糖心vlog

Key Points

Question听 Do plasma biomarkers glial fibrillary acidic protein (GFAP) and plasma neurofilament light chain (NfL) levels differ between frontotemporal lobar degeneration (FTLD) with tau (FTLD-tau) and TDP-43 (FTLD-TDP)?

Findings听 In this cross-sectional study including 31 controls and 141 patients in the training sample, patients with pathology-confirmed FTLD, GFAP/NfL ratio discriminated FTLD-tau from FTLD-TDP with excellent accuracy (area under the receiver operating characteristic curve鈥�=鈥�0.90), consistent across pathological subtypes and cognitive phenotypes. The ratio of GFAP/NfL performed better than either analyte alone, and diagnostic accuracy was replicated in an independent sample of 62 patients.

Meaning听 In this study, plasma GFAP/NfL ratio was a promising biomarker candidate to help discriminate FTLD-tau from FTLD-TDP, and these findings may be especially helpful to classify most patients with sporadic FTLD, whose pathology cannot be determined in life.

Importance听 Biomarkers are lacking that can discriminate frontotemporal lobar degeneration (FTLD) associated with tau (FTLD-tau) or TDP-43 (FTLD-TDP).

Objective听 To test whether plasma biomarkers glial fibrillary acidic protein (GFAP), neurofilament light chain (NfL), or their ratio (GFAP/NfL) differ between FTLD-tau and FTLD-TDP.

Design, Setting, and Participants听 This retrospective cross-sectional study included data from 2009 to 2020 from the University of Pennsylvania Integrated Neurodegenerative Disease Database, with a median (IQR) follow-up duration of 2 (0.3-4.2) years. The training sample was composed of patients with autopsy-confirmed and familial FTLD; nonimpaired controls were included as a reference group. The independent validation sample included patients with FTD with a clinical diagnosis of progressive supranuclear palsy syndrome (PSPS) associated with tau (PSPS-tau) or amytrophic lateral sclerosis (ALS) associated with TDP-43 (ALS-TDP). In patients with FTLD with autopsy-confirmed or variant-confirmed pathology, receiver operating characteristic (ROC) curves tested the GFAP/NfL ratio and established a pathology-confirmed cut point. The cut point was validated in an independent sample of patients with clinical frontotemporal dementia (FTD). Data were analyzed from February to July 2022.

Exposures听 Clinical, postmortem histopathological assessments, and plasma collection.

Main Outcomes and Measures听 ROC and area under the ROC curve (AUC) with 90% CIs evaluated discrimination of pure FTLD-tau from pure FTLD-TDP using plasma GFAP/NfL ratio; the Youden index established optimal cut points. Sensitivity and specificity of cut points were assessed in an independent validation sample.

Results听 Of 349 participants with available plasma data, 234 met inclusion criteria (31 controls, 141 in the training sample, and 62 in the validation sample). In the training sample, patients with FTLD-tau were older than patients with FTLD-TDP (FTLD-tau: n鈥�=鈥�46; mean [SD] age, 65.8 [8.29] years; FTLD-TDP: n鈥�=鈥�95; mean [SD] age, 62.3 [7.82] years; t_84.6鈥�=鈥�2.45; mean difference, 3.57; 95% CI, 0.67-6.48; P鈥�=鈥�.02) but with similar sex distribution (FTLD-tau: 27 of 46 [59%] were male; FTLD-TDP: 51 of 95 [54%] were male; 蠂²₁鈥�=鈥�0.14; P鈥�=鈥�.70). In the validation sample, patients with PSPS-tau were older than those with ALS-TDP (PSPS-tau: n鈥�=鈥�31; mean [SD] age, 69.3 [7.35] years; ALS-TDP: n鈥�=鈥�31; mean [SD] age, 54.6 [10.17] years; t_54.6鈥�=鈥�6.53; mean difference, 14.71; 95% CI, 10.19-19.23; P鈥�<鈥�.001) and had fewer patients who were male (PSPS-tau: 9 of 31 [29%] were male; ALS-TDP: 22 of 31 [71%] were male; 蠂²₁鈥�=鈥�9.3; P鈥�=鈥�.002). ROC revealed excellent discrimination of FTLD-tau from FTLD-TDP by plasma GFAP/NfL ratio (AUC鈥�=鈥�0.89; 90% CI, 0.82-0.95; sensitivity鈥�=鈥�0.73; 90% CI, 0.65-0.89; specificity鈥�=鈥�0.89; 90% CI, 0.78-0.98), which was higher than either GFAP level alone (AUC鈥�=鈥�0.65; 90% CI, 0.54-0.76) or NfL levels alone (AUC鈥�=鈥�0.75; 90% CI, 0.64-0.85). In the validation sample, there was sensitivity of 0.84 (90% CI, 0.66-0.94) and specificity of 0.81 (90% CI, 0.62-0.91) when applying the autopsy-derived plasma GFAP/NfL threshold.

