Subject characteristics and clinical trial data

Tables 1 and 2 summarize the demographic and neuropathological characteristics of six aducanumab clinical trial participants and nine untreated AD patients. Groups were matched for age, APOE genotype, and Braak neurofibrillary tangle stage. CAA severity was graded according to the scheme established by Love et al. [21], where 0 = absent, 1 = mild, 2 = moderate, and 3 = severe. All clinical trial subjects received the maximum dose of 10 mg/kg. Baseline Aβ PET Standardized Uptake Value ratio (SUVr) and Centiloid values, baseline MMSE and CDR-SB scores, trial participation, total number of infusions, cumulative dose, time between last dose and death, and clinical trajectories are provided in Table 3. Data from EMERGE and ENGAGE phase III trials show that 48% and 31% of patients treated with 10 mg/kg aducanumab, respectively, reached a PET composite SUVr of ≤ 1.10 after 78 weeks of treatment—approximately 18 infusions—a proposed threshold distinguishing Aβ-negative from Aβ-positive patients [9]. At the group level, patients enrolled in EMERGE and ENGAGE demonstrated a 71% and 59% in PET Aβ burden compared with baseline, respectively. All six immunotherapy patients studied here received between 19 and 76 monthly infusions, a treatment duration previously shown to produce substantial reductions in Aβ burden [9, 31]. The interval between final infusion and death ranged from 7.86 weeks to 4.97 years, enabling direct histological examination of both the consequences of long-term treatment and the sequelae following treatment discontinuation. Asymptomatic amyloid-related imaging abnormalities-hemosiderin (ARIA-H) and -edema (ARIA-E) were reported in four individuals during active treatment (Table 4). Cause of death for each participant was unrelated to treatment.

Table 3 Baseline Aβ PET and clinical trial participation and data of study subjectsTable 4 Amyloid-related imaging abnormalities (ARIA) in aducanumab clinical trial subjectsAducanumab-treated patients exhibit lower Aβ burden, with a positive correlation between withdrawal duration and Aβ load

Chronic aducanumab administration improves Aβ fluid and imaging biomarkers within 18 months of treatment initiation [9, 31]. To evaluate these effects histologically, we performed Aβ immunostaining in the MTL of six aducanumab clinical trial participants (Fig. 1a; Supplemental Figs. 1, 2). Consistent with prior PET studies, we observed a pronounced reduction in Aβ-immunopositive material in this brain region, with all treated subjects showing markedly lower Aβ levels than the untreated AD group average (Fig. 1b).

Fig. 1

Low-magnification micrographs illustrate a spectrum of Aβ pathology in the MTL revealed by Aβ immunolabeling (red). Images show an untreated AD case representing the “natural” AD neuropathological trajectory in the absence of anti-Aβ therapy (left), an aducanumab-treated subject with modest Aβ plaque deposition (~ 50% of the average untreated burden; Aducanumab 6, center), and an aducanumab-treated subject with almost no detectable Aβ plaques (~ 1% of the average untreated burden; Aducanumab 1, right). Micrographs were generated from fluorescent widefield images using an inverse background setting (scale bar = 1 mm). b, c Quantification of insoluble Aβ in the MTL revealed a significant reduction in total Aβ deposits in immunotherapy patients, as detected using antibodies D54D2 (p = 0.002) and 4G8 (p < 0.001). d Aβ area fraction, measured using D54D2 immunolabeling, was strongly associated with time since last aducanumab dose (r = 0.829; p = 0.03), but not with cumulative exposure (r = − 0.371; p = 0.25)

The Aβ antibody used in this initial analysis (clone D54D2) recognizes an N-terminal epitope overlapping with aducanumab’s binding site, which could potentially interfere with immunohistochemical labeling. To rule out this possibility, we applied two fluorophore-conjugated anti-human IgG antibodies to detect residual aducanumab bound to Aβ plaques. No IgG+ plaques were identified in any brain donor, including one patient who received their final dose just 7 weeks before death, indicating rapid antibody washout (Supplemental Fig. 3). Findings were confirmed using antibody clone 4G8, which recognizes the mid-segment (aa17–24) of the Aβ peptide (Fig. 1c), as well as antibodies targeting the C-termini of Aβ40 and Aβ42 (Supplemental Fig. 4a)—outside aducanumab’s binding domain—all of which confirmed lower Aβ deposition.

To assess changes in Aβ plaque structure and associated neuritic pathological changes, we quantified Aβ+ThioS+ dense-core plaques and SMI312+ dystrophic neurites. Although overall Aβ burden was considerably lower in treated individuals, the proportion of ThioS+ Aβ plaques and the number of plaque-associated dystrophic neurites were comparable between groups, suggesting that Aβ plaques may retain similar biochemical and neuropathological characteristics following treatment (Supplemental Fig. 4b).

