CSF Proteomics in Patients With Progressive Supranuclear Palsy

Abstract

Background and Objectives

Identification of fluid biomarkers for progressive supranuclear palsy (PSP) is critical to enhance therapeutic development. We implemented unbiased DNA aptamer (SOMAmer) proteomics to identify novel CSF PSP biomarkers.

Methods

This is a cross-sectional study in original (18 clinically diagnosed PSP-Richardson syndrome [PSP-RS], 28 cognitively healthy controls]), validation (23 PSP-RS, 26 healthy controls), and neuropathology-confirmed (21 PSP, 52 non-PSP frontotemporal lobar degeneration) cohorts. Participants were recruited through the University of California, San Francisco, and the 4-Repeat Neuroimaging Initiative. The original and neuropathology cohorts were analyzed with the SomaScan platform version 3.0 (5026-plex) and the validation cohort with version 4.1 (7595-plex). Clinical severity was measured with the PSP Rating Scale (PSPRS). CSF proteomic data were analyzed to identify differentially expressed targets, implicated biological pathways using enrichment and weighted consensus gene coexpression analyses, diagnostic value of top targets with receiver-operating characteristic curves, and associations with disease severity with linear regressions.

Results

A total of 136 participants were included (median age 70.6 ± 8 years, 68 [50%] women). One hundred fifty-five of 5,026 (3.1%), 959 of 7,595 (12.6%), and 321 of 5,026 (6.3%) SOMAmers were differentially expressed in PSP compared with controls in original, validation, and neuropathology-confirmed cohorts, with most of the SOMAmers showing reduced signal (83.1%, 95.1%, and 73.2%, respectively). Three coexpression modules were associated with PSP across cohorts: (1) synaptic function/JAK-STAT (β = −0.044, corrected p = 0.002), (2) vesicle cytoskeletal trafficking (β = 0.039, p = 0.007), and (3) cytokine-cytokine receptor interaction (β = −0.032, p = 0.035) pathways. Axon guidance was the top dysregulated pathway in PSP in original (strength = 1.71, p < 0.001), validation (strength = 0.84, p < 0.001), and neuropathology-confirmed (strength = 0.78, p < 0.001) cohorts. A panel of axon guidance pathway proteins discriminated between PSP and controls in original (area under the curve [AUC] = 0.924), validation (AUC = 0.815), and neuropathology-confirmed (AUC = 0.932) cohorts. Two inflammatory proteins, galectin-10 and cytotoxic T lymphocyte-associated protein-4, correlated with PSPRS scores across cohorts.

Discussion

Axon guidance pathway proteins and several other molecular pathways are downregulated in PSP, compared with controls. Proteins in these pathways may be useful targets for biomarker or therapeutic development.

Introduction

Progressive supranuclear palsy (PSP) is a sporadic neurodegenerative disease within the frontotemporal dementia (FTD) clinical spectrum that affects motor function, behavior, and cognition. Its classical phenotype, PSP-Richardson syndrome (PSP-RS), features oculomotor dysfunction and postural instability and has a median survival of 6.9 years.¹ PSP-RS and other PSP clinical variants are associated with PSP neuropathology, a form of frontotemporal lobar degeneration (FTLD) with tau aggregates that features accumulation of 4-repeat tau as intraneuronal neurofibrillary tangles in the dorsal midbrain, globus pallidus, and subthalamic nucleus.² With a prevalence of 2.3–10.6/100,000, PSP-RS is relatively rare, but it is also one of the most common FTD phenotypes.³ PSP-RS is highly predictive of the underlying FTLD-tau PSP neuropathology, and for this reason, it is often the focus of tauopathy research. PSP has no cure, and therapeutic development has been difficult because of the paucity of biomarkers to differentiate it from other forms of atypical parkinsonism or to detect the disease in early stages. Previous work supports the clinical value of several neurodegeneration biomarkers in PSP. High neurofilament light chain (NfL) and low CSF p-tau181 have been associated with faster clinical progression,^4-7 but they have limited sensitivity and specificity, and alternative fluid biomarkers are needed.

