Promising Diagnostic Accuracy of Plasma GFAP and NfL Within The AD Continuum

Background: Blood-based biomarkers may add a great benet in detecting the earliest neuropathological changes in patients with Alzheimer’s disease (AD). We examined the utility of neurolament light chain (NfL) and glial brillary acidic protein (GFAP) in plasma and cerebrospinal uid (CSF) regarding clinical diagnosis and amyloid positivity in an outpatient memory clinic - based cohort. Methods: In this retrospective analysis, we included a total of 185 patients, 141 patients along clinical the AD continuum, i.e. subjective cognitive decline (SCD, n=18), mild cognitive impairment (MCI, n=63), AD (n=60) and 44 age-matched healthy controls (HC). CSF and plasma concentrations of NfL and GFAP were measured with single molecule array (SIMOAâ) technology using the Neurology 2-Plex B kit from Quanterix. Amyloid-PET was performed in 75 patients and graded as amyloid positive and negative by visual rating. To assess the discriminatory potential of different biomarkers, age- and sex-adjusted receiver operating characteristic (ROC) curves were calculated and the area under the curve (AUC) of each model was compared using DeLong’s test for correlated AUC curves. Results: We constructed a panel combining plasma NfL and GFAP with known AD risk factors (age+sex+APOE4+GFAP+NfL panel). Using this panel, AUC was 91.6% for HC vs. AD, 81.7% for HC vs. MCI, 85% for SCD vs. AD, 81.3% for SCD vs. MCI, 77.7% for HC vs. SCD and 72.3% for MCI vs. AD. In terms of predicting amyloid PET status, we computed an AUC of 88.4%. Conclusion: The combination of plasma GFAP and NfL with well-established risk factors could contribute crucially to thereby

the neurodegenerative process 6,7 . Considering that positron emission tomography (PET) imaging is a cost-intensive diagnostic biomarker and lumbar puncture is an invasive examination, recent studies have been looking for the possibility to identify reliable uid biomarkers by conventional blood analysis. The establishment of new and sensitive analytical methods may facilitate this approach. In comparison to already established enzyme-linked immunosorbent assay (ELISA), the usage of ultrasensitive single molecule array (SIMOA→) has proven a superior value for several molecules 8 .
Neuro lament light chain (NfL), a subunit of speci c cytoskeletal proteins of neurons, represents a highly proposed biomarker regarding detection of neuronal loss. NfL is released through axonal damage into the cerebrospinal uid (CSF) and eventually into the blood 9 . Accumulating data have shown that plasma NfL could be used as a non-invasive biomarker for neurodegeneration, which correlates well with cognitive decline and brain atrophy [10][11][12] . Studies have found higher plasma NfL levels in patients with MCI and AD in comparison to cognitively unimpaired controls 13 , and estimated a beginning increase in plasma NfL levels about 10 years before AD diagnosis compared to those who remain dementia-free 14 . Another promising biomarker for tracking neurodegenerative changes could be glial brillary acidic protein (GFAP), an intermediate lament protein of astrocytes. Neuropathological data have shown a close spatial relationship between reactive astrocytes and amyloid plaques in brain tissue of patients with AD 15,16 . GFAP is known to be involved in the neuroin ammatory cascade of AD pathophysiology. High GFAP concentrations have been detected in the CSF of patients with dementia of various aetiologies [17][18][19] , and increased blood GFAP levels were also found in patients with AD in comparison to healthy controls 20,21 . Furthermore, an association between plasma GFAP levels and amyloid load in AD patients has been recently observed [20][21][22][23] , corroborating neuropathological ndings.
The aim of this study was to examine GFAP and NfL levels in CSF and plasma in various stages of the clinical AD continuum compared to healthy controls. Furthermore, we investigated the predictive value of these blood biomarkers in combination with well-established risk factors in relation to clinical diagnosis and amyloid positivity in an outpatient memory clinic setting, Methods Study population 185 patients were enrolled in this retrospective study at the Memory Clinic of the Department of Neurology, Medical University of Vienna. Using two existing registries, i.e. the Dementia Registry RDA MUV (EK 1323/2018) and the BIOBANK MUV (EK 2195/2016), we identi ed 141 patients with a diagnosis along the clinical spectrum of cognitive decline, i.e. SCD (n = 18), MCI (n = 63) and AD (n = 60).
Additionally, 44 age-matched healthy controls (HC) without signs of neurodegenerative disorder or cognitive decline were included.
All 141 patients with cognitive complaints (SCD, MCI, AD) underwent a thorough standardized diagnostic examination including physical and neurological evaluation, neuropsychological testing, magnetic resonance imaging (MRI) of the brain and basic laboratory testing. For a subset of patients, we extended our diagnosis with a biomarker-based approach. CSF analysis of established AD biomarkers were available (e.g. Ab42, tTau, pTau) in 75 patients, amyloid-PET imaging was performed in 75 patients, and 54 patients underwent both diagnostic methods.
Diagnosis of AD and MCI were based on the recommendation of the National Institute of Ageing and Alzheimer's Association (NIA-AA) 4,24 . The diagnosis of SCD was applied when no abnormalities on cognitive tests were observed and diagnosis criteria for MCI, AD or other major neurological or psychiatric disorders were not met 25,26 .
All 185 study participants were required to have a plasma EDTA sample stored in the Biobank MUV, for 103 study participants CSF samples were available as well.