To examine dose- and time-dependent responses to aducanumab, we correlated postmortem Aβ levels with total drug received and time since last administration (Fig. 1d). No clear association was found between cumulative dose and Aβ burden (Supplemental Table 2). By contrast, a statistically significant positive correlation emerged between withdrawal duration and Aβ burden, indicating that longer treatment gaps are associated with greater Aβ load (Fig. 1d; Supplemental Table 2). Notably, even individuals with treatment gaps of approximately 4–5 years had considerably less Aβ than the average untreated AD donor. This effect was not attributable to brain-antibody retention, as all plaques were IgG− (Supplemental Fig. 3).

We also confirmed these findings in Brodmann area 20 (BA20), which likewise exhibited reduced Aβ deposition in immunotherapy patients compared with untreated AD controls (Supplemental Fig. 4c). Baseline, pre-treatment Aβ Centiloid values did not correlate with postmortem Aβ levels in either BA20 or the MTL, indicating that Aβ burden at treatment initiation, as measured by PET, was likely reduced by therapeutic intervention (Supplemental Fig. 4d).

Aβ plaques are broadly categorized as diffuse or fibrillary/dense core based on ultrastructural and tinctorial characteristics. Given that aducanumab binds both oligomeric and fibrillar Aβ [14, 31], we asked whether specific plaque subtypes are more effectively impacted by immunotherapy. We quantified Aβ in the presubiculum, where diffuse ‘lake-like’ deposits predominate, and in the hippocampus proper—hereafter referred to as ‘hippocampus’—where cored and diffuse dense plaques accumulate (Supplemental Fig. 5a) [35]. Relative to untreated AD donors, immunotherapy patients exhibited 87% and 50% lower Aβ in the presubiculum and hippocampus, respectively (Supplemental Fig. 5b). A similar regional disparity was observed using the 4G8 antibody (Supplemental Fig. 5c).

Neuritic tau is reduced in association with Aβ burden, but neurofibrillary tangles remain

Antemortem biomarker studies show that Aβ removal by aducanumab corresponds with reductions in cerebrospinal fluid (CSF) phosphorylated tau (pTau) levels and improvements in tau PET [9]. To investigate whether analogous neuropathological changes occur in aducanumab-treated patients, we quantified several forms of pathological tau, including somatic tangles, neuropil threads, and plaque-associated neuritic dystrophies (Fig. 2a, b).

Fig. 2

Immunolabeling of pTau217 (green), PHF-1 (magenta), and AT8 (cyan) revealed abundant neurofibrillary tangles, neuropil threads, and Aβ plaque-associated neuritic dystrophies in untreated AD subjects, as well as in immunotherapy patients with Aβ plaque pathology (yellow) (scale bars = 100 µm in A; 200 µm in B). C Overall pTau217 burden did not differ between treatment groups (p = 0.12), whereas PHF-1 and AT8 area fractions were significantly lower following aducanumab treatment (PHF-1: p = 0.01; AT8: p = 0.03). D In contrast, the numbers of PHF-1+ and AT8+ neurofibrillary tangles were comparable between untreated and treated AD subjects (PHF-1: p = 0.34; AT8: p = 0.17), indicating that decreased total tau burden primarily reflects changes in plaque-associated neuritic pathology. E Super resolution imaging of Bassoon+ and Homer1+ pre- and post-synaptic terminals respectively in region CA1 of untreated and aducanumab-treated patients revealed no change in synapse abundance (Bassoon: p = 0.26; Homer1: p = 0.50) (scale bar = 50 µm)

Total pTau-217 did not differ between groups (Fig. 2c). However, total PHF-1 and AT8 percent area were markedly lower in treated individuals relative to untreated AD controls despite equivalent Braak stages (Fig. 2c). We also confirmed these findings in BA20, which likewise exhibited reduced AT8 percent area in immunotherapy patients compared with untreated AD controls (Supplemental Fig. 6). We considered whether resolution of dystrophic neurites and/or neurofibrillary tangles could account for this change. Tangle counts in the hippocampus were similar between groups (Fig. 2D), and no clusters of pTau+ dystrophic neurites were detected outside of clear association with an Aβ plaque core (Fig. 2a, b). Comparable results were obtained using a non-phospho-directed tau antibody (Supplemental Table 1). Post hoc power analysis indicated that, with the current sample sizes, we would have had 90% power to detect a 75% change in tangle number, and 60% power to detect a 50% change, assuming a one-sided test at α = 0.05. The observed group difference was only ~ 15%, suggesting that a biologically meaningful effect was unlikely to have been missed due to insufficient power. Together, these findings indicate that mature neurofibrillary tangles are largely stable and may be resistant to Aβ-directed therapy, whereas plaque-associated neuritic tau and neuropil threads were reduced in parallel with lower parenchymal Aβ burden. We also applied super-resolution microscopy to quantify pre- and post-synaptic boutons in region CA1 (Fig. 2e). Bassoon+ and Homer1+ pre- and post-synaptic densities were comparable between aducanumab-treated and untreated AD patients.

Aducanumab treatment is associated with increased Aβ on cortical microvessels

Active immunization with AN1792 has been associated with a shift in Aβ accumulation from plaques to microvessels [6, 29]. To evaluate whether this also occurs with passive immunization, we quantified vessel-associated Aβ in available tissue samples. Leptomeningeal vessels were excluded from analyses to mitigate potential artifacts from tissue sectioning.

