Diagnostic value of plasma p-tau181, NfL, and GFAP in a clinical setting cohort of prevalent neurodegenerative dementias

We retrospectively analyzed plasma and CSF samples from 316 consecutive FTD, PSP, CBS, DLB, AD, and SNAP patients and plasma samples from 60 healthy controls (HC) submitted to the Neuropathology Laboratory (NP-Lab) at the Institute of Neurological Science of Bologna, Italy, between 2005 and 2021 (2015–2020 for the AD group).

We reviewed the results of the diagnostic work-up for each patient, including clinical charts, neuropsychological testing, neuroimaging, and CSF AD core biomarkers. Neuropsychological evaluation was conducted as described [22]. According to the consensus criteria, clinical diagnoses were established by expert agreement [23‐31]. In the FTD, PSP, CBS, DLB, and AD groups, only patients with a “probable” diagnosis according to the internationally established criteria were included in the study cohort to pursue the highest likelihood between the clinical diagnoses and the underlying pathologic processes. Cases lacking thorough clinical information (n = 59) and those without sufficient plasma or CSF for the analyses (n = 152) were excluded. No individuals with severe systemic illnesses were included in the study cohort.

Patients belonging to the frontotemporal lobar degeneration (FTLD) spectrum comprised the largest group (n = 119) and included participants with FTD (n = 59), PSP (n = 31), and CBS (n = 29). In the FTD group, we also distinguished between those with “pure” cognitive phenotypes (n = 43) from those with associated motor features (“plus” phenotypes, n = 16), namely amyotrophic lateral sclerosis (ALS) and parkinsonism. Pure FTD phenotypes included the following clinical syndromes: behavioral variant FTD (bvFTD, n = 33), non-fluent variant primary progressive aphasia (nfvPPA, n = 6), and logopenic variant PPA (lvPPA, n = 2), while two subjects presented with mixed bvFTD/PPA features. The FTD+parkinsonism (n = 8) and FTD+ALS (n = 8) groups included patients who met the criteria for bvFTD and/or PPA but also showed either extrapyramidal signs (in the presence of a mixed phenotype or not fully satisfying the criteria for CBS or PSP diagnosis) or upper and lower motor neuron impairment, respectively [22, 25]. Forty-nine patients were diagnosed as probable DLB/MCI-LB. All 97 individuals fulfilling the clinical criteria for AD/MCI-AD had a characteristic AD CSF biomarker profile supporting the clinical diagnosis (i.e., pathological values of Aβ42/40, p-tau/Aβ42 and t-tau/Aβ42 ratios according to in-house cutoffs) [32]. SNAP patients (n = 51) were cognitively impaired at neuropsychological testing, had normal CSF Aβ42/40 ratio, positive neurodegeneration biomarkers (as defined by neuroimaging and/or CSF findings), and did not fulfill the criteria for “probable” FTD, PSP, CSB, AD, and DLB. Finally, controls included a group of healthy (i.e., medical history not relevant for significant diseases/medications) blood donors (HC). Before blood collection, all HC underwent medical evaluation, including a standardized interview to exclude neurological symptoms.

For each participant, EDTA plasma samples were collected, aliquoted, and stored at − 80 °C according to standard procedures. All blood analyses were performed on a SiMOA SR-X analyzer platform (Quanterix, Billerica, MA, USA). Plasma NfL, p-tau181, and GFAP were measured with the SiMOA NF-light advantage, SiMOA p-tau181 advantage V2, and SiMOA GFAP discovery kits (i.e., the same used for GFAP quantification in CSF), respectively. The mean intra-assay and inter-assay CVs were respectively 5% and 15% for NfL, 9% and 18.5% for p-tau181, and 7% and 19% for GFAP. The samples were analyzed randomly to avoid bias due to the effect of inter-assay variability on specific patient groups.

We screened all FTLD patients with a positive familial history of dementia defined by the presence of at least one dementia case among the first-degree relatives and/or those with a clinical history compatible with early-onset dementia (n = 57) for variants in 27 dementia-associated genes, including GRN, MAPT, TARDBP, and FUS, as previously reported [37]. In the same patient group, we also screened for the presence of the C9orf72 repeat expansion using the 2-step strategy with southern blotting confirmation, as previously described [38].

