Reversibility of cognitive worsening observed with BACE inhibitor umibecestat in the Alzheimer's Prevention Initiative (API) Generation Studies

Abstract

INTRODUCTION

The Alzheimer's Prevention Initiative (API) Generation Studies evaluated the BACE inhibitor umibecestat for Alzheimer's disease (AD) prevention. The studies were terminated early, and the reversibility of umibecestat's side effects was assessed.

METHODS

Cognitively unimpaired 60‐ to 75‐year‐old apolipoprotein E (APOE) ε4 homozygotes and heterozygotes (the latter with elevated brain amyloid deposition) (n = 1556) received umibecestat (50 or 15 mg daily) or placebo for 7 months on average and were followed for a median (interquartile range) of 4 (3 to 6) months after washout.

RESULTS

Compared to placebo, umibecestat‐treated participants had small, non‐progressive, but statistically significant decline in performance on certain cognitive batteries including Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) and API Preclinical Composite Cognitive test, but not Clinical Dementia Rating‐Sum of Boxes. RBANS differences were no longer significant at the end of follow‐up.

DISCUSSION

In people at genetic risk for AD, high‐dose beta‐site amyloid precursor protein cleaving enzyme (BACE) inhibition was associated with early mild cognitive worsening, which reversed shortly after washout, suggesting a symptomatic side effect not associated with neurodegeneration. Fully anonymized data, images, and samples are available upon request for further research on BACE inhibition.

Highlights

• This is the first trial with blinded assessment of reversibility of BACE inhibitor side effects.

• Umibecestat was tested in cognitively unimpaired persons at genetic risk for AD.

• Umibecestat led to early mild cognitive decline that reversed shortly after washout.

• This suggests a potentially manageable effect not associated with neurodegeneration.

• Further research may determine the future of BACE inhibition in AD prevention.

1. BACKGROUND

The number of people living with dementia, currently estimated at 52 million globally, is predicted to reach 153 million individuals in 2050. ¹ Alzheimer's disease (AD), the main cause of dementia, is also the most common cause of cognitive impairment in older adults. The earliest pathophysiological features of AD begin more than 20 years before the onset of cognitive impairment and are thought to include the accumulation of soluble and insoluble amyloid aggregates that are associated with amyloid plaques. ² While several amyloid‐modifying treatments have failed to demonstrate a clinical benefit in cognitively impaired patients, ³ recent pivotal trials have shown clinical benefit from amyloid plaque‐clearing antibody treatments in the early clinical stages of AD. ⁴ , ⁵ , ⁶

Inhibitors of the beta‐site amyloid precursor protein cleaving enzyme‐1 (BACE1), a β secretase involved in the production of amyloid β (Aβ), have the potential to reduce the accumulation of amyloid plaques. By interfering with the cleavage of APP into amyloid fragments, BACE1 inhibitors may be effective in slowing rather than reversing amyloid plaque deposition. However, all BACE1 clinical trials so far have been discontinued for futility or signs of cognitive worsening, non‐progressive brain shrinkage, or systemic toxic effects. ⁷ , ⁸ , ⁹ Since most BACE1 inhibitors also block the activity of BACE2, ¹⁰ it was not clear whether these adverse effects would be observed with a relatively selective BACE1 inhibitor like umibecestat.

Umibecestat, an oral BACE1 inhibitor with relative BACE1/BACE2 selectivity (ratio 2.7), reduced the levels of Aβ in the brain and cerebrospinal fluid (CSF) of rats and dogs compared to placebo and slowed the deposition of Aβ plaques in APP‐transgenic mice when given prior to symptom onset, without apparent toxicity. ¹¹ In a 3‐month phase 2a study conducted in healthy participants aged ≥60 years, 2 to 85 mg doses of umibecestat were found to be safe and well tolerated with no indication of impaired neurological function. The reduction of CSF Aβ_1‐40 observed after 3 months of treatment led to the selection of umibecestat doses of 15 and 50 mg, which were predicted to achieve about 70% and 90% BACE inhibition, respectively. ¹¹ We postulated that such treatment could have a more profound impact if initiated before the onset of cognitive impairmentwhen the Aβ plaque burden is already extensive. ¹²

The Alzheimer's Prevention Initiative (API) Generation Program enrolled cognitively unimpaired elderly individuals at high risk of developing clinical symptoms of AD based on the presence of two ε4 alleles of the apolipoprotein E (APOE) gene (APOE ε4 homozygotes) in Generation Study 1 and 2 (GS1 and GS2), or the presence of one ε4 allele (APOE ε4 heterozygotes) accompanied by evidence of Aβ deposition in GS2 only. The presence of one or two APOE ε4 alleles increases the risk of developing AD and lowers the average age of onset from 84 to 75 years with one allele and to 68 years with two alleles. ¹³ , ¹⁴ , ¹⁵ The Generation studies were originally designed to assess the effects of potential anti‐amyloid therapies including umibecestat and the active immunotherapy CAD106 on cognition, global clinical status, imaging, and fluid biomarkers of AD, in participants at genetic risk for the onset of clinical symptoms (mild cognitive impairment [MCI] or dementia) of AD. ¹⁶ However, treatment with umibecestat in both studies was terminated prematurely, mainly due to mild, early worsening in some measures of cognition among other side effects during a preplanned analysis conducted by the independent data monitoring committee (DMC). As prespecified in the protocol, subsequent assessments following treatment discontinuation were maintained, blinded to the original treatment allocation, in order to establish the reversibility of the observed adverse findings. Here, we report the worsening in cognition and other effects associated with umibecestat treatment and their reversibility following treatment discontinuation.

2. METHODS

2.1. Study design and early termination

Both API GS1 (ClinicalStudies.gov, number NCT02565511) and GS2 (ClinicalStudies.gov, number NCT03131453) were phase 2/3 randomized, double‐blind, placebo‐controlled, parallel‐group studies. Participants from 10 (GS1) and 25 (GS2) countries and 129 and 194 centers, respectively, received umibecestat (once‐daily capsule) or a matching placebo. In addition to the cohort of participants randomized to umibecestat or placebo (Cohort II), GS1 comprised another cohort (Cohort I) of participants randomized to the active Aβ immunotherapy CAD106 versus placebo, the results of which are reported separately. ¹⁷ Treatment was originally due to be administered for at least 5 years up to a maximum of 8 years, with subsequent follow‐up of participants planned for an additional 3 months without treatment. Both studies, initiated in February 2017 (GS1 Cohort II) and August 2017 (GS2) with recruitment on track (full recruitment was expected by the end of 2019), were terminated early (July 2019). ¹⁸ This decision followed a preplanned interim review by the DMC based on pooled data across umibecestat arms indicating that the participants treated with umibecestat had unexpected, mild, early worsening of cognition, increased brain volume loss, and greater mean body weight loss compared to those on placebo (see Lopez Lopez et al. ¹⁶ and Figure S1 for the designs of each study).

The description of the study protocols conforms to the 2013 Standard Protocol Items: Recommendations for Interventional Studies, ¹⁹ and the protocols were approved by the Institutional Review Board at each study site. The studies were designed, executed, and reported according to the International Council for Harmonization guidelines for good clinical practice and ethical principles outlined in the Declaration of Helsinki. ¹⁹

RESEARCH IN CONTEXT

1. Systematic review: We searched PubMed for clinical studies of beta‐site amyloid precursor protein cleaving enzyme (BACE) inhibitors in Alzheimer's disease (AD). Nearly all identified studies were conducted in patients with early AD, and none included a prospective follow‐up blinded to the treatment allocation after washout.