DNA aptamer-based proteomics is an emerging ultrasensitive tool with potential to support biomarker discovery in PSP. Aptamers are short single-stranded oligonucleotides with the dual ability to bind polypeptides and nucleic acids with high specificity. Tagged aptamer-protein complexes can be detected with fluorescent DNA hybridization microarrays⁸ and quantify relative abundance of thousands of proteins in a single assay. The specificity of the platform is being confirmed for a growing number of targets with standard protein quantification techniques and proteogenomic association studies.^9,10 SOMAmer proteomics has been successfully used in biomarker discovery in plasma or CSF in cancer,¹¹ aging,¹² and Alzheimer disease (AD).^13-15

In this study, we used SOMAmer proteomics for multiplexed analysis of the CSF proteome in 2 independent PSP-RS clinical cohorts, PSP neuropathology-confirmed and non-PSP FTLD neuropathology-confirmed disease cases, and cognitively healthy controls. We hypothesized that SOMAmer CSF proteomics would identify distinctive signatures in PSP, potentially providing insight into disease pathobiology and new candidate biomarkers and therapeutic targets.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

Research procedures were approved by the University of California, San Francisco (UCSF) Institutional Review Board, protocol 17-23473. All participants or companions provided written informed consent to participate in this research protocol.

Participants, Clinical Evaluation, and CSF Collection

In a cross-sectional design, an original cohort enrolled 20 participants with PSP through the UCSF Memory and Aging Center (MAC) observational study of FTD. A validation cohort recruited 34 participants with PSP through the 4-Repeat Tau Neuroimaging Initiative, a multisite project for the longitudinal clinical assessment of 4-repeat tauopathies. Neuropathology-confirmed FTLD participants (17 FTLD-tau PSP, 12 FTLD-corticobasal degeneration [CBD], 5 Pick disease, and 20 FTLD-TDP [7 type A and 9 type B, 2 FTLD-TDP unclassifiable, 2 amyotrophic lateral sclerosis]; eTable 1) were recruited through the UCSF MAC Neurodegenerative Disease Brain Bank (Table 1). Cognitively healthy volunteers (28 in the original cohort and 41 in the validation cohort) were recruited through the UCSF MAC Healthy Aging cohort. All patients with PSP met clinical diagnostic criteria for probable PSP-RS.¹⁶ Clinical disease severity was measured with the Clinical Dementia Rating Staging Instrument (CDR) plus behavioral and language domains of the National Alzheimer's Coordinating Center FTLD module (CDR + NACC/FTLD) global and sum of boxes scores¹⁷ and with the PSPRS.¹⁸ CSF was collected at baseline, following a previously described protocol detailed in eMethods.⁷

Table.

Demographic and Clinical Characteristics of Study Participants

	Original cohort			Validation cohort			Neuropathology-confirmed cohort
	Control (n = 28)	PSP-RS (n = 18)	p Value	Control (n = 26)	PSP-RS (n = 23)	p Value	PSP (n = 21)	TDP (n = 20)	p Value
Age at sample, y, median (IQR)	70.1 (5)	70.9 (10.4)	0.286	71 (13)	70 (6)	0.7495	72.2 (7.7)	62 (7.25)	0.4703
Sex, male/female	12/16	8/10	0.296	12/13	14/9	0.9119	8/11	11/9	0.6406
Disease duration, y, median (IQR)	0	5.6 (4.9)	<0.001	0	4 (4.5)	<0.001	5.27 (3.1)	4.58 (2.9)	0.4699
CDR + NACC FTLDsb, median (IQR)	0	7.5 (5)	<0.001	0	5 (6.3)	<0.001	4.5 (5.3)	7.75 (7.5)	0.4425
MMSE, median (IQR)	30	27 (7)	<0.001	29 (1)	26 (5)	<0.001	27 (6.25)	26 (6)	0.1717
PSPRS, median (IQR)	0	39 (33.5)	<0.001	0	35.5 (20.8)	<0.001	34 (21)	13 (6.5)	0.4271
Race, n (%)			0.065						0.3611
White	26 (92.9)	13 (72)		16 (65)	19 (82)	0.4544	19 (90.4)	17 (85)
Black	1 (3.6)	0		0	0		0
Other/unknown	1 (3.6)	5 (28)		9 (36)	4 (19)		2 (9.5)	3 (15)
APOE ε4 carrier, n (%)	18 (64.3)	2 (11.1)		NA	NA		4 (22)	5 (25)	0.6494
MAPT haplotype, n (%)			0.004	NA	NA				0.6369
H1/H1	10 (55.6)	18 (100)					21 (100)	17 (85)
H1/H2	4 (22.2)						0	1 (5)
H2/H2	4 (22.2)						0	2 (10)

