Anti‐amyloid immunotherapy and brain volume trajectories in early‐onset versus late‐onset Alzheimer's disease

Abstract

INTRODUCTION

Accelerated brain volume loss has been observed following trials of anti‐amyloid beta immunotherapy (AAT); however, there is limited understanding of this paradoxical phenomenon in early‐onset Alzheimer's disease (EOAD) versus late‐onset Alzheimer's disease (LOAD).

METHODS

We retrospectively analyzed brain volume changes over a 1‐year period following initiation of AAT. Annualized rates of brain volume alteration were calculated, and linear mixed‐effects models were performed to assess brain volume trajectories in EOAD versus LOAD.

RESULTS

One hundred fourteen participants (35 with EOAD; 79 with LOAD; mean baseline MMSE: 27) were analyzed. Ventricular expansion, measuring ∼15%/year, and whole‐brain atrophy occurred in both groups. LOAD showed ∼1.7× greater hippocampal atrophy compared with EOAD, independent of baseline cerebral amyloid burden, though global cortical amyloid burden predicted whole‐brain and hippocampal atrophy (p < 0.001).

DISCUSSION

Brain atrophy with prominent ventricular expansion occurs in early‐stage EOAD and LOAD following AAT. Additional studies are needed to elucidate the mechanisms and implications of this post‐immunotherapy effect.

Highlights

• Prominent ventricular expansion occurs in EOAD and LOAD after anti‐Aβ therapy.

• LOAD shows greater hippocampal atrophy than EOAD, independent of baseline amyloid burden.

• Global cortical amyloid burden predicts brain volume loss after anti‐Aβ therapy.

1. BACKGROUND

With the US Food and Drug Administration (FDA) approval of amyloid beta (Aβ)‐targeting monoclonal antibodies, a new therapeutic era for Alzheimer's disease (AD) has begun. After decades of experimentation and thousands of failed clinical trials, phase 3 studies of lecanemab and donanemab have shown amyloid clearance and an approximate 30% slowing of cognitive decline in AD. ¹ , ² Two adverse effects of these anti‐Aβ treatments (AAT) are well established: infusion‐related reactions and the more potentially serious amyloid‐related imaging abnormalities (ARIA). ³ , ⁴ , ⁵ Additional reported drug effects have received much less attention.

Accelerated brain atrophy has been described following administration of numerous AAT. ⁶ , ⁷ While the mechanism of this effect is currently a point of debate, some researchers have hypothesized its relation to Aβ plaque removal and related immune and cerebral fluid alterations as opposed to a detrimental etiology, that is, toxicity‐induced neuronal damage. ⁷ Nevertheless, brain tissue loss is an established biomarker of disease, and rates of brain atrophy are even more strongly correlated with cognitive decline in untreated AD. ⁸ , ⁹ , ¹⁰ , ¹¹ Thus, interrogation of this anti‐Aβ effect, referred to by some as pseudoatrophy, is warranted. ⁷ , ¹² While the phenomenon has been observed with multiple Aβ‐targeting immunotherapeutic agents, it remains poorly characterized, and there has been limited reporting on this effect in randomized AD clinical trials. ¹ , ² , ¹³ , ¹⁴ , ¹⁵

A recent meta‐analysis suggests that the FDA‐approved agents lecanemab and donanemab are associated with excess ventricular enlargement, along with slightly greater whole‐brain atrophy, when compared to placebo. ⁶ Clinical trials of these agents have largely enrolled patients within the typical age range for late‐onset AD (LOAD), with mean participant ages between 70 and 74 years. ¹ , ² , ¹⁵ Although the eligibility criteria allowed inclusion of individuals ≥50 years of age for aducanumab and lecanemab and ≥60 years of age for donanemab, early‐onset AD (EOAD) individuals were substantially underrepresented in these trial studies. ¹ , ² , ¹⁵ To date, we are not aware of any published investigation that compares rates of brain atrophy between EOAD and LOAD following treatment with AAT. This is despite known differences in baseline atrophy, cognitive trajectories, and biomarker progression patterns between these groups. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵

While both EOAD and LOAD are characterized by the cerebral aggregation of Aβ and tau proteins, the neuropathological literature reveals greater pathologic burden, particularly involving neocortical sites, in those affected earlier in life. ¹⁸ Recent work demonstrates broader neocortical tau deposition in EOAD, with faster tauopathy progression, even at early disease stages, whereas in LOAD, tauopathy remains centered in temporoparietal sites even in late disease. ²⁵ This is thought to be the driver of more rapid neurodegeneration and clinical deterioration and of the higher prevalence of atypical, non‐amnestic presentations in EOAD. ²⁵ Additionally, this evidence suggests that optimal therapeutic approaches for AD may differ according to age of disease onset.

The purpose of this real‐world study was to examine the progression of brain tissue atrophy over a 1‐year course of treatment with AAT in EOAD versus LOAD. Recruiting individuals from our institutional memory clinic, we analyzed structural magnetic resonance imaging (MRI) data to assess differences in brain volume trajectories between the groups. We hypothesized that EOAD would demonstrate greater rates of atrophy than LOAD after therapy, consistent with EOAD characterization as a more aggressive form of the disease. ¹⁶ , ¹⁷ , ²³ Specifically, we hypothesized that EOAD would show greater lateral ventricular expansion and whole‐brain tissue volume reduction relative to LOAD following AAT.