Conclusions and Relevance听 The plasma ratio of GFAP/NfL may discriminate FTLD-tau from FTLD-TDP.

Frontotemporal dementia (FTD) comprises a clincopathologic spectrum of progressive language, behavior, and motor dysfunction.^1,2 The 2 major pathologic types are frontotemporal lobar degeneration (FTLD) with misfolded tau (FTLD-tau) and FTLD with TAR DNA-binding protein of 43 kDa (FTLD-TDP).³ Biofluid biomarkers provide a reliable in vivo diagnosis to discriminate Alzheimer disease (AD) from FTLD,⁴ yet we currently lack biomarkers that can classify FTLD pathological subtypes in life. While some FTD phenotypes are strongly associated with either FTLD-tau (eg, progressive supranuclear palsy syndrome [PSPS]⁵) or FTLD-TDP (eg, amyotrophic lateral sclerosis [ALS]⁶), other phenotypes are weakly correlated with underlying pathology (eg, behavioral variant FTD [bvFTD]).^7,8 Moreover, there is significant symptomatic overlap across clinical phenotypes,^9,10 making antemortem diagnosis challenging. While variants in C9orf72, GRN, and TARDBP genes indicate underlying FTLD-TDP and MAPT variants indicate FTLD-tau, most cases of FTLD are considered sporadic, with no known genetic cause.¹¹ Without biomarkers to distinguish tau and TDP-43 proteinopathies, sporadic FTLD cannot be reliably differentiated in vivo, and definitive pathological diagnosis of FTLD-tau or FTLD-TDP is still available only postmortem. Consequently, clinical trials for disease-modifying agents must largely focus on familial disease, leaving most patients with FTLD with sporadic disease excluded from experimental treatments.¹² In this context, biomarkers that could distinguish between FTLD pathologic subtypes in life are needed.¹³

Previous studies have tested stratification of FTLD-tau from FTLD-TDP using cerebrospinal fluid (CSF) phosphorylated tau (p-tau)^14-16 and the CSF p-tau/total tau ratio (AUCs of 0.81 to 0.87).^17-19 Still, there has been limited success in their application²⁰; CSF tau levels in FTLD may be complicated by multiple biological influences, including age, variant status,^14,15 and concomitant AD neuropathologic change (ADNC).²¹ Technological advancements have made plasma biomarkers a less invasive alternative to CSF. Plasma biomarkers are sensitive to biological processes linked to neurodegeneration seen in FTLD,^22,23 including glial fibrillary acidic protein (GFAP) as a marker of astrogliosis^24,25 and neurofilament light chain (NfL) as a marker of axonal degeneration.²⁶ Still, it is unclear if these analytes differ between distinct tau and TDP-43 FTLD proteinopathies.

We evaluated plasma GFAP and NfL levels and their ratio to differentiate molecular subtypes of FTLD-TDP and FTLD-tau; cognitively unimpaired healthy controls were included as a reference group. We tested consistency across heterogenous pathological and clinical FTLD subtypes and related biomarkers to accumulations of postmortem tau and TDP-43 pathology. In a proof-of-concept validation, we tested our trained threshold in an independent living sample of FTD phenotypes highly predicative of pathology (PSPS and ALS).

This is a retrospective cross-sectional study that follows the Strengthening the Reporting of Observational Studies in Epidemiology () reporting guideline. Patients were selected using the University of Pennsylvania Integrated Neurodegenerative Disease Biobank and Database^27,28 (eAppendix 1 in the Supplement). Participants were recruited for observational research at the University of Pennsylvania FTD, ALS, and AD research centers; blood was banked as part of ongoing clinical research programs. Patients were autopsied at the University of Pennsylvania Center for Neurodegenerative Disease Research.²⁹ Written informed consent was obtained according to the Declaration of Helsinki and approved by the University of Pennsylvania Institutional Review Board.