Abbreviations: CDR + NACC FTLDsb = Clinical Dementia Rating Scale plus National Alzheimer's Coordinating Center Frontotemporal Lobar Degeneration Module Behavior and Language Domains sum of boxes score; IQR = interquartile range; MAPT = microtubule-associated protein tau; MMSE = Mini-Mental State Examination; PSP = progressive supranuclear palsy; PSPRS = PSP Rating Scale; PSP-RS = PSP-Richardson syndrome.

Unless otherwise specified, all variables are expressed as median values with IQRs.

SOMAmer Proteomics

CSF samples were analyzed using the SomaScan assay (SomaLogic, Boulder, CO) for multiplexed targeted SOMAmer proteomics. For a complete description, see eMethods.

Statistical Analyses

Data Visualization

The data analysis approach was the same for original and validation cohorts, but the original cohort was used as a discovery step and reference for analyses with the validation cohort. A quality control report based on array quality metrics run 4 separate times was used by investigators, blinded to patient identifiers and clinical group allocation, to visually identify outliers and confirm data homogeneity within the CSF biological matrix, biological variation, and technical quality. Log₂-transformed data for each SOMAmer were visualized using volcano plots to identify arithmetic between-group differences in relative fluorescent units, relative to their p values adjusted for multiple comparisons with the Benjamini-Hochberg method, using Spotfire (TIBCO Software, Palo Alto, CA). Prespecified group differences consisted of “disease vs control” for both cohorts. Proteins meeting the criteria |log₂(fold change)| ≥ 0.2 and p value <0.05 were considered biologically and statistically significant.

Biological System Pathway Analyses

Targets with significant differences were ranked by p value and entered in the online STRING database for known and predicted physical and functional protein-protein interactions. Functional enrichment within the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was tested against the whole human genome, and gene ontology (GO) analysis was tested using the SOMAmer list as the background. The enrichment strength was expressed as Log₁₀(observed/expected), and significant associations were ranked by p value corrected by false discovery rate (FDR). Top targets within the original rank and those converging in KEGG pathways were analyzed for their clinical value.

Consensus Weighted Gene Coexpression Analysis

Weighted gene coexpression analysis (WGCNA) was used to identify sets of proteins whose expression levels are correlated together in biologically meaningful protein networks (modules), when considering the entire CSF SOMAmer proteomics data set. Up to this point, the data from the different cohorts had not been combined. Consensus WGCNA helped overcome the challenge of analyzing data sets with significant interdataset variation.^19-21 See full description in eMethods.

Clinical Associations

Receiver-operating characteristic curves determined the diagnostic accuracy of targets for disease vs control groups. Areas under the curve (AUCs) were obtained for individual analytes and composites of targets within biological pathways. Associations between SOMAmer levels and disease severity were determined with age and sex-adjusted linear regressions with PSPRS and CDR + NACC/FTLD sum of boxes scores as dependent variables. Statistical analyses were completed in R.

Data Availability

The annotated menus for the SomaScan targets are shown in a separate supplementary file (eAppendix 1). The SomaScan data were obtained through an academia-industry partnership, which prevents publication of the entire SomaLogic proteomics report or data sharing by the UCSF outside a peer-reviewed scientific publication with relevant (i.e., non-single patient) data. Researchers interested in access to the entire proteomics data set for verification of reproducibility of results may request data access to SomaLogic, Inc.