Neuropsychological assessment
The Neuropsychological Test Battery Vienna (NTBV) was administered to assess cognitive function, including domains of attention, language, executive functioning and episodic memory and is described elsewhere 27,28 . Adequate normative data from cognitively unimpaired individuals were available and zscores for each variable were calculated, corrected for age, education and sex. Global cognition was computed by Mini Mental State Examination (MMSE) and Wortschatztest (WST), a standardized vocabulary test providing an estimate of premorbid IQ. Depressive symptoms were measured via Beck Depression Inventory (BDI-II) 29 .

MR Imaging
All patients underwent at least a T1-weighted MR sequence, a T2-weighted or a Fluid-attenuated inversion recovery (FLAIR) MR sequence and a diffusion-weighted MR sequence within the routine diagnostic setting for the evaluation of the extent and pattern of atrophy, the presence and degree of vascular lesions and to exclude other underlying pathologies causing cognitive decline (e.g. normal pressure hydrocephalus, subdural hematomas or brain tumors) and diffusion restricted areas (representing acute ischemia, in ammation or signal changes indicating Creutzfeldt-Jakob disease). where the tracer accumulation in the brain is reaching the maximum. Using the GE Advances PET scanner, a 3D ordered subset expectation maximization (OSEM) ltered backprojection (FBP) reconstruction was done (128x128 matrix, Hanning ler, cut-off 6.2mm). For [18F] utemetamol, using the Siemens Biograph 64 True Point the reconstruction was performed using a 3D OSEM with 4 iterations and 21 subsets into a 168x168 matrix with a Zoom of 2 (2 x 2 mm pixel size), and with a 5 mm full width at half-maximum (FWHM) Hann post reconstruction lter applied to the nal images. Subsequently, the image acquisition was performed for about 20 min following a computed tomography (CT) acquisition for attenuation correction using Siemens Biograph 64 True Point.

