Plasma glial fibrillary acidic protein in the visual and language variants of Alzheimer's disease

Abstract

INTRODUCTION

Glial fibrillary acidic protein (GFAP) in plasma is a proxy for astrocytic activity and is elevated in amyloid‐β (Aβ)‐positive individuals, making GFAP a potential blood‐based biomarker for Alzheimer's disease (AD).

METHODS

We assessed plasma GFAP in 72 Aβ‐positive participants diagnosed with the visual or language variant of AD who underwent Aβ‐ and tau‐PET. Fifty‐nine participants had follow‐up imaging. Linear regression was applied on GFAP and imaging quantities.

RESULTS

GFAP did not correlate with Aβ‐ or tau‐PET cross‐sectionally. There was a limited positive correlation between GFAP and rates of tau accumulation, particularly in the language variant of AD, although associations were weaker after removing one outlier patient with the highest GFAP level.

DISCUSSION

Among Aβ‐positive AD participants with atypical presentations, plasma GFAP did not correlate with levels of AD pathology on PET, suggesting that the associations between GFAP and AD pathology might plateau during the advanced phase of the disease.

1. BACKGROUND

Neuroinflammatory and immunological processes are increasingly being recognized as contributors to the pathogenesis of Alzheimer's disease (AD), ¹ although their role in the disease process is still debated and there is doubt on whether they cause or are caused by neurodegeneration. ² Neuroinflammation is characterized by the activation of both microglia and astrocytes. Reactive astrocytes overexpress proteins that are released extracellularly and can be measured in the blood plasma or cerebrospinal fluid (CSF). The glial fibrillary acidic protein (GFAP) is one of these proteins which is expressed in the cytoskeleton of astrocytes and provides a marker for astrocytic activation.

Autopsy studies have found that the concentration of GFAP is higher in areas surrounding amyloid‐β (Aβ) plaques. ³ Clinical studies have found that GFAP is elevated in Aβ‐positive individuals relative to Aβ‐negative. ⁴ , ⁵ , ⁶ Therefore, GFAP has the potential to be an inexpensive, non‐invasive, and widely available, but non‐specific to AD pathology, fluid biomarker for early diagnosis of AD, with the potential to track anti‐Aβ treatment effects. ⁷ , ⁸ The relationship between plasma GFAP and Aβ in AD appears to be related to disease stage. Positive associations between GFAP and Aβ observed in early disease stages ⁶ , ⁹ , ¹⁰ seem to plateau and then reverse as the disease progresses, eventually becoming negative, suggesting a dynamic relationship between these two biomarkers. ¹⁰ , ¹¹ In fact, Aβ‐PET burden and clinical disease stage, quantified with the Clinical Dementia Rating Score‐Sum of Boxes (CDR‐SB), ¹² demonstrated an interaction effect on plasma GFAP, with a positive association between Aβ and GFAP in individuals with CDR‐SB = 0 and a negative association in individuals with CDR‐SB > 4. ¹⁰ Similarly, in autosomal dominant variants of AD, astrocytosis on PET increased until individuals reached Aβ positivity, and then steadily declined while Aβ kept increasing; on the other hand, in sporadic AD, the decline of astrocytosis was not noted. ¹³

Astrocytosis might have a key role in triggering tau phosphorylation in the presence of Aβ plaques. In fact, among cognitively unimpaired individuals, Aβ was associated with increased plasma phosphorylated tau (p‐tau) only in those individuals with increased astrocyte reactivity, measured as elevated plasma GFAP levels. ¹⁴ Compared to individuals with lower plasma GFAP levels, those with higher plasma GFAP levels also had higher rates of tau‐PET accumulation, which were predicted by baseline Aβ burden. ¹⁴ However, inquiries on the associations between GFAP levels and tau pathology in AD have also led to conflicting findings. A positive correlation between tau‐PET or p‐tau and GFAP was either absent, ¹⁵ partially mediated by Aβ ⁴ or disappeared after adjusting for Aβ. ⁶ , ⁹ In contrast, another study reported a positive correlation between plasma GFAP and post mortem tau tangles, even after covarying for the amount of Aβ plaques, while the correlation between GFAP and plaques did not survive when covarying for the amount of tangles. ¹⁶