Results

CSF Protein Expression Profiles in Clinically Diagnosed PSP

There were 155 of 5,026 (3.0%) and 959 of 7,595 (12.6%) proteins differentially expressed in PSP compared with controls in the original and validation cohorts, respectively (Figure 1). Most of the differentially expressed targets in the 2 cohorts were downregulated in PSP compared with controls: 128 of 155 (82.6%) in the original cohort and 912 of 959 (95.1%) in the validation cohort. There were only 27 of 155 (17.5%) upregulated targets in the original cohort and 47 of 959 (5.0%) in the validation cohort. Eighty-nine targets were differentially expressed in PSP in both the original and validation clinically diagnosed PSP-RS cohorts, of which 85 were downregulated and 4 were upregulated (eFigures 1 and 2). In the validation cohort, it seemed unusual that approximately 1 in 8 proteins were dysregulated in PSP. To investigate the possibility that unidentified biological variables influence the covariance across proteins, we regressed the first principal component from the data and tested the differential expression between groups (eAppendix 2). Regressing the first principal component did not significantly affect the number of differentially abundant proteins. Eight hundred twenty-six proteins (10.8%) remained dysregulated after regressing the first principal component from the analysis, with 786 (95.1%) downregulated and 40 (4.8%) upregulated.

Figure 1. CSF SOMAmer Proteomic Signatures in PSP.

Volcano plots with CSF expression profiles of PSP compared with age-matched controls for the (A) original cohort, (B) validation cohort, (C) neuropathology-confirmed FTLD-tau PSP, and (D) neuropathology-confirmed FTLD-tau PSP vs FTLD TDP. Protein expression values are adjusted for age and sex. Markers significantly decreased are in blue, and markers significantly increased are in red. Vertical dotted lines represent 0.20 or −0.20 log fold change, for biological significance. Horizontal line is p value = 0.05. EPHA5 = ephrin receptor A5; FTLD = frontotemporal lobar degeneration; GFAP = glial fibrillary acidic protein; Log FC = log fold change; MAPT = microtubule-associated protein tau; NfL = neurofilament light chain; NPTX2 = neuronal pentraxin 2.

Previously Identified Proteins Associated With Neurodegeneration

CSF proteins expected to increase in PSP, compared with controls, including NfL,⁷ neurofilament heavy chain (NEFH),²² and glial fibrillary acidic protein,²³ showed consistent robust increases across all cohorts. Neuronal pentraxin-2 (NPTX2),²⁴ previously known to decrease in PSP, also showed significant decreases in all cohorts. Other proteins associated with neurodegeneration known to have no diagnostic value in PSP were not altered (eTable 2).

CSF Protein Expression Profiles in Neuropathology-Confirmed PSP and TDP

Compared with controls, the expression profile of neuropathology-confirmed FTLD-tau PSP cases showed 321 of 5,026 (6.4%) differentially expressed proteins, with 235 of 321 (73.2%) being downregulated and 86 of 321 (26.8%) upregulated. In turn, 69 of 5,026 (1.3%) proteins were differentially expressed in neuropathology-confirmed TDP cases compared with controls, with 34 of 69 (49.3%) being downregulated and 35 of 69 (50.7%) upregulated.

CSF Protein Expression Profiles in Neuropathology-Confirmed FTLD-Tau CBD or FTLD Tau Pick Disease

Compared with controls, there were no differentially expressed proteins in CBD cases after correcting for multiple comparisons (eAppendix 3). Running the analyses in a combined cohort of FTLD-tau CBD and Pick disease cases compared with controls resulted in only 4 proteins being differentially expressed after FDR correction (eAppendix 3). The 4 proteins were NEFH, YWHAZ, NEFL, and VEGFA.

Consensus WGCNA

A total of 3,888 proteins were included for consensus WGCNA as shared among original and validation clinical cohorts. Eighteen protein modules emerged, and 3 of them were significantly correlated with the diagnosis of PSP (eFigure 3). Proteins in module 1 (β = −0.044, p = 0.002) and module 2 (β = −0.032, p = 0.035) were decreased in PSP, whereas proteins in module 3 (β = 0.039, p = 0.007) were increased in PSP. Module 1 (202 proteins) was enriched for synaptic regulation with GO terms, GABAergic neurons for cell type, and SNARE complex regulation and JAK-STAT pathway with KEGG. Module 2 (77 proteins) was enriched for regulation of IL-17 production and vesicle cytoskeletal trafficking with GO terms and astrocyte/GABAergic neuron/dendritic cells for cell type. Module 3 (89 proteins) was enriched for response to muscle stretch/positive regulation of gluconeogenesis with GO terms, B-cell/CD4+ central memory T-cell/gamma-delta T-cell/glutamatergic neuron/dendritic cells with cell type, and cytokine-cytokine receptor interaction with KEGG (eFigure 4). Notably, the number of differentially expressed proteins by direct arithmetic contrast, as presented in the volcano plots (Figure 1), differed for different modules (eAppendix 1). Module 1 showed the largest proportion of proteins that were differentially expressed in PSP vs controls in the original (37%), validation (79.2%), and neuropathology-confirmed (53%) cohorts. By contrast, with a few exceptions, modules 2 and 3 mostly included proteins that showed no significant between-group differences by simple contrast, as reflected by the volcano plots (eTable 3 and eFigures 5–7).

