Introduction

In 2018, the FDA authorized the use of a blood test for glial fibrillary acidic protein (GFAP) and ubiquitin carboxy-terminal hydrolase L1 (UCH-L1) in mild traumatic brain injury (mTBI), crowning a long success story of CNS-driven blood biomarker development1,2,3. Initial efforts to identify fluid biomarkers for neurological diseases focused on the cerebrospinal fluid (CSF) as, compared with blood, CSF is closer to the brain extracellular space and contains higher concentrations of CNS-derived proteins4. The establishment of fourth-generation immune assays in the last decade3,5 brought the possibility of quickly obtaining rapid and robust protein biomarker measurements from blood samples, opening up new perspectives in the field of CNS-derived markers. For example, levels of classic CSF biomarkers of neuroaxonal damage, such as neurofilament light chain (NfL)5, phosphorylated tau 217 (ref.6), and UCH-L1 (ref.7) can now be readily quantified in blood, indicating that these markers hold potential for use in diagnosis and monitoring of disease activity, and as surrogate end points for treatment trials. The literature on the utility of blood GFAP as a biomarker is also growing, reinforcing the large body of published data on CSF GFAP3,8,9,10,11,12,13,14. The evaluation of blood levels of GFAP has the potential to enable the in vivo longitudinal evaluation of different aspects of the astrocytic response in several neurological disorders. Here, we provide an up-to-date review of the analytical aspects, current evidence, perspectives, and limitations of blood GFAP as a biomarker, with the purpose of outlining how to refine its application in the diagnosis and monitoring of neurological diseases.

GFAP biology and analysis

Astrocytes represent around 30–40% of the cells in the CNS15, form an integral part of the blood–brain barrier (BBB) and establish numerous interactions with other cells in the nervous system, including neurons. Astrocytes are central to the normal function of synapses and contribute to axonal metabolic maintenance through the regulation of ion homeostasis16 (Fig. 1). GFAP is the signature intermediate filament of astrocytes17. GFAP is a type-III intermediate filament and human GFAP comprises 432 amino acids, which are encoded by a gene on chromosome 17q21.1-q25. The filament is expressed in mature astrocytes in the grey and white matter, the cerebellum, the subventricular and subgranular zones, and Mueller cells in the retina18. GFAP is also expressed in the periphery by Schwann cells, mature glial cells in the gut, hepatic stellate cells, and other non-neural cells18,19. To date, evidence indicates that ten splice-isoforms of GFAP — α, β, δ, ζ, κ, ∆135, ∆164, ∆exon6 and ∆exon7 — are expressed in the nervous system20. The isoform that is most abundantly expressed and most often analysed in the literature is GFAPα21.

Fig. 1: Astrocytes have multiple physiological roles in the CNS.
figure 1

Astrocytic end-feet containing glial fibrillary acidic protein (brown circles) are an essential component of the blood–brain barrier and the glymphatic system157. Astrocytes are critical in maintaining axonal metabolic homeostasis158 and contribute to tripartite synapses159.

Full size image

Several sensitive enzyme-linked immunosorbent assays (ELISA), electrochemiluminescence (ECL), and fluorescence-based methods for the detection of GFAP in body fluids (for example, CSF, vitreous fluid and amniotic fluid) are commercially available3,22. However, detection of GFAP in the blood has historically been a challenge, as the available ELISA tests cannot reliably detect such low concentrations of GFAP. When high levels of GFAP in the CSF are observed (for example, in individuals with TBI23 or neuromyelitis optica exacerbation24), detection of GFAP in the blood is usually possible with ELISA3. The development of highly sensitive assays, such as single-molecule arrays (Simoa), has enabled the detection of GFAP in the blood of healthy individuals and individuals with different neurological diseases25. Furthermore, the detection of blood GFAP is now possible with a portable, point-of-care platform that can deliver results within 15 min26.

The mechanisms underlying drainage of GFAP and its breakdown products into the blood under pathological conditions seem to be complex and are a matter of continuing debate. Evidence indicates that drainage is likely to result from a combination of bulk flow into the blood via arachnoid villi, flow along the glymphatic system and the cervical lymph nodes, and continuous bidirectional fluid exchange at the barriers of the CNS (that is, the BBB and blood–CSF barrier)27,28,29. According to the available data, GFAP is stable in the blood (for at least five freeze–thaw cycles)30; however, a thorough characterization of pre-analytical confounders and an aggregation-related ‘hook effect’ remains to be completed. The hook effect is partially caused by the formation of protein aggregates that contribute to the extraordinary long-term stability of GFAP. These aggregates can last for millennia at ambient temperature, as exemplified by the Heslington brain31. The formation of pathological GFAP aggregates in vivo can accompany lethal neurological disorders such as Alexander disease2.