2. METHODS

2.1. Participants

Study participants were recruited from our academic memory clinic following AD diagnosis according to established criteria by multidisciplinary consensus. ²⁶ , ²⁷ Participants with symptomatology prior to age 65 were classified as EOAD. Our clinic began offering AAT with aducanumab in August 2021 and lecanemab in April 2023. Aducanumab was not offered after November 2024 due to its discontinuation by the manufacturer. Donanemab was offered beginning November 2024; however, none of our patients completed a full year of treatment with this agent at the time of analysis; thus, the current sample was limited to individuals who completed 1 year of treatment with aducanumab or lecanemab. In total, 13 patients who completed aducanumab and 101 who completed lecanemab treatment through November 2025 were included. Participants who transitioned to lecanemab from aducanumab were excluded unless they had completed 1 year of aducanumab treatment, and then only the year on aducanumab was examined. Participants provided written informed consent, and the study was performed in accordance with the principles of the Declaration of Helsinki.

2.2. Clinical measures

The following measures were used to characterize the sample at baseline: Mini‐Mental State Examination (MMSE) ²⁸ as an indication of general mental status, Functional Assessment Questionnaire (FAQ) ²⁹ as a measure of the degree of functional impairment, and the Neuropsychiatric Inventory‐Questionnaire (NPI‐Q) ³⁰ to measure degree of neuropsychiatric symptoms.

2.3. MRI acquisition

To measure brain volume, comparable high‐resolution T1‐weighted MRI scans were obtained on either a Siemens Magnetom Prisma (MPRAGE) or GE Architect (SPGR/BRAVO) 3T magnet with 0.5×0.5×1 or 1×1×1 mm³ voxel size.

Follow‐up MRI was defined as the scan obtained closest to 12 months after initiation of AAT. Because this was a real‐world cohort, follow‐up intervals varied across participants. Annualized rates of brain volume change were therefore calculated using the actual interscan interval. The supplemental table displays the frequency of different scanner/resolutions at baseline versus follow‐up. We controlled for these differences in our statistical analyses.

In addition, comparable three‐dimensional T2‐Fluid‐Attenuated Inversion Recovery (FLAIR) images were also obtained in the same scanning sessions to assess white matter hyperintensity (WMH) volume.

2.4. MRI processing

Following ComBat harmonization, brain and total intracranial (TICV) volumes were measured with the Computational Anatomy Toolbox (CAT12) using default settings and the Neuromorphometrics atlas as provided within CAT12. ³¹ Review of the CAT12 post‐processing image quality metrics and visual inspection of tissue segmentations confirmed the accuracy of all processed outputs. We included the following measures as they are most examined in the literature: hippocampal gray matter volume, whole‐brain (total supratentorial gray matter and white matter), and lateral ventricular cerebrospinal fluid (CSF) volumes. We also examined baseline T2‐FLAIR WMH volume, which was calculated using Lesion Segmentation Toolbox in milliliters (LST). ³²

RESEARCH IN CONTEXT

1. Systematic review: Numerous randomized controlled trials reveal paradoxical brain volume changes, including accelerated brain atrophy and ventricular expansion, in AD participants who receive anti‐Aβ immunotherapy compared to placebo. To date, no associated harmful clinical effects have been reported, but individual patient‐level trial data are unavailable, and this phenomenon is poorly characterized.

2. Interpretation: Our real‐world study reveals prominent ventricular expansion in both EOAD and LOAD participants following a 1‐year course of anti‐Aβ antibody treatment, adding to the consistency of this post‐treatment effect. Our analysis suggests that hippocampal volume loss is less pronounced following treatment of EOAD compared with LOAD participants. Baseline global cortical amyloid burden, measured by ¹⁸F‐florbetaben PET, predicts whole‐brain and hippocampal atrophy following anti‐Aβ treatment.

3. Future directions: Additional longitudinal investigations are needed to elucidate pathophysiological mechanisms and clinical consequences of Aβ immunotherapy‐induced brain volume changes and should evaluate trajectories in specific disease subsets.

2.5. Aβ measures

Amyloid positivity was determined by either CSF fluid analysis of Aβ₄₂ and phosphorylated tau at threonine 181 (p‐tau₁₈₁) through the Mayo reference laboratory, with p‐tau₁₈₁/Aβ₄₂ > 0.028, or by ¹⁸F‐florbetaben positron emission tomography (amyloid PET), using standard cutoff values of 1.18 (≤65 years of age) and 1.20 (>65 years of age) for mean whole‐brain standardized uptake value ratio (SUVr). ²¹ , ³³ , ³⁴ Amyloid PET scans were acquired with a Siemens Biograph scanner 90 min after injection of 8.0 mCI of ¹⁸F‐florbetaben. PET emission and concurrent CT data were co‐registered with T1‐weighted MRI scans and quantification of Aβ plaque levels was performed by brain cortical segmentation of T1‐weighted MRI data using SLANT. ³⁵ Average whole‐brain cortical SUVr was calculated using mean cerebellar cortical SUV as a reference. Regional hippocampal Aβ plaque burden was calculated by averaging of the bilateral hippocampal SUVr.