The axon guidance pathway, within module 1, was the top dysregulated pathway in PSP in original (strength = 1.71, p < 0.001), validation (strength = 0.84, p < 0.001), and PSP neuropathology-confirmed (strength = 0.78, p < 0.001) cohorts. In the original, validation, and neuropathology-confirmed cohorts, 7, 53, and 15 proteins, respectively, out of the 177 whole human genome axon guidance background proteins, were differentially expressed in PSP compared with controls. Axon guidance was the only identified KEGG pathway in the original cohort, but the validation and PSP neuropathology-confirmed cohorts had additional pathways involving glycoprotein synthesis, intracellular signaling, and cell-cell interactions that also showed dysregulation with strengths >0.5 (eTables 4–9).

Performance of SOMAmer Targets as Clinical Biomarkers

In the original cohort, the top 10 upregulated or downregulated targets were preponderantly markers of neurodegeneration, immune response, axon guidance, and glycoprotein synthesis (Figure 2 and eTable 10). In the validation and neuropathology-confirmed cohorts, the top 10 dysregulated targets included neurodegeneration markers and targets involved in gene expression and cell-cell interaction (eTables 11 and 12). From KEGG pathway and GO analyses, the 7 axon guidance differentially expressed proteins in the original cohort were plexin-A1 (PLXNA1), Unc-5 netrin receptor D, roundabout homolog 3, and 4 proteins of the Ephrin gene family (EPHA4, EPHA5, EPHB2, EPHB6). Individually, they showed fair to good between-group discrimination in all cohorts (eTable 13). When these 7 proteins were used together to discriminate PSP vs controls in the original, validation, and neuropathology-confirmed cohorts, the panel reached excellent diagnostic accuracy (original: AUC 0.92, 95% CI 0.848–1, p < 0.001; validation: AUC 0.81, 95% CI 0.656–0.973, p < 0.001, neuropathology-confirmed: AUC 0.93, 95% CI 0.8557–1, p < 0.001, Figure 3). The panel also offered good discrimination between neuropathology-confirmed PSP and neuropathology-confirmed TDP (AUC 0.895, 95% CI 0.791–0.999, p < 0.001).

Figure 2. Top 6 Targets Differentiating PSP vs Controls in Original, Validation, and Pathology-Confirmed FTLD Cohorts.

Box plots show individual data points with median, interquartile range, and range values for the top CSF protein targets with altered expression in PSP. FTLD = frontotemporal lobar degeneration; PSP = progressive supranuclear palsy; PSP-RS = PSP-Richardson syndrome.

Figure 3. Diagnostic Value of CSF Axon Guidance Pathway Proteins in PSP.

The receiver-operating characteristic curve graph shows the diagnostic accuracy of axon guidance pathway targets identified in the original cohort (EPHA4, PLXNA1, EPHB2, ROBO3, UNC5D, EPHA5, and EPHB6), entered as a panel for original, validation, and pathology-confirmed FTLD cohorts. FTLD = frontotemporal lobar degeneration; PSP = progressive supranuclear palsy; PSP-RS = PSP-Richardson syndrome.

In the original cohort, PLXNA1 had the highest single-analyte diagnostic accuracy to discriminate PSP from controls (AUC 0.85, 95% CI 0.733–0.963, p < 0.001). When considering all 53 axon guidance proteins identified in the validation cohort, C-X-C motif chemokine ligand 12 had the highest diagnostic accuracy to discriminate PSP from controls (AUC 0.81, 95% CI 0.692–0.935, p < 0.001). In the 15 axon guidance proteins identified in the neuropathology-confirmed cohort, semaphorin-3G (SEMA3G) had the highest diagnostic accuracy to discriminate neuropathology-confirmed PSP from controls (AUC 0.89, 95% CI 0.799–0.876, p < 0.001). SEMA3G also had good diagnostic accuracy in discriminating neuropathology-confirmed PSP from neuropathology-confirmed TDP cases (AUC 0.85, 95% CI 0.728–0.977, p < 0.001).

