Evaluation of plasma neurofilament and glial fibrillary acidic protein to predict or detect amyloid related imaging abnormalities

Abstract

INTRODUCTION

Amyloid‐targeting therapies (ATT) require frequent magnetic resonance imaging monitoring to detect vasogenic edema and hemorrhage, i.e., amyloid‐related imaging abnormalities (ARIA). Neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP) are potential blood‐based biomarkers (BBM) of ARIA.

METHODS

Plasma NfL and GFAP were longitudinally collected in an ATT clinic and natural log transformed. Baseline, pre‐ARIA, and trajectories were compared between patients with (N = 15) and without (N = 45) ARIA.

RESULTS

Baseline NfL and GFAP did not differ between patients with or without ARIA (ln[NfL] ECLIA 1.24 ± 0.38 pg/mL vs 1.25 ± 0.40 pg/mL, p = 0.77; ln[NfL] Simoa 2.87 ± 0.47 pg/mL vs 2.98 ± 0.40 pg/mL, p = 0.30, ln[GFAP] ECLIA 4.95 ± 0.36 pg/mL vs 4.88 ± 0.47 pg/mL, p = 0.69). No significant changes were seen at the timepoint preceding ARIA, and ARIA incidence made no detectable impact on biomarker trajectories.

DISCUSSION

Biobanking in treatment clinics can evaluate clinical utility of BBMs as monitoring tools. These findings do not support a strong association between NfL or GFAP and ARIA in the ATT clinic.

Highlights

• AD treatment clinic biobanks can inform real‐world utility of biomarkers.

• No significant association was observed between NfL and GFAP and ARIA.

• ARIA edema likely reflects transient vascular changes, not lasting neuronal injury.

• Relatively elevated occipital PET signal may reflect a higher risk of ARIA.

1. INTRODUCTION

Amyloid‐targeting therapies (ATTs) are being increasingly clinically implemented for the treatment of early stages of Alzheimer's disease (AD), but better tools are needed to predict and detect the treatment‐limiting and potentially life‐threatening side effects, namely, amyloid‐related imaging abnormalities (ARIA). Known risk factors for both ARIA with edema (ARIA‐E) and hemorrhage (ARIA‐H) include hetero‐ or homozygosity for the APOE ε4 gene allele (APOE ε4), the presence of cerebral amyloid angiopathy (CAA), as evidenced by pre‐existing cerebral microbleeds (CMBs) or superficial siderosis, hypertension (especially untreated), and amyloid plaque burden. ¹ , ² The burden of white matter disease, vascular malformations, and anticoagulation may additionally increase treatment‐related hemorrhagic risk. ³

Nevertheless, these predictors are insensitive and incomplete, necessitating the current approach of frequent interval magnetic resonance imaging (MRI) scans for safety monitoring, which introduces significant constraints due to MRI availability and cost, reducing both accessibility and scalability of ATT. ⁴ , ⁵ Blood‐based biomarkers (BBMs) hold promise as tools for prediction and detection, especially as frequent sampling can be performed prior to treatments at infusion centers or clinical labs. ⁶ , ⁷ Prior to treatment, BBMs could help stratify risk and allow for more nuanced, personalized treatment decisions. ⁸ During the course of treatment, if BBMs are available to accurately detect early changes in ARIA, they could guide MRI evaluation to facilitate earlier ARIA detection and limit unnecessary scans. Recent real‐world investigations focused on the utility of diagnostic biomarkers prior to ATT ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ and evaluating outcomes of patients receiving ATT. ¹⁵ , ¹⁶ , ¹⁷ However, diagnostic plasma biomarkers have largely not been evaluated for prognosis or monitoring, especially with respect to risk of ARIA and in a real‐world setting.

