Plasma NFL, GFAP, sTREM2, pTau231, and clinical severity in progressive supranuclear palsy Authors: Seungmin Lee, Heejung Kim, Ryul Kim, Bora Jin, Seoyeon Kim, Su Hyeon Ha, Jung Hwan Shin, Beomseok Jeon, Han-Joon Kim, Jee-Young Lee Published: 2025 ABSTRACT ObjectivesThis study aimed to investigate four plasma markers, pTau231, sTREM2, neurofilament light chain (NFL), and glial fibrillary acidic protein (GFAP), in relation to disease severity in a Korean cohort of progressive supranuclear palsy (PSP).MethodsBaseline data from patients with probable PSP enrolled in the SCANPSP cohort (NCT05579301) were compared with those of age-matched healthy controls (HC). Four plasma markers measured using the Simoa method, clinical findings, and volumetric 3 T MRI were investigated.ResultsThis study included 68 patients with PSP and 30 HC. Group level difference were found in NFL, correlating with the PSP-rating scale (r = 0.48, p = 2.0·10–4) and global cognitive scores (r = −0.34, p = 0.0061 for Mini-Mental Status Exam: r = −0.32, p = 0.0088 for Montreal Cognitive Assessment). Intercorrelations between NFL, pTau231, and GFAP were observed in patients with PSP but no intercorrelations in HC. The combination of pTau231, NFL, and GFAP yielded the highest ability to distinguish the PSP-Richarson syndrome (PSP-RS) from the PSP-subcortical, with an area under the curve of 0.80. The gray matter voxel-wise correlation of whole patient data showed that NFL correlated with third ventricle enlargement and frontal/cingulate atrophy, whereas sTREM2 correlated with third ventricle and thalamic volumes. White matter analyses revealed that the supplementary motor area and vermis correlated with NFL in whole PSP. In PSP-RS group, the superior frontal gyrus was associated with NFL, and the cerebellar crus with sTREM2.ConclusionThis study demonstrated the feasibility of using four plasma markers simultaneously to determine PSP disease severity. To verify these findings, longitudinal analyses are necessary.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00415-025-13570-7. Introduction Progressive supranuclear palsy (PSP) is a rapidly progressive primary 4R-tauopathy characterized by motor dysfunction, postural instability, vertical gaze palsy, and cognitive decline [1]. No disease-modifying therapy has been effective, and in its early stages, PSP can be difficult to differentiate from other Parkinsonian disorders [2, 3]. Fluid-based markers are attractive tools for monitoring disease progression and differential diagnosis of PSP. The neurofilament light chain (NFL) in the blood and cerebrospinal fluid (CSF) is one of the most widely investigated markers [4]. NFL is released in greater amounts during axonal damage or neuronal degeneration than under physiological conditions [5]. Meta-analyses and studies on PSP have demonstrated robust correlations between CSF and plasma NFL concentrations, particularly when using high-sensitivity assays such as a single-molecule array (Simoa) or electrochemiluminescence [6–10]. Plasma NFL has been reported to correlate with clinical severity and quality of life in patients with PSP [3, 7, 10–12], and high NFL is associated with poor survival [11]. NFL elevation has also been reported to be associated with brain atrophy [7, 8], which may differentiate between PSP subtypes [12]. Glial Fibrillary Acidic Protein (GFAP), a cytoskeletal intermediate filament, is mainly expressed in astrocytes and is elevated under neurodegenerative conditions [13]. Plasma GFAP levels are elevated in patients with PSP compared to healthy controls (HC) [3], correlating with disease severity [14, 15] and atrophy of the striatum and brainstem [3]. Regarding fluid-based tau markers, blood pTau181 showed no correlation with disease severity [3, 15, 16], but one study showed a correlation with the cortical tau burden measured with F18-flurzolotau [17]. Meanwhile, CSF pTau181 correlated with Schwab and England Activities of Daily Living (SEADL) scores and measures of executive function; it was not associated with the other indicators of PSP symptoms [8]. Plasma pTau231, a sensitive marker for Alzheimer’s disease (AD) [18], has not been well investigated in the PSP population and has shown limited value in other tauopathies, such as frontotemporal lobar degeneration [19, 20]. Alongside tau-related biomarkers, plasma soluble Triggering Receptor Expressed on Myeloid Cells 2 (sTREM2) plays a crucial role in microglial activation in AD, where it has been shown to attenuate tau phosphorylation and cognitive dysfunction [21, 22]. sTREM2 mRNA has been reported to be upregulated in the substantia nigra of patients with PSP [23], but blood sTREM2 and the clinical implications in the PSP population remain limited. This study aimed to investigate the clinical usability of four plasma biomarkers, NFL, GFAP, pTau231, and sTREM2, simultaneously measured in a PSP cohort, for