To our knowledge, GFAP levels in atypical non‐amnestic variants of AD have not yet been investigated. Atypical AD variants include predominant visual, language, executive, behavioral, or motor dysfunction. ¹⁷ While Aβ distribution does not differ between AD variants, in atypical phenotypes tau pathology occurs more extensively in the cortex, with relative sparing of the medial temporal lobe. ¹⁸ Additionally, atypical AD variants tend to have an earlier disease onset and a more aggressive disease course compared to amnestic AD. ¹⁷

We aimed to investigate the relationship between plasma GFAP and cross‐sectional and longitudinal Aβ‐PET and tau‐PET among individuals diagnosed with the visual or language AD variants. We hypothesized that plasma GFAP and Aβ‐PET would have a weak negative relationship ¹⁰ or no relationship, ⁹ and that there would be no differences in plasma GFAP values and associations with amyloid or tau between the two variants. We also hypothesized that plasma GFAP would be associated with faster rates of tau accumulation over 1 year. ¹⁴

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature on associations between glial fibrillary acidic protein (GFAP) and amyloid‐β (Aβ) and tau in Alzheimer's disease (AD). GFAP is higher in Aβ‐positive individuals and correlates with Aβ and tau in amnestic AD.

2. Interpretation: The absence of correlation between GFAP and Aβ and tau cross‐sectionally in the language and visual variants of AD suggests a possible plateau in these associations. Higher GFAP levels mildly correlate with faster tau accumulation.

3. Future directions: Future studies will include a larger sample of participants and more atypical AD variants at various disease stages.

2. METHODS

2.1. Participants

Seventy‐two participants were recruited from the Neurodegenerative Research Group (NRG) between 2016 and 2021 and underwent blood collection. Forty met clinical criteria for the visual variant of AD or posterior cortical atrophy (PCA) ¹⁹ and 32 for the language variant of AD or logopenic progressive aphasia (LPA). ²⁰ Participants underwent structural MRI, [¹¹C]Pittsburgh Compound B PET for Aβ, and [¹⁸F]flortaucipir PET for tau, and plasma GFAP levels were measured. The Aβ‐PET scans were analyzed to determine Aβ positivity, ²¹ and all participants were Aβ‐positive at baseline, except for one, who became positive at the follow‐up visit. Apolipoprotein E (APOE) genotyping was performed on all participants. Fifty‐nine participants (29 LPA, 30 PCA) underwent PET and structural MRI at 1‐year follow‐up. The study was approved by the Mayo Clinic Institutional Review Board (IRB), and all participants provided written informed consent to participate in this study. All participants underwent a clinical and neuropsychological evaluation, with details previously reported. ²² To quantify cerebrovascular pathology, white matter hyperintensities (WMH) were segmented and manually edited on the two‐dimensional fluid‐attenuated inversion recovery (FLAIR) images using a semi‐automated method. ²³ FLAIR images were available for 46 participants (21 LPA, 25 PCA).

2.2. Neuroimaging biomarkers

Details of neuroimaging acquisition and pre‐processing were previously reported. ²² Briefly, all PET scans were acquired using PET/CT scanners (GE Healthcare, Milwaukee, Wisconsin or Siemens Healthcare, Erlangen, Germany) operating in 3D mode. All participants also underwent a 3T head MRI protocol performed on GE or Siemens scanners with identical protocols. PET standard uptake value ratios (SUVRs) were calculated in voxels segmented as gray or white matter relative to the cerebellar crus gray matter. SUVR images were spatially normalized to the MCALT template and blurred with a 6‐mm full width at half maximum kernel, for voxel‐based analyses with SPM12. Tau‐PET SUVR was calculated in a meta‐region of interest (ROI) specific for atypical Alzheimer's disease (aty‐AD), including parietal, temporal, and occipital regions of the MCALT ADIR122 atlas. Tau‐PET SUVR were also calculated in each bilateral lobe of the MCALT lobar atlas (frontal, parietal, temporal, including the medial temporal structures, and occipital). SUVR values were calculated on images without partial volume correction (PVC). Tau‐PET SUVR annualized rate of change was calculated as the percent difference between follow‐up and baseline values divided by year difference. Additionally, regional SUVR were computed using a different, specialized pipeline to measure change over time in tau‐PET. ²⁴