APOE analysis was performed through PCR product digestion at 37 °C with the restriction enzyme HhaI (Thermo Fisher Scientific, Waltham, MA, USA) and visualized on 3.5% Metaphor agarose gel with GelStar nucleic acid gel stain or by targeted next-generation sequencing as previously reported [37].

Statistical analyses were performed using the software Graphpad Prism 8.4 and Stata SE 14.2.

In the descriptive analysis, continuous variables were presented with mean and standard deviation (SD) or median and interquartile range (IQR) depending on the data distribution. The Shapiro-Wilk test was used to evaluate the normal data distribution. The categorical variables were presented as absolute (n) and relative frequency (%). The Student t-test, Mann-Whitney U test, one-way analysis of variance (followed by Bonferroni post hoc test), and Kruskal-Wallis test (followed by Dunn’s post hoc analysis) were used to compare the continuous variables between the groups. The chi-square test was used to compare categorical variables between the groups.

The demographic characteristics of the study cohort and the results of genetic and CSF biomarker analyses are summarized in Tables 1 and 2 and Additional file 1: Table S1.

FTD and HC were significantly younger at biosample collection than the other diagnostic groups, but for SNAP (p ≤ 0.01 for all comparisons). There were no significant differences in the mean age between the AD, CBS, PSP, and DLB groups, except for DLB patients being older than AD and SNAP patients (p < 0.001). The time between disease onset and biosample collection differed between DLB and FTD (p < 0.001), PSP and FTD (p = 0.02), DLB and AD (p = 0.006), SNAP and PSP or DLB (p < 0.001 for both comparisons), and SNAP and AD groups (p = 0.03). Females were underrepresented in the DLB (vs. FTD, p = 0.002; vs. CBS, p = 0.004, vs. AD, p = 0.002, vs. SNAP p = 0.04) and PSP (vs. FTD, p = 0.046; vs. CBS, p = 0.039) groups.

In the FTD group, 18 patients had a monogenic disease linked to the most prevalent mutations in the GRN (n = 6) and C9orf72 (n = 7) genes. There was a slightly higher prevalence of genetic cases in the FTD “plus” than in the “pure” phenotype (37.5% vs. 27.9%), but the difference did not reach statistical significance. As expected, the AD group showed the highest prevalence of APOE ε4 carriers (Table 2). The mean MMSE scores were lower in AD than in FTD (p = 0.026), PSP (p = 0.028), and SNAP (p = 0.006) patients. We found no difference in the mean clinical dementia rating (CDR) between the groups except for a trend toward higher scores in AD patients than in those with PSP (p = 0.065) and SNAP (p = 0.105).

In plasma, age was associated with NfL levels in the controls (rho = 0.608, p < 0.001), with p-tau181 values in AD (rho = − 0.313, p = 0.002), and with GFAP concentrations in FTD (rho = 0.366, p = 0.005) and PSP (rho = 0.596, p < 0.001) groups. Sex showed no effect on blood and CSF biomarker values.

In line with previous studies, we found good correlations between CSF and plasma NfL values overall (rho = 0.668, p < 0.001) (Fig. 1). In particular, there was a strong association in FTD (rho = 0.749, p < 0.001), SNAP (rho = 0.722, p < 0.001), and DLB (rho = 0.720, p < 0.001) and a moderate correlation in CBS (rho = 0.635, p < 0.001).

In CSF, NfL levels were significantly increased in FTD compared to all other groups (vs. PSP, DLB, AD, and SNAP, p < 0.001; vs. CBS, p = 0.01) (Additional file 1: Table S2). We found similar trends in plasma, with significantly higher values in FTD compared to all other diagnostic groups (p < 0.001) (Fig. 1, Additional file 1: Table S3). HC showed the lowest plasma NfL levels, resulting in significant differences with each diagnostic group (p < 0.001). In the SNAP group, we detected higher plasma NfL levels in individuals presenting with prominent behavioral changes compared to those with “pure” cognitive impairment [26.7 (12.7–39.8) vs. 15.7 (9.9–22.9) pg/ml, p = 0.12] (Additional file 1: Fig. S1).

