Glial Fibrillary Acidic Protein in Serum is Increased in Alzheimer’s Disease and Correlates with Cognitive Impairment

INTRODUCTION

The measurement of the biomarkers Tau, pTau181, and Aβ₄₂ in cerebrospinal fluid (CSF) is well established in the differential diagnosis of Alzheimer’s disease (AD). Elevated Tau levels and reduced Aβ₄₂ levels are included in the diagnostic criteria [1] and are helpful to differentiate other types of dementia such as frontotemporal dementia (FTD) or Lewy body dementia from AD [2]. Established blood biomarkers for AD are not available so far but blood collection has some major advantages compared with CSF: it is more convenient for patients and clinicians, it is beneficial in follow-up studies also regarding patient’s withdrawal, and it is more cost effective than CSF regarding the time needed for sample collection.

Measurement of the classical AD biomarkers in blood is unsuitable since Aβ₄₂ is also expressed in peripheral tissues [3] and both Aβ₄₂ and tau levels in blood show low sensitivity and specificity [4, 5]. A substantial overlap of blood levels with non-AD patients is observed for other biomarker candidates such as neurofilament light chain (NfL) [6] which is why it can only be used for pre-selection of patients for subsequent CSF diagnostics. Assays measuring pathological forms of Aβ₄₂ in blood are promising [7], and a recent study showed a high correlation of Aβ peptides measured in blood by mass spectrometry with Aβ-burden in the brain [8], but robustness of these methods is unclear or the technology is not widely available.

Glial fibrillary acidic protein (GFAP) is a marker of astrogliosis and it is increased in AD brains [9] and CSF [10, 11], which is why it is a potential biomarker candidate for AD. Recent technological advances also enable the measurement of GFAP in serum and plasma. The aim of our study was to measure GFAP in serum and CSF of patients with AD and related neurodegenerative diseases to evaluate its diagnostic potential.

MATERIALS AND METHODS

Patients with AD, Parkinson’s disease without cognitive impairment (PD), PD dementia (PDD), dementia with Lewy bodies (DLB), and non-neurodegenerative controls were recruited at the Ulm University Hospital, Department of Neurology, and patients with behavioral variant frontotemporal dementia (bvFTD) were also recruited at LMU Munich, TU Munich, and University of Erlangen-Nuremberg. Patients or their legal representatives gave written informed consent to be included in this study. The study was approved by the Ethics Committees of the participating centers.

We used patients with a clear clinical diagnosis in this pilot study. AD patients fulfilled criteria for “probable or possible AD dementia with evidence of the AD pathophysiological process” (i.e., CSF Tau and Aβ₄₂) [12]. Patients with bvFTD were diagnosed according to Rascovsky et al. [13], diagnosis of PDD according to Emre et al. [14], DLB according to McKeith et al. [15], and PD according to Hughes et al. [16].

Cognitive impairment and severity of dementia was rated using the Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR). The results of the CDR are given as sum of boxes. Level of education was operationalized with the ISCED (International Standard Classification of Education) into seven levels (Organization for Economic Co-Operation and Development).

Serum and lumbar CSF was collected and stored at – 80°C within 2 h. GFAP in serum was measured with the Simoa technology using the Human GFAP Discovery Kit from Quanterix (Lexington, MA, USA). Intra-and inter-assay CV was <2% and <11%, respectively, and GFAP was stable in serum for at least five freeze/thaw cycles. CSF GFAP was measured with an ELISA from Biovendor (Brno, Czech Republic) and CSF Tau and CSF Aβ₄₂ were measured with ELISAs from Fujirebio Germany GmbH (Hanover, Germany).

Statistical analyses were performed with GraphPad Prism 5.00. Groups were compared by the Mann-Whitney test (2 groups) or Kruskal-Wallis test and Dunn’s post hoc test (>2 groups). Sex distribution was compared by the Chi-square test. Correlation analysis was performed with Spearman’s rank correlation coefficient and age-adjustment of the serum GFAP concentration was performed with a linear regression model. Diagnostic performance was tested using receiver operating characteristic (ROC) curve analysis and cut-offs were selected using the Youden index. A p-value < 0.05 was regarded significant.

RESULTS AND DISCUSSION

Patient characteristics are listed in Table 1. We observed a moderate correlation of serum GFAP but not CSF GFAP with age in the control patients (serum: r= 0.60, p = 0.0002; CSF: r= –0.003, p = 0.99). Therefore, the serum GFAP concentrations were age-adjusted for group comparisons. Sex distribution was significantly different between groups (p = 0.004) but we observed no difference of GFAP concentration (serum and CSF) between male and female patients in the control (serum: p = 0.53, CSF: p = 0.61) and AD (serum: p = 0.69, CSF: p = 0.15) groups.