2. Interpretation: The Alzheimer's Prevention Initiative (API) Generation studies evaluated umibecestat in cognitively unimpaired individuals at genetic risk of AD with a predefined blinded follow‐up after treatment discontinuation. Consistent with previous reports of BACE inhibitors, umibecestat was associated with mild worsening of cognitive function. The cognitive decline occurred early during treatment, did not progress over time, was not associated with changes in clinical dementia ratings, and was reversed shortly after treatment washout (as assessed within 3 to 6 months).

3. Future directions: Although negative, the trials’ results suggest a mechanism not related to neurodegeneration that might be mitigated once clarified with further research considering lower doses of BACE inhibitors. The whole dataset, including images and samples, is available upon request.

2.2. Participants

Participants were eligible for inclusion if they were aged 60 to 75 years (divided into two strata, with a maximum of 20% in the younger 60‐ to 65‐year stratum as determined by prior modeling ²⁰ ) and APOE ε4 homozygotes (GS1) or either APOE ε4 heterozygotes with elevated brain amyloid or APOE ε4 homozygotes (GS2). Participants were considered to have elevated brain amyloid based on either CSF or positron emission tomography (PET) measures. In addition, participants had to be cognitively unimpaired, defined by a Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) delayed memory index score of ≥85 and a Clinical Dementia Rating (CDR) global score of 0 (see Supplementary File for inclusion/exclusion criteria).

Written informed consent from all participants included consent to receive genetic counselling and disclosure of their risk estimates of developing clinical symptoms of AD (impact of APOE genotype and amyloid status disclosures will be published separately; see Consent statement). ²¹

2.3. Randomization and masking

All eligible participants were randomized via an interactive response technology to one of the available treatment arms and for GS1 across Cohort I versus Cohort II. The randomization was stratified based on age, region, genotype, and method (CSF or PET) used to determine amyloid status. The randomization scheme was reviewed and approved by the Novartis Randomization Office. Participants, site personnel, and data analysts remained blinded to the treatment identity up to the database lock, after completion of the preplanned follow‐up after treatment discontinuation.

2.4. Procedures

In GS1 Cohort II, participants were randomized to receive daily oral doses of umibecestat 50 mg or placebo in a 3:2 ratio. In GS2, participants were randomized to receive umibecestat 15 mg, umibecestat 50 mg, or placebo in a 1:2:2 ratio.

Pooled data were summarized on visits at Weeks 13, 26, and 52, one of them often being the visit at the end of treatment. After studying drug discontinuation, participants completed one or two additional visits including cognitive assessments within a 6‐month period. If the visit occurred within 31 days from the last dose (that is five times umibecestat terminal half‐life of 6.25 days, ie, before complete washout) the related cognitive assessment is further referred to as last visit on treatment; conversely, if the visit occurred more than 31 days from the last dose, it is referred to as last visit after washout.

2.5. Outcomes

The original study design was based on a dual endpoints approach, with two separate primary endpoints including time to event (with event defined as MCI or dementia due to AD) and a measure of cognitive decline (5‐year change in the API Preclinical Composite Cognitive test [APCC] score). ²⁰ The key secondary endpoints were change in Clinical Dementia Rating–Sum Of Boxes (CDR‐SOB), total RBANS score, and individual neurocognitive domain index scores, planned to be assessed from baseline to Month 60 (ie, 5‐year change).

Other secondary endpoints included change in (1) total scores of Everyday Cognition Scale (ECog); (2) AD‐related CSF/plasma biomarkers (Aβ₄₀ and Aβ₄₂), as well as biomarkers of neurodegeneration (CSF total tau and phosphorylated tau₁₈₁ [p‐tau181]), serum neurofilament light chain [NfL]); and (3) volume of brain regions (including the hippocampus and whole brain) as measured by volumetric MRI. Adverse events (AEs) including weight change were also monitored. Blood samples were collected before dosing on the day of the visit. See Supplementary File for details on the original endpoints and biomarker assessments (Section 2) and for descriptive statistics for all umibecestat concentrations (Table S1).

Due to study termination, treatment duration did not exceed 12 months in the majority of participants, and no data were collected at Month 60; therefore, both primary and most secondary endpoints could not be assessed. Here, efficacy and safety data collected on primary and secondary efficacy variables and comparing umibecestat and placebo are reported descriptively and pooled across GS1 Cohort II and GS2 populations and doses, and the analytic effort was revised to focus on assessing worsening in cognition based on the APCC, RBANS, and CDR‐SOB and its reversibility after treatment discontinuation.

2.6. Statistical analysis

GS1 Cohort II was planned to include 390 participants receiving umibecestat and 260 participants receiving matching placebo. GS2 was planned to include 800 participants in the umibecestat 50‐mg group, 400 in the umibecestat 15‐mg group, and 800 in the placebo group. The sample size calculations in both studies were driven by a target power of 80% for the first dual primary time‐to‐event endpoint, ²⁰ with an estimated event risk in 5 years of 30% to 40% based on longitudinal cohort studies, a dropout rate of 30% over 5 years, and a type 1 error rate of 4%. ¹⁶

Data were pooled across GS1 Cohort II and GS2 populations comparing umibecestat and placebo, where the umibecestat group comprised participants randomized to any active treatment (50 mg for GS1 Cohort II; 15 and 50 mg for GS2). This pooling was justified by the similarity in the participant population, study design, and operational implementation, largely at the same investigational sites.

Statistical analysis was performed using SAS Version 9.4 (SAS Institute Inc., Cary, North Carolina, USA).

All participants who received at least one dose of study medication formed the safety analysis set (SAF). For assessing worsening in cognition and its reversibility, participants with a minimum treatment exposure of 3 months (ie, 2 months of treatment and 31 days for washout to occur) formed the modified safety analysis set (mSAF). The data presented are pooled for the mSAF across the GS1 Cohort II and GS2 populations, except for participant demographics, duration of exposure, and the safety analyses that used the SAF.

Change from baseline to the last visit on treatment was used to evaluate the magnitude of the worsening during treatment; change from baseline to the last assessment after washout was used to assess reversal to baseline scores. Reversibility was further assessed by comparing the assessment at last visit on treatment with the assessment at last visit after washout (in participants who completed both). For the APCC and the RBANS total score and index scores, the treatment effect sizes of change from baseline (based on the Cohen's d formula) and 80% confidence interval (CI) for the effect size were reported as forest plots.

Exploratory, model‐based analyses including analyses of covariance (ANCOVA) were performed to investigate factors that may have affected a worsening in cognition in addition to treatment. Participants were retrospectively divided into categories based on their level of worsening/improvement in RBANS. See Supplementary File (“Worsening in cognition and reversibility” section) for further analysis details including effect size and CI calculation.

3. RESULTS

3.1. Participant disposition and baseline characteristics

Overall, 1560 participants were randomized to umibecestat or placebo across GS1 Cohort II (n = 415) and GS2 (n = 1145) (Figure 1). A total of 1556 participants received study medication (SAF) for a mean (SD) duration of 7.0 (4.4) months (exposure with umibecestat of 6.9 [4.4] months and with placebo of 7.1 [4.3] months; Table S2) and were included in this pooled analysis (umibecestat pooled, n = 937; umibecestat 50 mg, n = 704; umibecestat 15 mg, n = 233; placebo, n = 619; Figure 1).

FIGURE 1.