Among clinically available BBMs, neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP) have intriguing potential for predicting and detecting ARIA. NfL is a neurofilament subunit thought to reflect degree of neuronal injury and is elevated in many neurological diseases commensurate with rate of injury. ¹⁸ GFAP is an astrocytic protein associated with astrocytic activation and increases in the presence of amyloid plaque. ¹⁹ , ²⁰ , ²¹ Both NfL and GFAP been shown to be elevated in CAA, which has many overlapping features with ARIA. ¹ , ²² It is possible that heightened baseline levels of NfL may reflect increased axonal vulnerability or other neurodegenerative and neurovascular comorbidities, and heightened baseline levels of GFAP may reflect increased blood‐brain barrier dysregulation, each potentially indicative of an increased predisposition for the development of ARIA. Serial measures of NfL and GFAP as biomarkers have shown promise after traumatic brain injury and been used prospectively throughout a sports season. ²³ , ²⁴ Similarly, axonal stress from early edema and blood‐brain barrier dysfunction contributing to ARIA‐E could be reflected by transiently increasing levels of NfL and GFAP, respectively. In clinical trials of aducanumab, lecanemab, and donanemab, GFAP consistently demonstrated a treatment‐related decrease, although there does not appear to be a strong effect with NfL. ²⁵ , ²⁶ , ²⁷ However, it is unknown whether changes in NfL and GFAP, either at baseline or during treatment, are associated with ARIA.

The University of California, San Francisco (UCSF) Alzheimer's Infusion Clinic (AIC) is well positioned to investigate these questions by leveraging a treatment clinic biobank in which plasma is collected at every infusion from patients receiving ATT. Further understanding of ARIA risk could allow for more optimal risk‐benefit evaluation prior to treatment and increased detection of ARIA throughout treatment. To investigate these possibilities, plasma NfL and GFAP were measured using clinically available assays at baseline and throughout treatment to determine whether changes were associated with the development of clinically significant ARIA.

2. METHODS

2.1. Participants

The biobanking study was approved by the UCSF Institutional Review Board, and patients provided written informed consent at the time of enrollment. All patients were clinically approved for ATT by an institutional adjudication committee aligned with criteria from clinical trials for treatment with a planned course of 18 months of lecanemab or donanemab administered by modified dosing titration. ²⁸ All patients underwent comprehensive neurological evaluation as part of their standard clinical protocol in the treatment clinic, in addition to biobanking prior to each scheduled infusion. Clinical disease severity was assessed using the Mini‐Mental State Examination (MMSE), Clinical Dementia Rating (CDR) scale, and Functional Activities Questionnaire (FAQ), according to registry requirements. All patients received either lecanemab or donanemab at their discretion. For this study, ARIA was defined as significant edema, including sulcal effusions, with at least 1 cm of change on fluid‐attenuated inversion recovery (FLAIR) sequence, with equivocal cases and single microhemorrhages excluded from analysis. All ARIA‐negative cases lacked changes on MRI attributed to ARIA over the course of treatment, with all monitoring MRI scans performed in accordance with appropriate use guidelines and following society‐recommended protocols. ⁴ , ⁵ , ²⁹ All clinical MRI scans were reviewed by the study team, which included cognitive/behavioral neurologists and neuroradiologists experienced in ARIA detection. Renal impairment was defined as glomerular filtration rate (GFR) ≤ 60 at the available timepoint nearest the first infusion.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed prior studies on BBMs in AD, focusing on NfL, GFAP, and ARIA. Plasma NfL and GFAP are established markers of neuronal injury and astrocytic activation and have been evaluated in diagnostic and prognostic contexts. However, no studies have examined their utility for predicting or detecting ARIA in real‐world treatment settings, and evidence regarding longitudinal trajectories around ARIA remains limited.

2. Interpretation: In this real‐world treatment cohort, plasma NfL and GFAP at baseline showed no difference in subjects that later developed ARIA or not, did not change prior to ARIA, and did not have differing longitudinal trajectories. These findings suggest that these biomarkers are unlikely to be clinically robust tools for predicting or detecting ARIA, though larger cohorts may be needed to reveal modest differences.