APOE genotyping
Genomic DNA was extracted from peripheral blood leucocytes of individuals with available whole blood samples (n = 159) using standard DNA isolation methods. Apolipoprotein E (APOE) genotyping was performed using quantitative polymerase chain reaction (qPCR) with TaqMan probes (Thermo sher) evaluating two single nucleotide polymorphisms (SNPs) in the APOE gene (rs429358 and rs7412). Each sample was tested for both SNPs in triplicates using 20 ng DNA. Allelic discrimination analysis was used to determine the APOE genotype of the study participants.
CSF biomarkers CSF was obtained by lumbar puncture between the L3/L4, L4/5 or L5/S1 intervertebral space, collected in polypropylene tubes and further stored at − 20°C until biomarker analysis (as for Ab42, pTau 181 and tTau), or immediately at − 80°C for future research purposes 31,32 . Levels of Ab42, pTau 181 and tTau were measured with commercially available enzyme-linked immunosorbent assays (ELISA) (Innotest hTAU-Ag, Innotest phosphoTAU 181p, Innotest beta-amyloid 1-42) 33,34 . Plasma biomarkers EDTA plasma was collected through venepuncture and stored at -80°C in our local biobank. Concentrations of NfL and GFAP were quanti ed with ultrasensitive single molecule array (SIMOA) using the Neurology 2-Plex B kit from Quanterix in CSF and plasma. Detailed analyses are described elsewhere 35 . In short, equilibrated calibrators, samples and controls were diluted (1:4 for plasma and 1:100 for CSF) and incubated with detector and paramagnetic reagents provided by the manufacturer. Streptavidin ß-galactosidase was added to each well before samples were transferred to the Quanterix SR-X analyser for measurement of protein levels. All samples were analysed as duplicates and all assay materials were obtained from the same kit lot. Intra-assay CV was < 12% for GFAP and < 8% for NfL. Interassay CV for two samples measured repeatedly on 10 plates was well acceptable (< 13% for GFAP and 9% for NfL).
Statistical Analysis between groups was performed using chi-square test, the Mann-Whitney-U-test or the Kruskal-Wallis test. Correlation was assessed using Spearman's rank correlation coe cient. To evaluate the discriminatory performance of the biomarkers assessed herein, the cohort was split into pairs of two diagnosis (e.g. AD and HC) and the response variable was coded as existing for the more severe diagnosis (i.e. SCD when assessing SCD vs HC). Next, a baseline model consisting of sex, age and APOE4 status (increased risk for carriers of the APOE4 allele) was constructed using logistic regression. A receiver operating characteristic (ROC) curve was plotted and the area under the curve (AUC) was measured. Optimal cutoffs were calculated using Youden's J-Statistic 36 and sensitivity and speci city are reported as percentage. The baseline model was then supplemented by either levels of plasma GFAP, plasma NfL, or both and the AUC of each model was compared using DeLong's test for correlated AUC curves 37    Plasma GFAP showed a gradual increase along the four cohorts, with the lowest concentration in HC (median 79pg/ml, IQR 53.7, 120.6) and the highest in patients with AD (median 181.9pg/ml, IQR 129.6, 269.6) ( Table 2). Besides signi cantly higher levels of plasma GFAP in patients with MCI and AD compared to healthy controls (Fig. 1, p < 0.001), we observed a signi cant difference between patients with SCD and AD (p < 0.01) and patients with SCD and MCI (p < 0.05). The difference between HC and SCD, as well as between MCI and AD missed statistical signi cance (p = 0.092 and p = 0.098, respectively).
Plasma NfL performed similar to GFAP regarding the difference in concentrations between HC and MCI/AD (p < 0.001, Table 2 and Fig. 1). However, NfL levels showed a signi cant difference between HC and SCD (p < 0.01) as well as between MCI and AD (p < 0.05), but not between SCD and MCI/AD. Using Spearman correlation coe cient, the correlation of NfL and GFAP in CSF and plasma were analysed ( Fig. 2a and 2b). Correlation between NfL in CSF and plasma was calculated with R = 0.64 (p < 0.001, Fig. 2a) and GFAP in CSF and plasma with R = 0.4 (p < 0.001, Fig. 2b).
To assess the clinical utility of GFAP and NfL in plasma, particularly in distinguishing healthy controls from patients with cognitive complaints (e.g. SCD, MCI and AD) and potentially predicting cerebral amyloid status as measured by amyloid-PET imaging, ROC analyses were performed and adjusted for sex and age. We constructed a diagnostic panel, consisting of well -established risk factors such as age, sex (de ned as female > male), and APOE4 carriership (de ned as carrying at least one copy of the APOE4 allele) (i.e. age + sex + APOE4 panel) and compared it with a panel of age, sex, APOE4 carriership added by plasma NfL and plasma GFAP, called age + sex + APOE4 + GFAP + NfL panel (Fig. 3a-g). Additionally, we analysed in this model each biomarker separately to assess the potential bene t of GFAP or NfL alone (i.e. age + sex + APOE4 + GFAP panel and age + sex + APOE4 + NfL panel).