Participant flow charts. (A) CONSORT Generation Study 1 (GS1) Cohort II. (B) CONSORT Generation Study 2 (GS2). (C) Participant distribution across GS1 and GS2. (A) GS1: 482 participants completed screening, of whom 65 participants were randomized to Cohort I and two participants could not be randomized due to study termination, resulting in 415 participants randomized to Cohort II. *A total of 412 participants were treated, with 249 receiving umibecestat 50 mg and 163 receiving placebo (three not treated). (B) GS2: 1172 participants completed screening, of whom 1145 were randomized and 27 could not be randomized due to study termination. *A total of 1144 participants were treated, including 455 in the umibecestat 50 mg group (one not treated). (C) GS1 and GS2: percentage of mSAF uses SAF as denominator and percentage of "last visit on treatment and after washout" uses mSAF as denominator. Last visit on treatment and after washout: participants who had both a last visit on treatment and a last visit after washout; last visit on treatment is the last assessment (RBANS total score) before or at last day on study drug + 31 days; last visit after washout is the last assessment after last day on study drug + 31 days. mSAF, modified safety analysis set; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status; SAF, safety analysis set, which consists of all participants who received study medication; mSAF: consists of all participants of SAF with at least 3 months’ exposure duration.

The mSAF was composed of 1090 participants having at least 3 months of exposure including 647 (69% of SAF) on pooled umibecestat (umibecestat 50 mg, n = 493; umibecestat 15 mg, n = 154) and 443 participants (72% of SAF) on placebo. Among those, more than half (365 [56%] on umibecestat and 241 [54%] on placebo) had a cognitive assessment after washout in addition to their assessment at the last visit on treatment (Figure 1).

Although both GS1 and GS2 were terminated prior to completion of full recruitment, the demographic and baseline characteristics of participants from the SAF were well balanced across treatment groups. Participants had a mean (SD) age of 68.1 (4.1) years, with 39.8% of the participants aged ≥70 years. Overall, 61.1% were women and the majority (93.4%) were White (Table 1). Over 80% of participants had more than 12 years of education. Baseline cognitive/clinical measures were similar across genotype and amyloid status (Figure S2).

TABLE 1.

Participant demographics and baseline characteristics: pooled GS1 and GS2 studies (safety analysis set).

Characteristic	Umibecestat 15 mg N = 233	Umibecestat 50 mg N = 704	Umibecestat total N = 937	Placebo N = 619	Total N = 1556
Demographics
Age (years), mean (SD)	68.6 (4.1)	67.8 (4.1)	68.0 (4.1)	68.3 (4.2)	68.1 (4.1)
Age group (years), n (%)
<65	45 (19.3)	153 (21.7)	198 (21.1)	132 (21.3)	330 (21.2)
65−69	82 (35.2)	297 (42.2)	379 (40.4)	227 (36.7)	606 (38.9)
≥70	106 (45.5)	254 (36.1)	360 (38.4)	260 (42.0)	620 (39.8)
Female, n (%)	148 (63.5)	413 (58.7)	561 (59.9)	390 (63.0)	951 (61.1)
Race, n (%)
White	213 (91.4)	657 (93.3)	870 (92.8)	584 (94.3)	1454 (93.4)
Asian	12 (5.2)	22 (3.1)	34 (3.6)	22 (3.6)	56 (3.6)
Other	6 (2.6)	19 (2.7)	25 (2.7)	6 (1.0)	31 (2.0)
Unknown	2 (0.9)	6 (0.9)	8 (0.9)	7 (1.1)	15 (1.0)
Ethnicity, a n (%)
Hispanic/Latino	8 (3.4)	15 (2.1)	23 (2.5)	16 (2.6)	39 (2.5)
Japanese	6 (2.6)	13 (1.8)	19 (2.0)	11 (1.8)	30 (1.9)
Mixed ethnicity	17 (7.3)	57 (8.1)	74 (7.9)	51 (8.2)	125 (8.0)
Other	146 (62.7)	448 (63.6)	594 (63.4)	372 (60.1)	966 (62.1)
Unknown	18 (7.7)	59 (8.4)	77 (8.2)	54 (8.7)	131 (8.4)
Not reported	38 (16.3)	112 (15.9)	150 (16.0)	115 (18.6)	265 (17.0)
Years of education, n
≤12 years, n (%)	37 (18.4)	105 (16.2)	142 (16.7)	107 (19.2)	249 (17.7)
13 to 16 years, n (%)	89 (44.3)	269 (41.4)	358 (42.1)	227 (40.8)	585 (41.6)
≥17 years, n (%)	75 (37.3)	276 (42.5)	351 (41.2)	222 (39.9)	573 (40.7)
Family history of AD, n (%)
Mother/Father	136 (58.4)	436 (61.9)	572 (61.0)	373 (60.3)	945 (60.7)
Siblings	5 (2.1)	35 (5.0)	40 (4.3)	18 (2.9)	58 (3.7)
Grandparents	10 (4.3)	40 (5.7)	50 (5.3)	28 (4.5)	78 (5.0)
Other	10 (4.3)	30 (4.3)	40 (4.3)	35 (5.7)	75 (4.8)
None	72 (30.9)	163 (23.2)	235 (25.1)	165 (26.7)	400 (25.7)
APOE 𝛆4 genotype, n (%)
Homozygote	46 (19.7)	338 (48.0)	384 (41.0)	254 (41.0)	638 (41.0)
Heterozygote	187 (80.3)	366 (52.0)	553 (59.0)	365 (59.0)	918 (59.0)
Brain characteristics
Whole brain volume (cm3)	1062.0 (109.0)	1076.2 (108.4)	1072.6 (108.6)	1067.5 (110.3)	1070.6 (109.3)
Hippocampus volume (cm3)	8.0 (1.0)	8.1 (1.0)	8.1 (1.0)	8.0 (1.0)	8.0 (1.0)
Amyloid level—n (%)
Elevated (A+)	215 (94.3)	567 (83.0)	782 (85.8)	510 (84.4)	1292 (85.3)
Not elevated (A−)	13 (5.7)	116 (17.0)	129 (14.2)	94 (15.6)	223 (14.7)
Centiloid (no unit)	57.7 (30.5)	51.9 (33.1)	53.3 (32.6)	54.9 (34.8)	53.9 (33.5)
Centiloid level, n (%)
Centiloid < 10	5 (2.7)	67 (11.3)	72 (9.2)	51 (9.8)	123 (9.5)
10 ≤ Centiloid < 24.33	3 (1.6)	41 (6.9)	44 (5.6)	28 (5.4)	72 (5.5)
24.33 ≤ Centiloid ≤ 50	86 (46.0)	182 (30.6)	268 (34.3)	169 (32.5)	437 (33.6)
Centiloid > 50	93 (49.7)	304 (51.2)	397 (50.8)	272 (52.3)	669 (51.4)
Cognitive measures
APCC	75.8 (6.8)	75.3 (6.8)	75.4 (6.8)	75.2 (7.0)	75.3 (6.9)
RBANS total b	102.0 (12.1)	101.2 (12.1)	101.4 (12.1)	101.3 (12.4)	101.4 (12.2)
RBANS immediate memory index score	103.8 (13.8)	103.2 (13.9)	103.3 (13.8)	103.1 (14.2)	103.2 (14.0)
RBANS visuospatial/constructional index score	98.5 (16.2)	98.2 (16.4)	98.3 (16.4)	97.9 (16.2)	98.1 (16.3)
RBANS language index score	98.9 (10.9)	98.9 (11.4)	98.9 (11.2)	99.0 (11.7)	98.9 (11.4)
RBANS attention index score	104.0 (15.1)	103.4 (16.1)	103.6 (15.9)	104.0 (16.5)	103.8 (16.1)
RBANS delayed memory index score	103.3 (11.5)	102.0 (11.1)	102.3 (11.2)	101.9 (11.9)	102.1 (11.5)
MMSE total	28.9 (1.3)	29.0 (1.2)	29.0 (1.2)	29.0 (1.3)	29.0 (1.3)
CDR‐SOB	0.16 (0.4)	0.18 (0.4)	0.17 (0.4)	0.15 (0.4)	0.16 (0.4)
CDR‐global b
Score = 0, n (%)	202 (86.7)	612 (87.1)	814 (87.0)	548 (88.5)	1362 (87.6)
Score = 0.5, n (%)	31 (13.3)	91 (12.9)	122 (13.0)	71 (11.5)	193 (12.4)
Score > 0.5, n (%)	0	0	0	0	0
ECog‐participant	49.3 (8.8)	48.7 (9.2)	48.8 (9.1)	49.1 (8.6)	48.9 (8.9)
ECog‐informant	46.3 (9.1)	46.2(8.6)	46.2 (8.7)	46.4 (8.5)	46.3 (8.6)