3. Future directions: Larger, multicenter studies and evaluation of alternative biomarkers, including vascular and neuroinflammatory markers, are needed to improve ARIA risk stratification and monitoring.

2.2. Sample collection and processing

Blood was collected immediately prior to infusion by venipuncture into ethylenediaminetetraacetic acid (EDTA) tubes, centrifuged, and aliquoted into polypropylene tubes within an hour of collection, and plasma was stored at −80°C until analysis with no freeze/thaw cycles. Concentrations of NfL were measured by platforms based on electrochemiluminescence (ECLIA, Roche) and single‐molecule array (Simoa, Quanterix) in collaboration with Labcorp and Lucent Diagnostics. GFAP concentrations were measured by ECLIA only. Replicate testing and head‐to‐head platform comparisons were not part of the study design. Biomarker concentrations were transformed using a natural logarithm.

2.3. Statistical analyzes

Clinical characteristics, including BBMs, were compared at baseline using t‐test, U test, or chi‐squared as appropriate, with p < 0.05 considered statistically significant. Baseline BBMs in patients with and without ARIA were compared with Mann‐Whitney U tests, and change from baseline to timepoint prior to ARIA (falling within 30 days of ARIA) was evaluated with Wilcoxon signed‐rank tests. To evaluate longitudinal biomarker changes, linear mixed‐effects models were fit to multiply imputed trajectory data generated with predictive mean matching, controlling for age, sex, estimated GFR (eGFR), and baseline disease severity (CDR box score), allowing for random intercepts, with ARIA by time interaction the primary predictor of interest. All statistical analyses were conducted in R (version 4.5.2).

3. RESULTS

3.1. Clinical characteristics

Baseline demographics and clinical features are shown for patients with (N = 15) and without ARIA (N = 45) in Table 1. APOE genotype differed between cohorts, with ε4 carriers, especially homozygotes, at higher risk for ARIA, consistent with prior reports. ³⁰ No other baseline characteristic reached statistical significance in this small dataset, but comparatively fewer patients were on antihypertensive medications in the ARIA cohort (33% vs 42%) despite similar mean arterial pressures (MAP). Patients who developed ARIA tended to have more CMBs at baseline than those without ARIA (0.73 ± 1.39 vs 0.40 ± 0.91), and two out of 15 patients with ARIA had superficial siderosis before initiating treatment, compared with none in the ARIA negative group (data not shown). Although prior reports showed that ARIA risk varied based on amyloid plaque burden, no clear differences were seen in baseline amyloid PET Centiloid or plasma phosphorylated tau at threonine 217 (p‐tau217) in this small cohort of patients in early stages of the disease. ¹ , ³¹ Finally, whereas clinical trial results have suggested a higher rate of ARIA with donanemab than lecanemab, similar percentages of ARIA were observed between ATTs, potentially due to the adoption of modified titration dosing. ²⁸ , ³⁰

TABLE 1.

Baseline clinical characteristics.