differentiating patients from healthy controls and across different subtypes of PSP. The clinical and neuroimaging correlations of each biomarker were also analyzed. This study may help elucidate the underlying disease mechanisms and contribute to a more precise biomarker-based PSP characterization. Methods Study participants and data collection This study included patients from an ongoing multicenter cohort study in South Korea (Study of Comprehensive ANd multimodal marker-based cohort of PSP [SCAN-PSP, NCT05579301]), who were enrolled between January 2021 and December 2024. Clinical diagnosis and classification of PSP were performed by three movement disorder specialists (RK, JYL, and HJK) based on the 2017 MDS-PSP diagnostic criteria [24]. Only the patients for whom plasma samples were available were included in the analyses. Detailed demographic and clinical parameters, including PSP-rating scale (PSPRS), SEADL, Korean Mini-Mental State Examination (MMSE), and Korean Montreal Cognitive Assessment (MoCA) scores, were collected from medical records. HC with available plasma samples were included from previously established cohorts (Institutional Review Board (IRB) No. 20–2021-118). The study protocol was approved by the IRB of the two participating institutions (IRB No. 20‐2021‐121 and 2201–119–1294). All subjects provided written informed consent before enrollment in this study, which was conducted in accordance with the Declaration of Helsinki. Plasma biomarker assay Peripheral blood samples in EDTA tubes were collected from each participant. Samples were centrifuged for 15 min at room temperature, at 2500 × g, and stored in cryotubes at −4 °C. The concentration of plasma NFL, GFAP, sTREM2, and pTau231 was measured using ultrasensitive single-molecule array (Simoa®) technology (Quanterix, Billerica, MA, USA) on the automated Simoa HD-X platform. The average of the duplicate measurements was used as the final result. MRI acquisition and volumetric analysis All participants underwent 3.0 T MRI scanning (Ingenia 3 T CX), which included a prepared rapid gradient-echo sequence. Volumetric T1 images were obtained (repetition time [TR]: 8.3 ms, echo time [TE]: 4.6 ms, flip angle: 8.0°). The acquisition matrix was 222 × 220 × 180 with overlapping slice acquisition (slice thickness: 1 mm, spacing: 0.5 mm), which was reconstructed into a 360 × 480 × 480 matrix. The voxel size was 0.5 × 0.45 × 0.45mm3. We used the automated anatomical labeling 3 (AAL3) atlas (http://oxcns.org/aal3.html) to segment the frontal, parietal, temporal, and occipital cortices; thalamus; and total intracranial volume (TIV). The neuromorphometrics atlas (https://www.neuromorphometrics.com) was utilized for three striatal structures (caudate nucleus, pallidum, and putamen), and the automatic subcortical segmentation atlas (https://surfer.nmr.mgh.harvard.edu/fswiki/FreeSurferVersion3) was used to estimate brainstem and third ventricle volumes. Neuromorphometrics and AAL3 analyses were conducted using the Computation Anatomy Toolbox (CAT) 12 software (version 8.2 r2170) in MATLAB (version 9.14). Subcortical volume segmentation was performed using FreeSurfer (version 7.1.1). Voxel-based correlations between brain morphometry and plasma biomarkers To investigate the relationship between the four biomarkers and brain tissue volume (gray matter [GM] and white matter [WM]), we performed voxel-wise multiple regression analyses, including age, sex, and TIV as nuisance variables. Significant correlations between macroscopic brain atrophy and the four biomarkers were identified using the Gaussian random field (GRF) theory correction for multiple comparisons (voxel-level p value < 0.01, cluster-level p value < 0.05). Statistical analysis The normality of the data was tested using the Shapiro–Wilk test. As the four biomarkers did not follow a normal distribution, we applied a log10 transformation to achieve normality in all correlation analyses. However, raw plasma concentrations were used for group-wise biomarker comparisons. To compare two continuous variables, either the t test or the Wilcoxon rank-sum test was used, depending on the data distribution. Differences in categorical variables were assessed using the Chi-squared test. Group comparisons of plasma biomarker levels were performed using ANCOVA, adjusting for age, sex, and disease duration, and partial correlation analyses were performed with covariates (e.g., age, sex, and/or TIV) selected according to the specific context of each analysis. The resulting p values from the ANCOVA and partial correlation analyses adjusted for the specified covariates are hereafter referred to as adjusted p values. To evaluate the discriminative performance of individual plasma biomarkers and their combinations for differentiating PSP-RS from the PSP-subcortical group, receiver-operating characteristic (ROC) curve analyses were performed. Logistic