Note: Values are means (SD) unless otherwise stated.

Abbreviations: APCC, Alzheimer's Prevention Initiative Preclinical Composite Cognitive test; APOE ε4, apolipoprotein E ε4; CDR, Clinical Dementia Rating; CDR‐SOB, Clinical Dementia Rating‐Sum of Boxes; ECog, Everyday Cognition Scale; MMSE, Mini‐Mental State Examination; n, number of participants in related category; N, total number of participants in each treatment group; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status; SD, standard deviation.

^a Categories listed are among those the sites had to choose from.

^b Cognitively unimpaired at screening as defined by a score of 85 or greater on the RBANS delayed memory index score and CDR global score of 0 (with two exceptions; see inclusion criteria).

3.2. Cognitive outcomes

The RBANS total score, RBANS immediate memory, and RBANS delayed memory indicated a worsening in cognitive performance with umibecestat compared with placebo as early as Week 13, at Week 26, and up to the last visit on treatment (Table 2, Figure 2, Figure 3, Table S3). Cohen's d effect size for the RBANS total score in the pooled umibecestat group compared to placebo was −0.31; 80% CI: −0.40, −0.22 at Week 13 (Figure 2) and remained of similar magnitude at the last visit on treatment (Figures 2 and 3B). The worsening did not increase as a function of drug exposure (up to 500 days [1.4 years]; Figure 3A) and was only slightly different for the two umibecestat doses (Figures 2 and 3C), and baseline brain amyloid levels did not seem to impact the change from baseline in RBANS (Figure S3). Similarly, mild negative effects on cognition were observed for APCC at the last visit on treatment compared with placebo (Table 2, Figure 2).

TABLE 2.

APCC, RBANS, CDR‐SOB, and ECog (pooled GS1 and GS2, modified safety analysis set).

	Umibecestat 15 mg N = 154		Umibecestat 50 mg N = 493		Umibecestat total N = 647		Placebo N = 443
Time point statistics	Baseline	Change	Baseline	Change	Baseline	Change	Baseline	Change
APCC
Week 26
Mean (SD) n	75.9 (7.0) 74	−2.9 (5.1) 74	75.6 (7.1) 306	−3.2 (4.9) 306	75.6 (7.1) 380	−3.2 (4.9) 380	75.9 (6.8) 264	−1.4 (4.5) 264
Last visit on treatment
Mean (SD) n	76.4 (6.6) 115	−2.0 (4.8) 115	75.5 (7.1) 371	−1.8 (4.8) 371	75.7 (7.0) 486	−1.9 (4.8) 486	75.9 (6.8) 343	−0.7 (5.0) 343
Last visit after washout
Mean (SD) n	76.5 (7.0) 77	−0.5 (4.5) 77	75.8 (6.7) 299	−0.4 (4.8) 299	76.0 (6.7) 376	−0.4 (4.7) 376	74.5 (6.9) 243	−0.4 (4.5) 243
Change from last visit on treatment to last visit after washout
Mean (SD) n		2.2 (5.0) 42		1.9 (4.8) 195		2.0 (4.8) 237		0.2 (4.5) 158
RBANS total
Week 13
Mean (SD) n	103.1(11.7) 129	−0.3 (8.1) 129	100.6 (12.3) 342	0.6 (8.0) 342	101.2 (12.1) 471	0.4 (8.1) 471	101.4 (12.6) 321	2.9 (8.1) 321
Week 26
Mean (SD) n	103.9 (11) 76	−6.4 (8.8) 76	101.2 (12.5) 308	−5.0 (8.6) 308	101.7 (12.3) 384	−5.2 (8.6) 384	102.8 (12.8) 264	−3.1 (8.2) 264
Last visit on treatment
Mean (SD) n	103.3 (11.6) 152	−3.8 (9) 152	101.0 (12.5) 473	−2.8 (8.7) 473	101.5 (12.3) 625	−3.0 (8.7) 625	101.9 (12.8) 432	−0.2 (8.9) 432
Last visit after washout
Mean (SD) n	103.9 (10.1) 78	−2.8 (9.1) 78	101.8 (12.3) 304	−1.2 (8.9) 304	102.2 (11.9) 382	−1.5 (8.9) 382	100.7 (12.1) 246	−1.0 (8.7) 246
Change from last visit on treatment to last visit after washout
Mean (SD) n		0.7 (7.9) 77		1.0 (8.7) 288		0.9 (8.6) 364		−1.4 (8.2) 241
CDR‐SOB
Week 26
Mean (SD) n	0.14 (0.33) 73	0.21 (0.46) 73	0.17 (0.42) 305	0.10 (0.51) 305	0.16 (0.40) 378	0.12 (0.50) 378	0.13 (0.37) 262	0.06 (0.37) 262
Last visit on treatment
Mean (SD) n	0.15 (0.34) 116	0.08 (0.40) 116	0.18 (0.42) 370	0.11 (0.56) 370	0.17 (0.41) 486	0.10 (0.53) 486	0.14 (0.38) 335	0.06 (0.45) 335
Last visit after washout
Mean (SD) n	0.12 (0.30) 75	0.15 (0.38) 75	0.14 (0.32) 292	0.07 (0.53) 292	0.14 (0.32) 367	0.09 (0.51) 367	0.14 (0.39) 243	0.07 (0.50) 243
Change from last visit on treatment to last visit after washout
Mean (SD) n		0.01(0.34) 42		−0.01 (0.40) 191		−0.00 (0.40) 233		0.09 (0.42) 154
ECog participant total score
Week 26
Mean (SD) n	49.3 (8.1) 73	0.4 (5.5) 73	48.7 (9.6) 299	1.4 (7.1) 299	48.8 (9.2) 372	1.2 (6.8) 372	48.7 (8.2) 257	0.6 (6.7) 257
Last visit on treatment
Mean (SD) n	49.2 (8.6) 115	0.3 (5.7) 115	48.7 (9.7) 367	2.06 (8.2) 367	48.8 (9.5) 482	1.7 (7.7) 482	48.5 (7.9) 331	1.0 (7.8) 331
Last visit after washout
Mean (SD) n	49.6 (7.7) 75	−0.2 (5.4) 75	48.6 (8.5) 290	0.54 (6.4) 290	48.8 (8.3) 365	0.4 (6.3) 365	49.0 (8.5) 237	0.1 (7.3) 237
Change from last visit on treatment to last visit after washout
Mean (SD) n		0.7 (5.1) 44		−0.7 (5.7) 197		−0.5 (5.6) 241		−0.3 (7.0) 156
ECog informant total score
Week 26
Mean (SD) n	45.3 (7.4) 71	1.8 (8.0) 71	46.3 (9.0) 288	−0.2 (7.2) 288	46.1 (8.7) 359	0.2 (7.4) 359	46.6 (9.0) 245	−1.0 (7.2) 245
Last visit on treatment
Mean (SD) n	45.6 (8.1) 111	1.1 (6.8) 111	46.2 (9.0) 346	0.9 (8.6) 346	46.1 (8.8) 457	0.9 (8.2) 457	46.4 (8.7) 316	−0.1 (7.8) 316
Last visit after washout
Mean (SD) n	46.2 (7.2) 73	0.6 (6.8) 73	45.5 (7.6) 263	0.4 (7.4) 263	45.7 (7.5) 336	0.5 (7.3) 336	46.8 (8.8) 222	0.6 (8.3) 222
Change from last visit on treatment to last visit after washout
Mean (SD) n		−0.6 (8.7) 42		0.2 (7.9) 179		0.1 (8.0) 221		1.5 (7.0) 148