Acute CNS injury

Traumatic brain injury

TBI is a common cause of disability worldwide, mostly among young adults32. The current standard of care requires the prompt evaluation of TBI severity; however, this evaluation relies on physical (for example, the Glasgow Coma Scale (GCS)) and radiological (head CT) tools that have several limitations. For example, GCS scores cannot be used to assess severity of TBI in patients who are sedated for intubation. Moreover, head CT, which has long been the clinical standard for the radiographic detection of TBI in the emergency department, can result in unnecessary exposure to ionizing radiation if used indiscriminately, especially in young individuals with mTBI33. These limitations have led to the investigation of a range of astroglial and neuronal biomarkers, including S100β calcium-binding protein (S100B), GFAP and UCH-L1, with the aim of improving the accuracy of TBI diagnosis and the associated decision-making process34.

Diagnosis

The results of key studies of blood GFAP levels in TBI are summarized in Table 1. In a study in 584 participants with mild-to-moderate TBI, elevated levels of serum GFAP were detected within 1 h of injury, compared with levels in participants with non-TBI general trauma. GFAP levels in the group of participants with TBI peaked at 20 h after injury, and finally declined slowly until 72 h after injury35. Bazarian et al. performed a large multicentre observational study in more than 1,900 participants with mild-to-moderate TBI (the ALERT-TBI study), and found that a prespecified cut-off value of 22 pg/ml for serum GFAP, in addition to serum UCH-L1 levels above 327 pg/ml, was able to predict the presence of intracranial injuries on head CT with an area under the receiving operator characteristic curve (AUC) of 0.98 (ref.36). This finding contributed to the FDA authorization in 2018 to market the first blood-based test for the avoidance of unnecessary exposure to radiation from CT in individuals with suspected TBI1. In a more recent study, serum GFAP levels were used to discriminate participants with mTBI from age-matched participants without intracranial traumatic pathology (AUC 0.69). This study used a higher GFAP cut-off value (0.23 ng/ml) than the study by Bazarian et al.36, and involved concomitant assessment of serum S100B (AUC 0.84) and neurogranin (AUC 0.77)37. Another study compared the ability of serum and plasma GFAP to discriminate between participants with mTBI with and without acute abnormalities on head CT38. The AUC values were similar for serum (AUC 0.81) and plasma (AUC 0.79) GFAP although neither plasma nor serum levels were able to adequately predict functional outcomes at 1 week.

Table 1 Key studies of blood GFAP levels in traumatic brain injury
Full size table

Some studies have found that, in comparison with other serum biomarkers (for example, UCH-L1, S100B and NfL), GFAP is the best marker for discriminating individuals with TBI and abnormalities on head CT from individuals with TBI and normal head CT scans35,39,40,41. For example, the large prospective multicentre Collaborative European NeuroTrauma Effectiveness Research (CENTRE-TBI) study (with more than 2,800 participants) found an AUC of 0.89 for GFAP40; the second highest AUC (0.83) was for UCH-L1. Moreover, levels of GFAP scaled with clinical severity and care path intensity (emergency department < ward admission < intensive care unit) especially in individuals with mTBI40. In another example, blood GFAP levels were better than blood NfL levels at discriminating between participants with normal and abnormal head CT scans; GFAP had an AUC of 0.77, compared with 0.65 for NfL42. Similarly, blood GFAP outperformed blood S100B in discriminating between participants with and without lesions on CT, both at 0–8 h and at 12–32 h after injury (AUC 0.89 and 0.63 for GFAP and S100B, respectively, at 0–8 h; AUC 0.94 and 0.72 for GFAP and S100B, respectively, at 12–32 h)41.

Importantly, evidence suggests that blood GFAP levels are sensitive to subclinical intracranial pathologies that are not visible on head CT scans. Indeed, in the 18-centre Transforming Research and Clinical Knowledge in TBI (TRACK-TBI) study (study years 2014–2018), plasma GFAP was used to identify participants with MRI abnormalities from a prospective cohort of 450 participants with normal head CT scans after mTBI26. Plasma GFAP concentrations measured within 24 h of injury were significantly higher among the 120 participants with positive MRI scans than among the 330 participants with negative MRI scans, with an AUC of 0.78. In this cohort, participants with diffuse axonal injuries detected on MRI had higher levels of plasma GFAP than participants with other lesion types on MRI, suggesting that changes in GFAP levels are specific to some cellular pathologies. Moreover, participants with mTBI, and negative CT and MRI scans had higher median plasma GFAP concentrations than healthy control participants (74.0 pg/ml versus 8.0 pg/ml), suggesting the presence of subtle and/or microscopic glial damage not detectable with MRI26. These findings provide the impetus for future translational and clinical applications of GFAP to bridge the gap between molecular changes and the structural injury that is visible with neuroimaging tools. Furthermore, they suggest that GFAP could be used as a triage tool to obtain short-interval scans and much-needed follow-up care in patients with negative initial imaging results, as the subtle and/or subclinical effects of TBI are often missed in a field without formal guidelines for clinical follow-up.