regression models were constructed using each biomarker alone and in all possible combinations of the four plasma biomarkers, with age, sex, and disease duration included as covariates. For each model, the predicted probability was used to compute the area under the curve (AUC) and the 95% confidence interval (CI). The combination with the highest AUC was identified as the best-performing model. The differences in AUC between models with and without demographic variables were evaluated using DeLong’s test for correlated ROC curves. For cluster analysis, a Gaussian mixture model (GMM) was performed with the mclust package in R. Four plasma biomarkers were used as input, and the optimal number of clusters was selected based on the Bayesian Information Criterion. Multiple comparisons were corrected using the Benjamini–Hochberg (BH) method to obtain corrected p values. All statistical tests were two-tailed, with a significance threshold of 0.05. Statistical analyses were conducted using R software (version 4.3.3, https://www.r-project.org). Results Demographics and clinical characteristics of participants Sixty-eight patients with PSP and 30 healthy controls. The patient group had a significantly higher proportion of males, and lower MMSE scores than the HC group (Table 1). The patients were further classified into PSP-Richardson syndrome (PSP-RS, n = 44), PSP with predominant Parkinsonism (PSP-P, n = 15), PSP with progressive gait freezing (PSP-PGF, n = 7), PSP with speech/language disorder (PSP-SL, n = 1), and PSP with prominent ocular motor dysfunction (PSP-OM, n = 1). PSP-P and PSP-PGF were categorized as subcortical subtypes, while we excluded participants with PSP-SL and PSP-OM from subgroup analyses because of the small sample size. Age and disease duration did not differ between the PSP-RS and PSP-subcortical groups; however, the PSP-subcortical group had a significantly higher proportion of males. Participants with PSP-RS demonstrated lower MMSE scores and worse clinical parameters, as reflected by the SEADL and total PSPRS scores (Table 1). Table 1Demographics and characteristics of study participantsHC (n = 30)PSP (n = 68)p value (PSP vs HC)PSP-RS (n = 44)PSP-subcortical (n = 22)p value (R vs subcortical)Age at enroll#69.6 ± 6.272.1 ± 6.50.07571.8 ± 6.373.1 ± 6.90.49M:F###, ***7:2349:191.9·10–527:1720:20.019Disease duration−4.5 ± 2.24.5 ± 2.24.8 ± 2.30.55MMSE##, **28.6 ± 2.1 (n = 25)24.7 ± 4.82.6·10–523.8 ± 5.426.4 ± 3.10.049MoCA*−19.4 ± 6.318.4 ± 6.921.4 ± 5.00.052SEADL (%)**−64.9 ± 17.558.4 ± 16.376.4 ± 13.36.3·10–5Total PSPRS**−33.9 ± 13.138.4 ± 12.825.4 ± 8.71.7·10–5History**−7.3 ± 4.28.4 ± 4.45.1 ± 3.02.0·10–3Mentation**−3.4 ± 3.24.1 ± 3.62.1 ± 1.60.034Bulbar**−3.3 ± 1.63.7 ± 1.42.4 ± 1.52.9·10–3Ocular**−6.6 ± 2.97.4 ± 3.15.1 ± 2.02.0·10–3Limb**−4.5 ± 2.04.7 ± 2.24.0 ± 1.40.29Gait**−8.8 ± 3.910.2 ± 3.96.6 ± 2.81.8·10–4#Total PSP vs HC, p < 0.05, t test. ##Total PSP vs. HC, p < 0.05, Mann–Whitney U test; ###Total PSP vs. HC, p < 0.05, Pearson’s chi-square test**PSP-RS vs PSP-subcortical, p < 0.05, Mann–Whitney test, ***PSP-RS vs PSP-subcortical, p < 0.05 Fisher’s exact testData are presented with ± mean standard deviationF Female, HC Healthy Controls, H&Y Hoehn and Yahr, M Male, MMSE Mini-Mental State Examination, MoCA Montreal Cognitive Assessment, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, SEADL Schwab and England Activities of Daily Living Scale Table 1Demographics and characteristics of study participantsHC (n = 30)PSP (n = 68)p value (PSP vs HC)PSP-RS (n = 44)PSP-subcortical (n = 22)p value (R vs subcortical)Age at enroll#69.6 ± 6.272.1 ± 6.50.07571.8 ± 6.373.1 ± 6.90.49M:F###, ***7:2349:191.9·10–527:1720:20.019Disease duration−4.5 ± 2.24.5 ± 2.24.8 ± 2.30.55MMSE##, **28.6 ± 2.1 (n = 25)24.7 ± 4.82.6·10–523.8 ± 5.426.4 ± 3.10.049MoCA*−19.4 ± 6.318.4 ± 6.921.4 ± 5.00.052SEADL (%)**−64.9 ± 17.558.4 ± 16.376.4 ± 13.36.3·10–5Total PSPRS**−33.9 ± 13.138.4 ± 12.825.4 ± 8.71.7·10–5History**−7.3 ± 4.28.4 ± 4.45.1 ± 3.02.0·10–3Mentation**−3.4 ± 3.24.1 ± 3.62.1 ± 1.60.034Bulbar**−3.3 ± 1.63.7 ± 1.42.4 ± 1.52.9·10–3Ocular**−6.6 ± 2.97.4 ± 3.15.1 ± 2.02.0·10–3Limb**−4.5 ± 2.04.7 ± 2.24.0 ± 1.40.29Gait**−8.8 ± 3.910.2 ± 3.96.6 ± 2.81.8·10–4#Total PSP vs HC, p < 0.05, t test. ##Total PSP vs. HC, p < 0.05, Mann–Whitney U test; ###Total PSP vs. HC, p < 0.05, Pearson’s chi-square test**PSP-RS vs PSP-subcortical, p < 0.05, Mann–Whitney test, ***PSP-RS vs PSP-subcortical, p < 0.05 Fisher’s exact testData are presented with ± mean standard deviationF Female, HC Healthy Controls, H&Y Hoehn and Yahr, M Male, MMSE Mini-Mental State Examination, MoCA Montreal Cognitive Assessment, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, SEADL Schwab and England Activities of Daily Living Scale Demographics and characteristics of study participants #Total PSP vs HC, p < 0.05, t test. ##Total PSP vs. HC, p < 0.05, Mann–Whitney U test; ###Total PSP