For the training sample, inclusion criteria were a pathological diagnosis of FTLD-tau (n鈥�=鈥�46) or FTLD-TDP (n鈥�=鈥�95) and available plasma data. Pathological diagnosis was determined either at autopsy by expert neuropathologists or by a pathogenic variant associated with FTLD-tau (eg, MAPT) or FTLD-TDP (eg, C9orf72, GRN, and TARDBP).³⁰ Cognitively unimpaired healthy controls were included as a reference group (n鈥�=鈥�31); they had no reported neurologic history or cognitive impairment, a Mini-Mental State Examination of 27 or greater; autopsy data were not available for controls. Exclusion criteria were a confounding neurological condition (eg, brain tumor or hydrocephalus) or plasma collection prior to symptom onset; CSF biomarkers excluded likely AD pathology in individuals without autopsy data.^4,31

Demographic characteristics were age at onset (earliest reported symptom), age at plasma collection, plasma collection to death interval, disease duration (onset to plasma collection), and sex. Patients were clinically diagnosed with ALS/ALS-FTD, bvFTD, corticobasal syndrome, dementia with Lewy bodies, semantic variant primary progressive aphasia (PPA), nonfluent/agrammatic PPA, logopenic variant PPA, or PSPS based on clinical diagnostic criteria.^27,28 Race was recorded by self-report.

Patients were genotyped for C9orf72 hexanucleotide repeated expansion using repeat-primed polymerase chain reaction and for pathogenic variants in MAPT, GRN, TARDBP using a custom targeted sequencing panel based on structured pedigree analysis of familial risk.³²

Autopsied individuals (47 with FTLD-TDP and 39 with FTLD-tau) had full neuropathological diagnostic evaluation for FTLD,^33,34 ADNC,³⁵ and 伪-synuclein positive Lewy body disease (LBD).³⁶ FTLD-tau included Pick disease, argyrophilic grain disease, PSP, tauopathy unclassifiable, corticobasal degeneration, and globular glial tauopathy.³³ FTLD-TDP included types A to E^34,37 or ALS with untypable TDP-43. Patients with FTLD were assessed for cooccurring ADNC³⁵ or LBD in the brainstem, limbic, or neocortical regions.³⁶ If copathology was negligible and not clinically meaningful (ie, no or low ADNC, no or amygdala-predominant LBD), patients with FTLD were considered pure (30 with FTLD-tau and 38 with FTLD-TDP).

Severity of pathologic tau and TDP-43 burden was scored at autopsy according to standardized methods³⁵ using a semiquantitative 5-point scale (ie, 0 indicates none; 0.5, rare; 1, low; 2, intermediate; 3, high). Regional sampling was randomized between hemispheres. Tau and TDP-43 burdens were calculated as the mean severity score across 2 frontal (middle frontal and cingulate), 2 temporal (superior/middle temporal and amygdala), and 2 brainstem regions (medulla and pons).

For the independent validation sample, inclusion criteria were clinical syndromes ALS and PSPS, which are highly predictive of TDP-43⁶ (n鈥�=鈥�31) and tau⁵ (n鈥�=鈥�31), respectively. Exclusion criteria were presymptomatic plasma and CSF biomarkers indicative of likely AD.⁴ Demographic characteristics were recorded and consent was obtained, as described above.

Plasma was collected according to standardized procedures.³⁸ Samples were analyzed in duplicate for NfL using the Quanterix single-molecule array (Simoa) NF-Light Advantage kit reagents³⁹ using the Uman antibody reagents⁴⁰ and in duplicate for GFAP using the Quanterix Simoa Discovery kit reagents,⁴¹ both on the Quanterix HD-X automated immunoassay platform. For patients with more than 1 plasma time point, the earliest (baseline) time point was selected to test early discrimination (eAppendix 1 in the Supplement).

For patients and controls without autopsy data, CSF amyloid-尾 peptide 1-42 (A尾42) less than 192⁴² or A尾42/A尾40 ratio less than 0.075⁴³ excluded A尾-positive individuals⁴ (eAppendix 1 in the Supplement). Luminex xMAP quantified CSF A尾42⁴⁴; Fujirebio Lumipulse platform using Lumipulse kits quantified CSF A尾42 and A尾40.

Some demographic variables were not normally distributed; Mann-Whitney-Wilcoxon and Kruskal-Wallis tests compared continuous variables and 蠂² compared categorical variables. Plasma biomarkers were log-transformed for a normal distribution in parametric models; unadjusted Mann-Whitney-Wilcoxon comparisons are summarized in the figures. Spearman correlations tested associations between biomarkers. Statistical tests were 2-tailed with a significance threshold of 伪鈥�=鈥�.05. Analyses were conducted using R version 4.1.2 (The R Foundation) and the cutpointr⁴⁵ and effectsize⁴⁶ packages.