In one study, blood GFAP levels were higher in 73 participants with acute orthopaedic trauma than in 93 participants with mTBI and a negative head CT scan (P = 0.026) on arrival; however, no differences between the two groups were observed during the following days43. In a large prospective study, initial (4 h after injury) blood GFAP levels were also able to discriminate participants (both children and adults) with body and head trauma with concussion from participants with non-concussive trauma with AUCs of 0.80 and 0.76, respectively; the highest GFAP levels were observed in participants with concussive head trauma44. An early increase in blood GFAP levels after head concussive trauma could reflect a mechanical disruption of the BBB that occurs in parallel with the observed transient and chronic GFAP over-expression following single and multiple concussions, respectively45,46. The temporal cascade of plasma GFAP following TBI was investigated in a small subset of 34 participants from the Citicoline Brain Injury Treatment Trial (COBRIT): plasma GFAP was maximal on day 0, remained elevated 30 and 90 days after TBI, and was excellent in discriminating participants with complicated (for example, CT-positive) mTBI from healthy control participants (AUC 0.94)47.

Age-related differences in the ability of GFAP to detect TBI should be considered when interpreting these findings. The ability of GFAP to identify participants with intracranial trauma on CT from a group of participants with mTBI declined with increasing age in a subset of 169 participants from the three-centre TRACK-TBI pilot cohort (2010–2014; AUC 0.73 in participants aged >60 years; AUC 0.93 in participants aged <40 years)48. The study also examined the ability of plasma levels of phosphorylated tau and total tau to identify participants with intracranial trauma on CT; however the AUC for these markers did not differ greatly according to age. These observations support evidence that other glial biomarkers (for example, S100B) have reduced specificity in older individuals with TBI compared with specificity in younger individuals with TBI49. This effect of age could result from incipient neurodegeneration, different anatomical locations and types of injury in older individuals, or differences in the sensitivities of the assays or imaging methods used48.

Prognosis

In one study, the prognostic utility of serum GFAP in mTBI was studied by recording return to work status and Glasgow Outcome Scale Extended (GOSE) scores at 6 months after injury50. Participants with incomplete return to work had higher GFAP levels at hospital admission than participants with complete return to work. However, in multivariate analysis, GFAP was not predictive of outcome determined by GOSE or complete return to work. The BIO-ProTECT study found an association between serum levels of GFAP and S100B and poor outcomes, as defined by GOSE scores at 1, 4 and 6 months after injury51. The prognostic capacity of a model containing participant variables (age, sex, GCS score) and CT score was consistently improved by the incorporation of biomarker data, which were available for 566 out of 882 participants51. In another study of 243 participants with moderate to severe TBI, the addition of blood GFAP, UCH-L1 and microtubule-associated protein 2 measurements to known clinical predictors (age, sex, GCS score) improved the prediction of a favourable outcome at 6 months compared with the known clinical predictors alone52. GFAP was the most promising of the blood markers: the AUC for GFAP and clinical predictors was 0.78 and the AUC for clinical predictors alone was 0.69. A comprehensive longitudinal assessment identified a biphasic profile of serum GFAP in participants with moderate and severe TBI. At enrolment, GFAP levels were elevated in participants with TBI compared with levels in control participants. This initial increase was followed by a reduction over the following 6 months after injury, but an increase over the subsequent years53. A study in 155 veterans with a mean age of 79 years also demonstrated the reliability of GFAP combined with other biomarkers (phosphorylated tau, NfL, IL-6 and TNFα) for the differentiation of participants with a past medical history of TBI (up to decades before blood sampling) and cognitive impairment from those with a history of previous TBI but no cognitive impairment (AUC 0.85)54. The same panel of biomarkers was able to differentiate participants with cognitive impairment and TBI from those with cognitive impairment and no TBI (AUC 0.88).

In summary, GFAP might represent a reliable proxy for small and diffuse structural damage that is not easily assessed with CT, and even MRI, and therefore, it could be employed as a surrogate marker of intracranial pathology. We postulate that adding measurement of blood GFAP (among other biomarkers) to the diagnostic process might provide a more accurate definition of ‘mild’, ‘moderate’, and ‘severe’ TBI than clinical classification alone, which is frequently hampered by the caveat of impaired consciousness. A point-of-care analysis platform for serum GFAP55 in ambulances might guide the triage of patients with TBI.

Traumatic spinal cord injury

A few studies have tested serum GFAP as a marker of the presence and severity of traumatic spinal cord injury (SCI) in patients who have had a traumatic injury56,57. GFAP levels seem to correlate with the severity of SCI, suggesting that the use of GFAP as a biomarker of neurological outcome (segmental motor recovery) is clinically feasible. Specifically, a biochemical model that included both CSF and serum S100B, GFAP and IL-8, measured 24 h after injury, was able to correctly predict the American Spinal Injury Association (ASIA) grade — an accurate predictor of neurological outcome — in a consistent proportion (89%) of 27 patients with SCI57. Remarkably, the combined evaluation of these three biomarkers outperformed the ASIA grade in predicting segmental motor recovery at 6 months. Following these results, another cohort study found significantly higher serum levels of GFAP in individuals with severe (ASIA grade A) SCI than in individuals with moderate SCI (grade B; P < 0.05), mild SCI (grade C; P < 0.01) or controls (individuals with vertebral fractures but without neurological symptoms; P < 0.01)56. In addition, individuals with SCI who died postoperatively had significantly higher serum GFAP levels in the first 24 h after injury than individuals with SCI who survived (P < 0.05)56. Last, following complex surgery of the thoracic aorta, serum GFAP levels were considerably higher in participants with SCI than in participants without SCI; however, these comparisons failed to reach statistical significance after adjusting for multiple testing, probably owing to the limited number (n = 3) of participants with SCI included in the study58. Although evidence is so far very limited, the measurement of serum GFAP levels could provide an avenue to determine the ‘biological’ severity of injury and predict neurological outcome in patients with SCI, thereby supporting clinical decision-making regarding the identification of patients who are likely to benefit from surgery.