vs. HC, p < 0.05, Pearson’s chi-square test **PSP-RS vs PSP-subcortical, p < 0.05, Mann–Whitney test, ***PSP-RS vs PSP-subcortical, p < 0.05 Fisher’s exact test Data are presented with ± mean standard deviation F Female, HC Healthy Controls, H&Y Hoehn and Yahr, M Male, MMSE Mini-Mental State Examination, MoCA Montreal Cognitive Assessment, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, SEADL Schwab and England Activities of Daily Living Scale Plasma biomarkers in PSP compared to HC In the overall PSP group, plasma NFL was significantly elevated compared with HC (corrected-p = 2.1·10–5), whereas the other three markers did not exhibit significant differences (corrected p > 0.05, Fig. 1a–d). None of the four plasma biomarkers showed significant correlations with disease duration. In the PSP group, sTREM2 exhibited a significant positive correlation with age, whereas in the HC group, NFL and pTau231 were significantly correlated with age (Supplementary Fig. 1). No sex differences were observed in any of the biomarkers in either the PSP or HC group (Supplementary Fig. 2).Fig. 1Boxplots comparing plasma biomarker levels between PSP and HC (a–d), and subgroup analysis of NFL was performed among PSP-RS, PSP-sub, and HC (e), after adjusting for age, sex, and disease duration. Statistical significance is indicated as follows: ns = not significant; *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively. HC healthy control, GFAP glial fibrillary acidic protein, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSP-sub PSP-subcortical group, sTREM2 soluble triggering receptor expressed in myeloid cells 2 Boxplots comparing plasma biomarker levels between PSP and HC (a–d), and subgroup analysis of NFL was performed among PSP-RS, PSP-sub, and HC (e), after adjusting for age, sex, and disease duration. Statistical significance is indicated as follows: ns = not significant; *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively. HC healthy control, GFAP glial fibrillary acidic protein, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSP-sub PSP-subcortical group, sTREM2 soluble triggering receptor expressed in myeloid cells 2 Significant group effects for NFL (adjusted-p = 4.0·10⁻10) and GFAP (adjusted-p = 0.043) were observed in the subgroup analysis. Specifically, NFL levels were significantly elevated in the PSP-RS group relative to both the PSP-subcortical and HC groups in post hoc comparisons (Fig. 1e). For GFAP, no significant intersubgroup differences were detected in the post hoc analysis. The overall group effect for pTau231 and sTREM2 was also insignificant (adjusted-p = 0.13 pTau231, adjusted-p = 0.75, sTREM2). Intercorrelations of four plasma biomarkers and its discriminative abilities Correlation analysis among the four biomarkers revealed that NFL exhibited significant positive correlations with pTau231 and GFAP, whereas GFAP was positively correlated with the other two biomarkers in the overall PSP group. In the PSP-RS group, significant positive correlations were observed between NFL and GFAP levels, whereas in the PSP-subcortical group, pTau231 and GFAP levels were significantly correlated positively. There were no significant intercorrelations among the four biomarkers in the HC group (Fig. 2a–d).Fig. 2Correlation matrix of four biomarkers in (a) the total PSP group, b the PSP-RS group, c the PSP-subcortical group, and d HC. e ROC curve illustrating the discriminatory performance of the optimal biomarker combination (NFL, GFAP, and pTau231) with three clinical parameters (age, sex, and disease duration) for differentiating PSP-RS from PSP-subcortical. All biomarkers were log10-transformed and correlations were assessed using partial correlation with age and sex as covariates. The color scale within each cell represents the strength of the correlation coefficients. The Benjamini–Hochberg correction was applied for multiple comparisons. *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively. GFAP glial fibrillary acidic protein, HC healthy control, NFL neurofilament light chain, PSP, progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, ROC receiver-operating characteristic, sTREM2 soluble triggering receptor expressed on myeloid cells 2 Correlation matrix of four biomarkers in (a) the total PSP group, b the PSP-RS group, c the PSP-subcortical group, and d HC. e ROC curve illustrating the discriminatory performance of the optimal biomarker combination (NFL, GFAP, and pTau231) with three clinical parameters (age, sex, and disease duration) for differentiating PSP-RS from PSP-subcortical. All biomarkers were log10-transformed and correlations were assessed using partial correlation with age and sex as covariates. The color scale within each cell represents the strength of the correlation coefficients. The Benjamini–Hochberg correction was applied for multiple comparisons. *, **, and *** denote p < 0.05, p < 0.01, and