Discussion
In this outpatient memory clinic-based study we examined the performance of two promising biomarkers of neurodegeneration and neuroin ammation, e.g. NfL and GFAP, for the diagnostic work-up of patients along the continuum of AD-related cognitive decline. We aimed to develop a practical and reproducible model for a quick and accurate patient diagnosis in an easy-to-handle way for an outpatient memory clinic setting. As age, sex and APOE4 status are the most well-known risk factors for AD 39,40 and are already part of an established diagnostic work -up in many memory clinic settings, we contemplated an extended risk model that included these risk factors in addition to plasma GFAP and NfL levels (age + sex + APOE4 + GFAP + NfL panel). To assess the best performance of each single biomarker, we furthermore analysed the impact of adding GFAP (age + sex + APOE4 + GFAP panel) or NfL alone (age + sex + APOE4 + NfL panel).
The combined age + sex + APOE4 + GFAP + NfL panel could differentiate healthy controls from patients with MCI or AD and signi cantly outperformed the age + sex + APOE4 panel alone. For discriminating the disease states of MCI and AD, the age + sex + APOE4 + GFAP + NfL panel could not add a signi cant bene t to the age + sex + APOE4 panel.
We further investigated the potential of the age + sex + APOE4 + GFAP + NfL panel in regard of discriminating patients with SCD from the other cohorts and found promising results, especially in differentiating SCD from patients with an objective cognitive decline (MCI and AD), representing therefore a potential useful complementary diagnostic tool in an outpatient setting. In terms of predicting amyloid positivity on PET-imaging the age + sex + APOE4 + GFAP + NfL panel performed signi cantly better than the age + sex + APOE4 panel alone and showed a good distinction between Ab + and Ab-patients. When analysing the performance of a combination of each biomarker alone, the age + sex + APOE4 + GFAP panel outperformed the age + sex + APOE4 + NfL panel in all examined patient cohorts, except for MCI vs.