2.3. Plasma biomarkers

Plasma GFAP level was measured on all participants. Venous blood, including EDTA plasma, was collected, centrifuged, and stored in freezers in the Biospecimens Accessioning and Processing (BAP) laboratory at Mayo Clinic in Rochester, MN. Frozen plasma samples were sent to the University of Minnesota. Samples were thawed at room temperature and centrifuged at 10,000 × g for 5 min prior to dilution with assay specific sample diluent. Measurements of GFAP were performed using the Simoa HD‐X analyzer (Quanterix). ²⁵ , ²⁶ GFAP levels were analyzed using Neurology 4‐plex assay E kit. All analyses were performed in duplicate, and the average value was used in the statistical analyses. All sample concentrations had coefficients of variance (CV) < 10%. The mean ± standard deviation CV% for all included samples was 3.03% ± 2.02%.

2.4. Statistical analyses

Baseline characteristics were compared between LPA and PCA with Fisher's exact test for categorial variables and Wilcoxon rank sum test for continuous variables.

2.4.1. Linear regression models

Linear regression models were run to investigate the association between plasma GFAP and age, disease duration, WMH volume, neurological and neuropsychological scores (Table 1), cross‐sectional global Aβ‐PET SUVR, aty‐AD meta‐ROI tau‐PET SUVR, lobar tau‐PET SUVR, SUVR percent change. Covariates included age, sex, and also baseline regional tau‐PET SUVR and global Aβ‐PET SUVR when testing longitudinal SUVR change. Analyses were run on: full sample; LPA participants; PCA participants; participants with CDR‐SB > 4 (n = 4 LPA, n = 15 PCA); participants with CDR‐SB < = 4 (n = 26 LPA, n = 23 PCA). ¹⁰ Results are reported as β estimates ± standard errors, with p values. GFAP and SUVR values were log‐transformed in the models. Statistical significance was set at p < 0.05. Analyses were run in R version 4.2.2.

TABLE 1.

Demographic, clinical, and biomarker characteristics of all participants.

	All aty‐AD (n = 72)	PCA (n = 40)	LPA (n = 32)	p‐Value PCA vs. LPA
Demographics
Male, n (%)	23 (32%)	12 (30%)	11 (34%)	0.8
Education, year	16 (3)	16 (3)	16 (2)	0.4
Age at scan, year	64 (7)	61 (6)	68 (7)	0.005
Disease duration, year	3.5 (2.5)	4.3 (2.5)	2.9 (2.4)	0.04
APOE ε4 carriers, n (%)	31 (43%)	19 (48%)	12 (38%)	0.5
Neuroimaging biomarkers
Aty‐AD tau‐PET SUVR	2.2 (0.5)	2.4 (0.5)	2.1 (0.4)	0.01
Global Aβ‐PET SUVR	2.4 (0.4)	2.5 (0.3)	2.4 (0.5)	0.4
Plasma biomarkers
GFAP, pg/mL	266 (110)	265 (107)	273 (115)	0.9
Vascular pathology
WMH/TIV%	0.68 (0.9)	0.75 (0.9)	0.61 (1.0)	0.09
Neurological and neuropsychological tests
MoCA	18 (7)	18 (6)	19 (7)	0.8
CDR‐SB	2.5 (3.5)	3.0 (3.7)	1.5 (3.1)	0.01
BNT	11.0 (3.7)	12.0 (3.1)	9.5 (4.2)	0.05
BDAE repetition	7.0 (2.3)	8.0 (2.1)	7.0 (2.4)	0.01
VOSP cubes	4.0 (4.1)	1.0 (2.5)	9.0 (3.1)	<0.001
VOSP letters	16.0 (7.0)	9.0 (6.9)	19.0 (1.9)	<0.001

Note: Data are shown as n (%) or median (standard deviation). For continuous variables, p‐values are from Wilcoxon rank sum test. For categorical variables, p‐values are from Fisher's exact test. p‐values < 0.05 are in bold. WMH was available for 46 participants. CDR‐SB was missing in 2 PCA and 2 LPA participants.