Note: A decrease in APCC indicates a worsening in cognition. A decrease in RBANS indicates a worsening in cognition. An increase in CDR‐sum of box score indicates a higher probability of being diagnosed with dementia. An increase in Ecog indicates a worsening in cognition. Last visit on treatment is the last assessment before or on last day on study drug + 31 days. Last visit after washout is the last assessment after last day on study drug + 31 days. Scheduled visits at Weeks 13 and 26 (as well as Week 52 as shown in Table S2) occurred before study termination, so most participants at these visits were on treatment as per protocol (unless the drug was discontinued for other reasons). Participants who had their last visit on treatment at Week 26 were counted in both “Week 26” and “last visit on treatment” statistics. At each visit, only participants with a non‐missing value at both baseline and that visit are included.

Abbreviations: APCC, Alzheimer's Prevention Initiative Preclinical Composite Cognitive; CDR‐SOB, clinical dementia rating‐sum of boxes; ECog, Everyday Cognition scale; n, number of participants with non‐missing value; N, total number of participants in each treatment group; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status; SD, standard deviation.

FIGURE 2.

Treatment effect size of change from baseline in RBANS and APCC (modified safety analysis set). A decrease in RBANS and APCC indicates worsening condition. Effect size is calculated using Cohen's d formula (raw, not model‐based, mean to standard deviation [SD] ratio) as the difference between active treatment (umibecestat) and placebo in the mean change from baseline divided by the pooled SD of the change (negative effect size indicates worsening in cognitive score with umibecestat vs placebo). For further details on effect size, CI calculation, and methodology, see Supplementary File. Cohen's d = 0.2 is considered a small, 0.5 a medium, and 0.8 a large effect size. Last visit on treatment is the last assessment before washout (ie, last day on study drug + 31 days). Last visit after washout is the last assessment off treatment after washout. Scheduled visits at Weeks 13 and 26 (as well as Week 52, as shown in Table S3) occurred before study termination; therefore, most participants at these visits were on treatment as per protocol (unless the drug was discontinued for other reasons). Participants who had their last visit on treatment at Week 26 were counted in both Week 26 and last week on treatment analyses. APCC, Alzheimer's Prevention Initiative Preclinical Composite Cognitive test; n, number of participants with non‐missing value; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status.

FIGURE 3.

Change in RBANS total from baseline to last visit on treatment (A) as a function of umibecestat exposure (decline in RBANS score detectable but non‐progressive throughout the treatment period); (B) by neurocognitive domain; (C) proportion of participants in each category of change (modified safety analysis set). Locally estimated scatterplot smoothing (LOESS) estimates were derived using LOESS regression that uses local weighted regression to fit a smooth curve through points in a scatterplot of RBANS total change at last visit on treatment versus exposure. LOESS smooth curve is presented with degree = 1 and smoothing parameter = 0.65. For details on effect size and CI calculation, see figures 2 and 5 legends and Supplementary File. Last visit on treatment is the last assessment before washout (ie, last day on study drug + 31 days). n, number of participants with non‐missing value; N, total number of participants in each treatment group; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status.

We did not observe any worsening at any time point for CDR‐SOB (mean [SD] change at last visit on treatment 0.10 [0.53] for umibecestat vs 0.06 [0.45] for placebo) (Table 2). The mean (SD) changes from baseline in ECog participant scores for umibecestat and placebo were 1.7 (7.7) and 1.0 (7.8), respectively, at the last visit on treatment and 0.4 (6.3) and 0.1 (7.3), respectively, after washout (last visit). These scores were in conformation with those observed by the study partners at any time point. Note that, contrary to RBANS and APCC, an increase in CDR‐SOB or ECog score indicates a worsening in cognition.

Following treatment termination, a median (interquartile range) duration of 3.5 months (2.8 to 5.6 months) from the last visit on treatment to the last visit after washout was observed for both the umibecestat and the placebo groups (Table S2). The mean (SD) change in RBANS total score from baseline to the last visit after washout was −1.5 (8.9) with umibecestat versus −1.0 (8.7) with placebo. These changes were observed after a worsening of −3.0 (8.7) versus −0.2 (8.9), respectively, at the last visit on treatment (Table 2), reflecting the reversibility of the worsening in cognition after washout in the umibecestat group. Cognition as measured by the RBANS total score indeed increased from the last visit on treatment to the last visit after washout by +0.9 (8.6) in the umibecestat group versus a continuing decrease of −1.4 (8.2) for the placebo group. Cohen's d effect size for the RBANS total score with umibecestat compared to placebo indicated the absence of difference between the two groups in the change from baseline to the last visit after washout: −0.06 (80% CI: −0.16, +0.04) (Figure 2). Baseline RBANS and the interaction of baseline RBANS with treatment impacted the change from baseline to Week 26 in RBANS across all the ANCOVA models investigated. No other factors or interactions were identified that impacted this change.

3.3. Change in body weight

The mean (SD) weight change from baseline to the last visit on treatment was −2.2 (3.3) kg in the pooled umibecestat group compared with −0.7 (2.9) kg in the placebo group (Figure 4) with greater weight loss with umibecestat 50 mg than with the 15‐mg dose. At the last visit after washout, weight was regained in the umibecestat group, with a mean (SD) weight change from baseline of −1.1 (3.7) kg with umibecestat compared with −0.7 (3.4) kg for placebo. The change in weight from the last visit on treatment to the last visit after washout was +0.9 (2.8) kg in the umibecestat group compared with −0.2 (2.6) kg in the placebo group.

FIGURE 4.

Change in body weight (kilograms) by visit (safety analysis set). Last visit on treatment is the last assessment before washout (ie, last day on study drug + 31 days). Last visit after washout is the last assessment off treatment after washout. Scheduled visits at Weeks 13 and 26 occurred before study termination; therefore, most participants at these visits were on treatment as per protocol (unless the drug was discontinued for other reasons). Participants who had their last visit on treatment at Week 26 were counted in both Week 26 and last week on treatment analyses. n, number of participants with a non‐missing value; N, total number of participants in each treatment group; at each visit, only participants with a non‐missing value at both baseline and that visit are included; SE, standard error.