The incidence of Aβ+ and ThioS+ deposits colocalizing with GLUT1+ endothelial cells was almost threefold greater in aducanumab-treated individuals compared with untreated AD patients (Fig. 3a, b). While the vast majority of Aβ (mean 93.5%) was concentrated in parenchymal plaques in untreated AD donors, up to 75% of Aβ was vessel-associated in immunotherapy patients (Fig. 3c). Vessel-type analysis showed an almost fivefold increase in the average number of Aβ+ capillaries (≤ 10 μm diameter) in clinical trial subjects, with no corresponding change in larger vessels (Fig. 3d). Capillary Aβ was also detected with the 4G8 antibody (Supplemental Fig. 7a), but, notably, not with Aβ1-40- or Aβ1-42-directed antibodies. The proportion of Aβ40/42 deposited on larger vessels was similar between treatment groups (Supplemental Fig. 3e).

Fig. 3

b GLUT1+ blood vessels (cyan) exhibited substantially more Aβ+ and ThioS+ accumulation (yellow) in the MTL of aducanumab-treated vs. untreated AD patients (Aβ: p = 0.01; ThioS: p = 0.03). Arrows denote sites of vascular Aβ deposition (scale bar = 50 µm). c Despite an overall decrease in parenchymal Aβ following immunotherapy, the percentage of total Aβ localized to blood vessels was nearly fourfold higher in treated participants (p = 0.02). d This shift was driven by an increase in the number of affected capillaries (p = 0.04), whereas the frequency of affected vessels larger than 10 μm did not differ between groups (p = 0.33). e The ratio of Aβ40/42 deposited on larger vessels was similar between treatment groups

Although aducanumab-treated individuals showed considerably lower overall cerebral Aβ burden, a disproportionate fraction of Aβ pathology was localized to the brain vasculature following immunotherapy. Importantly, this change in distribution appears to be a lasting and/or progressive phenomenon, as most subjects had discontinued treatment years before autopsy. One subject whose last dose was only 7 weeks prior to death had a similar pattern of loss of Aβ40/42 on capillaries. A trend toward a significant positive correlation between time since last dose and vascular Aβ was observed (Supplemental Fig. 7a), though this fell just below the threshold for significance, suggesting that redistribution of Aβ toward cerebral blood vessels may occur during treatment and persist beyond the treatment period.

Whole-slide assessment of the MTL revealed a pronounced negative spatial correlation between GLUT1+-Aβ and areas of persistent parenchymal Aβ plaques: vascular Aβ was most prevalent in regions where plaque burden was minimal, as observed in the four cases with detectable plaque pathology (Supplemental Figs. 8–11). No such relationship was found in untreated AD donors (Supplemental Figs. 12–14).

Local extravasation of albumin, fibrinogen, and human IgG in the MTL lobe showed no treatment-dependent alterations. Prussian blue staining for hemosiderin deposits likewise revealed no treatment-related differences in microhemorrhage frequency (Supplemental Fig. 7c), consistent with the absence of ARIA detected by MRI in the MTL of any analyzed subject during the study period (Table 4).

Microglia and astrocytes do not show overt reactivity after immunotherapy

Anti-Aβ antibodies with full effector function activate microglial Fc receptors to trigger phagocytosis and cytokine production. In a recent study using chimeric murine aducanumab in a transgenic mouse model harboring both Aβ and tau neuropathologies, microglia remained in proximity to plaques at least one week after drug washout [39], accompanied by pronounced astrocytic GFAP expression in both plaque-adjacent and more distal cortical regions.

To determine whether analogous cellular responses persist in patients following aducanumab treatment and discontinuation, we quantified IBA1+ microglia and GFAP+ astrocytes in the MTL. Despite reduced Aβ burden in aducanumab-treated cases, there was only a nonsignificant trend towards increased IBA1 coverage (Fig. 4a, b). Though only sparse plaques were visible in some immunotherapy patients (Supplemental Fig. 15), microglial association with plaques was comparable to that observed in the untreated AD group (Fig. 4c). A modest trend towards expanded GFAP expression was similarly observed, but did not reach statistical significance (Fig. 4d, e). These results may reflect variability in treatment duration, withdrawal interval, and limited sample size. Given the increase in vascular Aβ+ and ThioS+ deposits, we also assessed perivascular astrogliosis by quantifying vessel-associated GFAP expression [18]. No treatment-dependent differences were detected (Fig. 4f).

Fig. 4

b Total microglial IBA1+ area fraction (green) showed no significant treatment-related differences (p = 0.06). C IBA1 coverage of Aβ plaques (magenta), reflecting plaque-associated microglial recruitment, was also comparable between treatment groups, potentially due to drug washout and prolonged intervals between last dose and death in most patients (p = 0.38). D, E Total astrocytic GFAP+ area fraction (cyan) was largely unchanged in aducanumab-treated cases (p = 0.09). F Vascular astrogliosis was not observed (p = 0.19) (scale bar = 100 µm)