Alzheimer’s disease (AD) is characterized by deposition of amyloid-β (Aβ) plaques and neurofibrillary tangles in the brain, accompanied by synaptic dysfunction and neurodegeneration. Antibody-based immunotherapy against Aβ to trigger its clearance or mitigate its neurotoxicity has so far been unsuccessful. Here we report the generation of aducanumab, a human monoclonal antibody that selectively targets aggregated Aβ. In a transgenic mouse model of AD, aducanumab is shown to enter the brain, bind parenchymal Aβ, and reduce soluble and insoluble Aβ in a dose-dependent manner.
In patients with prodromal or mild AD, one year of monthly intravenous infusions of aducanumab reduces brain Aβ in a dose- and time-dependent manner. This is accompanied by a slowing of clinical decline measured by Clinical Dementia Rating—Sum of Boxes and Mini Mental State Examination scores. The main safety and tolerability findings are amyloidrelated imaging abnormalities. These results justify further development of aducanumab for the treatment of AD. Should the slowing of clinical decline be confirmed in ongoing phase 3 clinical trials, it would provide compelling support for
the amyloid hypothesis.
The amyloid hypothesis posits that Aβ-related toxicity is the primary cause of synaptic dysfunction and subsequent neurodegeneration that underlies the progression characteristic of AD1,2. Genetic, neuropathological, and cell biological evidence strongly suggest that targeting Aβcould be beneficial for patients with AD3,4. So far, attempts at therapeutically targeting Aβ have not been successful5–7, casting doubt on the validity of the amyloid hypothesis. However, the lack of success may have been due to the inability of the antibodies to adequately engage their target or the proper target in the brain, or selecting the wrong
patient population. We describe the development of an antibody-based immunotherapeutic approach by selecting human B-cell clones triggered by neo-epitopes present in pathological Aβ aggregates. The screening of libraries of human memory B cells for reactivity against aggregated Aβ led to molecular cloning, sequencing, and recombinant expression of aducanumab (BIIB037), a human monoclonal antibody that selectively reacts with Aβ aggregates, including soluble oligomers and insoluble fibrils. In preclinical studies, we show that an analogue of aducanumab is capable of crossing the blood–brain barrier, engaging its target, and clearing Aβ from plaque-bearing transgenic mouse brains. These results prompted the start of clinical trials8.
We report here interim results from a double-blind, placebocontrolled phase 1b randomized trial (PRIME; ClinicalTrials.gov identifier NCT01677572) designed to investigate the safety, tolerability, pharmacokinetics, and pharmacodynamics of monthly infusions of aducanumab in patients with prodromal or mild AD with brain
Aβ pathology confirmed by molecular positron emission tomography (PET) imaging. Together, our data support further development of aducanumab as an Aβ-removing, disease-modifying therapy for AD.
Removal of brain Aβ plaques in patients with AD
In the PRIME study, 165 patients were randomized and treated between October 2012 and January 2014 at 33 sites in the United States. Patients with a clinical diagnosis of prodromal or mild AD and visually positive
Aβ PET scan9 were given monthly intravenous infusions of placebo or aducanumab at doses of 1, 3, 6 or 10 mg kg−1 for one year. Of these patients, 125 completed and 40 discontinued treatment, most commonly
due to adverse events (20 patients) and withdrawal of consent (14 patients): 25% of the placebo group discontinued compared with 23%, 19%, 17%, and 38% of the 1, 3, 6 and 10 mg kg−1 aducanumab dose
groups, respectively (Extended Data Fig. 1). Baseline characteristics, including cognitive measures, were generally well-balanced across the groups, although the 1 mg kg−1 dose group included a higher proportion
of patients with mild AD, and the aducanumab treatment groups tended to have a higher Clinical Dementia Rating—Sum of Boxes (CDR-SB) score (Table 1).