As expected, we found the highest CSF p-tau181 levels in the AD group (p < 0.001 for all comparisons). Additionally, CBS patients showed higher biomarker levels than those with FTD, PSP and SNAP (p < 0.001 for all comparisons), and DLB (p = 0.006). In line with CSF findings, plasma p-tau181 levels were significantly higher in AD than in the other groups (p < 0.001 for all comparisons) and in DLB compared with FTD (p = 0.002). HC had significantly lower plasma p-tau181 levels than AD and DLB (p < 0.001 for both comparisons) patients but comparable to those in the FTD, PSP, CBS, and SNAP groups.

Among the biomarkers analyzed, GFAP showed the weakest correlation between plasma and CSF levels (rho = 0.325, p < 0.001). Accordingly, we found only a weak to fair correlation in the FTD (rho = 0.299, p = 0.021), SNAP (rho = 0.368, p = 0.010), and DLB (rho = 0.452, p = 0.001) participants. In CSF, GFAP did not differ among the diagnostic groups. Still, plasma levels were significantly higher in AD than in FTD, PSP, and SNAP (p < 0.001 for all comparisons) (Fig. 1). Finally, HC had the lowest plasma GFAP values compared with the other groups (p < 0.001 for all comparisons).

After stratifying FTD patients according to the phenotype, there were no significant differences in either CSF or plasma biomarker values (Additional file 1: Table S4). Although both plasma and CSF NfL showed the highest levels in the FTD+ALS, they did not reach statistical significance, likely because of the few cases analyzed and the variability within the group itself.

A monogenic disease was most represented in the FTD group (Table 2); therefore, we also compared pathogenic mutations in the former group, besides evaluating biomarker levels between genetic and sporadic cases. We found higher CSF NfL values in the genetic cohort than in the sporadic group (p = 0.013) and, within genetic FTD, a higher increase in plasma GFAP levels in GRN mutation carriers than in individuals with C9orf72 or other mutations (p = 0.029 and p = 0.036, respectively) (Additional file 1: Table S5).

In the discrimination between the HC and disease groups, ROC curve analysis demonstrated high accuracy for both plasma NfL and GFAP, with an area under the curve (AUC) values ranging from 0.948 (vs. CBS) to 0.778 (vs. SNAP) for NfL and from 0.942 (vs. CBS) to 0.740 (vs. SNAP) for GFAP (Additional file 1: Fig. S2). Plasma p-tau181, instead, showed the overall highest accuracy in distinguishing HC from AD (AUC 0.971), but had a lower performance than NfL and GFAP in separating HC from the other groups (AUC range 0.533–0.661) (Table 3).

In the distinction between the disease groups, plasma NfL showed a moderate accuracy in differentiating FTD from the other disease groups (AD+PSP+CBS+DLB) (cutoff > 31.3 pg/ml, sensitivity 72.9%, specificity 74.3%, AUC 0.761) with a similar diagnostic performance against each group (Fig. 2). As for the distinction from HC, plasma p-tau181 showed the highest accuracy in discriminating AD from the other disease groups (FTD+PSP+CBS+DLB: cutoff > 1.98 pg/ml, sensitivity 86.6%, specificity 80.0%, AUC 0.889), in particular from FTD (AUC 0.964) and PSP (AUC 0.916), while its diagnostic value was lower for CBS (AUC 0.854) and DLB (AUC 0.806).

The comparison of biomarker performance between CSF and plasma revealed an almost identical accuracy for NfL in the discrimination between FTD and AD, PSP, or CBS, with CSF NfL performing slightly better only in the distinction between FTD and DLB (AUC 0.831 vs. 0.756, p = 0.020) (Additional file 1: Fig. S3). According to the ROC curves, CSF p-tau181 showed a greater accuracy than plasma p-tau181 only in the distinction between AD and DLB (AUC 0.951 vs. 0.806, p < 0.001). In contrast to p-tau181, GFAP showed higher overall diagnostic accuracy in plasma than in CSF (AUC 0.703 vs. 0.584, p = 0.005) (Table 3), especially in the distinction between AD and FTD (AUC 0.632 vs. 0.818, p < 0.001) and between AD and PSP (AUC 0.575 vs. 0.765, p = 0.008).