p < 0.001, respectively. GFAP glial fibrillary acidic protein, HC healthy control, NFL neurofilament light chain, PSP, progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, ROC receiver-operating characteristic, sTREM2 soluble triggering receptor expressed on myeloid cells 2 To evaluate the discriminative ability of plasma biomarkers for distinguishing PSP-RS from PSP-subcortical subtypes, ROC curve analyses were performed. Biomarker-only models achieved moderate discriminative performance, with AUCs around 0.70. The inclusion of demographic variables (age, sex, and disease duration) further increased AUC values across most biomarker combinations, although these improvements did not reach statistical significance (DeLong p > 0.05). Only models using pTau231, sTREM2, and their combination exhibited significant improvements (Fig. 2e, Supplementary Table 2). Among the individual biomarkers, NFL showed the strongest contribution to the classification performance. Cluster analysis was performed using GMM, and identified two distinct groups: one with more severe disease (n = 16) and another with less severe disease (n = 52). The severe group showed significant elevations in all three biomarkers except sTREM2 (Supplementary Table 3). Clinical correlates of plasma biomarkers in PSP In the PSP group, NFL levels exhibited significant correlations with the four clinical measures of disease severity (negative correlation with MMSE, MoCA, SEADL, and positive correlation with total PSPRS) after multiple comparison correction (corrected p < 0.01 all) (Supplementary Table 1). A detailed analysis of the PSPRS subdomain scores revealed that NFL levels were significantly associated with all subdomain scores, except the limb score, and that GFAP demonstrated a significant positive correlation with the mentation domain scores in the overall PSP group (Table 2). pTau231 showed a positive tendency to correlate with mentation score (adjusted p = 0.025, Table 2). Table 2Correlations between plasma markers and PSPRS subscoresBiomarkerPSPRS subscoreTotalPSP-RSPSP-subcorticalrprprppTau231History0.190.13 − 0.0100.950.480.033Mentation0.280.0250.220.160.270.24Bulbar − 0.0210.87 − 0.150.34 − 0.110.64Ocular0.160.190.0370.810.160.49Limb0.0330.79 − 0.0430.79 − 0.0420.86Gait0.110.37 − 9.6·10−41.00.180.45sTREM2History − 0.0490.7 − 0.0910.57 − 0.0120.96Mentation0.0420.740.100.52 − 0.180.45Bulbar − 0.120.34 − 0.0990.53 − 0.150.52Ocular0.0670.590.00960.950.190.43Limb0.0710.570.0440.780.140.55Gait − 0.220.08 − 0.240.13 − 0.280.24NFLHistory0.37*0.00210.200.20.370.11Mentation0.48*3.9·10−50.50*6.6·10−40.110.66Bulbar0.33*0.00710.240.130.00160.99Ocular0.32*0.00870.290.067 − 0.140.55Limb0.210.090.140.390.00960.97Gait0.31*0.0130.270.0840.0810.73GFAPHistory0.160.19 − 0.00910.950.260.27Mentation0.33*0.0070.340.029 − 0.0160.95Bulbar0.110.37 − 0.0170.91 − 0.170.46Ocular0.120.320.0910.57 − 0.230.33Limb − 0.0210.87 − 0.190.240.230.34Gait0.0990.43 − 0.120.430.340.14All biomarkers were log10-transformed and correlations were assessed using partial correlations with age and sex as covariates*p < 0.05, after Benjamini–Hochberg correctionGFAP Glial fibrillary acidic protein, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, sTREM2 soluble triggering receptor expressed on myeloid cells 2 Table 2Correlations between plasma markers and PSPRS subscoresBiomarkerPSPRS subscoreTotalPSP-RSPSP-subcorticalrprprppTau231History0.190.13 − 0.0100.950.480.033Mentation0.280.0250.220.160.270.24Bulbar − 0.0210.87 − 0.150.34 − 0.110.64Ocular0.160.190.0370.810.160.49Limb0.0330.79 − 0.0430.79 − 0.0420.86Gait0.110.37 − 9.6·10−41.00.180.45sTREM2History − 0.0490.7 − 0.0910.57 − 0.0120.96Mentation0.0420.740.100.52 − 0.180.45Bulbar − 0.120.34 − 0.0990.53 − 0.150.52Ocular0.0670.590.00960.950.190.43Limb0.0710.570.0440.780.140.55Gait − 0.220.08 − 0.240.13 − 0.280.24NFLHistory0.37*0.00210.200.20.370.11Mentation0.48*3.9·10−50.50*6.6·10−40.110.66Bulbar0.33*0.00710.240.130.00160.99Ocular0.32*0.00870.290.067 − 0.140.55Limb0.210.090.140.390.00960.97Gait0.31*0.0130.270.0840.0810.73GFAPHistory0.160.19 − 0.00910.950.260.27Mentation0.33*0.0070.340.029 − 0.0160.95Bulbar0.110.37 − 0.0170.91 − 0.170.46Ocular0.120.320.0910.57 − 0.230.33Limb − 0.0210.87 − 0.190.240.230.34Gait0.0990.43 − 0.120.430.340.14All biomarkers were log10-transformed and correlations were assessed using partial correlations with age and sex as covariates*p < 0.05, after Benjamini–Hochberg correctionGFAP Glial fibrillary acidic protein, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, sTREM2 soluble triggering receptor expressed on myeloid cells 2 Correlations between plasma markers and PSPRS subscores All biomarkers were log10-transformed and correlations were assessed using partial correlations with age and sex as covariates *p < 0.05, after Benjamini–Hochberg correction GFAP Glial fibrillary acidic protein, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-RS