Abbreviations: Aβ, amyloid‐β; APOE, apolipoprotein; Aty‐AD, atypical Alzheimer's disease; BDAE, Boston diagnostic aphasia examination; BNT, 15‐item Boston Naming Test; CDR‐SB, Clinical Dementia Rating Scale ‐ Sum of Boxes; GFAP, glial fibrillary acid protein; LPA, logopenic progressive aphasia; MoCA, Montreal cognitive assessment; PCA, posterior cortical atrophy; SUVR, standard uptake value ratio; TIV, total intracranial volume; VOSP, visual object and space perception.; WMH, white matter hyperintensity.

2.4.2. Voxel‐based analyses

SPM12 one‐sample t‐tests were performed to evaluate the effect of plasma GFAP on cross‐sectional and annualized percent rates of change of Aβ‐PET and tau‐PET SUVR images, covarying for age and sex. Baseline global Aβ‐PET and aty‐AD tau‐PET SUVR were included as covariates in the analyses of longitudinal SUVR change. Statistical significance was set at p < 0.001.

3. RESULTS

3.1. Participants

PCA participants were younger than LPA (p = 0.005), had longer disease duration (p = 0.04), higher aty‐AD tau‐PET SUVR (p = 0.01), and differed from LPA on most tests (Table 1). Median plasma GFAP levels were similar between PCA and LPA (median 265 vs 273 pg/mL; p = 0.9) (Table 1).

3.2. Biomarkers analyses

Plasma GFAP was not associated with age, WMH volume nor any neurological or neuropsychological scores, but there was a trend between higher GFAP and longer disease duration (0.03 ± 0.02, p = 0.06). In linear regression models adjusted for age and sex, plasma GFAP levels were not associated with Aβ‐PET SUVR (Figure 1A), aty‐AD meta‐ROI tau‐PET SUVR (Figure 1B), or lobar tau‐PET SUVR in the full sample or any of the investigated subsets. In the full sample, a 10% increase in global Aβ‐PET SUVR was associated with a 2% ± 3% (p = 0.5) decrease in GFAP, while a 10% increase in Aty‐AD tau‐PET SUVR was associated with a 0.2% ± 3% (p = 0.9) increase in GFAP. In the CDR‐SB > 4 sub‐group, a 10% increase in global Aβ‐PET SUVR was associated with a 3% ± 5% (p = 0.5) reduction in GFAP (Figure 1A). Individuals with higher rates of change of global Aβ‐PET SUVR did not have significantly higher baseline GFAP; in the full sample, a 5‐percentage point faster annual increase in global Aβ‐PET was associated with a 2% ± 4% (p = 0.7) lower baseline GFAP. Individuals with faster rates of increase in Aty‐AD tau‐PET SUVR did not have significantly higher baseline GFAP; in the full sample, a 5‐percentage point faster rate of SUVR increase was associated with a 4% ± 6% (p = 0.5) higher GFAP (Figure 2). Not covarying for age did change the associations. Among LPA participants, the effect of GFAP on rates of tau accumulation was significant in the aty‐AD meta‐ROI, while covarying for age, sex, and baseline tau and Aβ pathology (Figure 2): a 5‐percentage point increase in the annualized percentage change in tau‐PET was associated with about a 30% increase in GFAP (0.28 ± 0.10; p < 0.01). However, the association was not significant (0.18 ± 0.11; p = 0.12) after removing the LPA participant with the highest GFAP value of 658 pg/mL and the highest tau‐PET SUVR change of 15% in the aty‐AD meta‐ROI (female, age = 53, disease duration = 9.7 years, CDR‐SB = 0.5, WHM/TIV % = 0.54). Similar positive associations between GFAP and tau rates of change were found in the left parietal (p = 0.04), left temporal (p < 0.01), and left occipital (p = 0.02) lobes in LPA. Removing the covariates did not change this association.

FIGURE 1.