Treatment with aducanumab reduced brain Aβ plaques as measured by florbetapir PET imaging in a dose- and time-dependent fashion (Fig. 1, 2a). The mean PET standard uptake value ratio (SUVR) composite
score at baseline was 1.44. After 54 weeks of treatment, this had decreased significantly (P < 0.001) in the 3, 6 and 10 mg kg−1 dose groups; whereas change for the placebo group was minimal (Fig. 2a, Extended Data Table 1). In the 10 mg kg−1 dose group, the SUVR composite score was 1.16 after 54 weeks of treatment, a value near thepurported quantitative cut-point of 1.10 that discriminates between positive and negative scans (Fig. 2b)10. The adjusted mean changes in SUVR composite scores in the 6 and 10 mg kg−1 groups treated for 26 weeks were similar in magnitude to the dose group below (3 and 6 mg kg−1, respectively) treated for 54 weeks (Fig. 2a). Reductions in amyloid PET SUVR composite score in aducanumab-treated patients were similar in patients with mild and prodromal AD, and apolipoprotein E (ApoE) ε4 carriers and non-carriers (Extended Data Fig. 2a, b). Pre-specified regional analyses of SUVR changes demonstrated statistically significant dose-dependent reductions in all brain regions, except for the pons and sub-cortical white matter, two areas in which
Aβ plaques are not expected to accumulate (Extended Data Fig. 3).
Effect on clinical measures
Clinical assessments were exploratory as the study was not powered to detect clinical change. The test of dose response was the pre-specified primary analysis for the clinical assessments. Analysis of change from baseline on the CDR-SB (adjusted for baseline CDR-SB and ApoE ε4 status) demonstrated dose-dependent slowing of clinical progression with aducanumab treatment at one year (dose-response, P < 0.05), with the greatest slowing for 10 mg kg−1 (P < 0.05 versus placebo) (Fig. 3a, Extended Data Table 1). Sensitivity analysis using a mixed model for repeated measures (MMRM) also showed a trend for slowing of decline on the CDR-SB at one year (P = 0.07 with 10 mg kg−1 aducanumab versus placebo). A dose-dependent slowing of clinical progression on the Mini Mental State Examination (MMSE) with aducanumab treatment was also observed at one year (dose-response, P < 0.05), with the greatest effects at 3 and 10 mg kg−1 aducanumab (P < 0.05 versus placebo) (Fig. 3b, Extended Data Table 1). On sensitivity analysis using MMRM, the greatest difference was retained for 10 mg kg−1 aducanumab (P < 0.05 versus placebo), with a smaller difference at 3 mg kg−1 (P = 0.10 versus placebo). No changes from baseline after one year were found on the composite europsychological test battery (NTB) or the Free and Cued Selective Reminding Test (FCSRT) free recall (Extended Data Table 1), but skewed non-normal (floor) effects at baseline were observed. The floor effects on the NTB were seen in
the individual tests; specifically, in the two most clinically relevant components given the stage of the population enrolled: Wechsler Memory Scale-Fourth Edition Verbal Paired Associates II (WMS-IV VPA II)
and Rey Auditory Verbal Learning Test (RAVLT) delayed recall of the NTB memory domain.