Linear models compared biomarkers across pathology (FTLD-tau vs FTLD-TDP), covarying for factors that might affect plasma concentrations, including disease duration and sex, using the equation log(Biomarker)鈥�=鈥壩�₀ + (尾₁鈥壝椻�塒athology) + (尾₂鈥壝椻�塂isease Duration) + (尾₃鈥壝椻�塖ex) + 蔚. 95% CIs are provided for 尾 estimates. Effect sizes with 95% CIs were calculated using generalized 畏² (畏²_G interpretation: 鈮�0.01, small; 鈮�0.06, medium; 鈮�0.14, large⁴⁷). Models then compared FTLD groups with controls, covarying for age and sex.

Linear models tested how biomarkers associated with pathologic tau and TDP-43 burden, covarying for ADNC, LBD, sex, and, because additional pathological changes may occur between plasma collection and death, plasma-to-death interval, using the equation log(Biomarker)鈥�=鈥壩�₀ + (尾₁鈥壝椻�塗au Burden) + (尾₂鈥壝椻�塗DP-43 Burden) + (尾₃鈥壝椻�堿DNC) + (尾₄鈥壝椻�塋BD) + (尾₅鈥壝椻�塈nterval to Death) + (尾₆鈥壝椻�塖ex) + 蔚. Within-group models tested how biomarkers related to postmortem tau within individuals with pure FTLD-tau and TDP-43 within those with pure FTLD-TDP, covarying for plasma-to-death interval and sex.

Receiver operating characteristic (ROC) analyses with bootstrapping (500 iterations) tested diagnostic accuracy; we tested patients with pure FTLD-tau (n鈥�=鈥�30) and pure FTLD-TDP (n鈥�=鈥�38) pathology to ensure copathologies (eg, ADNC) did not influence thresholds.⁴⁸ Area under the ROC curve (AUC) with 90% (5% to 95%) CIs were reported. The Youden index determined threshold that maximized sensitivity and specificity.

To test application in sporadic cases, ROC analyses were repeated excluding variant carriers (n鈥�=鈥�68); 54 patients overlapped between the pure and sporadic samples. Finally, ROC analyses were repeated in the full FTLD sample, including familial and mixed pathology cases (n鈥�=鈥�141).

A linear model compared GFAP/NfL ratio across PSPS-tau and ALS-TDP, covarying for disease duration and sex, using the equation log(GFAP/NfL)鈥�=鈥壩�₀ + (尾₁鈥壝椻�塆roup) + (尾₂鈥壝椻�塂isease Duration) + (尾₃鈥壝椻�塖ex) + 蔚. Training sample鈥揹erived threshold for GFAP/NfL was applied to the validation sample, with sensitivity and specificity reported.

Table 1 outlines patient characteristics for training and validation samples. In the training sample, patients with FTLD-tau were older than patients with FTLD-TDP (FTLD-tau: n鈥�=鈥�46; mean [SD] age, 65.8 [8.29] years; FTLD-TDP: n鈥�=鈥�95; mean [SD] age, 62.3 [7.82] years; t_84.6鈥�=鈥�2.45; mean difference, 3.57; 95% CI, 0.67-6.48; P鈥�=鈥�.02) but with similar sex distribution (FTLD-tau: 27 of 46 [59%] were male; FTLD-TDP: 51 of 95 [54%] were male; 蠂²₁鈥�=鈥�0.14; P鈥�=鈥�.70). In the validation sample, patients with PSPS-tau were older than those with ALS-TDP (PSPS-tau: n鈥�=鈥�31; mean [SD] age, 69.3 [7.35] years; ALS-TDP: n鈥�=鈥�31; mean [SD] age, 54.6 [10.17] years; t_54.6鈥�=鈥�6.53; mean difference, 14.71; 95% CI, 10.19-19.23; P鈥�<鈥�.001) and had fewer patients who were male (PSPS-tau: 9 of 31 [29%] were male; ALS-TDP: 22 of 31 [71%] were male; 蠂²₁鈥�=鈥�9.3; P鈥�=鈥�.002). Spearman correlations (eAppendix 2 in the Supplement) showed that plasma GFAP and NfL levels were significantly associated across all FTLD (蟻鈥�=鈥�0.35; P鈥�<鈥�.001), within FTLD-tau (蟻鈥�=鈥�0.64; P鈥�<鈥�.001), and within FTLD-TDP (蟻鈥�=鈥�0.42; P鈥�&濒迟;鈥�.001).