	ARIA+ (N = 15)	ARIA− (N = 45)	Total (N = 60)
Age, years (mean ± SD)	72.2 ± 5.9	73.8 ± 7.3	73.4 ± 6.9
Sex, female (N, %)	10 (67)	28 (49)	38 (63)
Education, years (mean ± SD)	18 ± 2	17 ± 2	17 ± 2
Race/ethnicity (N, %)
Asian or Pacific Islander	–	2 (4)	2 (3)
Black or African American	1 (7)	1 (2)	2 (3)
Hispanic	–	3 (7)	3 (5)
Non‐Hispanic White	13 (87)	38 (84)	51 (85)
Other or unknown	1 (7)	1 (2)	2 (3)
MCI (N, %)	12 (80)	33 (73)	45 (75)
Mild dementia (N, %)	3 (20)	12 (27)	15 (25)
MMSE (mean ± SD)	26 ± 2	25 ± 3	26 ± 3
CDR, global (mean ± SD)	0.6 ± 0.2	0.6 ± 0.2	0.6 ± 0.2
CDR, box score (mean ± SD)	2.5 ± 1.1	2.7 ± 1.4	2.6 ± 1.3
FAQ (mean ± SD)	8 ± 5	9 ± 6	9 ± 6
BMI, kg/m2 (mean ± SD)	23.4 ± 2.7	25.1 ± 3.8	24.6 ± 3.6
MAP, mmHg (mean ± SD)	95 ± 13	94 ± 12	94 ± 12
HTN medicine (N, %)	5 (33)	19 (42)	24 (40)
APOE genotype, (N, %) *
ε2/ε3	–	1 (2)	1 (2)
ε3/ε3	3 (20)	25 (56)	28 (47)
ε3/ε4	6 (40)	16 (36)	22 (37)
ε4/ε4	6 (40)	3 (7)	9 (15)
Fazekas (mean ± SD)	1.1 ± 0.5	1.1 ± 0.7	1.1 ± 0.6
Cerebral microbleeds (mean ± SD)	0.73 ± 1.39	0.40 ± 0.91	0.48 ± 1.05
Amyloid PET, Centiloids (mean ± SD)	84 ± 37	83 ± 30	83 ± 32
P‐tau217 (Lumipulse), pg/mL (mean ± SD)	0.67 ± 0.40	0.59 ± 0.35	0.61 ± 0.36
P‐tau217 (Simoa), pg/mL (mean ± SD)	0.12 ± 0.05	0.11 ± 0.06	0.11 ± 0.05
NfL (ECLIA), pg/mL (mean ± SD)	3.72 ± 1.49	3.76 ± 1.42	3.75 ± 1.42
NfL (Simoa), pg/mL (mean ± SD)	20.5 ± 11.8	21.9 ± 8.9	21.6 ± 9.4
GFAP (ECLIA), pg/mL (mean ± SD)	152 ± 68	146 ± 69	148 ± 68
Lecanemab (N, % total)	9 (25)	27 (75)	36 (100)
Donanemab (N, % total)	6 (25)	18 (75)	24 (100)

Abbreviations: APOE, apolipoprotein E; BMI, body mass index; CDR, Clinical Dementia Rating scale; ECLIA, electrochemiluminescence immunoassay; FAQ, Functional Activities Questionnaire; GFAP, glial fibrillary acidic protein; HTN, hypertension; MAP, mean arterial pressure; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; NfL neurofilament light chain; PET, positron emission tomography, Simoa, single‐molecule array.

^* p < 0.05.

3.2. ARIA cases

Clinical characteristics are reported for N = 15 patients who developed ARIA (Table 2). Even excluding equivocal FLAIR change and single CMBs, ARIA was frequently asymptomatic (47%), and ARIA‐E typically resolved, whereas ARIA‐H stabilized, though more than one follow‐up scan was required in 67% (10/15) of cases. ARIA typically occurred within the first 6 months of treatment and recurred in 58% (7/12) of patients in whom treatment was resumed after resolution (no recurrences of ARIA‐E were graded higher in severity). The most common symptoms were headache and mild confusion. Two cases were severe and symptomatic (though mild), and intravenous methylprednisolone was promptly initiated upon detection. Clinical symptoms improved with treatment, completely resolving in one patient, but ATT was discontinued permanently in both. Interestingly, both severe symptomatic cases were ε3/ε3, and retrospective chart review identified no known risk factors.

TABLE 2.

Case characteristics of ARIA.