2.6. Statistical analysis

Normality of continuous variables was assessed using the D'Agostino–Pearson omnibus test. Group comparisons of demographic and clinical characteristics were performed using independent‐sample t‐tests when both groups met normality assumptions; otherwise, the Mann–Whitney U test was applied. Chi‐squared tests were used for comparisons of categorical variables. Baseline brain volume measurements between the two groups were examined by analysis of covariance (ANCOVA) with adjustment for sex, TICV, and whole‐sample mean‐centered age.

To facilitate comparisons with the literature, ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ whole‐brain, hippocampal, and ventricular volume trajectories were assessed using annualized percentage brain volume change. Normalizing for TICV, this value was calculated as the percentage change between baseline and follow‐up volume measurements divided by the interscan interval in years. Comparisons between EOAD and LOAD were performed using independent‐sample t‐tests or Mann–Whitney U tests, as appropriate.

Longitudinal changes in brain volumes were also analyzed using linear mixed‐effects models (LMMs). Separate models were fitted for whole‐brain, hippocampal, and lateral ventricular volumes. In each model, Time (baseline and follow‐up), Group (EOAD vs LOAD), and their interaction (Time × Group) were specified as fixed effects to evaluate longitudinal group differences. Age, sex, and TICV were included as covariates. Although EOAD and LOAD were defined according to age at symptom onset, chronological age at MRI acquisition varied within each group. Therefore, age was included as a continuous covariate to account for within‐group age‐related variability in brain volume, while onset group captured differences related to disease‐onset classification. A random intercept for subject was included to account for within‐subject correlations due to repeated measurements; no random slopes were specified. Subsequent models with addition of apolipoprotein E (APOE) ε4 carrier status and clinical diagnosis of dementia were also performed. To evaluate the influence of baseline amyloid burden on longitudinal structural change, additional models incorporated baseline amyloid PET measures as covariates. Global cortical Aβ SUVr was included in the whole‐brain, hippocampus, and lateral ventricle volume models, whereas hippocampal Aβ SUVr was included only in the hippocampus volume model. Amyloid burden analyses were restricted to participants with available amyloid PET data (n = 86; EOAD = 27, LOAD = 59). Models were estimated using restricted maximum likelihood (REML), and the Kenward–Roger approximation was applied to obtain denominator degrees of freedom for fixed effects. Type III tests of fixed effects were used to assess statistical significance. Model convergence was evaluated using log‐likelihood, parameter, and Hessian convergence criteria as implemented in SPSS.

To avoid overadjustment, WMH volume was not included as a covariate given that its brain volume associations were found to be driven by age. For example, baseline WMH volume was strongly associated with hippocampal volume change in unadjusted analysis (Spearman's correlation r = −0.199, p = 0.037); however, this association was markedly attenuated after controlling for age (partial correlation = −0.079, p = 0.415), suggesting that age accounts for a portion of the variance.

All tests were two‐sided, and a p value < 0.05 was considered statistically significant. Statistical analyses and graphical visualizations were generated using GraphPad Prism version 10.6.1 (GraphPad Software, La Jolla, CA, USA), except for ANCOVA and LMM analyses, which were performed using IBM SPSS Statistics for Windows version 31.0.1.0 (IBM Corp., Armonk, NY, USA).

3. RESULTS

3.1. Participant characteristics

The study cohort comprised 114 participants, 35 with EOAD and 79 with LOAD (Table 1). As expected, the EOAD group was significantly younger than the LOAD group (57.09 ± 5.85 vs 71.68 ± 3.98 years, p < 0.001). Sex distribution did not differ between groups, with females representing 45.7% and 54.4% of the EOAD and LOAD groups, respectively (p = 0.390). The proportion of APOE ε4 carriers was comparable between the groups (51.4% vs 58.2%, p = 0.500). Additionally, education was similar between the groups (14.54 ± 2.56 vs 15.20 ± 2.83, p = 0.285).

TABLE 1.

Demographics, clinical characteristics, and baseline brain volume measurements.

Characteristics	Early‐onset Alzheimer's disease mean ± SD	Late‐onset Alzheimer's disease mean ± SD	p value
N	35	79
Age (years)	57.09 ± 5.85	71.68 ± 3.98	<0.001 a
Sex, N (%)
Female	16 (45.71)	43 (54.43)	0.390 b
Education (years)	14.54 ± 2.56	15.20 ± 2.83	0.285 c
APOE ε4 carriers, N (%)	18 (51.43)	46 (58.23)	0.500 b
Baseline MMSE score	27.71 ± 3.16	27.38 ± 2.30	0.209 c
Baseline FAQ score	3.18 ± 3.40	4.82 ± 4.40	0.075 c
Baseline WMH volume (cm3)	1.57 ± 1.91	4.03 ± 3.72	0.748 d
Baseline global Aβ SUVr (N = 86)	1.35 ± 0.28	1.59 ± 0.25	<0.001 c
Baseline hippocampal Aβ SUVr (N = 86)	1.36 ± 0.17	1.34 ± 0.11	0.213 c
Baseline CSF pTau181/Aβ42 (N = 28)	0.063 ± 0.056	0.047 ± 0.036	0.492 c
Clinical diagnosis, N (%)
Mild cognitive impairment	32 (91.43)	64 (81.01)	0.160 b
Dementia	3 (8.57)	15 (18.99)
Anti‐amyloid treatment, N (%)
Aducanumab	6 (17.14)	7 (8.86)	0.199 b
Lecanemab	29 (82.86)	72 (91.14)
ARIA, N (%)	5 (14.29)	8 (10.13)	0.519 b
ARIA‐E	3 (8.57)	2 (2.50)
ARIA‐H	2 (5.71)	5 (6.33)
ARIA‐E and ARIA‐H	0 (0)	1 (1.27)
Follow‐up time (years)	1.30 (0.91–2.11)	1.42 (0.92–2.50)	0.042 c
Total intracranial volume (cm3)	1,424.00 ± 174.60	1,409.00 ± 145.70	0.656 c
Baseline brain volume measurement (cm3)
Whole‐brain volume	1,127.96 ± 127.40	1,070.11 ± 97.93	0.167 e
Hippocampus	6.42 ± 0.76	5.91 ± 0.68	0.010 e
Lateral ventricle	19.99 ± 11.43	28.24 ± 14.20	0.301 e

Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; ARIA, amyloid‐related imaging abnormalities; ARIA‐E, amyloid‐related imaging abnormalities‐edema; ARIA‐H, amyloid‐related imaging abnormalities‐hemorrhage; CSF, cerebrospinal fluid; FAQ, Functional Activities Questionnaire; MMSE, Mini‐Mental State Examination; SUVr, standardized uptake value ratio; WMH, white matter hyperintensity.

^a T‐test.

^b Chi‐squared test.

^c Mann‐Whitney test.

^d ANCOVA, adjusted by age.

^e ANCOVA, volume adjusted by sex, total intracranial volume, and age.

Disease stage at baseline was comparable between groups, with most participants classified as mild cognitive impairment (MCI) due to AD (EOAD, 91.4%; LOAD, 81.0%) and a smaller proportion meeting criteria for dementia (8.6% vs 19.0%; p = 0.160). Baseline global cognitive and functional performance was similar across groups, as reflected by MMSE (27.71 ± 3.16 vs 27.38 ± 2.30, p = 0.209) and FAQ scores (3.18 ± 3.40 vs 4.82 ± 4.40, p = 0.075).

Cerebral amyloid burden differed by region and group. Global Aβ SUVr was significantly higher in LOAD than in EOAD (1.59 ± 0.25 vs 1.35 ± 0.28, p < 0.001), whereas the hippocampal Aβ SUVr did not differ between groups (1.34 ± 0.11 vs 1.36 ± 0.17, p = 0.213). CSF p‐tau₁₈₁/Aβ₄₂ ratios were available in a subset of participants (N = 28) who did not have Aβ PET imaging data and were similar between the groups (p = 0.492).

3.2. Baseline brain volume measurements

TICV did not differ between the EOAD and LOAD groups (1,424.00 ± 174.60 vs 1,409.00 ± 145.70 cm³, p = 0.656). Baseline MRI measures demonstrated smaller hippocampal volumes in LOAD compared with EOAD (5.91 ± 0.68 vs 6.42 ± 0.76, p = 0.010), whereas whole‐brain and lateral ventricular volumes did not differ significantly at baseline between groups (p = 0.167 and p = 0.301, respectively). Baseline WMH volume was significantly greater in LOAD than in EOAD (4.03 ± 3.72 vs 1.57 ± 1.91 cm³, p < 0.001); however, when controlling for age, the difference was no longer significant (p = 0.748), indicating that the association was largely driven by age.

3.3. AAT exposures

AAT exposure was comparable between the groups. Most participants received lecanemab (EOAD: 82.9%, LOAD: 91.1%), whereas a minority received aducanumab. The incidence of ARIA did not differ between the groups (14.3% in EOAD vs 10.1% in LOAD, p = 0.519). The subtypes of amyloid‐related imaging abnormalities‐edema (ARIA‐E) and amyloid‐related imaging abnormalities‐hemorrhage (ARIA‐H) were infrequent and similarly distributed (Table 1).

3.4. Brain volume trajectories in EOAD versus LOAD exposed to AAT

Figure 1 illustrates the longitudinal change in whole‐brain, hippocampal, and lateral ventricular volumes (raw values) over the study period in the EOAD and LOAD groups. Over the 12‐month treatment course following initiation of AAT, both groups experienced progressive brain atrophy. However, absolute reductions in whole‐brain and hippocampal volumes and the increase in ventricular volume (in milliliters [mL]) were greater in the LOAD group.

FIGURE 1.

Absolute changes in longitudinal brain volume measurements in early‐ and late‐onset Alzheimer's disease (EOAD and LOAD) following anti‐amyloid therapy. At baseline, EOAD exhibited larger whole‐brain and hippocampal volumes and smaller lateral ventricle volumes compared with LOAD. At follow‐up, both groups demonstrated significant brain atrophy, characterized by declines in whole‐brain and hippocampal volumes and expansion of the lateral ventricles. However, the changes in volume differed by disease onset, with LOAD showing greater longitudinal reductions in whole‐brain and hippocampal volumes and more ventricular enlargement relative to EOAD. Note: Follow‐up scans were obtained at varying intervals centered around approximately 12 months after treatment initiation. Values shown represent observed measurements and are not adjusted for differences in follow‐up duration.