Only 2 proteins, cytotoxic T-lymphocyte–associated protein 4 (CTLA-4) and galectin-10 (CLC) showed consistent high correlations with PSPRS scores across cohorts, after correction for multiple comparisons (Figure 4). CLC was positively correlated with PSPRS in original (β = 0.57, 95% CI 0.045–1.08, p = 0.037), validation (β = 0.48, 95% CI 0.03–0.96 p = 0.035), and neuropathology-confirmed (β = 0.54, 95% CI 0.11–0.97, p = 0.018) cohorts. By contrast, CTLA-4 showed consistent negative correlations with PSPRS scores in the original (β = −0.64, 95% CI −1.11 to −0.199, p = 0.009) and validation (β = −0.47, 95% CI −0.89 to −0.05, p = 0.030) cohorts, but not in the neuropathology-confirmed cohort (β = 0.02, 95% CI −0.371 to 0.778 p = 0.463). After correcting for multiple comparisons, none of the axon guidance proteins were correlated with disease severity assessed with the CDR + NACC FTLD sum of boxes or PSPRS scores.

Figure 4. Correlations of CSF Protein Targets With Measures of Disease Severity.

Plots display correlations corrected for multiple comparisons of representative CSF targets with PSPRS in original and validation cohorts with galectin-10 (CLC) and cytotoxic T-lymphocyte–associated protein 4 (CTLA-4). AUC = area under the curve; PSPRS = Progressive Supranuclear Palsy Rating Scale.

Discussion

We identified CSF proteins that were consistently downregulated in 2 PSP clinical cohorts and 1 neuropathologically confirmed PSP cohort, compared with controls. A significant proportion (∼3%–12%) of quantified CSF proteins showed alterations in PSP, with most being decreased relative to controls. Dysregulated protein networks included axon guidance, intracellular signaling, immune function, glycoprotein synthesis, and cell-cell interaction pathways. The axon guidance pathway, which includes members of the ephrin, netrin, and semaphorin signaling pathways, was consistently altered in all 3 PSP data sets. We identified proteins that discriminated PSP from controls with high accuracy, whether analyzed individually or as a panel. Known markers of neurodegeneration were among the targets with the largest differences between PSP and controls, including upregulation of NEFH and NfL and downregulation of NPTX2.^7,22-24 We also identified inflammatory proteins (CLC and CTLA-4) that consistently correlated with measures of disease severity, which makes them candidate CSF biomarkers of PSP progression. The SOMAmer PSP proteomic signature was different from that of the non-PSP neuropathologically confirmed FTLD cohorts. An independent confirmatory weighted protein coexpression analysis that allowed merging data from the 3 cohorts identified a protein module enriched for axon guidance and synaptic transmission pathway proteins as the top module correlated with the diagnosis of PSP. This remarkable consistency suggests that axon guidance proteins have potential as biomarkers and/or therapeutic targets in PSP.

Prior proteomic studies using mass spectrometry and other targeted techniques have identified proteins with clinical value in FTLD, such as YKL-40²⁵ and NPTX2.²⁶ In this study, we introduce aptamer-based proteomics for CSF biomarker discovery in PSP. This tool is technically efficient and robust to preanalytical and analytical factors, which may be well suited for biomarker discovery on clinical grounds. We observed a dramatic reduction in the CSF levels of multiple proteins in PSP compared with cognitively healthy controls. The mechanisms leading to decreased CSF protein levels in PSP are unclear. Although the changes could reflect loss of cells that produce and/or secrete these proteins, this seems unlikely, because the brain regions with the most severe neuropathologic 4-repeat tauopathy burden in PSP are discrete and relatively small subcortical regions. The downregulation pattern is also not likely to be explained by the sensitivity of the analytical method, because it detected molecules with low concentrations, reductions were not observed in TDP cases, and we replicated changes (both increases and decreases) in previously reported PSP-associated CSF proteins from the literature. Molecular signatures detected by SOMAmer proteomics have been shown to be consistent with and complementary to those detected by immunoassay-based proteomics or mass spectrometry in AD.¹⁵ Elucidating the cause of the prominent CSF protein reductions requires further work, but our data align with potential mediating mechanisms, including suppression of the JAK-STAT pathway,²⁷ bioenergetic deficiencies,²⁸ CNS lymphatics alterations,²⁹ and blood-brain barrier permeability disruptions,³⁰ all of which have been implicated in tauopathies.