Cerebrovascular accidents

Diagnosis

Biomarkers reflecting the underlying pathophysiological changes associated with cerebrovascular brain injury could improve the management and prognostic assessment of patients with acute stroke59. The results of previous studies suggest that serum GFAP could be employed as a biomarker of glial injury indicative of intracerebral haemorrhage in patients presenting with acute stroke symptoms60,61,62,63,64,65. As a result of sudden BBB disruption and subsequent brain injury, GFAP becomes rapidly detectable in blood during the hyperacute phase of intracerebral haemorrhage. Accordingly, studies have found serum levels of GFAP to be substantially higher in patients with intracerebral haemorrhage than in patients with ischaemic stroke60,65. In a multicentre cohort study, the analysis of plasma GFAP levels in 205 participants using electrochemiluminometric immunoassays within 4.5 h of symptom onset differentiated participants with intracerebral haemorrhage from participants with ischaemic stroke and stroke mimics66. Specifically, the use of a GFAP cut-off value of 0.29 µg/l enabled intracerebral haemorrhage from acute ischaemic stroke to be differentiated from stroke mimics with a sensitivity of 84.2% and a specificity of 96.3% (AUC 0.92). Interestingly, in the BE FAST II study, serum levels of GFAP obtained upon hospital admission were about 16 times higher in participants with intracerebral haemorrhage than in participants with acute ischaemic stroke. In the same study, participants with a large lobar intracerebral haemorrhage had a higher median serum GFAP concentration than participants with a small, deep intracerebral haemorrhage61.

Prognosis

A number of studies investigated the role of GFAP as a predictor of functional outcomes after acute ischaemic stroke67,68. In one such study, serum GFAP levels were measured using ELISA in 286 participants with ischaemic stroke on the first day of admission and participants were followed up for a year68. After adjusting for all the established predictors (for example, stroke severity and infarct volume), multivariate analysis showed that elevated GFAP levels on the first day of admission independently predicted poor functional outcomes during the 1-year follow-up. Furthermore, a robust body of evidence suggests that GFAP is a sensitive indicator of injury and a predictor of outcome in patients with subarachnoid haemorrhage. In one study, GFAP levels in 67 participants with subarachnoid haemorrhage were measured at hospital admission69. The mean GFAP serum concentration in these participants was 1.8-fold higher than the upper limit of the normal laboratory reference range. In addition, participants in a coma at the time of hospital admission had higher serum GFAP levels than conscious participants. In another study, serum GFAP levels remained high from day 1 to day 6 after subarachnoid haemorrhage70. Similar to ischaemic stroke, blood GFAP concentration at admission could significantly predict poor outcomes after subarachnoid haemorrhage, as observed by Zheng et al. at 6 months after the event71. In another study, a secondary rise in CSF GFAP levels on about day 7 after subarachnoid haemorrhage was related to complications, including the development of hydrocephalus and cerebral vasospasm72. Nevertheless, longitudinal data regarding blood GFAP dynamics following subarachnoid haemorrhage are still lacking.

Overall, these findings indicate a valuable prognostic role for blood GFAP in patients with stroke, although an important limitation of the diagnostic use of blood GFAP could be a low specificity for differentiating among stroke subtypes. In particular, in the setting of acute stroke symptoms, distinguishing between ischaemic stroke and intracerebral haemorrhage and between stroke and stroke mimics is essential, especially for the correct identification of patients who could be eligible for time-dependent reperfusion therapies (intravenous thrombolysis, mechanical thrombectomy for large-vessel occlusion). In this diagnostic context, the available data do not strongly support an imminent application of serum GFAP.

Inflammatory CNS diseases

Multiple sclerosis

Multiple sclerosis (MS) is the disease that led to the discovery of GFAP by Eng et al. in 1971 (ref.73). MS is a complex inflammatory and neurodegenerative disorder that affects more than two million people worldwide74. Therefore, efforts to identify a reliable and readily available biomarker that reflects disease severity and progression in MS are paramount in improving the clinical work-up and guiding the therapeutic approach. A reliable blood biomarker for use in MS will need to show an association with clinical severity, disease activity, worsening disability and treatment effectiveness.

Studies using ELISA or ECL assays to detect GFAP failed to identify significant differences between blood GFAP levels in participants with MS and participants with non-inflammatory neurological diseases75,76; these studies included a relatively small number of participants. However, subsequent studies using the more sensitive Simoa assay found evidence of higher serum GFAP levels in participants with MS than in healthy control participants and participants with non-inflammatory neurological diseases25,77,78. In particular, higher serum GFAP levels than in controls were consistently reported in participants with progressive MS (PMS), whereas the results for the relapsing–remitting MS (RRMS) phenotype differed between studies25,78. In one study, samples collected after a recent clinical relapse (RRMS+) had a higher concentration of GFAP than samples from healthy control participants, but no significant difference in GFAP levels was observed between participants with stable MS (RRMS−) and healthy control participants77. The same study reported higher levels of GFAP in participants with RRMS+ than in participants with RRMS− (129.8 pg/ml and 112.9 pg/ml, respectively; P < 0.012), but with a substantial overlap between the two groups.

