activity assay

A typical BACE1 HTRF reaction with the 27–amino acid peptide substrate (Bi27) was initiated by the addition of 2 μl of 600 nM Bi27 (Biotin-KTEEISEVNLDAEFRHDSGYEVHHQKL, American Peptide Company) to 6 μl of 2.7 nM rhBACE1 ECD preincubated with anti-BACE1 antibody in BACE reaction buffer [50 mM sodium acetate (pH 4.4) and 0.1% CHAPS] in a 384-well ProxiPlate (Perkin Elmer). The proteolytic reaction with short FRET peptide (Rh-EVNLDAEFK-Quencher, Invitrogen) was carried out with 75 nM rhBACE1 ECD. Detailed methods can be found in the Supplementary Material. The data were analyzed with GraphPad Prism 5.

BACE1, BACE2, and cathepsin D activity assay with Caliper LabChip 3000

BACE1, BACE2, and cathepsin D reactions were carried out in a 384-well microplate. A standard enzymatic reaction contained 1 μM FAM-KTEEISEVNLDAEFRWKK-CONH2, 12 nM enzyme, and inhibitor in reaction buffer. Detailed methods can be found in the Supplementary Material.

Cellular Aβ1–40 assays

1–40 production was measured in 293-HEK cells stably expressing wild-type human APP(695) complementary DNA (cDNA) (293-hAPP). Cells were seeded overnight at 3 × 104 cells per well in a 96-well plate. Fresh media [Dulbecco’s modified Eagle’s medium (DMEM) + 10% fetal bovine serum (FBS)] containing anti-BACE1, control IgG1 (anti-gD), or BACE1 small-molecule inhibitor were incubated with the 293-hAPP cells for 24 hours. The cellular media were harvested and assayed for the presence of Aβ1–40 with an Aβ1–40 HTRF assay (CisBio) according to the manufacturer’s instructions. Dissociated neuronal cultures (plated at 2.5 × 104 cells per well in a 96-well plate) were prepared from embryonic mice as follows: embryonic day 13.5 (E13.5) CD1 mice for DRG, and E16.5 C57BL/6J for cortical and hippocampal neurons. Neurons were grown for 5 days in vitro (DIV). Fresh media containing anti-BACE1, control IgG1 (anti-gD), or BACE1 small-molecule inhibitor were incubated with the neurons for 24 hours. Media were harvested and assayed for Aβ1–40 with a sensitive mouse Aβ1–40 ELISA developed in house (see the Supplementary Material for details). For all cell types, Aβ1–40 values were normalized for cell viability, as determined with the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Experiments were performed at least three times, and each point in each experiment was repeated in duplicate. Data were plotted with a four-parameter nonlinear regression curve-fitting program (KaleidaGraph, Synergy Software).

Western blot analysis of sAPPβ and sAPPα

Conditioned media from 293-hAPP cells or DIV5 cortical neurons were collected as described above and then run on 10% Novex bis-tris gels (Invitrogen) for SDS–polyacrylamide gel electrophoresis analysis. Anti-sAPPβ and anti-sAPPα antibodies were obtained from Covance. Imaging and quantification were performed on the Bio-Rad VersaDoc gel imaging system.

ELISA to assess competitive binding of BACE1 inhibitors

Nunc 96-well Maxisorp immunoplates were coated overnight at 4°C with anti-BACE1 (1 μg/ml) and blocked for 1 hour at room temperature with blocking buffer PBST [PBS and 1% bovine serum albumin (BSA) and 0.05% Tween 20]. Next, serial dilution of anti-BACE1 or BACE1-binding peptides was incubated with a predetermined amount of biotinylated rhBACE1 at room temperature for an hour. The antibody-rhBACE1 or peptide-rhBACE1 mixture was then added to the anti-BACE1–coated plate and incubated for 30 min at room temperature. Subsequently, the plates were washed with wash buffer (PBS with 0.05% Tween 20) and incubated for 30 min with horseradish peroxidase (HRP)–conjugated streptavidin. Last, the plates were washed and developed with tetramethylbenzidine (TMB) substrate. Absorbance was measured spectrophotometrically at 630 nm.

To determine the optimal concentration of biotinylated target protein used for the above competition ELISA assay, we coated and blocked an immunoplate the same way as described above. Then, serial dilution of biotinylated target was incubated with the antibody-coated plate for 30 min at room temperature. The plate was washed with PBST, followed by incubation with HRP-conjugated streptavidin for 30 min at room temperature. Detection of binding signal was as described above. Data were plotted with a four-parameter nonlinear regression curve-fitting program (KaleidaGraph, Synergy Software). The subsaturating concentration of biotinylated rhBACE1 was determined from the curve fitting and applied to the competition ELISA from above.

Generation of anti-BACE1/BACE1 crystals and structural resolution

Coordinates and structure factors of anti-BACE1 and BACE1 complex structure have been deposited in the Protein Data Bank under accession code 3R1G. Detailed methods can be found in the Supplementary Material.

Anti-BACE1 internalization studies and immunocytochemistry

E16.5 dissociated cortical neurons cultured for 14 DIV were fixed with 4% paraformaldehyde and then either permeabilized with 0.1% Triton X-100 or not. Cells were immunostained with anti-BACE1. To examine antibody internalization, we added media containing 1 μM anti-BACE1 to cultures for 10 min to 3 hours and incubated at either 4°C or 37°C. Unbound antibody was washed out thoroughly with Hanks’ balanced salt solution (HBSS) after treatment. Cultures were fixed with 2% paraformaldehyde for 10 min at room temperature and then permeabilized or not. Anti-BACE1 signal was detected with an Alexa 488–conjugated anti-human IgG secondary antibody (Molecular Probes).