The PSP CSF proteomic signature showed good, albeit imperfect replication between cohorts. One potential explanation is the use of different SomaScan platforms in each clinical cohort. The v4.1 platform used in the replication cohort has a different data normalization process and a higher number of analytes than the v3.0 platform used in the original cohort. Because group differences are corrected by multiple comparisons, the number of analytes may have affected their between-group difference rankings in each cohort analysis. Lack of identical replication could have been due to biological cohort differences because controls in the original cohort had a relatively high prevalence of APOE ε4 carriership (∼64%), compared with the general population, whereas these data were not available in the validation cohort. Although patients with PSP in the validation cohort were slightly less impaired than those in the original cohort, as measured by the PSPRS, patients in both cohorts had relatively mild disease severity. Between-cohort differences could also be influenced by differences in cerebrovascular, α-synuclein, or AD copathology, which may occur in 20%–52% of PSP cases.³¹

Altered expression of proteins implicated in axon guidance has been previously observed in neurodegeneration. For example, recent neurobiology studies highlight the dynamic structural role of the axonal cytoskeleton and its vulnerability to mechanical stress, injury, and neurodegeneration, and this may be relevant for neurologic disorders in which injury to long tracks is involved. Constitutive reduction of ephrin-A5 (EFNA5) brain levels in SOD1 mutant mice induced faster progression, and lower CSF EFNA5 levels measured by mass spectrometry correlated with faster progression in patients with amyotrophic lateral sclerosis.³² Upregulation of semaphorin-3G has been observed in the frontal cortex and hippocampus of mice with deletion of the RNA-binding protein fused in sarcoma, which is associated with FTD and amyotrophic lateral sclerosis.³³ A common polymorphism in the UNC5C gene is associated with vulnerability to age-related neuropathologies, greater cognitive decline, and accelerated hippocampal atrophy in cognitively healthy older adults, independently of the APOE genotype.³⁴ Finally, CTLA-4 and CLC, both inflammatory proteins related to PSP disease severity in this study, have been implicated in T-cell regulation, and CTLA-4 has reduced expression in FTD, compared with AD.³⁵

This study has several limitations. PSP is a rare disease, and the cohorts had thus small sample sizes. Future collaborative efforts to combine CSF data from multiple cohorts including clinical trials are warranted. To confirm the specificity of the findings, it will be important to compare proteomic changes in PSP with the secondary tauopathy AD, as well as parkinsonism caused by synucleinopathies and other neurodegenerative diseases. The validity of the protein targets has not been established for the entire SomaScan catalog. Orthogonal methodologies, such as mass spectrometry, single-analyte assays, immunohistochemical analyses in neuropathology specimens, and proteogenomic associations, will be needed to confirm the disease-associated proteomic signatures identified here. The SomaScan platform only targets approximately one-third of the proteome and useful biomarkers may exist that are not currently measurable. Targets with post-translational modifications, such as amyloid β_1-42 or p-tau181, are not detected. The platform is heavily biased toward membrane and secreted proteins and may miss other major intracellular pathways implicated in neurodegeneration. The large difference in the amount of differentially abundant proteins in the original and validation cohorts could have been influenced by the use of 2 different versions of the SomaScan platform, as well as additional factors that required further control such as the prevalence of APOE ε4 carriership, cerebrovascular disease, or α-synuclein copathology. We did not study an ethnically, racially, or socioeconomic diverse population, and the data may not be generalizable. Future studies should assess the sensitivity of this analytical platform to longitudinal changes and disease progression, specifically in other 4R-tauopathies. CSF is relatively difficult to assess, and characterization of SOMAmer proteomics in plasma is the scope of a follow-up study. Despite these limitations, our study relied on cohorts with deep phenotypical characterization, and the main results were replicated in 3 independent cohorts.