PSP-Richardson syndrome, PSPRS Progressive Supranuclear Palsy Rating Scale, sTREM2 soluble triggering receptor expressed on myeloid cells 2 In the PSP-RS subgroup, NFL was significantly correlated with the total PSPRS and ADL scores, whereas no significant correlations were observed in the PSP-subcortical group (Supplementary Table 4). In the subdomain analysis, NFL was significantly correlated with mentation domain scores in the PSP-RS group, but not in the PSP-subcortical group. pTau231 tended to correlate with history domain scores in the PSP-subcortical group (adjusted p = 0.025, corrected p for multiple comparisons > 0.05) (Table 2). Neuroimaging correlates of plasma biomarkers in PSP In the overall PSP group, atlas-based volumetric analyses revealed significant correlations between plasma biomarkers and regional brain atrophy. Third ventricular enlargement was positively correlated with NFL levels (adjusted p = 0.010, corrected p = 0.038) and negatively correlated with sTREM2 levels (adjusted p = 0.024, corrected p = 0.049) (Fig. 3a–b). Higher NFL levels were associated with frontal cortical (r = − 0.37, adjusted p = 0.0030, corrected p = 0.029), midbrain (r = − 0.32, adjusted p = 0.0090, corrected p = 0.036), and thalamic atrophies (r = − 0.32, adjusted p = 0.01, corrected p = 0.036). Trends toward associations were also observed between NFL and temporal (r = − 0.28, adjusted p = 0.023, corrected p = 0.052), occipital (r = − 0.28, adjusted p = 0.026, corrected p = 0.052), and pontine atrophy (r = − 0.27, adjusted p = 0.028, corrected p = 0.052). There was also a tendency for thalamic volume to increase with higher sTREM2 levels (r = 0.32, adjusted p = 0.010, corrected p = 0.052).Fig. 3Associations between plasma biomarkers and brain structural measures in patients with PSP, including third ventricular volume. a-b. Scatter plots showing the partial correlations between third ventricular volume and plasma biomarker levels for NFL (a), sTREM2 (b) in the overall PSP group. c-i. Voxel-based morphometry (VBM) results corrected using GRF theory, illustrating brain regions associated with plasma biomarker levels. For the total PSP group, gray matter correlates are displayed for NFL (c) and sTREM2 (d), and white matter correlates are displayed for NFL (e- f). For the PSP-RS group, gray matter correlates are displayed for NFL (g), and white matter correlates are displayed for NFL (h) and sTREM2 (i). All biomarkers were log10-transformed and partial correlation analyses were performed, adjusting for age, sex, and TIV as covariates. The Benjamini–Hochberg correction was applied for multiple comparisons. A voxel-level threshold of p < 0.01 was applied, followed by a cluster-level threshold of p < 0.05 for statistical significance. Abbreviations: GFAP glial fibrillary acidic protein, GRF Gaussian random field correction, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-OM PSP with prominent ocular motor dysfunction, PSP-RS PSP-Richardson’s syndrome, PSP-SL PSP with speech/language disorder, TIV total intracranial volume, sTREM2 soluble triggering receptor expressed on myeloid cells 2, VBM voxel-based morphometry Associations between plasma biomarkers and brain structural measures in patients with PSP, including third ventricular volume. a-b. Scatter plots showing the partial correlations between third ventricular volume and plasma biomarker levels for NFL (a), sTREM2 (b) in the overall PSP group. c-i. Voxel-based morphometry (VBM) results corrected using GRF theory, illustrating brain regions associated with plasma biomarker levels. For the total PSP group, gray matter correlates are displayed for NFL (c) and sTREM2 (d), and white matter correlates are displayed for NFL (e- f). For the PSP-RS group, gray matter correlates are displayed for NFL (g), and white matter correlates are displayed for NFL (h) and sTREM2 (i). All biomarkers were log10-transformed and partial correlation analyses were performed, adjusting for age, sex, and TIV as covariates. The Benjamini–Hochberg correction was applied for multiple comparisons. A voxel-level threshold of p < 0.01 was applied, followed by a cluster-level threshold of p < 0.05 for statistical significance. Abbreviations: GFAP glial fibrillary acidic protein, GRF Gaussian random field correction, NFL neurofilament light chain, PSP progressive supranuclear palsy, PSP-OM PSP with prominent ocular motor dysfunction, PSP-RS PSP-Richardson’s syndrome, PSP-SL PSP with speech/language disorder, TIV total intracranial volume, sTREM2 soluble triggering receptor expressed on myeloid cells 2, VBM voxel-based morphometry In the PSP-RS subgroup, NFL levels tended to correlate with thalamic (r = − 0.39, adjusted p = 0.013, corrected p = 0.10) and frontal cortical atrophy (r = − 0.37, adjusted p = 0.018, corrected p = 0.10). In the PSP-subcortical group, sTREM2 showed a trend toward association with third