Safety and tolerability
The most common adverse effects were amyloid-related imaging abnormalities (ARIA), headache, urinary tract infection, and upper respiratory tract infection (Table 2). Using the most specific description description of ARIA by magnetic resonance imaging (MRI), ARIA-vasogenic oedema (ARIA-E) abnormalities occurred in no patients receiving placebo compared with 1 (3%), 2 (6%), 11 (37%), and 13 (41%) patients receiving 1, 3, 6 and 10 mg kg−1 aducanumab, respectively (Extended Data Table 2). ARIA-E was generally observed early in the course of treatment, MRI findings typically resolved within 4–12 weeks, and of the 27 patients who developed ARIA-E, 15 (56%) continued treatment (Supplementary Information). All cases of symptomatic required to be reported as medically important serious adverse effects. No patients were hospitalised for ARIA. The only serious adverse effects (by preferred term) that occurred in more than one patient in any treatment group were ARIA (0, 1 (3%), 1 (3%), 4 (13%), and 5 (16%) of patients receiving placebo, and 1, 3, 6 and 10 mg kg−1 aducanumab, respectively) and superficial siderosis of the central nervous system (0, 1 (3%), 0, 2 (7%), and 3 (9%) of patients receiving placebo and 1, 3, 6 and 10 mg kg−1 aducanumab, respectively). Owing to the requirement for repeated MRI assessments of those patients who developed ARIA, these individuals were partially unblinded to treatment. Other adverse effects and serious adverse effects were consistent with the
patient population. There were no drug-related deaths (Supplementary Information).
The pharmacokinetics of aducanumab (maximum concentration (Cmax) and cumulative area under the concentration curve (AUC)) were linear across the dose range in patients who received all 14 planned
doses (Extended Data Table 3). The median plasma half-life was 21 days. In total, 3 of 118 evaluable patients (3%) in the combined aducanumab groups developed treatment-emergent anti-aducanumab antibodies within the first year of treatment. Antibody responses were transient, with minimal titres, and had no apparent effect on aducanumab pharmacokinetics or safety.
Brain penetration and binding to Aβ plaques
In the preclinical studies which preceded PRIME, systemically administered aducanumab (single dose, 30 mg kg−1 intraperitoneally (i.p.)) bound to diffuse and compact Aβ plaques in the brains of 22-month-old female Tg2576 transgenic mice (‘Target engagement study’; Extended Data Fig. 4a–d). Cmax in plasma was 181 μg ml−1, with a terminal elimination half-life (t1/2) of 2.5 days. The Cmax in brain was 1,062 ng g−1 of tissue, and approximately 400–500 ng g−1 of drug was measured 3 weeks after dosing, suggesting long-term retention. Consequently, the brain:plasma AUC ratio of 1.3% was higher than the 0.1% frequently reported for systemically administered antibodies11,12. Administration of a single dose of aducanumab did not affect plasma
Reduction of brain Aβ in transgenic mice
Exposure in plasma and brain correlated linearly with dose after chronic dosing in plaque-bearing transgenic mice (Extended Data Fig. 5) (Supplementary Information). chaducanumab, a murine IgG2a/κ chimaeric analogue, dose-dependently reduced Aβ measured in brain homogenates by up to 50% relative to the vehicle control in the diethylamine (DEA) fraction that extracted soluble monomeric and oligomeric forms of Aβ40 and Aβ42, and in the guanidine hydrochloride (GuHCl) fraction that extracted insoluble Aβ fibrils (Fig. 4a, b). Quantitative 6E10 immunohistochemistry showed significant reductions in all forms of Aβ deposits by up to 70% (Fig. 4c, d). Thioflavin S (ThioS) staining of compact Aβ plaques showed dose-dependent and
statistically significant reductions in the cortex and hippocampus by up to 63% (Fig. 4c, d). Quantitative histology indicated that chaducanumab significantly reduced the number of plaques of all sizes, including plaques >500 μm2 and plaques <125 μm2 (Extended Data Fig. 6a–c). Quantification of ThioS-positive vascular and parenchymal Aβ plaques separately showed that chaducanumab did not affect vascular
Aβ in either cortex or hippocampus (Fig. 4e–h). To identify the mechanism of Aβ clearance, we analysed the involvement of microglia which are known to display enhanced phagocytic activities through binding to the Fc region of an antibody13,14. chaducanumab significantly increased recruitment of Iba-1-positive microglia to Aβ plaques, suggesting FcγR-mediated phagocytosis of antibody–Aβ complexes as a possible clearance mechanism (Extended Data Fig. 7a–c and Supplementary Information).