AD.
Focusing on plasma GFAP alone, its levels showed a gradual increase along the four cohorts, with the lowest concentration in HC and the highest in patients with AD, thereby allowing a good biological interpretation of a gradual rise of this biomarker along the progressing neuropathological process.
Plasma GFAP alone achieved the most prominent discrimination between HC and patients with an objective cognitive decline (MCI and AD). Moreover, the SCD group could be discriminated signi cantly from MCI and AD, however, MCI and AD as well as HC and SCD could not be differentiated by the use of plasma GFAP alone.
Plasma NfL alone could best discriminate between HC and MCI/AD patients and showed also promising demarcation between the HC and SCD group and even more interesting between the MCI and AD group.
Intriguingly, we found the highest concentration of plasma NfL in patients with SCD, which might not be solely explained by the small sample size. Therefore, these results must be interpreted cautiously.
Diagnosis of early phases of AD is crucial in regard of detecting patients at-risk as early as possible in the development of the neuropathological cascade. Besides the abnormal aggregation of Ab peptide and tau protein, neuroin ammation and neurodegeneration represent major components in the pathophysiology of AD 4 . Due to the recent evolvement of more sensitive analytical methods, non-invasive blood-based biomarkers could serve as a reliable approximation to this early pathogenic process. In recent years, the role of neuroin ammation in the pathogenesis of AD has been increasingly focused on in the literature. Neuropathological data have shown a close spatial relationship between Ab plaques and reactive astrocytes, which along with microglia, may trigger a pro-in ammatory cascade and eventually lead to neurodegeneration, which in turn activates astrocytes and microglia 41,42 . As a cytoskeletal component of astrocytes, GFAP could serve as a promising biomarker re ecting astrocytic activation and proliferation during the neurodegenerative processes, including AD, particularly in its earliest stages 43,44 . On the other hand, NfL represents a rather unspeci c biomarker for neurodegeneration, as it is released by axonal damage in multiple neurological disorders 45,46 . While the importance of NfL as a blood-based biomarker has been already reported in several studies 12,13,47,48 , the signi cance of GFAP is currently still evolving. To our knowledge, only few studies have evaluated the combination of GFAP with other biomarkers so far, and presented the utility of plasma GFAP not just in discriminating healthy controls from patients with AD but also in distinguishing Ab + from Ab-individuals [21][22][23] . Furthermore, higher GFAP levels have been associated with an increased risk for future progression to dementia and a steeper cognitive decline 43,49 . Our data encourage the application of a blood-based biomarker model for an outpatient memory clinic setting, with even more promising results for GFAP than NfL for the discrimination of early stages along the AD continuum, whereas NfL might be more helpful in discriminating later stages. This could be in line with data, in which NfL has been reported to be rather a marker for progression of disease, whereas GFAP could be superior in identifying patients at-risk for developing dementia 43 . Our data may support this hypothesis. Regarding the heterogeneity of AD pathology, a panel of well-combined blood-based biomarkers could aid in the early detection as well as disease monitoring in the future. The failure to show a discriminatory superiority of the age + sex + APOE4 + GFAP + NfL panel in regard of differentiation of MCI and AD, while showing a slightly better performance of plasma NfL in this matter, might further underline this concept.
While recent studies increasingly focus on the evaluation of biomarkers in blood, especially in terms of feasibility, CSF biomarkers in AD have partially faded from the spotlight. In our cohort, we found the highest concentration of CSF GFAP in patients with AD followed by the HC group and the lowest concentration in MCI.
CSF NfL demonstrated a gradual increase over the three cohorts with the lowest levels in HC and the highest in AD and concentration of NfL in CSF and plasma correlated well with each other, which is in line with already published data 9,12,50−52 , suggesting that plasma levels might be considered as an acceptable proxy of CSF levels.
We found the lowest concentration of CSF GFAP in patients with MCI, followed by the HC and AD group, therefore results must be interpreted cautiously. In contrast to NfL, levels of GFAP in CSF and plasma showed a lower correlation, which was already described by another study 21 . While other studies analysed levels of GFAP with ELISA 18,19,21,53 , we found no comparable results for measurement of CSF GFAP with SIMOA→ technology in patients along the AD continuum. Thus, further investigations are needed to better determine the role of CSF GFAP and its correlation with GFAP levels in blood in these patient cohorts.
Nevertheless, both measurements-CSF GFAP and CSF NfL-allowed a good discrimination between HC and AD as well as MCI and AD, with better results for CSF NfL.

Limitations
A limitation of our study is its retrospective and cross-sectional nature. Therefore, we lack longitudinal follow-up results and miss data in some of the patients. Positive amyloid-PET and APOE4 carriership were signi cantly less common in the MCI group, which might lead to the notion, that at least some of the MCI patients are not along the AD continuum. Neuropathological data could not be attained in our in vivo patient cohort. The fact that the highest concentration of plasma NfL was found in patients with SCD could be interpreted as a sign of inclusion of a too diverse population, caused by the known heterogeneity of this diagnostic entity. Still, since we focused on patients with putative AD pathology, we excluded patients with other causes of dementia or high vascular burden, which led to a smaller sample size.
Moreover, healthy controls were signi cantly younger than the patient groups, which might in uence results of these biomarkers and corresponding analysis. To counteract this potential bias in our data, ROC analyses were adjusted for sex and age.

Conclusion
Blood-based biomarkers for AD may represent a valuable complementary tool for the clinical diagnosis and patient management in the near future. As to the increasing incidence of patients suffering from neurodegenerative disorders, developing minimally invasive and affordable biomarkers might confer great bene t in a quick and accurate diagnostic work -up. The outstanding bene ts of blood-based biomarkers are the good availability and the potential of repetitive analysis in an individual patient without major efforts. In a time, where clinical trials are increasingly focusing on early stages of AD, it is crucial to develop a reliable diagnostic method that can be used as an easy-to-apply screening tool. We suggest that plasma GFAP could aid in a better distinction of patients along different predementia stages and that the combination of GFAP and NfL plasma levels with conventional risk factors could serve as a good "at-risk" model for early patient selection and assessment of progression to AD. Regarding the heterogeneity of AD pathology, the implementation of a panel of well combined blood-based biomarkers could serve as a valuable adjunct during the diagnostic process in an outpatient memory clinic setting.