ventricular enlargement (r = − 0.46, adjusted p = 0.049, corrected p = 0.20), and GFAP exhibited a trend toward association with pallidal atrophy (r = − 0.57, adjusted p = 0.011, corrected p = 0.122). Voxel-wise VBM analyses further supported these findings. GM-VBM analysis (voxel-level p < 0.01, cluster-level p < 0.05, GRF corrected) showed that NFL levels were negatively associated with atrophy in the thalamus, parahippocampal region, and cingulate gyrus (Fig. 3c), while sTREM2 demonstrated a significant positive correlation with thalamic GM volume (Fig. 3d). WM-VBM identified negative associations between NFL and the supplementary motor area (SMA) and vermis (Fig. 3e–f). In the PSP-RS subgroup, GM-VBM demonstrated negative correlations between NFL and the thalamus, right hippocampus, and bilateral medial orbitofrontal gyri (Fig. 3g), while WM-VBM revealed a negative association between NFL and the superior frontal gyrus (Fig. 3h) and a positive association between sTREM2 and the cerebellar crus I (Fig. 3i). Discussion This study demonstrated the utility of four plasma markers (NFL, sTREM2, pTau231, and GFAP) measured simultaneously in a prospective PSP cohort sample. In addition to replicating previous studies demonstrating the utility of NFL in PSP, this study is the first to show that sTREM2 is linked to atrophic changes in the brain regions specific to PSP pathology and to the clinical severity of PSP. GFAP exhibited correlations with PSPRS sub-items. In PSP, higher serum NFL levels are associated with more severe disease, cognitive impairment, and a shorter survival [3, 11]. Additionally, we showed that NFL correlated with five of the six PSPRS subdomain scores, reaffirming its usefulness in assessing disease severity. The third ventricle volume has been identified as a robust marker of PSP disease severity, showing significant correlations with the PSPRS and SEADL, and demonstrating a large effect size in the previous studies [25, 26]. In our study, NFL levels did correlate with the third ventricular volume, supporting its role in monitoring disease progression. We also found a correlation between NFL and frontal volume in patients with PSP, which was in line with the previous reports indicating the highest effect size on frontal lobe volume in PSP [25]. Through imaging correlation analysis, we found that NFL and sTREM2 levels in PSP may reflect neurodegeneration specifically involved in PSP. For example, atrophy in the cingulate cortex, supplementary motor area, parahippocampal gyrus, thalamus, and vermis correlated well with NFL levels. The cingulate cortex, which has extensive anatomical connections with the frontal and parietal lobes, striatum, amygdala, and hypothalamus [27], has previously been linked to gait impairment and vertical gaze limitation [28, 29]. More severe atrophy has been associated with a greater downgaze limitation [26]. Pathological changes in the parahippocampal gyrus and hippocampus have also been reported in PSP [30, 31]. Moreover, higher NFL levels have been associated with cognitive impairment in patients with PSP [7]. Furthermore, alterations in the dentatorubrothalamic tract (DRTT) have been demonstrated in multiple imaging and pathological studies of patients with PSP [32]. Our study showed that NFL levels were inversely correlated with thalamic volume. The supplementary motor area (SMA) has previously been reported to be atrophied in PSP, and longitudinal data have revealed that greater SMA atrophy is associated with a higher baseline PSPRS [33]. The SMA plays a crucial role in motor initiation, gait regulation, and other processes that are prominently affected in PSP [28]. Previous studies have implicated the cerebellar vermis in PSP, reporting associations with language–semantic performance, increased mean diffusivity, and significant volume loss compared with controls [34–36], highlighting the need for further investigation of its role in PSP pathophysiology. In PSP-RS, the medial orbitofrontal cortex emerged as a volumetric correlate of the NFL, with previous studies reporting an association between atrophy in this region and impulsivity in PSP-RS [37]. The medial orbitofrontal cortex is affected relatively early in PSP, whereas the lateral orbitofrontal cortex is affected later and may be associated with more severe phenotypes [26, 38]. The right superior frontal gyrus (SFG) plays a key role in conflict anticipation and proactive inhibitory control [39]. In a study comparing PSP-RS with behavioral variant frontotemporal dementia and healthy controls, PSP-RS showed more severe gray matter loss in the frontal regions, including the superior frontal gyrus, highlighting the stronger involvement of these regions in PSP pathology than in other frontotemporal degenerations [40]. In a murine study, sTREM2 knockout mice exhibited enhanced seeding and spreading of neuritic plaque tau aggregates, suggesting that