The apparent affinities of aducanumab and chaducanumab for aggregatedAβ42, with half maximal effective concentration (EC50) values of 0.1 nM, were comparable to 3D6 (ref. 13) (Fig. 5a). Neither aducanumab
nor chaducanumab bound monomeric soluble Aβ40 at concentrations up to 1 μM, indicating >10,000-fold selectivity for aggregated Aβ over monomer, whereas 3D6 bound soluble Aβ40 with an EC50 of 1 nM
(Fig. 5b). In contrast to 3D6, which immunoprecipitated both monomeric and aggregated Aβ, chaducanumab bound soluble Aβ42 oligomers and insoluble Aβ42 fibrils prepared in vitro, but not Aβ42 monomers (Fig. 5c). Histological staining of autopsy tissue from patients with AD or aged amyloid precursor protein (APP) transgenic mice confirmed binding of aducanumab to bona fide human Aβ fibrils (Fig. 5d, e).
The PRIME study shows that aducanumab penetrates the brain and decreases Aβ in patients with AD in a time- and dose-dependent manner. Within 54 weeks of treatment, 3, 6 and 10 mg kg−1 doses of aducanumab significantly decreased the amyloid PET SUVR. Patients receiving placebo showed virtually no change in their mean PET SUVR composite scores over one year, indicating that Aβ pathology had already reached an asymptote of accumulation. Considering that it may have taken up to 20 years for Aβ to have accumulated to the levels in these patients at study entry15, the observed kinetics of Aβ removal within a 12-month time period appears encouraging for a diseasemodifying treatment for patients with AD.
The cognitive results for CDR-SB and MMSE provide support for the clinical hypothesis that reduction of brain Aβ confers a clinical benefit. Post hoc analysis showed that those aducanumab-treated patients who had decreased SUVR scores >1 standard deviation unit relative to placebo-treated patients after one year of treatment experienced a stabilization of clinical decline on both CDR-SB and MMSE scores; whereas, those patients with a smaller or no decrease experienced clinical decline similar to placebo patients (Fig. 2c). The apparent clinical benefit observed in PRIME could also be explained by the binding of aducanumab to oligomeric forms of Aβ, which would not be directly detected by PET imaging. The reductions in SUVR scores may be surrogates for reductions in toxic soluble Aβ oligomers which may have had a more functionally relevant impact on cognition. Whereas significant Aβ reduction was detectable by 6 months, clinical effects were not and slowing of disease progression is not altogether surprising.
The main safety finding, ARIA-E, was dose-dependent and more common in ApoE ε4 carriers, consistent with findings with other anti-Aβ monoclonal antibodies7,16,17. Although the underlying cause of ARIA is not well understood, it is likely that the MRI signal of ARIA is due to increased extracellular fluid. This may be a result of underlying CAA, changes in perivascular clearance and vascular integrity, or local inflammatory processes associated with Aβ-targeting therapies17–20 (see Supplementary Information for further discussion).