Scatter plots depicting the associations between plasma glial fibrillary acidic protein (GFAP) levels and cross‐sectional global amyloid‐β positron emission tomography standard uptake value ratio (Aβ‐PET SUVR) (A), and cross‐sectional tau‐PET SUVR the atypical Alzheimer's disease (aty‐AD) meta‐region of interest (ROI) (B) in the full sample, posterior cortical atrophy (PCA) participants, logopenic progressive aphasia (LPA) participants, participants with Clinical Dementia Rating Scale‐Sum of Boxes (CDR‐SB) score above 4, participants with CDR‐SB score below or equal to 4. The solid line indicates the regression line, and the gray band indicates the 95% confidence intervals. GFAP and SUVR data are plotted on a log scale.

FIGURE 2.

Scatter plots depicting the associations between plasma glial fibrillary acidic protein (GFAP) levels and annualized percent rate of change in the atypical Alzheimer's disease (aty‐AD) meta‐region of interest (ROI) tau positron emission tomography (PET) standard uptake value ratio ( SUVR) in the full sample, posterior cortical atrophy (PCA) participants, logopenic progressive aphasia (LPA) participants, participants with Clinical Dementia Rating Scale‐Sum of Boxes (CDR‐SB) score above 4, participants with CDR‐SB score below or equal to 4. The solid line indicates the regression line, and the gray band indicates the 95% confidence intervals. GFAP is plotted on a log scale.

3.3. Voxel‐based analyses

At the voxel‐level, there were no associations between plasma GFAP and cross‐sectional Aβ‐PET, longitudinal Aβ‐PET, and cross‐sectional tau‐PET. GFAP was mildly positively correlated with annualized percent change in tau‐PET SUVR in the left temporoparietal cortex, driven by LPA participants, and by participants with CDR‐SB < = 4, at p < 0.001 without multiple comparisons correction (Figure 3A). PCA participants exhibited some positive correlations between GFAP and rates of change in tau‐PET in frontal regions, while LPA exhibited some negative correlations in the frontal regions (Figure 3A). When the LPA patient with the highest GFAP was removed, the same trends were noticed at p < 0.01 (Figure 3B), but no voxels survived at p < 0.001.

FIGURE 3.

Voxel‐based maps of the effect of glial fibrillary acidic protein (GFAP) on annualized tau positron emission tomography (PET) standard uptake value ratio (SUVR) percent rate of change, covarying for age, sex, baseline global amyloid‐β (Aβ)‐PET SUVR and baseline atypical Alzheimer's disease (aty‐AD) meta‐region of interest (ROI) tau‐PET SUVR. Maps are shown for the full sample, posterior cortical atrophy (PCA) participants, logopenic progressive aphasia (LPA) participants, participants with Clinical Dementia Rating Scale‐Sum of Boxes (CDR‐SB) score above 4, participants with CDR‐SB score below or equal to 4. Maps are showed at p < 0.001 (A). Maps obtained removing the LPA (CDR‐SB = 0.5) patient with the highest GFAP value (653 pg/mL) are shown at p < 0.01 (B).

4. DISCUSSION

Among participants with the visual (i.e., PCA) and language (i.e, LPA) atypical AD variants, plasma GFAP did not vary systematically with Aβ‐ and tau‐PET burden cross‐sectionally nor with rates of change in Aβ longitudinally. However, plasma GFAP levels mildly correlated with rates of tau‐PET accumulation over 1 year.

The absence of a relationship between Aβ‐PET SUVR and plasma GFAP measures suggests that, although plasma GFAP is typically elevated in Aβ‐positive individuals, its levels do not closely track Aβ levels throughout the course of the disease. This warrants attention in the use of plasma GFAP as a biomarker for tracking AD progression. This finding is consistent with Pereira et al., ⁹ who reported a plateau in the association between GFAP and amyloid in cognitively impaired Aβ‐positive individuals, and with Chattertjee et al., ¹¹ who reported no associations between plasma GFAP and Aβ levels in symptomatic mutation carriers of autosomal dominant AD. Our findings are also somewhat in agreement with Asken et al. ¹⁰ who reported a negative association in individuals with CDR‐SB > 4; however, the negative relationship we found among these participants was not significant.