sTREM2 may play a regulatory role in tau pathology [41]. We found that sTREM2 showed significant negative correlations with third ventricle expansion and atrophy of the thalamus and cerebellar crus in patients with PSP, whereas its correlation with clinical measures was not significant. Additionally, sTREM2 was associated with age but not with disease duration, and showed no differences according to sex or PSP subgroup. In a study of patients with PD and atypical Parkinsonism, including PSP, CSF sTREM2 levels did not show significant differences or correlation with disease severity, which is consistent with our finding [42]. GFAP showed a limited correlation with disease severity in our cohort. Previous studies have reported inconsistent findings; one study found correlations with brainstem structures, but little association with clinical measures, whereas another reported association with MoCA and UPDRS motor scores [3, 15]. Taken together, these observations suggest that GFAP may not reliably track disease severity in patients with PSP, underscoring the need for further studies to clarify its role as a biomarker. Our study did not observe a significant difference in plasma GFAP levels in the between PSP and HC. Previous studies have reported elevated GFAP in patients with PSP compared to HC [15, 17]. However, GFAP concentration and its variability have been shown to increase with advancing age [43]. Because our control group was relatively older than those in previous PSP studies, age-related effects may have influenced the group comparison. Further studies in larger and age-matched cohorts are warranted to validate these findings. The intercorrelations among plasma biomarkers may provide further insight into the underlying pathological processes across PSP subtypes. In our cohort, GFAP and NFL were coupled in the PSP-RS group, whereas NFL showed a stronger association with pTau231 in the subcortical variant. These patterns suggest that different pathological mechanisms may differentially contribute to neurodegeneration in different clinical phenotypes. Although GFAP and pTau231 were not significantly associated with disease severity in our sample, their correlations with NFL indicate that these markers still track biologically relevant processes. It is therefore possible that with a larger sample size, inclusion of more patients with subcortical phenotypes, or broader representation of disease severity, additional significant associations might be found. Taken together, the observed coupling among plasma markers implies that multiple pathological pathways interact in PSP, and the relative contribution of each pathway may vary across phenotypes. This study also revealed that the combination of three PSP biomarkers (NFL, GFAP, and pTau231) provided the most robust discrimination between the PSP-RS and PSP-subcortical groups (AUC = 0.80, 95% CI = 0.68–0.91). In contrast, NFL alone—the biomarker showing the strongest correlation with disease severity—yielded an AUC of 0.78 (95% CI = 0.67–0.89). Consistently, the GMM analysis demonstrated that patients with elevated NFL concentrations were predominantly classified into the severe cluster, reinforcing the interpretation that NFL is the principal driver of disease severity, while GFAP and pTau231 provide complementary but secondary information. This study has several limitations. First, the diagnosis of PSP was not confirmed pathologically, which may have introduced misclassification. However, our findings were largely consistent with those of previously reported pathological and clinical correlation studies. Second, the insufficient number of PSP-cortical subtypes prevented us from analyzing the data for this subpopulation. Additionally, the relatively small sample size in the PSP-subcortical group may have limited the detection of significant correlations, emphasizing the need for further studies with larger cohorts. Finally, data on the longitudinal changes in biomarker levels and their effects on clinical or imaging progression were not available for the current analysis. However, our ongoing cohort will collect follow-up data, and we hope to analyze the prognostic value of plasma markers in future studies. In conclusion, our study showed that NFL levels do correlate with clinical and neuroimaging markers associated with disease progression. Additionally, the correlation of sTREM2 with the third ventricular volume and thalamic atrophy suggests a potential role of this marker in PSP. The correlation patterns observed in the PSP subgroups may aid in clinical diagnosis and indicate underlying differences in pathomechanisms among the subgroups. To verify the value of plasma biomarkers for assessing PSP, long-term longitudinal studies are required. Supplementary Information Below is the link to the electronic supplementary material.Supplementary file1 (PDF 736 KB) Supplementary file1 (PDF 736 KB)