Study limitations of the PRIME phase 1b study included staggered parallel-group design, small sample sizes, limited region (USA only), and possible partial unblinding due to ARIA-E. Measures were taken to maintain blinding to adverse effects: raters of given tests were not permitted to perform other clinical assessments, and were blinded to other assessments (for example, MMSE and CDR raters were required to be different and neither were permitted to perform other study assessments). Post hoc analyses of change from baseline PET SUVR composite score and cognition by presence/absence of ARIA suggested no apparent difference in treatment effect when comparing patients with and without ARIA-E (Extended Data Table 4). There was overlap in enrolment in Arms 1–3 (aducanumab 1 and 3 mg kg−1, placebo) and Arms 4 and 5 (aducanumab 10 mg kg−1, placebo) but Arms 6 and 7 (aducanumab 6 mg kg−1, placebo) were initiated after enrolment in Arms 1–5 was complete. This was a small study designed for assessment of safety and tolerability, and for detecting a pharmacological effect on brain Aβ levels measured by PET imaging. The trial was not powered for the exploratory clinical endpoints, thus the clinical cognitive results should be interpreted with caution. Primary analyses were based on observed data with no imputation for missing values, nominal P values were presented with no adjustments for multiple comparisons, and they were supported by sensitivity analyses using a MMRM. The initiation of the PRIME study and its results are supported by extensive preclinical data. Detection on parenchymal Aβ plaques following a single systemic administration confirmed that aducanumab
penetrates the brain to a sufficient extent to allow accumulation on Aβ plaques. This is consistent with earlier findings showing that, in the presence of significant Aβ deposition, plaque-binding antibodies can be detected bound to the target over an extended period14,21. The minimal effective dose upon repeated systemic administration of chaducanumab in transgenic mice was 3 mg kg−1 (corresponding to minimally effective concentrations of 13.8 } 1.9 μg ml−1 in plasma and 99.8 } 30.0 ng g−1 in brain) with reductions of Aβ42 in soluble and insoluble brain fractions of approximately 50%, and reductions in Aβ plaque of approximately 40%. Since exposure at 3 mg kg−1 in animals and humans is approximately equivalent, the observed dose-response in the model was consistent with the clinical doses that led to reductions in amyloid PET SUVR. chaducanumab cleared plaques of all sizes, suggesting that aducanumab triggered clearance of pre-existing Aβ plaques and prevented formation of new plaques. In transgenic mice, aducanumab preferentially bound to parenchymal Aβ over vascular Aβ deposits, consistent with the lack of effect on vascular Aβ following chronic dosing. The effect of anti-Aβ antibody therapies on the vascular Aβ compartment could be related to micro-haemorrhages or oedema in transgenic mice, and may relate to ARIA in clinical trials22. Nevertheless, the preferential binding of aducanumab to parenchymal versus vascular Aβ may have been critical in allowing the use of relatively high doses in the clinical study so as to achieve robust target engagement in the brains of patients with AD. Several mechanisms may be involved in aducanumab’s Aβ-lowering activity. The clearance of Aβ deposits was accompanied by enhanced recruitment of microglia. Together with the reduced potency of the aglycosylated form of chaducanumab (data not shown), and the ex vivo phagocytosis data, this suggests that FcγR-mediated microglial recruitment and phagocytosis played an important role in Aβ clearance in these models. Activated microglia appeared to encapsulate the remaining central dense core of plaques in treated animals, possibly isolating them from the surrounding neuropil. It is commonly thought
that soluble Aβ oligomers, rather than monomers or plaques, may be the primary toxic species23,24. Considering that Aβ plaques might be a source of Aβ oligomers25–28, this suggests that treatment with
aducanumab might slow their release into the neuropil, thereby limiting their toxic effect on neurons29. In fact, chronic dosing of 18-monthold Tg2576 transgenic mice with chaducanumab led to normalization of neuritic calcium overload in the brain30. Other studies have linked calcium dyshomeostasis in neurons and microglia to binding of Aβ oligomers to metabotropic receptors31–33. Aducanumab binding to soluble Aβ oligomers may prevent their interaction with those receptors, thereby preventing the detrimental effect of membrane depolarization. Restoration of this functional endpoint suggests that aducanumab treatment may lead to beneficial effects on neuronal network function underlying cognitive deficits. Together, the clinical and preclinical data support continued development of aducanumab as a disease-modifying treatment for AD. The
clinical study results provide robust support to the biological hypothesis that treatment with aducanumab reduces brain Aβ plaques and, more importantly, to the clinical hypothesis that Aβ plaque reduction confers
clinical benefit. This concurs with preclinical data demonstrating brain penetration, target engagement, and dose-dependent clearance of Aβplaques in transgenic mice. The clinical effects of aducanumab need to be confirmed in larger studies. Both the long-term extension (LTE) phase of this study and phase 3 development are ongoing.
Published online 24 August 2016.