Tx​	Age/sex	Stage	APOE	Dose	ARIA‐E severity	ARIA‐H  severity​	Location​	Symptoms​	Scans to resolution	Outcome
Lec​	71 M	Mild dem​	ε3/ε4​	6	Mild​	None​	R frontal	Confusion	1	Tx resumed, no recur
Lec​	76 F	Mild dem​	ε3/ε4​	6	Mild​	None​	L frontal	Headache	3	Tx resumed, no recur
Lec​	62 F	MCI​	ε3/ε4​	4	Mild​	Mild​	L occipital and temporal	Headache​	1	Tx resumed, recur ×2
Lec​	72 M	MCI​	ε4/ε4​	6	Moderate​	Mild​	R frontal and L occipital	None​	3	Tx resumed, recur ×1
Lec​	70 F	Mild dem​	ε3/ε4​	6	Moderate​	Mild​	L > R occipital	None​	4	Tx resumed, recur ×2 (Tx discontinued)
Lec​	67 F	MCI​	ε4/ε4​	13	Moderate​	Mild​	L occipital and BL frontal	None​	1	Tx resumed, no recur
Lec​	70 F	MCI​	ε4/ε4​	6	Moderate​	Mild​	R parietal (multifocal)	None​	3	Tx resumed, no recur
Lec​	70 M	MCI​	ε3/ε4​	6	Moderate​	Moderate​	L occipital and R frontal (multifocal)	None​	1	Tx resumed, recur ×1
Lec​	68 M	MCI​	ε3/ε3​	2	Severe​	Severe​	Multifocal diffuse	Headache, confusion	5	Tx discontinued
Don	71 F	MCI​	ε4/ε4​	6	Mild​	Moderate​	L temporal and cerebellum	Headache​	1	Tx resumed, recur ×1 (Tx discontinued)
Don	73 F	MCI​	ε4/ε4​	3	Moderate​	Mild​	Multifocal diffuse	Headache, Confusion	2	Tx Resumed, recur ×1
Don​	67 F	MCI​	ε4/ε4​	2	Moderate​	Moderate​	Multifocal diffuse	Confusion, photopsia	2	Tx resumed, recur ×1
Don​	80 M	MCI​	ε3/ε4​	4	Moderate​	Moderate​	R > L frontal (multifocal)	None​	2 b	Tx discontinued
Don	87 F	MCI​	ε3/ε3​	3	Moderate​	None​	R occipital and R temporal	None​	2	Tx resumed, no recur
Don	75 F	MCI​	ε3/ε3​	1	Severe​	Severe​	Multifocal diffuse	Headache, confusion, imbalance	3+ a	Tx discontinued

Abbreviations: BL, bilateral; Don, donanemab; L, left; Lec, lecanemab; MCI, mild cognitive impairment; R, right; Recur, recurrence; Tx treatment.

^a Residual FLAIR change present on last available scan.

^b ARIA first detected on out‐of‐protocol MRI reviewed after patient received infusion.

3.3. Plasma NfL and GFAP do not predict or detect ARIA

Baseline NfL and GFAP levels immediately prior to ATT initiation were compared between patients who later developed ARIA to patients without ever any ARIA (Table 1, Figure 1A–C). Baseline NfL and GFAP did not differ between groups (ln[NfL] ECLIA 1.24 ± 0.38 pg/mL vs 1.25 ± 0.40 pg/mL, p = 0.77 [Figure 1A]; ln[NfL] Simoa 2.87 ± 0.47 pg/mL vs 2.98 ± 0.40 pg/mL, p = 0.30 [Figure 1B], ln[GFAP] ECLIA 4.95 ± 0.36 pg/mL vs 4.88 ± 0.47 pg/mL, p = 0.69 [Figure 1C]). To evaluate whether there was a detectable change in NfL or GFAP preceding ARIA, percentage change from baseline in natural log transformed NfL and GFAP levels at the closest timepoint to ARIA within 30 days prior to ARIA was evaluated. Using a 30‐day cutoff removed one patient from the analysis with a closest blood collection 78 days prior to ARIA, with other patients’ biomarkers collected 11.2 ± 8.7 days (median 11.5 days) prior to ARIA. No significant change was seen in either NfL or GFAP (Figure 1D). Patients with symptomatic ARIA did not have significantly different baseline NfL or GFAP levels or significant change in baseline prior to ARIA compared to patients with asymptomatic ARIA (Table S1). ARIA severity did not affect concentrations of NfL or GFAP, though the presence of renal impairment likely increased concentrations, as reported previously. ³² , ³³

FIGURE 1.