Plasma GFAP levels in the atypical AD participants were in the same range as previously reported for Aβ‐positive cognitively impaired individuals. ⁵ , ⁹ , ¹⁶ Plasma biomarkers, including GFAP, are more abnormal in early‐onset than in late‐onset AD relative to controls, ²⁷ and atypical AD participants more often experience an early disease onset. This aspect, coupled with the fact that Aβ‐PET rates of accumulation plateau and start decreasing at 2 SUVR reaching zero at baseline SUVR of 2.7, ²⁸ might explain the absence of association between GFAP and amyloid in our participants, with a median global Aβ‐PET SUVR of 2.4.

Unlike other studies, ⁶ , ⁹ we did not find a positive association between tau‐PET burden and plasma GFAP that disappeared after adjusting for Aβ. This discrepancy can most likely be attributed to the absence of Aβ‐negative controls in our study; it could also be attributed to the higher levels of cortical tau‐PET SUVR in atypical AD relative to amnestic AD. Plasma GFAP levels have been associated with a memory composite score but not an executive composite score, ¹⁵ suggesting a possible association with medial temporal lobe structures that subserve memory. The medial temporal lobe is often burdened with less tau pathology than the cortex in atypical AD, and this might explain the absence of associations between tau and GFAP in our participants.

The mild positive association between baseline plasma GFAP level and subsequent tau accumulation is consistent with a recent study reporting higher rates of tau‐PET accumulation in individuals with higher plasma GFAP values relative to those with lower GFAP values. ¹⁴ This association was stronger among LPA participants in the left temporoparietal cortex. However, even though limited to a few voxels in the frontal regions, the positive correlation between rates of tau accumulation and GFAP existed also within PCA participants alone. The fact that the correlation between GFAP and tau accumulation was more pronounced in LPA is not surprising, since we recently demonstrated that in this cohort, LPA accumulate tau faster than PCA, ²² making it easier to detect associations. However, we must stress that this correlation was somewhat driven by one LPA participant with the highest GFAP level and the highest levels of tau accumulation. Removing this patient significantly weakened the correlation: in fact, LPA participants also exhibited a negative association between tau accumulation rates and GFAP in frontal regions.

The main limitation of our study is the absence of amyloid‐negative controls to provide a comparison to atypical AD GFAP levels. Nevertheless, the goal of our study was to investigate GFAP associations to AD pathology in atypical AD. Limitations also include having only two atypical AD clinical phenotypes and inadequate racial and ethnic diversity.

In conclusion, plasma GFAP was not associated with cross‐sectional Aβ‐ and tau‐PET burden in the visual and language variants of AD. A mild positive correlation between GFAP and tau rates of accumulation was noted. However, more studies with participants at different disease stages and with different atypical AD variants are needed to validate our results. Our findings have implications for the choice of optimal blood‐based biomarkers in trials for Alzheimer's disease modifying therapies.

CONFLICT OF INTEREST STATEMENT

I.S., N.A.S., and D.L. have no disclosures to report. J.L.W., K.A.J., J.G.R., M.M.M., C.G.S., and C.R.J. reported receiving research funding from the NIH. J.G.R. is DSMB for NINDS STROKENET. M.M.M. (Dr. Mielke) has served on scientific advisory boards and/or has consulted for Biogen, LabCorp, Lilly, Merck, Siemens Healthineers, and Sunbird Bio and receives grant support from the NIH and Department of Defense. M.L.S. reported holding stock in Gilead Sciences, Inc., Inovio Pharmaceuticals, Medtronic, Oncothyreon, Inc., and PAREXEL International. V.J.L. reported consulting for Bayer Schering Pharma, Piramal Life Sciences, Life Molecular Imaging, Eisai Inc., AVID Radiopharmaceuticals, and Merck Research and receiving research support from GE Healthcare, Siemens Molecular Imaging, AVID Radiopharmaceuticals and the NIH (NIA, NCI). Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants gave written informed consent to participate in this study.

Supporting information

Supporting Information

Sintini I, Singh NA, Li D, et al. Plasma glial fibrillary acidic protein in the visual and language variants of Alzheimer's disease. Alzheimer's Dement. 2024;20:3679–3686. 10.1002/alz.13713