We compared plasma biomarkers by group (Figure 1). Covarying for disease duration and sex, plasma GFAP/NfL ratio was lower in patients with FTLD-TDP than patients with FTLD-tau (尾鈥�=鈥夆垝0.67; 95% CI, 鈭�0.90 to 鈭�0.43; P鈥�<鈥�.001) with large effect size (畏²_G鈥�=鈥�0.24; 95% CI, 0.14-1.00). GFAP levels were higher in patients with FTLD-tau than patients with FTLD-TDP (尾鈥�=鈥夆垝0.20; 95% CI, 鈭�0.38 to 鈭�0.02; P鈥�=鈥�.03) with medium effect size (畏²_G鈥�=鈥�0.07; 95% CI, 0.02-1.00). NfL levels were higher in patients with FTLD-TDP than those with FTLD-tau (尾鈥�=鈥�0.47; 95% CI, 0.20-0.73; P鈥�<鈥�.001) with medium effect size (畏²_G鈥�=鈥�0.10; 95% CI, 0.03-1.00).

Disease duration was associated with GFAP levels (尾鈥�=鈥�0.04; 95% CI, 0.01-0.07; P鈥�=鈥�.007) but not GFAP/NfL ratio (尾鈥�=鈥�0.03; 95% CI, 鈭�0.003 to 0.07; P鈥�=鈥�.07) or NfL levels (尾鈥�=鈥�0.01; 95% CI, 鈭�0.04 to 0.05; P鈥�=鈥�.80). Sex was associated with GFAP levels (尾鈥�=鈥夆垝0.28; 95% CI, 鈭�0.44 to 鈭�0.12; P鈥�<鈥�.001) and NfL levels (尾鈥�=鈥夆垝0.48; 95% CI, 鈭�0.72 to 鈭�0.25; P鈥�<鈥�.001), but not GFAP/NfL ratio (尾鈥�=鈥�0.20; 95% CI, 鈭�0.01 to 0.41; P鈥�=鈥�.06). We calculated 尾 dfs for high leverage points: 0 points were greater than the threshold of 0.17 for all models.

While our main comparison of interest was FTLD-TDP with FTLD-tau, we also compared biomarker levels with controls. After covarying for age and sex, GFAP/NfL ratio was lower in FTLD-tau (尾鈥�=鈥夆垝0.64; 95% CI, 鈭�0.93 to 鈭�0.35; P鈥�<鈥�.001) and FTLD-TDP (尾鈥�=鈥夆垝1.36; 95% CI, 鈭�1.63 to 鈭�1.09; P鈥�<鈥�.001) than controls. GFAP levels were higher in patients with FTLD-tau (尾鈥�=鈥�0.33; 95% CI, 0.12-0.55; P鈥�=鈥�.003) than controls but not patients with FTLD-TDP (尾鈥�=鈥�0.12; 95% CI, 鈭�0.088 to 0.32; P鈥�=鈥�.27). NfL levels were higher in patients with FTLD-tau (尾鈥�=鈥�0.97; 95% CI, 0.66-1.29; P鈥�<鈥�.001) and those with FTLD-TDP (尾鈥�=鈥�1.47; 95% CI, 1.18-1.77; P鈥�<鈥�.001) than controls.

Age was significantly associated with GFAP (尾鈥�=鈥�0.02; 95% CI, 0.01-0.03; P鈥�<鈥�.001) but not NfL levels (尾鈥�=鈥�0.01; 95% CI, 0 to 0.03; P鈥�=鈥�.06) or GFAP/NfL ratio (尾鈥�=鈥�0.01; 95% CI, 鈭�0.01 to 0.02; P鈥�=鈥�.34). Sex associated with GFAP levels (尾鈥�=鈥夆垝0.27; 95% CI, 鈭�0.41 to 鈭�0.13; P鈥�<鈥�.001) and NfL levels (尾鈥�=鈥夆垝0.43; 95% CI, 鈭�0.63 to 鈭�0.23; P鈥�<鈥�.001) but not GFAP/NfL ratio (尾鈥�=鈥�0.16; 95% CI, 鈭�0.026 to 0.35; P鈥�=鈥�.09).

To investigate biological correlates of GFAP/NfL ratio in patients with FTLD (eAppendix 2 in the Supplement), linear models tested how biomarkers related to postmortem tau and TDP-43 pathological severity in autopsied patients, covarying for ADNC, LBD, plasma-to-death interval, and sex (eAppendix 2 in the Supplement). Plasma GFAP/NfL ratio (eAppendix 2 in the Supplement) was associated with greater tau burden (尾鈥�=鈥�0.28; 95% CI, 0.15-0.41; P鈥�<鈥�.001) with large effect size (畏²_G鈥�=鈥�0.41; 95% CI, 0.27-1.00) and inversely associated with TDP-43 (尾鈥�=鈥夆垝0.22; 95% CI, 鈭�0.38 to 鈭�0.07; P鈥�=鈥�.006) with medium effect size (畏²_G鈥�=鈥�0.10; 95% CI, 0.02-1.00), confirming group-level differences.