Plasma NfL and GFAP do not predict or detect ARIA. Concentrations of plasma biomarkers are shown for each patient color‐coded by severity (renal impairment denoted with a plus), including (A) baseline ln(NfL) by ECLIA, (B) baseline ln(NfL) by Simoa, (C) baseline ln(GFAP) by ECLIA, and (D) change from baseline at the timepoint immediately prior to ARIA. No differences were statistically significant.

3.4. ARIA does not alter plasma NfL and GFAP trajectories during treatment

To visualize if a pattern of change in NfL or GFAP occurred in relation to incidence of ARIA, spaghetti plots of natural log‐transformed NfL and GFAP values were generated with t = 0 defined by time of diagnosis of ARIA (Figure 2A,C,E). Similar to prior analysis, no pattern of change was observed in NfL or GFAP. Of note, the patient with severe ARIA with NfL increase after detection had worsening eGFR from 62 to 48 mL/min/1.73 m², suggesting fluctuation may have been secondary to decreased renal clearance, as this change was not observed in other patients with ARIA. Additionally, another patient had an elevation in NfL at a single timepoint directly preceding ARIA in only one of the assays (Simoa), which could represent a measurement error. To evaluate if NfL and GFAP trajectories were impacted by ARIA, linear mixed‐effects models were constructed, controlling for age, sex, eGFR, and clinical severity, with no significant change found in either biomarker over the course of treatment (Figure 2B,D,E). Sensitivity analyzes looking at impact of ARIA on disease progression also did not show differences in this small dataset.

FIGURE 2.

ARIA does not impact plasma NfL or GFAP either immediately preceding ARIA or over the course of treatment. Spaghetti plots for each patient with ARIA color‐coded by severity (renal impairment denoted with a plus) and model‐predicted concentrations over time for (A and B) ln(NfL) by ECLIA, (C and D) ln(NfL) by Simoa, and (E and F) ln(GFAP) by ECLIA. Asterisk (*) denotes a change in eGFR associated with worsening chronic kidney disease.

3.5. Case vignette: severe symptomatic ARIA does not impact NfL or GFAP concentrations

A 68‐year‐old, APOE ε3/ε3 male with mild cognitive impairment (MCI) due to AD and no CMBs or superficial siderosis on pre‐treatment MRI developed severe symptomatic ARIA after two infusions of lecanemab associated with progressive encephalopathy, mild aphasia, and low‐grade headache. Urgent MRI revealed multifocal, diffuse bilateral cortical T2‐hyperintensities, numerous microhemorrhages, and contrast extravasation, consistent with severe ARIA‐E and severe ARIA‐H (Figure 3A). Baseline NfL and GFAP were not elevated compared to controls or other cases with ARIA (Figure 2), and from the blood draw immediately prior to the second infusion, absolute NfL and GFAP were only mildly increased from pre‐treatment (13% and 18%, respectively), within the range of inter‐test variability and not predictive of the severe ARIA that developed days after the infusion. He was treated with methylprednisolone 1 g for 5 days followed by a 1‐month oral steroid taper (one other patient was treated with a similar course of steroids after severe ARIA). Lecanemab was permanently discontinued, and ARIA‐E resolved and ARIA‐H stabilized in tandem with resolution of symptoms over 6 months (Figure 3B). Retrospectively, no clear ARIA risk factors were identified, though review of his baseline amyloid PET demonstrated relatively high posterior occipital signal (80 Centiloids, Figure 3C).