Patients in the AD group showed the highest CSF t-tau and p-tau181 values and the lowest mean Aβ42/40 ratio (Additional file 1: Table S2). CSF Aβ40 was positively associated with age (rho = 0.187, p < 0.001). Analyzing the whole disease cohort, we found a significant negative correlation between Aβ42/40 ratio and plasma p-tau181 (rho − 0.631, p < 0.001, rho − 0.197, p = 0.017 after excluding the T+ cases) and between Aβ42/40 ratio and plasma GFAP (rho − 0.471, p < 0.001, rho − 0.189, p = 0.022 after excluding the T+ cases).

As previously demonstrated, the diagnostic accuracy of both plasma p-tau181 and GFAP was lower in distinguishing AD from CBS and DLB than FTD and PSP. Notably, the former groups had greater prevalence of amyloid co-pathology as disclosed by the A/T/N classification (A+: CBS vs. FTD, p < 0.001; vs. PSP, p = 0.004; DLB vs. FTD, p < 0.001; vs. PSP, p = 0.023) (Table 2). Therefore, we evaluated plasma biomarkers in CBS and DLB groups after stratifying individuals according to their amyloid status (A+ vs. A−) (Table 4, Additional file 1: Fig. S4). In both groups, plasma p-tau181 and GFAP showed significantly higher levels in A+ cases. In contrast, there was no association between plasma NfL and amyloid status. We then repeated the ROC analyses for p-tau181 and GFAP after excluding the patients with CSF A+ and found a significant improvement in the diagnostic accuracy for AD vs. CBS (p-tau181 AUC from 0.854 to 0.972; GFAP AUC from 0.616 to 0.758) and AD vs. DLB (p-tau181 AUC from 0.806 to 0.905; GFAP AUC from 0.578 to 0.707).

The RT-QuIC revealed a positive α-syn seeding activity in the CSF of 47 out of 49 (95.9%) patients in the DLB group, in 15/97 (15.5%) in AD, and a single subject also showing a CSF AD profile (A+T+N+) in the CBS group. Additionally, we found α-syn seeding activity in the CSF of a SNAP patient presenting with major (amnestic multiple domain) cognitive decline and visual hallucinations, fulfilling the criteria for “possible” DLB. In contrast, the α-syn RT-QuIC assay was invariably negative in FTD and PSP subjects.

Within the DLB group, most cases showed a full (4/4) response (n = 41, 87.2%), 4 a 3/4 positivity (8.5%), and 2 a 2/4 response (4.2%). We found a significantly lower percentage of 4/4 (n = 5, 33.3%, p < 0.001) response and a higher prevalence of 3/4 (n = 6, 40%, p = 0.009) and 2/4 (n = 4, 26.7%, p = 0.026) responses in AD patients compared to those with DLB.

After stratifying the AD subgroup according to α-syn co-pathology, we found no significant difference in demographic features or plasma biomarker levels (Table 5).

In the overall disease cohort, all plasma biomarkers were associated with CDR score (p < 0.001) and impairment of daily life activities (p < 0.001). Additionally, we found that both p-tau181 and GFAP values were negatively associated with MMSE score (rho = − 0.269, p < 0.001, and rho = − 0.294, p < 0.001, respectively). The BBDM score was correlated only with GFAP (rho = − 0.194, p = 0.008), while p-tau181 was associated with the APOE ε4 status (p = 0.001). In the SNAP group, plasma NfL levels were higher in individuals with impairment of daily life activities (p = 0.054) and correlated with the MMSE (rho = − 0.327, p = 0.028) and CDR scores (rho = 0.304, p = 0.016). Of note, after stratifying the DLB group according to the amyloid status, we found a lower mean MMSE score in the DLB A+ group than in DLB A− (Table 4), indicating a possible contribution of AD co-pathology worsening cognitive performance.