LMMs, adjusted for age, sex, and TICV, indicated whole‐brain volume declined significantly over time in the cohort (p < 0.001); however, a significant difference was not observed between EOAD and LOAD (p = 0.163) (Table 2A). Hippocampal volume decreased significantly in both groups (p < 0.001), with greater atrophy seen in LOAD (p = 0.036) (Table 2B). Lateral ventricular volume increased significantly over time in both groups (p < 0.001). After covariate adjustment, the difference in longitudinal ventricular expansion between groups was not significant (p = 0.102) (Table 2C). Controlling for scanner resolution revealed no differences in results (Supplemental Table). Overall, these results suggest differential brain volume trajectories with greater hippocampal atrophy in individuals with LOAD compared to those with EOAD following initiation of AAT. Exploratory analyses further suggested that APOE ε4 carrier status and clinical diagnosis of dementia were each associated with greater hippocampal volume reduction over time (p = 0.019 and p = 0.014, respectively; Table 3B). Dementia diagnosis was additionally associated with greater longitudinal whole‐brain volume loss (p < 0.001; Table 3A).

TABLE 2.

Linear mixed‐effects model of brain volume measurements.

Variable	Estimate	95% CI	p value
A. Linear mixed‐effects model of whole‐brain volume
Time	18.820	13.280 to 24.360	<0.001
Group (EOAD vs LOAD)	36.626	−1.373 to 74.625	0.059
Time × Group	−7.079	−17.077 to 2.920	0.163
Sex	−17.473	−42.057 to 7.110	0.162
Age	−1.467	−3.636 to 0.701	0.183
TICV	0.616	0.536 to 0.695	<0.001
Intercept	198.458	90.895 to 306.021	<0.001
B. Linear mixed‐effects model of hippocampus volume
Time	0.275	0.217 to 0.333	<0.001
Group	0.613	0.190 to 1.036	0.005
Time × Group	−0.112	−0.216 to −0.007	0.036
Sex	0.068	−0.206 to 0.342	0.623
Age	0.002	−0.022 to 0.027	0.846
TICV	0.002	0.001 to 0.003	<0.001
Intercept	2.746	1.548 to 3.944	<0.001
C. Linear mixed‐effects model of lateral ventricle volume
Time	−4.774	−5.481 to −4.066	<0.001
Group	−4.775	−13.319 to 3.769	0.271
Time × Group	1.062	−0.215 to 2.339	0.102
Sex	2.589	−2.970 to 8.148	0.358
Age	0.382	−0.108 to 0.872	0.126
TICV	0.039	0.021 to 0.057	<0.001
Intercept	−24.245	−48.564 to 0.074	0.051
D. Linear mixed‐effects model of whole‐brain volume covariate with baseline global Aβ SUVr
Time	17.369	10.948 to 23.790	<0.001
Group (EOAD vs LOAD)	29.175	−9.282 to 67.632	0.135
Time × Group	−6.341	−17.801 to 5.119	0.274
Sex	−17.805	−42.008 to 6.397	0.147
Age	−0.012	−6.660 to 6.637	0.997
TICV	0.633	0.559 to 0.706	<0.001
Baseline global Aβ SUVr	−86.318	−126.125 to −46.512	<0.001
Age × Baseline global Aβ SUVr	−0.726	−4.991 to 3.539	0.736
Intercept	313.968	199.637 to 428.299	<0.001
E. Linear mixed‐effects model of hippocampus volume covariate with baseline global Aβ SUVr
Time	0.263	0.196 to 0.330	<0.001
Group	0.504	0.062 to 0.946	0.026
Time × Group	−0.133	−0.253 to −0.013	0.030
Sex	0.020	−0.258 to 0.299	0.886
Age	0.002	−0.074 to 0.079	0.950
TICV	0.002	0.001 to 0.003	<0.001
Baseline global Aβ SUVr	−0.880	−1.338 to −0.422	<0.001
Age × Baseline global Aβ SUVr	0.003	−0.046 to 0.052	0.897
Intercept	4.011	2.695 to 5.328	<0.001
F. Linear mixed‐effects model of hippocampus volume covariate with baseline hippocampal Aβ SUVr
Time	0.263	0.196 to 0.330	<0.001
Group	0.627	0.170 to 1.084	0.008
Time × Group	−0.133	−0.253 to −0.013	0.030
Sex	−0.010	−0.304 to 0.283	0.944
Age	0.106	−0.020 to 0.232	0.097
TICV	0.002	0.001 to 0.003	<0.001
Baseline hippocampal Aβ SUVr	0.386	−0.651 to 1.422	0.461
Age × Baseline hippocampal Aβ SUVr	−0.078	−0.171 to 0.014	0.096
Intercept	2.248	0.371 to 4.125	0.020
G. Linear mixed‐effects model of lateral ventricle volume covariate with baseline global Aβ SUVr
Time	−4.893	−5.695 to −4.090	<0.001
Group	−7.627	−17.490 to 2.236	0.128
Time × Group	1.731	0.299 to 3.163	0.018
Sex	1.797	−4.460 to 8.054	0.569
Age	−0.488	−2.207 to 1.231	0.573
TICV	0.034	0.015 to 0.054	<0.001
Baseline global Aβ SUVr	9.443	−0.848 to 19.735	0.072
Age × Baseline global Aβ SUVr	0.449	−0.653 to 1.552	0.420
Intercept	−33.007	−62.558 to −3.456	0.029