The median serum GFAP level in AD was 376 pg/mL (interquartile range, IQR 294–537 pg/mL) and was significantly higher in comparison to controls (median 157 pg/mL, IQR 126–218 pg/mL, p < 0.001), to bvFTD patients (median 211 pg/mL, IQR 166–263 pg/mL, p < 0.001) as well as PD patients (median 186 pg/mL, IQR 115–225 pg/mL, p < 0.001, Fig. 1A). Evidence for increased GFAP in blood of AD patients also comes from another study with a small AD cohort (n = 4) [17] and from a study showing increased levels of GFAP autoantibodies [18], both in support of our observation. In addition, we could confirm the previously reported increased GFAP levels in CSF of AD patients [10, 11] compared with controls (AD: median 1396 pg/mL, IQR 1007–2842 pg/mL; controls: median 826 pg/mL, IQR 628–1041 pg/mL, p < 0.001, Fig. 1B) which is consistent with our observation in serum samples. CSF and serum concentration correlated weakly in controls (r= 0.39, p = 0.02) and AD patients (r= 0.40, p = 0.04), but there was no significant correlation in the other groups (PD: r= 0.30, p = 0.37; bvFTD: r= –0.13, p = 0.45; DLB/PDD: r= 0.19, p = 0.45).

OPEN IN VIEWER

Fig.1

Different profile of GFAP in serum and CSF of neurodegenerative diseases. GFAP was measured in (A) serum and (B) CSF of non-neurodegenerative controls (Con) and patients with Parkinson’s disease (PD), behavioral variant frontotemporal dementia (bvFTD), Alzheimer’s disease (AD), and Lewy body dementias (i.e., dementia with Lewy bodies, DLB and PD dementia, PDD). C) CSF-to-serum ratio of GFAP concentrations. D) Correlation analysis of serum GFAP concentration and the MMSE score (Spearman r= –0.42, p < 0.001). Boxes are median and interquartile range, whiskers are minimum and maximum. Groups were compared by Kruskal-Wallis test and Dunn’s post-hoc test.

Our data in living individuals further support an important role of astrogliosis in AD which is also evident from the more pronounced GFAP-immunoreactivity in postmortem AD brains [9] and from AD model systems [19]. Increased GFAP levels in CSF of PD (median 1244 pg/mL, IQR 1071–1550 pg/mL, not significant), bvFTD (median 1365 pg/mL, IQR 1041–2295 pg/mL, p < 0.001), and DLB/PDD patients (median 1381 pg/mL, IQR 1108–2566 pg/mL, p < 0.001) support astrogliosis also in these diseases. However, we observed a disagreement of CSF and serum GFAP levels in some of the investigated neurodegenerative diseases. Whereas GFAP in CSF is increased to a similar extent in all diseases investigated here (Fig. 1B), serum GFAP levels are only elevated in AD and DLB/PDD (Fig. 1A) also resulting in a different CSF-to-serum ratio of GFAP compared with the other neurodegenerative diseases (Fig. 1C). This could originate from the different brain regions affected in the diseases or the location of astrogliosis in different compartments. For instance, the vascular accumulation of Aβ, called cerebral amyloid angiopathy (CAA), is frequently observed in AD brains and Aβ plaques are surrounded by reactive astrocytes [19]. The proximity of reactive astrocytes to the microvasculature in AD could be an explanation for the strong increase of serum GFAP in AD but needs further confirmation. In agreement with this hypothesis, CAA is increased in DLB/PDD compared with PD as well [20]. We used different assays for the measurement of CSF and serum GFAP who might recognize different GFAP isoforms [21]. In this context, our results could indicate different types of astrogliosis between the neurodegenerative diseases. Further studies are needed to clarify different astrogliosis between diseases and also whether these differences are cause or effect of the various pathophysiologies.

ROC curve analysis of serum GFAP levels showed high sensitivity (77%, 95% CI: 59–89%) and high specificity (93%, 95% CI: 77–99%) for the discrimination between AD and controls (AUC 0.91) at a cut-off of 219 pg/mL (Fig. 2A). This is only slightly lower than the established CSF marker Tau (AUC 0.99, Fig. 2B) and even better than CSF Aβ₄₂ (AUC 0.87, Fig. 2C). In addition, serum GFAP discriminated AD from bvFTD patients with 89% (95% CI: 73–97%) sensitivity and 79% (95% CI: 59–92%) specificity (AUC 0.85) at a cut-off of 289pg/mL (Fig. 2A) which is in the desired range for a diagnostic biomarker (>85% sensitivity and specificity) according to the Task Force on Biological Markers in Psychiatry [22]. The value might reach the 85% threshold in a larger patient cohort with a more exact ROC curve. In addition, serum and CSF GFAP showed a significantly negative correlation with the MMSE score (r= –0.42, p < 0.001 and r= –0.24, p < 0.05), a measure of cognitive impairment, and splitting of the AD group at the median MMSE showed higher serum GFAP levels in patients with a low MMSE (MMSE < 23: median 423 pg/mL, IQR 344–604 pg/mL; MMSE ≥23: median 317 pg/mL, IQR 227–438 pg/mL, p = 0.03) which is consistent with the observation of Fukuyama et al. in CSF [11]. We observed no correlation of serum and CSF GFAP with the CDR (r= 0.18, p = 0.08 and r= 0.16, p = 0.13).