In agreement with the proposed pathological role of astrocytes in MS79,80,81,82, multiple studies have found a correlation between blood GFAP concentration and severity of disability, as assessed by the expanded disability status scale (EDSS)8,25,77,78,83,84,85,86 (Table 2). Only a single study found a positive correlation between blood GFAP concentration and disease duration78. Notably, higher GFAP levels were associated with a greater lesion load on MRI in most of the reported studies25,78,86; blood GFAP levels also correlated with other markers of neurodegeneration (for example, NfL) and brain atrophy8,25,77,78,84. One study found an association between disease-modifying treatment (DMT) and reduced levels of GFAP78, whereas all other studies found no change in GFAP levels associated with such treatment8,25,78,86. Another study assessed blood GFAP levels in patients receiving autologous haematopoietic stem cell transplantation, and identified a paradoxical increase compared with baseline after the initiation of treatment85. A possible explanation for this finding could be the transient worsening of CNS inflammation following the administration of the chemotherapeutic agent busulfan, which constitutes part of the haematopoietic stem cell transplantation procedure, and might cause intrinsic neurotoxicity85. A similar increase in GFAP levels was observed in the context of neurotoxicity following immune effector cell-associated neurotoxicity syndrome after chimeric antigen receptor T cell therapy87. The potential value of GFAP as a predictor of future relapses and disability progression over time has scarcely been explored in individuals with MS and in populations with heterogeneous characteristics. Indeed, although a study including fewer than 50 participants with MS77 failed to identify a prognostic value of blood GFAP levels, preliminary results from a larger trial (EXPAND) identified a higher risk (HR 1.96) of reaching an EDSS of 7.0 in participants with secondary PMS who had higher GFAP levels (>80th percentile) at baseline88.

Table 2 Key studies of blood glial fibrillary acidic protein levels in multiple sclerosis
Full size table

Neuromyelitis optica spectrum disorder

Neuromyelitis optica spectrum disorder (NMOSD) is a classic autoimmune inflammatory astrocytopathy89. Aquaporin-4 antibodies, among other mechanisms, induce astrocytic damage in NMOSD lesions and subsequently cause neuroaxonal damage90. Data regarding GFAP concentrations in NMOSD are limited but promising3. Even using the standard ELISA, which is less sensitive than the ECL or Simoa assays, higher CSF and serum GFAP levels were reported in participants with NMOSD than in healthy control participants or participants with MS76. These findings are supported by more recent results obtained using more sensitive assays3,77,84 and suggest that blood GFAP could be used to distinguish between NMOSD and MS.

Furthermore, in one study, GFAP levels were higher in 33 participants with NMOSD than in 16 participants with myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD), two diseases with overlapping clinical and radiological findings83. Similar to the findings in MS, evidence indicates that serum GFAP levels in patients with NMOSD are higher shortly before (within 1 week) and during acute clinical relapses than during stable disease77,91. Data also indicate that serum GFAP levels correlate with EDSS score, most notably in younger patients77,84. The GFAP to NfL ratio increased during NMOSD relapses and decreased during MS relapses (AUC = 0.78)77, suggesting that this combination of markers could be used to distinguish between the two diseases. In contrast to MS and NMOSD, a correlation between GFAP levels and clinical severity was not observed in participants with MOGAD83. Finally, immunomodulatory therapies (corticosteroids, azathioprine, tacrolimus, methotrexate, cyclophosphamide, cyclosporine) did not seem to influence serum GFAP in patients with NMOSD77, most probably owing to the timing and relative ineffectiveness of the DMT used in this study. Additionally, evidence indicates that the main extent of astrocytic loss occurs during acute inflammatory exacerbation3,24,90. Nevertheless, the N-MOmentum study91 demonstrated that GFAP levels between NMOSD attacks were associated with risk of relapse and, therefore, could still be informative. Additionally, serum GFAP levels decreased by 12.9% from baseline in inebilizumab-treated participants with NMOSD, who did not show relapse over the follow-up period of the study91. The potential of serum GFAP as a possible treatment marker in NMOSD, including AQP4-seronegative disease, remains to be addressed in further studies.

In summary, GFAP might not be the most suitable marker for the differentiation of disease phenotypes in MS, or the monitoring of disease activity or treatment effectiveness, as blood levels of the marker in different subgroups seem to overlap substantially. However, several studies have found an association between high GFAP concentrations and PMS25,78,86. The consistent correlation between GFAP concentrations and clinical severity metrics suggest promising applications of the marker for exploring and monitoring relapse-independent progression in RRMS and PMS. However, in astrocytopathies, GFAP levels could be useful for the identification of patients with the highest relapse risk. Nevertheless, sufficiently powered prospective multicentre trials that aim to identify clear cut-off values are warranted to clarify some of these open questions.