For the colocalization studies, cells were costained with anti-EEA1 (BD Biosciences, catalog no. 610456) or anti-LAMP1 (BD Pharmingen, catalog no. 553792), followed by detection with Alexa 594–conjugated secondary antibodies (Molecular Probes).

In vivo mouse studies

PK study

Ten- to 12-week-old C57BL/6J mice (n = 9 per group) or 10- to 12-week-old BACE+/+ (C57BL6/J), BACE1+/−, or littermate BACE1−/− mice (Jackson Laboratories) (n = 9 per group) were dosed with anti-BACE1 (1 or 10 mg/kg) via intravenous delivery. Serum (n = 3 per time point) was collected by periorbital bleed or cardiac puncture (at euthanasia) and analyzed for anti-BACE1 levels.

Wild-type mouse efficacy studies

Eight-week-old female C57BL/6J mice (n = 10 per group) were dosed intraperitoneally with anti-BACE1 or control IgG at doses indicated in the test, for either a single dose or every 4 days, for three doses (that is, days 1, 5, and 9). At defined times after the last dose (see main text), animals were deeply anesthetized and blood was collected by cardiac puncture to isolate plasma. After PBS perfusion, the brain was harvested and the left forebrain was homogenized in 1 ml of PK buffer (1% NP-40 in PBS, with Roche Complete protease inhibitors), whereas the right forebrain was homogenized in 10 volumes of 5 M Gu-HCl and 50 mM tris (pH 8.0) and further diluted 1:10 in Casein Blocking Buffer [0.25% casein, 0.05% sodium azide, aprotinin (20 μg/ml), 5 mM EDTA (pH 8.0), leupeptin (10 μg/ml) in PBS] for Aβ1–40 analysis.

Cynomolgus monkey PK and PD study

Animals (male cynomolgus macaques aged 2.5 to 4 years; five animals per group) were surgically prepared with indwelling cannulae inserted into the cisterna magna and connected to a subcutaneous access port to permit CSF sampling. Control IgG or anti-BACE1 was delivered at 30 mg/kg via an intravenous bolus injection into a cannulated peripheral vein. CSF and blood samples (for plasma and serum) were collected at the same time of day.

All animal protocols were approved by the Institutional Animal Care and Use Committee of Genentech and were conducted in accordance with the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals.

PK and PD assays in mice and cynomolgus monkeys

Detailed methods can be found in the Supplementary Material.

Supplementary Material

www.sciencetranslationalmedicine.org/cgi/content/full/3/84/84ra43/DC1

Materials and Methods

Fig. S1. Anti-BACE1 inhibits BACE1 in a short peptide BACE1 FRET assay.

Fig. S2. Anti-BACE1 internalization can be blocked by cold incubation and occurs specifically via binding to BACE1.

Fig. S3. Anti-BACE1 dosing in two APP-transgenic mouse models.

Table S1. Crystallography data statistics.

Table S2. Antibody concentrations from APP-Tg mouse studies.

References

Footnotes

  • Citation: J. K. Atwal, Y. Chen, C. Chiu, D. L. Mortensen, W. J. Meilandt, Y. Liu, C. E. Heise, K. Hoyte, W. Luk, Y. Lu, K. Peng, P. Wu, L. Rouge, Y. Zhang, R. A. Lazarus, K. Scearce-Levie, W. Wang, Y. Wu, M. Tessier-Lavigne, R. J. Watts, A Therapeutic Antibody Targeting BACE1 Inhibits Amyloid-β Production in Vivo. Sci. Transl. Med. 3, 84ra43 (2011).

References and Notes

  1. Acknowledgments: We thank S. Prabhu, L. de Forge, and S. Fisher for scientific discussions and guidance; J. Maloney for providing neuronal cultures; and H. Solonoy for assistance with in vivo experiments. We acknowledge the Genentech protein expression and purification groups. Portions of this research were carried out at the Stanford Synchrotron Radiation Laboratory, a national user facility operated by Stanford University on behalf of the U.S. Department of Energy. Funding: Supported by Genentech. Author contributions: R.J.W., M.T.-L., and J.K.A. designed the project. Y.C. and C.C. generated the phage display antibodies and performed in vitro binding experiments and competitive ELISA experiments. Y.W. designed multiple strategies on naïve panning and affinity maturation; oversaw the panning, binding, and competitive ELISA experiments; and analyzed the related data. Y.Z. and R.A.L. contributed to the phaging strategy, and R.A.L. contributed to analysis of the exosite. P.W. and L.R. purified and crystallized the BACE1–anti-BACE1 Fab complex, and W.W. oversaw the crystallization experiments, solved and analyzed the structure, and generated the corresponding figures and text. Y. Liu and C.E.H. performed and analyzed in vitro biochemical assays. J.K.A. performed and analyzed all cell-based experiments. J.K.A., D.L.M., W.J.M., and K.S.-L. designed, performed, oversaw, and analyzed in vivo experiments. K.H., W.L., Y. Lu, and K.P. ran PK and PD assays, and Y. Lu developed anti-BACE1 and mouse Ab1–40 assays. J.K.A. and R.J.W. wrote the manuscript, with valuable input from M.T.-L., R.A.L., and D.L.M. Data and materials access: Data and materials are available from Genentech under a materials transfer agreement. Competing interests: All authors are paid employees of Genentech. Genentech has filed a patent application related to this work on using an antibody derived from a human synthetic antibody library to inhibit β-secretase.
View Abstract

Navigate This Article