OPEN IN VIEWER

Fig.2

Receiver operating characteristic (ROC) curve analysis of GFAP, Tau, and Aβ₄₂. ROC curve analysis for (A,D) serum GFAP, (B) CSF Tau, and (C) CSF Aβ₄₂ in non-neurodegenerative controls (Con) and patients with Parkinson’s disease (PD), behavioral variant frontotemporal dementia (bvFTD), Alzheimer’s disease (AD), and Lewy body dementias (i.e., dementia with Lewy bodies, DLB and PD dementia, PDD). AUC, area under the curve; CI, 95% confidence interval.

The diagnostic performance of serum GFAP in AD patients in our study could be of high clinical relevance. About 10–30% of FTD patients are misdiagnosed and rather suffer from AD [23] and new biomarkers might help to increase accuracy of clinical diagnosis especially in non-specialized centers. Blood-based biomarkers for AD are highly appreciated due to their convenience for patients and clinicians and their cost effectiveness compared with CSF analysis or PET [22]. Other candidate biomarkers for AD in blood such as Tau are unsuitable as diagnostic markers due to a substantial overlap with controls [5]. Promising results have been reported for the diagnostic performance of blood NfL to distinguish AD from controls [6] but blood NfL is also increased in bvFTD [24]. Thus, NfL in blood might be used to identify patients but subsequent CSF analysis is necessary for differential diagnosis. Serum GFAP had a higher diagnostic power in our study compared with previous studies on Tau (AUC 0.91 versus 0.78) and NfL (AUC 0.91 versus 0.87) [6, 25]. It might be the first blood biomarker to differentiate AD and bvFTD without additional CSF analysis. Yet, this proposal needs further confirmation in larger cohorts.

Patients with Lewy body dementias, DLB and PDD, also showed significantly increased GFAP in serum (median 343 pg/mL, IQR 234–539 pg/mL) compared with controls (p < 0.001), bvFTD (p < 0.01) and PD (p < 0.01), but showed no difference to AD patients (Fig. 1A). Data on GFAP in CSF of non-AD dementias are sparse but in agreement with our findings (Fig. 1B); also Ishiki and colleagues observed increased CSF GFAP levels in DLB [26]. In contrast to our study, they observed higher CSF GFAP in bvFTD compared with DLB and AD which likely originates in a few bvFTD patients with very high GFAP levels. The absolute GFAP levels measured by Ishiki et al. were slightly higher compared with our study although they used the same ELISA. This could be due to preanalytical differences or Lot-to-Lot variations.

ROC curve analysis in our study yielded an AUC of 0.87 for DLB/PDD versus controls (sensitivity 100%, 95% CI: 82–100%; specificity 59%, 95% CI: 41–75%; cut-off 168 pg/mL), AUC of 0.79 for DLB/PDD versus bvFTD (sensitivity 68%, 95% CI: 44–89%; specificity 89%, 95% CI: 73–97%; cut-off 291 pg/mL) and an AUC of 0.88 for DLB/PDD versus PD (sensitivity 100%, 95% CI: 72–100%; specificity 74%, 95% CI: 49–91%; cut-off 269 pg/mL) (Fig. 2D). Thus, serum GFAP might also help in the differential diagnosis of bvFTD and DLB/PDD, being of additional advantage over blood NfL which is increased in both conditions [24–27]. Serum GFAP also showed good discriminatory power for PD and DLB/PDD. This might be helpful for stratification of patients in clinical trials or as a follow-up marker. Longitudinal studies are necessary to clarify whether higher serum GFAP levels in PD patients are predictive of conversion to PDD which would be of high clinical relevance regarding individual treatment and drug discovery.

In this pilot study, we used a small-to-medium sized cohort of patients to investigate differences and the diagnostic potential of serum and CSF GFAP. Larger studies are needed to confirm the results and calculate more precise values for sensitivity, specificity and cut-offs.

In conclusion, our data provide evidence that GFAP in serum has the diagnostic power to be used as a CSF-independent marker to discriminate AD patients from controls and especially from bvFTD patients. The correlation of serum GFAP with cognitive impairment could be helpful as a follow-up marker in clinical trials. Assays for GFAP measurement are commercially available. This offers rapid implementation into clinical routine of the first blood biomarker for the differential diagnosis between AD and bvFTD.