We next tested within-group pathological associations (eAppendix 2 in the Supplement); models tested analyte associations with tau burden (eAppendix 2 in the Supplement) and TDP-43 burden (eAppendix 2 in the Supplement), covarying for plasma-to-death and sex. Plasma GFAP levels were not associated with tau burden within patients with pure FTLD-tau (尾鈥�=鈥�0.22; 95% CI, 鈭�0.31 to 0.76; P鈥�=鈥�.41) nor with TDP-43 within patients with pure FTLD-TDP (尾鈥�=鈥�0.13; 95% CI, 鈭�0.098 to 0.35; P鈥�=鈥�.26). Plasma NfL was associated with TDP-43 burden within patients with pure FTLD-TDP (尾鈥�=鈥�0.36; 95% CI, 0.11-0.61; P鈥�=鈥�.007) with medium effect size (畏²_G鈥�=鈥�0.13; 95% CI, 0.01-1.00) but not tau burden (尾鈥�=鈥�0.56; 95% CI, 鈭�0.003 to 1.12; P鈥�=鈥�.05).

Given the clinicopathological heterogeneity of FTLD, we evaluated consistency of biomarkers across pathological subtypes (eAppendix 3 in the Supplement) and clinical phenotypes (eAppendix 3 in the Supplement). All FTLD-tau pathological subtypes had a higher median GFAP/NfL ratio than all FTLD-TDP pathological subtypes. Within bvFTD, GFAP/NfL ratio was significantly lower in patients with FTLD-TDP than patients with FTLD-tau (尾鈥�=鈥夆垝0.37; 95% CI, 鈭�0.72 to 鈭�0.02; P鈥�=鈥�.04) with medium effect size (畏²鈥�=鈥�0.08). Within patients with PPA, GFAP/NfL ratio was significantly lower in patients with FTLD-TDP than patients with FTLD-tau (尾鈥�=鈥夆垝0.50; 95% CI, 鈭�0.91 to 鈭�0.09; P鈥�=鈥�.02) with large effect size (畏²鈥�=鈥�0.67; 95% CI, 0.16-1.00).

ROC curves compared GFAP/NfL ratio performance in the pure pathology, sporadic, and full FTLD samples (Figure 2). Plasma GFAP/NfL ratio had excellent performance in the pure pathology sample (AUC鈥�=鈥�0.89; 90% CI, 0.82-0.95) and sporadic sample (AUC鈥�=鈥�0.88; 90% CI, 0.81-0.95); GFAP/NfL ratio performance was good, although lower, in the full sample (Table 2).

We examined individual misclassifications for GFAP/NfL ratio when applying the pure-derived threshold (2.73; Table 2) to the total sample, and found 85% of false-positive FTLD-TDP cases were variant carriers (eAppendix 3 in the Supplement).

In the validation sample, plasma GFAP/NfL ratio (Figure 3) was significantly lower in patients with ALS-TDP than patients with PSPS-tau (尾鈥�=鈥夆垝0.85; 95% CI, 鈭�1.16 to 鈭�0.54; P鈥�<鈥�.001) with large effect size (畏²_G鈥�=鈥�0.47; 95% CI, 0.32-1.00), covarying for disease duration (尾鈥�=鈥�0.11; 95% CI, 0.05-0.17; P鈥�<鈥�.001) and sex (尾鈥�=鈥夆垝0.12; 95% CI, 鈭�0.42 to 0.19; P鈥�=鈥�.45). Likewise, GFAP/NfL ratio had good overall discrimination accuracy (AUC鈥�=鈥�0.87; 90% CI, 0.75-0.93). When applying the trained threshold (2.73; Table 2), GFAP/NfL ratio had sensitivity of 0.84 (90% CI, 0.66-0.94) and specificity of 0.81 (90% CI, 0.62-0.91).