3.4. Fluid biomarkers

Post‐baseline blood samples were analyzed for plasma Aβ₄₀ and NfL in a subset of 66 and 209 participants, respectively. Plasma Aβ₄₀ levels were lower for umibecestat 15 and 50 mg versus placebo at Weeks 26 and 52 and up to the last visit on treatment (mean [SD] change from baseline to last visit on treatment: −44.4 [11.3] pmol/L for the pooled umibecestat doses versus +2.5 [18.6] pmol/L for placebo; p < .001). Plasma Aβ₄₂ was not assessed due to the lack of adequately sensitive assays at the time of the study. Serum NfL levels remained stable in both groups (Table S4).

CSF samples were collected at baseline and Year 2 in the 27 participants (including 18 in the umibecestat arm) consenting to the two optional lumbar punctures. Small numerical treatment differences in CSF Aβ₄₂ and tau were seen in this small sample (Table S4). CSF p‐tau in particular remained stable on placebo versus a minor decrease on umibecestat (p = .06 at completion of treatment epoch, not adjusted for multiplicity). It should be noted that unlike blood measurements, most CSF measurements were performed after washout (the median [min, max] number of days from last dose to lumbar puncture was 72 [22, 105]).

3.5. Volumetric MRI

Decreased whole brain and hippocampal volumes were observed at Week 26 with umibecestat (Figure 5); this reduction was dose independent and non‐progressive up to the last visit on treatment. The mean (SD) annualized percentage volume change from baseline with umibecestat versus placebo at Week 26 was −1.00 (1.01) versus −0.52 (1.14) for the whole brain and 1.90 (2.64) versus −1.09 (2.49) for the hippocampus and remained of the same magnitude at the last visit on treatment (Table S5). Group differences were reduced at the last assessment after washout, with respective volume loss from baseline for umibecestat and placebo, respectively, of −0.76 (0.78) versus −0.53 (0.73) for whole brain (p = .017) and −1.47 (1.97) versus −1.16 (1.77) (p = .196) for the hippocampus.

FIGURE 5.

Treatment effect size of percentage volume loss from baseline (Week 26) (modified safety analysis set). Effect size is calculated using Cohen's d formula (raw, not model‐based mean to standard deviation [SD] ratio) as difference between active and placebo in mean change from baseline divided by pooled SD of change (negative effect size indicates worsening in cognitive score with umibecestat vs placebo). For further details on effect size, CI calculation, and methodology, see Supplementary File. Cohen's d = 0.2 is considered a small, 0.5 a medium, and 0.8 a large effect size. For all participants, the scheduled visit at Week 26 occurred before study termination; therefore, most participants at this visit were on treatment as per protocol (unless the drug was discontinued for other reasons).

3.6. Safety

The proportion of participants with at least one treatment‐emergent AE was 58.4% in the umibecestat total group and 53.3% in the placebo group (Table 3). This proportion was slightly greater with umibecestat 50 mg versus umibecestat 15 mg (60.4% vs 52.4%, respectively). The incidence of AEs leading to discontinuation (4.2% with umibecestat vs 1.6% with placebo) and SAEs (3.4% with umibecestat vs 3.6% with placebo) was low across both treatment groups.

TABLE 3.

Treatment emergent adverse events by primary system organ class (regardless of the study treatment relationship, safety analysis set).

Primary system organ class	Umibecestat 15 mg N = 233 n (%)	Umibecestat 50 mg N = 704 n (%)	Umibecestat total N = 937 n (%)	Placebo N = 619 n (%)
AE leading to discontinuation	10 (4.3)	29 (4.1)	39 (4.2)	10 (1.6)
Participants with at least one serious AE	9 (3.9)	23 (3.3)	32 (3.4)	22 (3.6)
Cardiac disorders	2 (0.9)	3 (0.4)	5 (0.5)	2 (0.3)
Gastrointestinal disorders	2 (0.9)	1 (0.1)	3 (0.3)	1 (0.2)
General disorders and administration site conditions	0	0	0	1 (0.2)
Hepatobiliary disorders	1 (0.4)	1 (0.1)	2 (0.2)	0
Infections and infestations	0	3 (0.4)	3 (0.3)	1 (0.2)
Injury, poisoning, and procedural complications	0	5 (0.7)	5 (0.5)	3 (0.5)
Investigations	0	1 (0.1)	1 (0.1)	1 (0.2)
Metabolism and nutrition disorders	0	1 (0.1)	1 (0.1)	0
Musculoskeletal and connective tissue disorders	1 (0.4)	2 (0.3)	3 (0.3)	1 (0.2)
Neoplasms benign, malignant, and unspecified	1 (0.4)	3 (0.4)	4 (0.4)	5 (0.8)
Nervous system disorders	3 (1.3)	3 (0.4)	6 (0.6)	7 (1.1)
Psychiatric disorders	1 (0.4)	1 (0.1)	2 (0.2)	0
Renal and urinary disorders	0	0	0	1 (0.2)
Respiratory, thoracic and mediastinal disorders	0	0	0	2 (0.3)
Vascular disorders	2 (0.9)	0	2 (0.2)	0
Participants with at least one AE a	122 (52.4)	425 (60.4)	547 (58.4)	330 (53.3)
Gastrointestinal disorders	25 (10.7)	79 (11.2)	104 (11.1)	46 (7.4)
General disorders and administration site conditions	16 (6.9)	38 (5.4)	54 (5.8)	30 (4.8)
Investigations	11 (4.7)	72 (10.2)	83 (8.9)	46 (7.4)
Metabolism and nutrition disorders	8 (3.4)	37 (5.3)	45 (4.8)	20 (3.2)
Musculoskeletal and connective tissue disorders	21 (9.0)	93 (13.2)	114 (12.2)	67 (10.8)
Nervous system disorders	36 (15.5)	114 (16.2)	150 (16.0)	80 (12.9)
Psychiatric disorders b	25 (10.7)	123 (17.5)	148 (15.8)	41 (6.6)
Respiratory, thoracic and mediastinal disorders	14 (6.0)	40 (5.7)	54 (5.8)	23 (3.7)
Skin and subcutaneous tissue disorders	31 (13.3)	99 (14.1)	130 (13.9)	73 (11.8)
Vascular disorders	11 (4.7)	25 (3.6)	36 (3.8)	14 (2.3)

Note: A participant with multiple AEs within a system organ class was counted only once. Percentages are calculated using N as denominator.

Abbreviations: AE, adverse event; n, number of participants with an event; N, total number of participants in each treatment group.

^a Only most frequent AEs are shown (defined as >2.5% in any treatment group for each system organ class).

^b Most frequent psychiatric disorders with umibecestat include abnormal dreams (4.7%) and anxiety (3.8%) versus 1.3% for each with placebo (also see Table S6 for most frequent AE details by preferred terms and severity).

The higher incidence of AEs observed with umibecestat was mainly driven by disorders of the nervous system, psychiatric disorders, disorders of the skin and subcutaneous tissue, musculoskeletal and connective tissue, and gastrointestinal disorders (Table 3). Most treatment‐emergent AEs (listed by preferred term in Table S6) were mild or moderate in severity, including skin reactions (most commonly pruritus: 5.8% vs 3.4% for umibecestat vs placebo, respectively), nasopharyngitis (4.4% vs 2.1%), and abnormal dreams (4.7% vs 1.3%). Details on skin and subcutaneous tissue disorders are shown in Table S7. There were no clinically meaningful abnormalities in hematology, biochemistry, or vital signs during the study period. The frequency of increases in liver enzymes that fulfilled the criteria for either liver laboratory trigger or liver event was similar across the pooled umibecestat group and the placebo group (Table S8). There was one unrelated death in the placebo group due to metastatic neoplasm.