FIGURE 3.

Neither plasma NfL nor GFAP changes prior to severe symptomatic ARIA. (A) MRI during ARIA event showing diffuse cortical FLAIR hyperintense changes (left), CMBs on SWI (center), and contrast extravasation (right). (B) FLAIR resolution after IV methylprednisolone, with fluctuating subtle residual occipital FLAIR signal (red arrow) through 7 months (mo = months). (C) Color‐scale amyloid PET (florbetapir) demonstrating predominantly posterior occipital signal.

4. DISCUSSION

Overall, no discernible differences were found in plasma NfL or GFAP either at baseline or immediately prior to the development of ARIA in this cohort, including subsequent trajectories over the course of treatment. Of note, ARIA cases in this study were enriched for patients with unambiguous changes and single CMBs were excluded. The lack of appreciable increase in NfL suggests no detectable neurological injury occurs during ARIA, with the possible exception of severe symptomatic ARIA, whereas the stability of GFAP suggests that astrocytic activation is not a component of ARIA pathobiology. These findings are consistent with the understanding that T2 hyperintensity changes observed in ARIA likely reflect transient vascular changes, mediated in part by microglial activity by either direct clearance of amyloid from vessel walls or secondary inflammatory from vascular deposition of amyloid:antibody complexes. ² , ³⁴ Altogether, given the lack of substantial change either at baseline or immediately prior to ARIA in this study, plasma NfL and GFAP were not supported as robust biomarkers to predict or detect ARIA, though larger cohorts may be needed to reveal modest differences and these biomarkers may have other roles in monitoring treatment response.

The patients who developed severe ARIA despite lacking traditional risk factors reflect the urgent need for improved prediction and monitoring of ARIA prior to and throughout treatment. Aside from APOE genotype, no other baseline characteristics were significantly different between ARIA positive and negative groups. An observed trend of relatively fewer patients receiving anti‐hypertensives in the ARIA positive group is consistent with reports showing untreated hypertension may increase ARIA risk. ³¹ , ³⁵ Additionally, increased baseline CMBs and presence of siderosis in the ARIA‐positive group parallels associations of CAA burden with ARIA risk. ¹ Interestingly, one of the patients who developed severe ARIA without known risk factors had relatively high occipital amyloid PET signal, as described in the case vignette. CAA pathology, which is thought to share common mechanistic pathways with ARIA, has been shown to preferentially affect the occipital lobe. ³⁶ Multiple studies evaluating amyloid PET distribution suggested a relative predilection for occipital signal in patients with CAA rather than AD. ³⁷ , ³⁸ , ³⁹ Recently, retrospective evaluation of amyloid PET scans from patients in donanemab clinical trials identified an association with occipital amyloid PET elevation and ARIA, independent of APOE genotype. ⁴⁰ Additional evaluation of occipital amyloid PET patterns in real‐world settings will be valuable to determine if PET can be used in ARIA risk stratification, especially as accurate and sensitive biomarkers of CAA are a priority for the field due to poor sensitivity of current MRI‐based methods.

In addition to the evaluation of the clinical utility of NfL and GFAP in the treatment clinic, this study demonstrated the potential for treatment centers to leverage ATT clinics to establish biobanks that will help inform real‐world knowledge of biomarker utility for multiple contexts of use. Moreover, serial biobanking of plasma throughout treatment benchmarked to clinical data, including MRI scans and clinical diagnoses, may provide a powerful tool to evaluate longitudinal biomarker changes related to clinical outcomes and the biological changes that occur during amyloid removal in AD. Harmonized clinical protocols and national registries like the Alzheimer's Association's ALZ‐NET provide a platform by which biobanks could potentially be integrated, articulated, and scaled across the AD research landscape, providing an enduring resource for biomarker development and clinic‐based research.