Note: A–C Linear mixed‐effects model of brain volume measurements, with fixed effect of Time, Group (EOAD vs LOAD), Time×Group interaction, sex, age, WMH, TICV with random intercept for subject. Equation for LMMs; volume ij ​ = β0 ​+ β1​(Time​) + β2​(Groupi​) + β3​(Timeij​×Groupi​) + β4​(Sexi​) + β5​(Agei​) + β6​(TICVi​) + b0i​ + εij. D–G Linear mixed‐effects model of brain volume measurements with addition to covariate with baseline global or hippocampal Aβ SUVr and Age×Aβ SUVr interaction. Equation for LMMs; volume ij ​ = β0 ​+ β1​(Time​) + β2​(Groupi​) + β3​(Timeij​×Groupi​) + β4​(Sexi​) + β5​(Agei​) + β6​(TICVi​) + β7​(Baseline global/hippocampal Aβ SUVRi​) ​) + β8​(Age i×Aβ SUVRi​) + b0i​ + εij.

TABLE 3.

Linear mixed‐effects model of brain volume measurements covariate with APOE ε4 carrier status and clinical diagnosis.

Variable	Estimate	95% CI (lower, upper)	p value
A. Linear mixed‐effects model of whole‐brain volume covariate with APOE ε4 carrier status and clinical diagnosis
Time	18.820	13.280 to 24.360	<0.001
Group (EOAD vs LOAD)	28.274	−8.347 to 64.896	0.129
Time × Group	−7.079	−17.077 to 2.920	0.163
Sex	−18.497	−41.940 to 4.946	0.121
Age	−1.700	−3.781 to 0.381	0.108
TICV	0.607	0.530 to 0.683	<0.001
APOE ε4 (non‐carrier vs carrier)	4.656	−15.536 to 24.847	0.649
Diagnosis (MCI vs AD)	48.110	20.327 to 75.892	<0.001
Intercept	171.954	68.237 to 275.672	0.001
B. Linear mixed‐effects model of hippocampus volume covariate with APOE ε4 carrier status and clinical diagnosis
Time	0.275	0.217 to 0.333	<0.001
Group	0.557	0.152 to 0.962	0.007
Time × Group	−0.112	−0.216 to −0.007	0.036
Sex	0.051	−0.208 to 0.311	0.696
Age	0.002	−0.021 to 0.025	0.836
TICV	0.002	0.001 to 0.003	<0.001
APOE ε4	0.269	0.046 to 0.492	0.019
Diagnosis	0.389	0.082 to 0.696	0.014
Intercept	2.412	1.265 to 3.559	<0.001
C. Linear mixed‐effects model of lateral ventricle volume covariate with APOE ε4 carrier status and clinical diagnosis
Time	−4.774	−5.481 to −4.066	<0.001
Group (EOAD vs LOAD)	−4.019	−12.662 to 4.625	0.359
Time × Group	1.062	−0.215 to 2.339	0.102
Sex	2.675	−2.893 to 8.244	0.343
Age	0.404	−0.090 to 0.898	0.108
TICV	0.039	0.021 to 0.057	<0.001
APOE ε4	−0.266	−5.062 to 4.530	0.913
Diagnosis	−4.314	−10.913 to 2.285	0.198
Intercept	−21.947	−46.578 to 2.684	0.080

Note: Linear mixed‐effects model of brain volume measurements, with fixed effect of Time, Group (EOAD vs LOAD), Time×Group interaction, Sex, Age, WMH, TICV, APOE ε4 carrier status and Clinical diagnosis with random intercept for subject. Equation for LMMs; volume ij ​ = β0 ​+ β1​(Time​) + β2​(Groupi​) + β3​(Timeij​×Groupi​) + β4​(Sexi​) + β5​(Agei​) + β6​(TICVi​)+ β7​(APOEε4​) + β8​(Diagnosis) + b0i​ + εij.

Figure 2 illustrates the annualized rate of change in whole‐brain, hippocampal, and lateral ventricular volumes 12 months after the initiation of AAT. Both EOAD and LOAD demonstrated reductions in whole‐brain and hippocampal volumes over 12 months, accompanied by ventricular expansion. The rate of whole‐brain atrophy was greater in LOAD than in EOAD (mean whole‐brain atrophy rate: −1.19 ± 1.42 vs −0.77 ± 2.06%/year); however, this did not reach statistical significance (p = 0.278). The rate of hippocampal atrophy was greater in LOAD than in EOAD (mean hippocampal atrophy rate: −3.42 ± 3.84 vs −2.01 ± 2.23%/year; p = 0.017). The ventricular expansion rate was similar between groups and reflected the greatest brain volume change in both groups (mean ventricular expansion rate: 14.35 ± 10.29 vs 15.71 ± 15.51%/year in LOAD vs EOAD; p = 0.381).

FIGURE 2.

Annualized brain atrophy and ventricular expansion rates in early‐ and late‐onset Alzheimer's disease (EOAD and LOAD) following anti‐amyloid therapy. LOAD showed greater (∼1.5×) whole‐brain and (∼1.7×) hippocampal volume reduction rates compared with EOAD, with differential hippocampal effects reaching statistical significance, while the ventricular expansion rate was similar between the groups.