Neurodegenerative diseases

Several studies have found increased levels of CSF GFAP in the most common neurodegenerative diseases, including Alzheimer disease (AD), prion diseases, frontotemporal lobar degeneration (FTLD), Parkinson disease (PD), PD dementia (PDD), and dementia with Lewy bodies (DLB)9,12,13,30. In contrast, only a few studies have explored levels of blood GFAP in these proteinopathies92,93,94,95, supporting the notion that more extensive investigations are needed to address this topic in detail.

Alzheimer disease

In a study by Oeckl et al, blood GFAP levels were higher in participants with AD and in participants with DLB or PDD than in control participants, participants with behavioural variant frontotemporal dementia (bvFTD) or participants with PD30, whereas blood biomarker levels did not differ between control participants, participants with PD and participants with bvFTD30. Interestingly, CSF levels of GFAP were similar across all participants with neurodegenerative disease; the presence of higher blood GFAP levels (that is, higher CSF to serum ratio) in participants with AD only was attributed to the heterogeneous topographical involvement of neuroinflammation and/or distinct types and patterns of astrogliosis occurring among neurodegenerative diseases13,30,92,95. Another study also found increased blood GFAP levels in individuals with AD compared with levels in cognitively healthy individuals96. Most interestingly, one study reported a correlation between plasma GFAP levels and cortical Aβ deposition in individuals with symptomatic AD97. Linear, positive associations were observed early in disease and diverged during more severe disease stages. These findings suggest that astrocytic damage or activation begins in the presymptomatic phase of AD and is associated with brain Aβ load98.

The FTLD spectrum

Compared with data in AD, the data on GFAP in FTLD spectrum diseases are inconsistent30,92,93,95. In the large, multicentre Genetic FTD Initiative (GENFI) study, including 469 participants with genetic FTD, plasma GFAP levels were elevated in symptomatic carriers of GRN mutations, but not in carriers of other FTD mutations, compared with levels in controls93. Moreover, biomarker changes were associated with the appearance of clinical symptoms and were not detectable in presymptomatic mutation carriers93. In support of these findings, two other studies found no changes in serum GFAP levels in participants with sporadic30 and genetic bvFTD95 compared with levels in control participants without neurodegenerative diseases. However, in a large Italian cohort, serum GFAP levels were elevated in participants with all FTLD clinical syndromes (sporadic and genetic) compared with those in healthy control participants92. The one exception was the group of participants with progressive supranuclear palsy, who had similar serum GFAP levels to healthy control participants. In this study, the two FTLD groups with the most elevated GFAP were those with bvFTD and those with agrammatic variant PPA; these groups included an unusually high percentage of participants with GRN mutations (13% and 25%, respectively). However, no significant difference in serum GFAP levels was observed between these two groups of participants and participants with sporadic FTLD. Several studies are ongoing in this field, the results of which might help clarify the discrepancies between the studies discussed here.

Alexander disease

Blood GFAP levels are of particular interest in specific genetic neurodegenerative diseases, such as Alexander disease. Alexander disease is caused by various dominant heterozygous mutations in the gene encoding GFAP99. The pathological hallmark of the disease is the formation of cytoplasmic aggregations in astrocytes100. These aggregates contain mainly GFAP, along with other cytoplasmic proteins. In a mouse model, the degree of GFAP expression in the brain showed a clear, negative correlation with survival100. Owing to the rarity of the disease, studies investigating GFAP levels in the blood of individuals with Alexander disease are limited. One study found a modest, elevation of GFAP levels in the serum of participants with infantile and juvenile Alexander disease, but not in adult participants with the disease, compared with levels in healthy controls101. This finding contrasts with the high concentrations of GFAP found in the CSF of participants with Alexander disease101,102,103. A possible explanation for this divergence is the hook effect mentioned above, whereby GFAP aggregate formation might limit its detection in the blood3. Blood GFAP might still serve as a promising treatment outcome parameter for future trials in Alexander disease (for example, in trials of antisense oligonucleotide therapies), but further studies are necessary.

Other neurodegenerative diseases

Data on blood GFAP in other neurodegenerative diseases are scarce. In one study, blood GFAP concentrations were not significantly elevated in participants with genetic or sporadic amyotrophic lateral sclerosis compared with levels in healthy control participants95. Another study found higher blood GFAP levels in participants with PD than in healthy control participants104. Blood GFAP levels were also elevated in participants with neurological manifestations of Wilson disease compared with levels in healthy controls and participants with pure hepatic manifestations of Wilson disease105. We found only one study that assessed blood GFAP levels in individuals with vascular cognitive impairment — no significant difference between healthy control participants and participants with vascular cognitive impairment was observed106. Notably, the studies discussed in this section used a range of analytical methods, including immunoassays with relatively low sensitivity (that is, standard ELISA).