No confirmed case of amyloid‐related imaging abnormality‐edema (ARIA‐E) was reported in the umibecestat group compared with two (0.3%) of moderate severity reported in the placebo group. Six cases (0.6%) of amyloid‐related imaging abnormality‐hemorrhages (ARIA‐H) were reported in the umibecestat group and two cases (0.3%) in the placebo group (Table S9). All reported ARIA cases were asymptomatic.

4. DISCUSSION

The API Generation Program included the first National Institutes of Health (NIH)‐ and industry‐supported AD prevention trials of potential Aβ‐modifying treatments in cognitively unimpaired persons at genetic risk for late‐onset AD. The study population included APOE ε4 heterozygotes with elevated brain amyloid levels and APOE ε4 homozygotes with or without elevated brain amyloid levels, thereby exploring prevention across all stages of preclinical AD. Two different anti‐amyloid drugs were tested ¹⁶ , ¹⁷ ; here, we report the results obtained with umibecestat, an oral BACE1 inhibitor with a higher BACE1/BACE2 selectivity than previous candidates, based on pooled data of over 1500 at‐risk participants. As predicted by results from the previous 3‐month study in healthy elderly, ¹¹ umibecestat significantly lowered plasma Aβ₄₀ levels compared with placebo at both 6 and 12 months. However, as observed with other BACE inhibitors, ⁷ , ⁸ , ⁹ umibecestat was also associated with a worsening in cognition that led to the premature termination of the trials. The API Generation Program was the first to evaluate the reversibility of cognitive worsening and brain shrinkage following discontinuation of a BACE inhibitor treatment with preplanned follow‐up assessments blinded to original randomized treatment assignment.

The observed cognitive decline as measured by RBANS and APCC was mild and occurred as early as Week 13 at a group level. However, no individual report of AEs related to worsening in cognition were reported, and CDR‐SOB – which encompasses both cognitive and functional status – was not affected, suggesting that the observed cognitive worsening may not have been clinically relevant. This cognitive decline remained stable until the last visit on treatment and was reversed (both RBANS and APCC returned to baseline levels) within 3 to 6 months after treatment washout.

The worsening in cognition previously reported with other BACE inhibitors was dose dependent and persisted with longer exposure. ⁸ , ²² , ²³ The early (and possibly manageable/not clinically significant) cognitive effects detected with umibecestat did not appear to be dose dependent, did not progress with longer treatment, and were no longer apparent after treatment washout. Our finding strengthens the results of the EARLY trial of BACE inhibitor atabecestat, which provided preliminary evidence for the reversibility of cognitive worsening after 6 months of a non‐blinded off‐treatment follow‐up of 175 at‐risk participants (vs 1057 in our studies). ⁹

As also reported with other BACE inhibitors, ²⁴ treatment with umibecestat was associated with volumetric MRI reductions, including early, dose‐independent, and non‐progressive reductions in the hippocampal and whole brain volume, which was partially reversed after washout. The fact that brain shrinkage did not progress and did not seem to be permanent suggests that accelerated neurodegeneration is unlikely to be the cause, which is also consistent with the stable NfL and total tau levels observed in our study. A slower rate of amyloid accumulation with BACE inhibition is probably also insufficient to account for the loss of brain volume. As an alternative explanation, Sur et al. proposed that the volumetric MRI changes observed with BACE inhibitors could be due to changes in the amyloid plaque microenvironment, possibly involving inflammatory processes. ²⁴

Finally, treatment with umibecestat was associated with weight loss, which again started to reverse following treatment discontinuation. This effect on weight could potentially be explained by the impact that lowering BACE1 activity has on leptin sensitivity. ²⁵ Interestingly, in line with Meakin et al.’s findings in male mice, the majority of weight loss events occurring in male participants were reported in the highest weight category (results not shown).

The various physiological roles of BACE1, APP, and Aβ in a normally functioning brain are not fully understood. While higher Aβ load is associated with the pathological pathways leading to the accumulation of amyloid plaques and the impairment of memory and cognitive function characteristic of AD, depletion of soluble Aβ below a certain level has been correlated with reduced short‐term and long‐term memory capability. ²⁶ , ²⁷ The protective or toxic impact of Aβ, depending both on concentration and isoform, might explain some of the side effects of anti‐Aβ treatments. ²⁷ , ²⁸

Another hypothesis for the observed cognitive worsening may be related to the lack of APP substrate specificity. ²⁶ , ²⁹ , ³⁰ BACE1 may inhibit other substrates involved in neuroplasticity and synaptic pathways, and complete suppression of BACE1 enzymatic activity could substantially impair hippocampal neurogenesis, which is essential for memory formation and learning. ¹⁰ , ²⁹ A 70% to 90% inhibition as expected with the studied doses may still have been too high. All of these hypotheses align with our observations, including hippocampal volume loss and decrease in immediate and delayed memory indices driving cognitive worsening as measured by RBANS.

In any case, the adverse effects from umibecestat treatment were reversed, at least partially, soon after washout, which suggests that umibecestat did not worsen neurodegeneration but involved processes different from AD‐related pathological mechanisms.

Based on these putative mechanisms, together with the evidence that umibecestat affects BACE1 substrates in a dose‐dependent manner, ³⁰ it may be possible to define a therapeutic window leading to incomplete suppression of BACE1 activity for safe and efficient chronic Aβ lowering. ²⁹ , ³⁰ Determining the optimal level of BACE1 inhibition would require a better understanding of the mechanisms and substrates involved in the observed side effects and the dose–response relationships between Aβ lowering and the processing of these substrates. McDade et al. expanded on this approach by describing how the results of trials such as ours may inform future clinical trials and pave the way to achieve this goal. Some additional insights may be yielded by studies on genetic variants and the associated levels of Aβ reduction that are shown to have a protective effect (e.g., 40% in Jonsson et al.). ³¹ Carefully adjusted BACE inhibition may have further benefits in individuals with a genetic predisposition to early‐onset brain amyloid deposition, such as those with Down syndrome. ³² , ³³ Thus, despite these negative findings, we believe that there are reasons to investigate further the modification of BACE activity for AD prevention.

4.1. Strengths and limitations

The recruitment of an unprecedented number of persons at genetic risk for late‐onset AD – including 638 cognitively unimpaired APOE ε4 homozygotes (who represent 1% to 2% of the population ¹⁴ ) and over 918 APOE ε4 heterozygotes with biomarker evidence of amyloid plaques – was achieved thanks to the Alzheimer's Prevention Registry, GeneMatch, and other research participant engagement and recruitment campaigns around the world. ¹⁸ The similarities of the trial designs allowed the pooling of two studies for DMC preplanned review, to promptly enable detection of a cognitive worsening. Furthermore, despite premature termination, a fully blinded follow‐up was achieved with a good retention rate in a large sample size (more than 50% of participants [n = 606] attending a visit after washout) and a sufficient duration to establish reversal at 3 to 6 months after treatment discontinuation. The fact remains that we are missing after‐washout outcomes for close to half of the participants (n = 484), which may have had an impact on the overall reversal results. For the other half who were followed up, being made aware of the termination of the study due to cognitive side effects may also have biased the results. However, since the participants were still blinded at follow‐up, this impact would be similar across treatment groups, rendering the comparison still valid. Most importantly, the early trial termination precluded the achievement of any of the preplanned objectives. Consequently, the long‐term effects of umibecestat remain unknown. Consistent with results from the EARLY trial, ⁹ no further cognitive decline in this preclinical population was noted after the observed onset at 3 months and up to 500 days (1.4 years; Figure 3A), but we do not know the effect of longer drug exposure. Further, despite data pooling, the impact of dose and amyloid status on drug‐induced cognitive impairment could not be assessed considering the specific designs of the two studies. Finally, although we partnered with an advocacy group in the United States in an effort to engage and recruit traditionally underrepresented communities, ¹⁸ this partnership was not in place from the start of the trial and study termination hindered our effort. In addition, our participant recruitment was supported by registries that lacked racial and ethnicity diversity at the time of recruitment. ³⁴ , ³⁵ As a result, the majority who engaged in genotyping for the prescreening to enter in the two Generation studies were non‐Hispanic White, preventing us from generalizing these findings to other races and ethnicities. We are currently investigating facilitators and barriers for understudied populations to join participant recruitment registries.