3.5. Amyloid burden predicts longitudinal whole‐brain and hippocampal atrophy after AAT

Amyloid burden was assessed for its potential influence on longitudinal brain atrophy following initiation of AAT by including baseline amyloid PET measures as covariates in the mixed‐effects models. After adjustment for age, sex, TICV, and onset group, baseline global cortical Aβ SUVr was significantly associated with longitudinal change in whole‐brain (p < 0.001) and hippocampal (p < 0.001) volumes, but not with lateral ventricular volume (p = 0.072), although a trend toward greater ventricular expansion with higher Aβ burden was observed (Table 2D,E,G). In contrast, baseline hippocampal Aβ SUVr was not significantly associated with longitudinal hippocampal volume loss (p = 0.461; Table 2F). Importantly, differential longitudinal trajectories between EOAD and LOAD remained even after inclusion of Aβ burden in the models. Group differences in the degree of longitudinal hippocampal volume loss persisted (p = 0.030), indicating that baseline Aβ burden did not fully account for the observed group differences in structural decline. These findings reveal distinct associations of regional and global Aβ burden with brain volume loss, with global cortical Aβ burden predicting whole‐brain and hippocampal atrophy, whereas regional hippocampal Aβ burden was not independently associated with hippocampal volume loss over the study period.

Supplemental material displays associations of annualized brain volume changes with Aβ SUVr and additional covariates (Supplemental Figures S1–S5).

4. DISCUSSION

This real‐world study examined differences in brain structural trajectories between EOAD and LOAD participants following a 1‐year course of AAT. We found that prominent ventricular expansion occurs along with whole‐brain atrophy in both EOAD and LOAD. Differential regional hippocampal volume loss occurred, with LOAD showing greater hippocampal atrophy compared with EOAD over the study period, independently of baseline cerebral amyloid burden. However, global cortical amyloid burden predicted both whole‐brain and hippocampal structural loss following AAT. Overall, these findings add to the consistency of ventriculomegaly as an AAT‐induced effect and suggest the need for further study of brain volume changes following AAT in specific AD subsets.

Excess ventricular expansion has been demonstrated in numerous randomized placebo‐controlled AD trials, including studies of aducanumab, lecanemab, and donanemab. ⁶ , ⁷ , ¹² , ¹⁵ In aducanumab trials, a dose‐dependent increase in ventricular volume was observed in all treatment groups, compared to placebo. ⁷ , ¹⁵ In our analysis, and in accordance with clinical trial findings, ventricular expansion was the most pronounced brain volume effect and was similar in magnitude, with ∼15% annual expansion rate on average, in both EOAD and LOAD. While control participants were not available in our analysis, this expansion rate is prominent when accounting for the early disease stage of our sample. Prior literature demonstrated that the ventricles expand by only 1.5% to 3.0% per year in healthy elderly individuals and by 5% to 16% per year in persons with untreated AD. ⁹ Furthermore, the rate of ventricular expansion is known to accelerate as the disease progresses. ³⁶ , ³⁷ Our study cohort consisted predominantly of individuals in the MCI stage with very early cognitive symptoms (mean MMSE was 27), differing substantially from many previously studied trial cohorts. ² , ¹³ Future analysis with age‐matched AD controls who have not received AAT will be needed to better quantify the ventricular effects observed in our study.

Parenchymal atrophy was less pronounced than ventricular volume effects in our study cohort. Our findings suggest differential brain structural loss in EOAD and LOAD following AAT, which reached statistical significance in the hippocampus. We found that LOAD had a 1.2% whole‐brain volume reduction rate following AAT, whereas this measured 0.8%, on average, in EOAD. Rates of whole‐brain atrophy are estimated at 1.4% to 2.2% per year in untreated AD and do not usually exceed 0.7% per year during normal aging (for a mean age of 70 years). ⁹ Regarding hippocampal structure, we found lower baseline volume and a 1.7× greater atrophy rate (3.4% compared to 2.0% structural loss per year) following AAT in LOAD versus EOAD. In untreated mild AD (MMSE score > 20), hippocampal atrophy rates are reported at 3% to 6% per year, compared with 0.3% to 2.2% per year in normal aging. ⁹ While there is no consistent evidence for excess hippocampal volume loss across anti‐Aβ immunotherapies, phase 3 trials of lecanemab and donanemab interestingly demonstrated less hippocampal volume reduction in treated participants compared to placebo‐treated individuals. ⁷ Future controlled analysis will be needed to determine whether this finding is replicated and reflects a therapy‐induced disease modification that associates with cognitive benefits in EOAD and LOAD.