Diagnosis and prognosis

Regarding the potential for diagnostic use, serum GFAP has shown promising performance in neurodegenerative diseases. In the study by Oeckl et al., mentioned above, serum GFAP allowed a better distinction between participants with AD and control participants than CSF Aβ1–42 (AUC 0.91 and 0.87, respectively). In the same study, serum GFAP distinguished between participants with AD and participants with bvFTD with an AUC of 0.85 (ref.30). Moreover, blood GFAP was able to discriminate participants with PDD or DLB from control participants (AUC 0.87), participants with PD (AUC 0.88) and participants with bvFTD (AUC 0.79)30. In two other studies, plasma GFAP seemed to perform similarly to plasma Aβ1–42 to Aβ1–40 ratio for the identification of amyloid PET positivity in participants with AD107,108. Plasma GFAP level predicted amyloid PET positivity with an accuracy of 88% (when combined with Aβ1–42 to Aβ1–40 ratio, age and APOE genotype), and AD CSF biomarker profile with an accuracy of 79–80%. These findings might be relevant to the early identification of candidates for clinical trials.

Remarkably, blood GFAP levels correlated negatively with Mini-mental State Examination (MMSE) score and performance in the major cognitive domains in participants with AD or FTD30,92,107. Accordingly, in participants with presymptomatic GRN-related FTD, higher plasma GFAP levels were associated with lower MMSE scores and brain volumes93. Higher GFAP concentrations correlated with faster rates of atrophy in the temporal lobes of participants with symptomatic GRN-related FTD. Therefore, elevated GFAP levels might be a characteristic of the late presymptomatic phase and relate to disease severity93. Even in cognitively healthy older adults at risk of cognitive impairment, blood GFAP levels were higher than in control participants and were associated with a higher risk of dementia98,107,109, conversion to AD108,110, a faster rate of cognitive decline109, and decline in hippocampal volume110. However, the prognostic value of GFAP levels in other neurodegenerative diseases has been poorly analysed; we could only find a cohort of participants with sporadic CJD94 and a cohort with FTD92. In both studies, blood GFAP levels failed to predict survival.

In summary, the implementation of blood GFAP as a biomarker in neurodegenerative diseases, especially in combination with other markers, is a promising approach for improving the precision of differential diagnosis. The association of higher blood GFAP concentrations with faster cognitive decline, higher incidence of dementia and a greater likelihood of conversion to symptomatic cognitive impairment in the presence of amyloid pathology and in carriers of GRN mutations indicates potential prognostic applications. Nevertheless, blood GFAP levels might be affected by the heterogeneity of a disorder, the stage of disease and abnormal GFAP aggregation formation. This raises concerns about the practical usefulness of the marker and must be considered during data interpretation. More research is needed to clarify the effects of these possible confounders. Furthermore, in older individuals with neurodegenerative diseases, the coexistence of large and small cerebral vessel comorbidities might further complicate inferences based on measurements of brain-derived proteins in the blood.

Brain tumours

Similar to other structural neurological diseases, a large body of evidence indicates that blood GFAP levels are elevated in individuals with brain tumours. Some studies have found blood GFAP levels to be higher in participants with glioblastoma multiforme (GBM) than in healthy control participants, participants with other non-glial primary tumours and participants with brain metastasis111,112,113,114,115, whereas, in other studies, a statistically significant difference in blood GFAP levels between participants with high-grade (that is, GBM) and low-grade brain tumours was not detected116,117. In participants with GBM, blood GFAP concentration correlated with preoperative tumour volume111,112,114,118,119, volume of necrosis111,119 and GFAP expression levels in tumour tissue111,119. In one study, individuals with systemic metastasis of myxopapillary ependymoma, a brain tumour with high GFAP expression, had very high blood GFAP concentrations compared with those in healthy controls120.

Data regarding the prognostic value of blood GFAP levels in individuals with brain tumours seem to be inconsistent. Evidence indicates that blood GFAP levels rise shortly after operative treatment compared with preoperative levels121, before ultimately decreasing122. One study found that blood GFAP levels did not correlate with the amount of malignant tissue that remained postoperatively123. In several studies, blood GFAP levels did not help predict postoperative tumour recurrence or overall survival112,116,122. However, two studies found an association between high blood GFAP levels and poor progression-free survival113,114. The major limitations of the studies discussed here are that they used immunoassays with lower sensitivity and lower readout resolution than highly sensitive bead-based assays such as Simoa, and that they included a relatively small number of participants. Overall, additional, sufficiently powered studies with newer immunoassays are a major unmet need for the evaluation of the diagnostic and prognostic application of GFAP in brain tumours.

In addition to the conditions discussed in this Review, changes in blood GFAP levels have been observed in various other neurological and systemic conditions; a summary of the available evidence is provided in Table 3.

Table 3 Other CNS and systemic diseases associated with changes in blood concentration of GFAP
Full size table

Challenges facing clinical use of GFAP

In addition to the disease-specific limitations mentioned in each section of this Review, the accurate implementation of blood GFAP measurement and the correct interpretation of the results faces other challenges (Box 1). Evidence indicates that the expression of GFAP by astrocytes increases with age in healthy individuals124, so the correlation between GFAP levels and age needs to be explored in more extensive studies to enable the definition of age-specific normal ranges. Sex-specific normal ranges might also be required. Additionally, the mechanisms that underlie the release of intermediate filaments such as GFAP following astrocytic activation remain unclear.