Our study has limitations, notably the relatively few patients who developed significant ARIA, limiting the ability to detect smaller effect sizes. The lack of significance of NfL and GFAP in this study does not exclude the possibility that this study was underpowered to detect modest differences in biomarkers. While we analyzed “significant” ARIA cases with at least 1 cm of edema, blood may still have limited detection for the many cases of ARIA‐E, which are relatively localized or mild in the brain. Accordingly, larger multicenter real‐world studies would greatly facilitate the goal of finding better ways to predict and detect ARIA, particularly by providing for a more diverse population and allowing for further exploration of factors that may influence ARIA and biomarkers. We accounted for renal impairment present at treatment initiation, although we had limited ability to capture fluctuations and worsening renal function throughout treatment that may have affected biomarker levels in this real‐world population. Other factors that may contribute to biomarker elevations, such as stroke and neuropathy, were also not accounted for and may be important to consider in real‐world settings. ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ Additional factors that could affect biomarkers include therapeutic strategies for ARIA such as steroids. One patient had a single blood draw 1 month after steroid pulse while completing steroid taper without a stark effect on NfL or GFAP. Larger studies will be better equipped to analyze the effect of steroids and other treatment‐related effects on biomarkers.

Another limitation is that relatively few plasma samples were available in the period surrounding ARIA. As our sampling paradigm was centered around infusions and not ARIA events, sample timing may have been insensitive to changes in NfL or GFAP, which may mirror the transient nature of ARIA. Future collection paradigms can be designed to increase sample collection during events. A biomarker that substantially changes only after ARIA is detectable on MRI would have limited use for prognosis or monitoring but might help better understand ARIA pathobiology. An additional limitation is that ARIA detection is defined by the time of detection of MRI, which may be an insensitive tool, though it remains the gold standard at present. MRI was performed at timepoints following the label and appropriate use recommendations, but this approach may miss interval development of ARIA outside screening windows. As this study was designed to explore biomarker trajectories in a real‐world cohort with significant ARIA, formal analytic assay performance was not evaluated. Platform characteristics such as sensitivity and inter‐assay variability should be evaluated in larger, multi‐site studies, particularly where real‐world timing and sample handling may affect performance.

In conclusion, plasma NfL and GFAP were not shown to be robust biomarkers of ARIA in a single‐center academic treatment clinic cohort, but we have demonstrated that real‐world treatment biobanks can facilitate rapid evaluation of the clinical utility of BBMs in this setting. Benchmarking blood collection with amyloid PET and measures of disease severity will permit evaluation of whether other plasma biomarkers can predict or detect clinical or biological treatment responses, including amyloid clearance. For improved ARIA detection, ongoing work is focusing on evaluating other candidate biomarkers, including improved cerebrovascular imaging and certain biomarkers of neuroinflammatory changes. However, given the relatively low prevalence of ARIA, multisite collaborative studies are critical for assessing the potential of ARIA biomarkers, especially to enrich for severe cases, which may have the largest biological signal.

CONFLICT OF INTEREST STATEMENT

Dr. Schwartz was funded by a UCSF Resource Allocation Program (RAP) grant through the UCSF Alzheimer's Disease Research Center grant from the National Institutes of Health (NIH) (P30AG062422) and UCSF Behavioral Neurology Fellowship Program (2T32AG023481). Dr. Dage was funded by the Indiana University School of Medicine‐Purdue University TREAT‐AD Center (U54AG065181), the National Centralized Repository for Alzheimer's and Related Dementias (U24AG021886), and the Indiana Alzheimer's Disease Research Center (P30AG072976). Dr. Rojas has been funded by the John D. French Alzheimer's Foundation. Dr. VandeVrede was funded by grants from the NIH (K23AG073514), the American Academy of Neurology, the Alzheimer's Association, the Tau Consortium, and the Shenandoah Foundation. Other authors have no relevant disclosures. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All patients provided written informed consent.

Supporting information

Supporting Information

Supporting Information