Post‐immunotherapy volumetric effects, or Aβ‐related pseudoatrophy, is considered a paradoxical outcome. ⁷ , ¹² Improved cognitive trajectories despite brain structural loss have led to the assumption of underlying non‐toxic etiologies. ⁷ Belder et al. surmised that this imaging sequela reflected clearance of parenchymal Aβ plaques and associated inflammatory cellular reactivity along with alterations in cerebral fluid dynamics. Treatment‐induced toxicity remains a possibility, however, as definitive studies excluding accelerated neurodegeneration are lacking. ⁶ , ⁷ In support of an Aβ‐related response, excess brain volume changes are characteristic of only those immunotherapies that achieve Aβ reduction, and it has been noted that the magnitude of volumetric alterations appears related to the extent of amyloid removal. ⁷ Our data lend support to the notion that amyloid pathology is linked with post‐immunotherapy brain volume effects, in that pre‐treatment global cortical amyloid burden was found to be predictive of both whole‐brain and hippocampal atrophy following AAT. It is important to emphasize that our EOAD and LOAD groups showed differences in baseline amyloid burden, and while we controlled for this, Aβ burden may play a role in the differential volume changes observed between the groups. Larger‐scale longitudinal studies with inclusion of temporal amyloid PET and MRI biomarker analysis and long‐term clinical outcomes assessment will be needed to draw conclusions on this and to understand whether our findings hold over longer treatment durations. Additionally, retrospective analyses of the placebo‐controlled clinical trial datasets are necessary to elucidate the mechanistic underpinnings of the observed MRI phenomena.

In parallel with these needed studies, additional work is necessary to understand longitudinal biomarker trajectories in untreated EOAD. ²¹ , ²² Limited reports specifically assess brain volume changes in sporadic EOAD. ⁹ , ²¹ , ²² While a non‐linear pattern of brain atrophy is known to occur over the disease course in LOAD and in normal aging, ¹¹ , ³⁸ studies examining such patterns in EOAD are lacking. Additionally, given heterogeneous measures of brain atrophy rates in typical late‐onset sporadic AD, ⁹ understanding atrophy trajectories and clinical outcomes with various APOE genotypes, comorbidities, for example, diabetes, hypertension, and other vascular diseases, and other risk factors, will also be relevant to interpreting data in this field and in personalizing AAT recommendations. Our data suggest that APOE ε4 status and clinical dementia diagnosis are each associated with greater brain structural volume reduction following AAT. Furthermore, given results of this study that differ from our initial hypothesis, the effects of lower baseline brain volumes and extent of AD‐related copathologies, such as additional proteinopathies that may differ between LOAD and EOAD, should be analyzed in future work.

There are several limitations of this work. As mentioned earlier, we did not include an untreated AD control group. As brain atrophy is known to follow a non‐linear course across the lifespan, future controlled studies will be required. The primary aim of this study was not to determine whether excess atrophy occurred over time but rather to assess whether the magnitude of brain volume effects differed between EOAD and LOAD over the first year of AAT. For diagnosis of EOAD, we did not perform genetic testing, so it is possible that a few individuals may have carried a gene known to impart familial disease. In addition, imaging data were acquired using different scanners and resolutions. While we statistically controlled for these differences in scanner and sequence parameters and used data harmonization techniques, we cannot exclude the possibility that imaging differences may contribute to the results. Nevertheless, this limitation reflects the practicality of real‐world data; with more than 200 patients receiving monoclonal antibodies in our clinic, it is not realistic to obtain all imaging data on a single scanner. Analysis of larger diverse cohorts is needed to assess the generalizability of our findings. Finally, our study was limited to participants who received lecanemab or aducanumab, which is no longer on the market. Future studies are needed to evaluate differences with donanemab.

In conclusion, we find differential brain structural trajectories between EOAD and LOAD with greater hippocampal atrophy in LOAD following AAT. Prominent ventricular expansion occurs in both groups, further establishing the consistency of this post‐AAT effect. Larger long‐term longitudinal studies with individual patient‐level and controlled analyses are needed to discern the mechanisms and implications of the observed brain morphologic changes. A clearer understanding of these anti‐Aβ immunotherapy effects and their relationships with clinical outcomes will allow for more informed therapeutic decision‐making, and potential optimized therapeutic approaches, in individuals with varied AD progression risk.

AUTHOR CONTRIBUTIONS

Conceptualization: Rashi I. Mehta, Methasit Jaisa‐aad, Marc W. Haut. Methodology: Rashi I. Mehta, Methasit Jaisa‐aad, Marc W. Haut. Investigation: Rashi I. Mehta, Patrick D. Worhunsky, Joseph E. Malone, Melanie Ward, Cierra M. Keith, Holly E. Phelps, Stephanie Pockl, Nafiisah Rajabalee, Gary D. Marano, Lauren Bojarski, Samantha Smith, Kayla M. Massey, Marcus Hatfield, Marc W. Haut. Formal Analysis: Methasit Jaisa‐aad, Patrick D. Worhunsky, Marc W. Haut. Writing‐original draft: Rashi I. Mehta, Methasit Jaisa‐aad, Marc W. Haut. Writing‐review and editing: Rashi I. Mehta, Methasit Jaisa‐aad, Camila Vieira Ligo Teixeira, Patrick D. Worhunsky, Marc W. Haut. Resources: Ali R. Rezai. Data curation: Pierre‐François D'Haese. All authors read and approved the manuscript.

FUNDING INFORMATION

This research did not receive any specific grant from funding agencies in the public, commercial, or not‐for‐profit sectors.

CONFLICT OF INTEREST STATEMENT

All authors declare no relevant conflicts of interest with respect to the research, authorship, and/or publication of this article. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants provided written informed consent. The study was conducted in accordance with the principles of the Declaration of Helsinki.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

DATA AVAILABILITY STATEMENT

The data supporting this study will be made available from the corresponding author upon reasonable request from qualified investigators.