With the exception of the work on TBI, most of the studies discussed in this Review were single-centre, retrospective, or had methodological limitations such as small sample sizes. Furthermore, as different platforms and methods are available to detect GFAP in blood, it is essential to note that many of the studies are not directly comparable with each other and that a general agreement on a ‘gold standard’ detection method is currently lacking. Also, the epitopes targeted by GFAP antibodies are mostly unknown or proprietary, raising some concerns about the GFAP isoforms detected by different assays. Therefore, the identification of a reference method for detecting blood GFAP (for example, by mass spectrometry) is highly recommended. Furthermore, whether the antibody pairs used in existing assays detect the full-length GFAP protein or proteolytic fragments is unclear. GFAP also undergoes various post-translational modifications (for example, phosphorylation, citrullination and acetylation) and is vulnerable to the proteolytic activity of calpain and caspase 6 (ref.17). Site-specific phosphorylation can be disease-relevant and has been reported to be associated with disease severity — for example, in Alexander disease125 and following hypoxic injury126. The effects of post-translational modifications on the analytical performance and clinical utilization of different GFAP assays, which use different proprietary antibody pairs, has been poorly characterized.

Similarly, compared with levels in healthy participants, blood levels of GFAP breakdown products seem to be elevated following TBI and follow a diagnostic and prognostic pattern similar to that of the blood levels of the full GFAP protein127,128,129. If we consider GFAP breakdown products as a product of activated calpain proteolysis following TBI, these products might be a better marker of astrocyte damage, but not necessarily astrocyte activation, when compared with standard GFAP assays. However, the added value of assays that measure levels of GFAP breakdown products, compared with conventional GFAP assays, should be investigated further. One hint that some antibodies in GFAP assays do not recognize (proteolytic) protein fragments is the observation that subjecting blood or CSF samples to several freeze–thaw cycles results in a significant decrease in the detected GFAP concentration, especially in the CSF8. In addition, the effect of inhibitory matrix effects has not yet been completely clarified. For example, circulating GFAP autoantibodies have been reported in the blood of individuals with AD and following traumatic CNS injury125,130,131,132. The effect of these autoantibodies on the measurement of circulating GFAP levels with the different commercial platforms is poorly characterized.

Interpreting the meaning of elevated GFAP concentrations in CNS chronic diseases could be challenging. Indeed, GFAP expression accompanies astrocytic activation, which is a ‘double-edged sword’ in neurological diseases133. Although some subclasses of astrocytes (for example, neurotoxic astrocytes) are toxic to neurons and oligodendrocytes, other subclasses promote CNS repair133,134. Data suggest that harmful pan-activated astrocytes at the rim of MS lesions are GFAP-positive, whereas direct neurotoxic astrocyte subpopulations are not79,82,135. So far, the expression of GFAP over the spectrum of astrocyte subclasses remains poorly characterized, and more specific markers are needed to investigate the different subclasses of activated astrocytes and the different isoforms of GFAP82 (Box 2).

Furthermore, the dynamics of blood GFAP levels depend on the underlying pathology. In acute events without major astrogliosis and gliotic scar formation, such as mTBI136, the half-life of GFAP in blood is around 24–72 h35,137. In this context, GFAP could be merely a marker of structural damage to the CNS. In less acute events, such as inflammatory relapses, GFAP remains elevated for weeks after clinical onset138. In such cases, blood GFAP levels are likely to reflect ongoing astrocytic activation in addition to the possible astrocytic damage, as has been shown in NMOSD91. In chronic neuroinflammatory (for example, PMS) and neurodegenerative diseases, the levels of GFAP in the blood are expected to increase with accumulating astrogliosis. However, whether GFAP levels continue to climb, become stable or even decrease over time remains unclear, as the counterbalance between GFAP release and clearance is still not well defined.

Conclusions

Unprecedentedly, the FDA recently authorized a panel test for blood-derived brain protein biomarkers, including GFAP, for clinical use in a neurological diseases. GFAP is a well-established marker of astrocyte injury and activation in CNS diseases and is a valuable addition to the expanding panel of CNS-based blood biomarkers. The potential for clinical application of blood GFAP is encouraging, especially in the field of TBI, where robust data show that the marker has discriminatory ability for CNS injuries evident on CT and MRI head scans. Importantly, historical data on the diagnostic performance of GFAP has been validated in multicentre prospective studies using point-of-care assays, which might facilitate the triage of patients with TBI in pre-hospital and acute hospital settings if integrated into standard care. In inflammatory neurological diseases, blood GFAP has promising applications in PMS, as the marker could reflect and predict long-term disability worsening and, therefore, contribute to the treatment decision algorithm. In the older population, GFAP seems to predict the rate of cognitive decline and conversion to overt dementia, which makes it an attractive marker to recognize individuals at risk and enable rapid initiation of future preventive, and eventually therapeutic measures. Finally, recent insights suggest that blood GFAP has the potential to track even subtle structural CNS involvement in various neurological and systemic diseases. Academic collaborations could significantly accelerate efforts to fill current knowledge gaps and facilitate the implementation of blood GFAP as a biomarker on a wide scale.