This study performed a systematic assessment of plasma GFAP and NfL levels in a large diverse cohort of patients with pathology-confirmed FTLD. Plasma GFAP/NfL ratio showed excellent discrimination of FTLD-tau from FTLD-TDP, particularly in autopsy-confirmed, criterion-standard FTLD with pure or minimal copathology (AUC鈥�=鈥�0.89; 90% CI, 0.82-0.95). In a test of real-world application, GFAP/NfL ratio also showed excellent discrimination in patients with sporadic FTLD (AUC鈥�=鈥�0.88; 90% CI, 0.81-0.95). Rigorous analysis of other contributing factors found disease duration, age, variant status, and sex were important sources of biological variation. When the pure-derived autopsy threshold was applied to an independent sample, plasma GFAP/NfL ratio had excellent performance (sensitivity鈥�=鈥�0.84; 90% CI, 0.66-0.94; specificity鈥�=鈥�0.81; 90% CI, 0.62-0.91). In sum, plasma GFAP/NfL ratio may be a candidate biomarker to distinguish sporadic FTLD-tau and FTLD-TDP during life.

NfL is a structural component of the neural cytoskeleton and is increased in plasma following axonal injury and degeneration.²⁶ While plasma NfL is a nonspecific marker of degeneration, showing elevated levels in AD and other neurodegenerative disorders, it is consistently highest in FTD and ALS disorders.^49-52 Even so, most studies lack sufficient autopsy-confirmed samples to rigorously test associations between FTLD proteinopathy subtypes. Here, NfL was elevated in both patients with FTLD-tau and FTLD-TDP compared with controls and was associated with more severe TDP-43 burden in those with FTLD-TDP. Nonetheless, we find that plasma NfL levels were significantly higher in patients with FTLD-TDP than those with FTLD-tau, consistent across pathological subtypes. Importantly, this difference has been observed by others²³ despite extensive white matter disease in FTLD-tau.⁵³ Further histopathological exploration is needed to interrogate the biological underpinnings of elevated plasma NfL levels in those with FTLD-TDP compared with those with FTLD-tau; it is possible that distinct patterns of degeneration^53,54 underlie the different plasma signatures.

Neuroinflammation and reactive astrogliosis are part of FTD pathogenesis,^55-57 and plasma GFAP levels are elevated in those with FTD compared with controls.⁵⁸ In clinical FTD, elevated GFAP levels have been linked to late-stage disease and variant carriers.^22,58,59 Still, while elevated GFAP is well-studied in AD,^48,60-62 there are limited data examining the full breadth of clinical, genetic, and pathological subtypes of FTLD. Here, we found greater plasma GFAP levels in patients with FTLD-tau compared with those with FTLD-TDP and with controls after controlling for age and sex. However, GFAP levels were not associated with pathological burden within those with pure FTLD-tau nor within those with pure FTLD-TDP. To better understand the pathological correlates of plasma GFAP in FTLD, future studies should relate to measures of gliosis and inflammation in neocortical and white matter regions of FTLD-tau and FTLD-TDP.

The GFAP/NfL ratio had the best discrimination of FTLD-tau from FTLD-TDP. In addition to biological processes underlying FTLD-tau and FTLD-TDP, our findings indicate that other factors may influence GFAP/NfL levels. Studies show that plasma levels differ by variant status,^22,63,64 which may affect diagnostic accuracy. In the full training sample, most FTLD-TDP false-positive errors were variant carriers, while diagnostic performance was excellent in the sporadic training sample and the validation sample (also sporadic). These findings emphasize the importance of genetic testing in patients with FTD, especially when interpreting biofluid levels. It will be crucial to validate our findings in larger autopsy-confirmed and sporadic samples as they become available. In addition, plasma analytes may be sensitive to disease severity in FTLD,^58,65 and GFAP/NfL ratio was positively associated with disease duration in both training and validation samples. This may explain, in part, the improved discrimination when NfL and GFAP are combined in a ratio, thus partially controlling for patient factors. Indeed, GFAP and NfL levels were positively correlated in our sample. Thus, longitudinal studies tracking plasma alterations over time are needed to confirm that these thresholds generalize to different stages of disease severity.