5. CONCLUSION

In cognitively unimpaired participants at risk of progression to MCI or dementia due to AD, treatment with umibecestat impacted performance in several measures of cognition compared with placebo. This effect was detected early at a group level but not reported individually, was not associated with change in a measure of global cognitive/functional status, did not increase with treatment duration, and reversed after washout. The findings from the Generation studies may provide useful information about the potential benefits (in terms of safety and tolerability), risks, and roles of BACE inhibitors in the development of AD prevention therapies, including those that could be used for the primary prevention of AD in cognitively unimpaired individuals at genetic risk.

The subtle, non‐progressive and reversible worsening in indicators of cognitive performance and brain shrinkage suggests that umibecestat's side effects are not related to neurodegeneration or other AD‐related processes. Further research informed by additional insights on BACE1 mode of action, substrates, and dose–response relationships is warranted to confirm this finding with other BACE inhibitors and clarify the underlying mechanism. There may be a safe approach, potentially considering lower doses, to modify BACE activity to protect against AD progression. If so, BACE1 inhibition may still find its place in a prevention treatment paradigm. ²⁹

AUTHOR CONTRIBUTIONS

Pierre N. Tariot, Marie‐Emmanuelle Riviere, Cristina Lopez Lopez, Vissia Viglietta, Angelika Caputo, Jessica B. Langbaum, Eric M. Reiman, Fonda Liu, Marie‐Laure Rouzade‐Dominguez, and Ana Graf contributed to the design of the studies. Angelika Caputo and Yihan Sui contributed to the data analysis, Marie‐Catherine Mousseau to the writing, and Beth Borowsky to study termination. All authors were involved in data interpretation and critically reviewed and provided their final approval to the manuscript. All named authors meet the ICMJE criteria for authorship, agreed to be accountable for the work as a whole, and are responsible for its intellectual content and data accuracy.

CONFLICT OF INTEREST STATEMENT

Stephen Salloway has received grants from Biogen, Eisai, Genentech, Roche, Novartis, and Eli Lilly; personal fees from Biogen, Eisai, Eli Lilly, Novartis, Genentech, NovoNordisk, Prothena, AbbVie, Kisbee, Acumen, CognitionRX, and Roche; non‐financial support from Biogen, Lilly, Acumen, Abbvie, and Roche; and was a site investigator for the Generation studies and co‐chair of the investigator steering committee for the ENGAGE study. Jeffrey M. Burns has received research support to conduct clinical trials (paid to his institution) from Eli Lilly, Amylyx Pharmaceuticals, Biogen, AbbVie, AstraZeneca, and Roche and has served as a consultant for Renew Research, Amylyx Pharmaceuticals, Eisai, Eli Lilly, Labcorp, New Amsterdam Pharma, and Renew Biotechnologies. Jón G. Snaeda has received consulting fees from Novartis. Vissia Viglietta was a full‐time employee of Amgen at the time of the study and is currently an employee of Nido Biosciences, which was not involved in this study. Pierre N. Tariot, Eric M. Reiman, and Jessica B. Langbaum are full‐time employees of Banner Health. Banner Health received financial support from Novartis Pharma AG and Amgen for the conduct of the API Generation Program, from Eli Lilly for the conduct of another Alzheimer's prevention trial, and from Genentech/Roche for another Alzheimer's prevention trial. Pierre N. Tariot reports receiving grants from the National Institute on Aging (NIA) (UF1AG046150, RF1 AG041705, R01AG055444, and 1R01AG058468) and received consulting fees from AbbVie, AC Immune, Acadia, Axsome, Biogen, BioXcel, Cortexyme, Eisai, Genentech, Lundbeck, Otsuka & Astex, Merck & Co., Novo Nordisk, Syneos, and T3D Therapeutics. Eric M. Reiman reports receiving grants from the NIA (1UF1AG046150, RF1AG041705, R01AG05544, 1R01AG058468, P30AG19610, and P30AG072980). He is a compensated scientific advisor for Alzheon, Aural Analytics, Denali, Retromer Therapeutics, and Vaxxinity and an uncompensated advisor to Biogen and Eli Lilly. He is a co‐founder, advisor, and shareholder in ALZPath. Jessica B. Langbaum reports receiving grants from NIA (1UF1AG046150, RF1AG041705, R01AG05544, 1R01AG058468, and P30AG072980) and received consulting fees from Alector, Biogen, Denovo Biopharma, and Provoc. Marie‐Emmanuelle Riviere, Beth Borowsky, Fonda Liu, Marie‐Laure Rouzade‐Dominguez, Pilar Cazorla, Marie‐Catherine Mousseau, Michal Arkuszewski, Javier Ricart, Yihan Sui, Angelika Caputo, and Ana Graf are employees of and shareholders in Novartis. This research was sponsored by Novartis Pharma AG, Basel, Switzerland. Cristina Lopez Lopez was a full‐time employee of Novartis Pharma AG at the time of the study and is currently an employee of Roche, which was not involved in this study. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provided written informed consent before screening. They also consented to receive counseling and disclosure of their APOE genotype and associated risk estimates to develop MCI or dementia due to AD (GS1 and GS2) and, if heterozygous, evidence of elevated brain amyloid (GS2). ²¹ The impact of APOE genotype and amyloid status disclosures was assessed and will be presented in a separate paper.

Supporting information

ICMJE Disclosure Form

Supplementary Materials

Tariot PN, Riviere M‐E, Salloway S, et al. Reversibility of cognitive worsening observed with BACE inhibitor umibecestat in the Alzheimer's Prevention Initiative (API) Generation Studies. Alzheimer's Dement. 2024;20:7745–7761. 10.1002/alz.14237

DATA AVAILABILITY STATEMENT

The fully anonymized data set, including MRI and PET images (amyloid and tau or FDG in a subset) and most fluid samples (blood and CSF) from participants who consented to additional research in both studies and all treatment arms (active and placebo), is available on demand for access by the external scientific community. Further independent analyses of this rich set of data may generate further hypotheses of use for designing future studies in AD prevention. For both studies and both active and placebo arms, the clinical data and images from participants who consented to additional research were anonymized for upload to the Laboratory of NeuroImaging (LONI) (https://www.loni.usc.edu) server, and most of the samples were provided to the National Centralized Repository for Alzheimer's Disease and Related Dementias (NCRD) (https://ncrad.iu.edu/). These data are now available with a signed data access agreement upon request to these centers for external researchers to access and use for additional analysis. Study protocols and statistical analysis plans are available from the clinicaltrials.gov website at GS1 and GS2 (see “Drug and device information, study documents, and helpful links”). Clinical study reports will be provided upon request from marie-emmanuelle.riviere@novartis.com.