There are limitations to our findings. Foremost, the validation sample was composed of clinical phenotypes highly predictive of pathology (PSPS and ALS) to increase confidence in tau and TDP groupings. While ALS is associated with TDP-43 pathology, it typically has minimal TDP-43 pathological burden that is limited to the motor cortex and thus often differs from other forms of TDP-43. One interpretation for these observations is that topographical differences in regional neurodegeneration could influence GFAP/NfL ratio. An optimal test of GFAP/NfL ratio would therefore be within a clinically homogeneous group, such as bvFTD. Subanalyses of clinical and pathological subgroups (eAppendix 3 in the Supplement) found GFAP/NfL ratio was consistently higher in patients with FTLD-tau compared with those with FTLD-TDP, even in the small subset of patients with sporadic bvFTD and PPA. Future studies should confirm findings with larger sample sizes, using expanded clinical phenotypes and sporadic cases; studies should also test if GFAP/NfL ratio is influenced by topography of disease using longitudinal structural imaging with postmortem validation. Second, previous studies have found that plasma GFAP level was elevated in patients with AD compared with FTLD.^48,60 Our models covaried for ADNC and indicated that it did not confound results. Moreover, ROC analyses were performed in pure FTLD to exclude patients with concomitant pathologies, including ADNC. Results confirmed the utility of GFAP/NfL ratio in FTLD, showing excellent diagnostic performance, even when excluding concomitant ADNC. Third, given clinical overlap between FTD and AD syndromes,^31,66-68 a 2-step algorithm that excludes AD pathology may be necessary.¹⁴ However plasma p-tau may not be a good candidate, as it is elevated in ALS.⁶⁹ Here, we use CSF A尾42 and A尾42/A尾40 levels to exclude patients with likely AD. Fourth, most of our sample self-identified as White, and we were underpowered to test differences in analyte levels across racial and ethnic groups; results may not generalize to other populations. Fifth, it is unknown how other forms of FTLD (eg, fused in sarcoma) would be classified by GFAP/NfL ratio. Sixth, our results suggest that plasma levels may increase with disease severity, corroborated by other studies.^23,59 Thus, a single threshold, broadly applied, may perform less well than a model that accounts for disease stage.

Substantial strengths of this study were the pathology-confirmed training sample that was large enough to allow thorough examination within the extensive clinicopathologic spectrum of FTLD, and validation in an independent test sample. Our findings provide strong support for plasma GFAP/NfL ratio as a candidate biomarker to help distinguish FTLD-tau from FTLD-TDP in life.

Accepted for Publication: August 5, 2022.

Published Online: October 10, 2022. doi:10.1001/jamaneurol.2022.3265

Corresponding Author: Katheryn A. Q. Cousins, PhD, Penn Frontotemporal Degeneration Center, Department of Neurology, Richards Medical Research Laboratories, 3700 Hamilton Walk, Ste 600B, Philadelphia, PA 19104 (katheryn.cousins@pennmedicine.upenn.edu).

Author Contributions: Drs Cousins and Irwin had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Cousins, Grossman, Irwin.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Cousins, Shaw, Irwin.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Cousins, Shaw.

Obtained funding: Cousins, Shaw, Chen-Plotkin, Wolk, Lee, Grossman, Irwin.

Administrative, technical, or material support: Shaw, Van Deerlin, Lee, Irwin.

Study supervision: Irwin.

Conflict of Interest Disclosures: Dr Cousins has received grants from the Alzheimer鈥檚 Association and University of Pennsylvania鈥檚 Alzheimer鈥檚 Disease Research Center during the conduct of the study. Dr Shaw has received grants from the National Institute on Aging during the conduct of the study and has received personal fees from Biogen and Fujirebio outside the submitted work. Dr Chen-Plotkin has received grants from the National Institutes of Health, Michael J. Fox Foundation/Alzheimer鈥檚 Association, American Heart Association/Allen Institute, and Chan Zuckerberg Initiative and support from an endowed chair from the Parker Family outside the submitted work. Dr Wolk has received grants from the National Institute on Aging during the conduct of the study; grants from Biogen and Merck; personal fees from Functional Neuromodulation, Qynapse, Eli Lilly, Neuronix, and GE Healthcare outside the submitted work. Dr Lee has received grants from the National Institutes of Health during the conduct of the study. Dr Grossman has received grants from the National Institutes of Health, Department of Defense, and Samuel Newhouse Foundation during the conduct of the study. Dr Irwin has received grants from the National Institutes of Health during the conduct of the study and is a member of the scientific advisory board of Denali Therapeutics outside the submitted work. No other disclosures were reported.

Funding/Support: This work is supported by funding from the National Institute of Aging (grants P01-AG066597 and P30-AG072979; former grant P01-AG017586), the National Institute of Neurological Disorders and Stroke (grant R01-NS109260-01A1), the Penn Institute on Aging, and the Wyncote Foundation. Dr Cousins is supported by the Alzheimer鈥檚 Association (grants AARF-D-619473 and AARF-D-619473-RAPID).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank the patients and families for contributing to our research and for participating in the brain donation program. We also thank and acknowledge the contributions of our late colleague, John Q. Trojanowski, MD, PhD, without whom this work would not be possible. Dr Trojanowski鈥檚 leadership in program projects in Alzheimer鈥檚 Disease Research Center at University of Pennsylvania and development of infrastructure for pathological and biomarker resources were essential for this study.