Unbiased aptamer-based proteomics detected prominent downregulation of CSF proteins in PSP in 3 independent cohorts, including neuropathology-confirmed cases. Axon guidance proteins were significantly altered, and individual proteins in this and other identified pathways have clinical correlates, suggesting potential value as biomarkers for diagnosis and tracking disease severity.

Glossary

AD

Alzheimer disease

AUC

area under the curve

CBD

corticobasal degeneration

CDR

Clinical Dementia Rating Scale

CLC

galectin-10

CTLA-4

cytotoxic T-lymphocyte-associated protein 4

EFNA5

ephrin-A5

FDR

false discovery rate

FTD

frontotemporal dementia

FTLD

frontotemporal lobar degeneration

GO

gene ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

MAC

Memory and Aging Center

NACC

National Alzheimer's Coordinating Center

NEFH

neurofilament heavy chain

NfL

neurofilament light chain

NPTX2

neuronal pentraxin-2

PLXNA1

plexin-A1

PSP

progressive supranuclear palsy

PSP-RS

PSP–Richardson syndrome

PSPRS

PSP rating scale

SEMA3G

semaphorin-3G

UCSF

University of California, San Francisco

WGCNA

weighted gene coexpression analysis

Appendix. Authors

Name	Location	Contribution
Amy Wise, BA	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Jingyao Li, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Mai Yamakawa, MD	Department of Neurology, University of California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Joseph Loureiro, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Brant Peterson, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Kathleen Worringer, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; additional contributions: project supervision
Rajeev Sivasankaran, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; additional contributions: project supervision
Jose-Alberto Palma, MD, PhD	Novartis Institutes for Biomedical Research, Inc., Cambridge, MA	Drafting/revision of the manuscript for content, including medical writing for content
Laura Mitic, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco; The Bluefield Project to Cure FTD	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design
Hilary W. Heuer, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design
Argentina Lario-Lago, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; additional contributions: data and biospecimens management
Adam M. Staffaroni, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Annie Clark, MS	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Jack Taylor, MS	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content
Peter A. Ljubenkov, MD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Lawren Vandevrede, MD, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Lea T. Grinberg, MD, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Salvatore Spina, MD, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
William W. Seeley, MD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Bruce L. Miller, MD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Bradley F. Boeve, MD	Department of Neurology, Mayo Clinic, Rochester, MN	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Bradford C. Dickerson, MD	Department of Neurology, Massachusetts General Hospital and Harvard Medical School, Boston	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Murray Grossman, MDCM, FAAN	Department of Neurology, University of Pennsylvania, Philadelphia	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Irene Litvan, MD	Department of Neurology, University of California, San Diego	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Alexander Pantelyat, MD	Department of Neurology, Johns Hopkins University, Baltimore, MD	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Maria Carmela Tartaglia, MD	Department of Neurology, University of Toronto, Ontario, Canada	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Zihan Zhang, BS	Departments of Mathematics and Statistics, University of California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Anne-Marie A. Wills, MD	Department of Neurology, Massachusetts General Hospital and Harvard Medical School, Boston	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Jessica Rexach, MD, PhD	Department of Neurology, University of California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Julio C. Rojas, MD, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Adam L. Boxer, MD, PhD	Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; additional contributions: project supervision

Study Funding

R01AG038791 (ALB), U19AG063911 (ALB) Rainwater Charitable Foundation (ALB), NIA/NIH K23AG59888 (JCR), AlzOut-Alzheimer's Research (JCR), John Douglas French Alzheimer's Foundation (JCR), P01AG019724 (FTD PPG), P30AG062422 (ADRC), National Institute of Neurological Disorders and Stroke/NIH K08NS105916 (JER), NIH/NIA K23AG059891, NIH/National Institute of Neurological Disorders and Stroke U01NS102035 and NIH/NIA R01AG038791 (AP), NIH K23AG073514 (LV), The Bluefield Project to Cure FTD (LM).

Disclosure

L. Mitic is supported by the Bluefield Project to Cure FTD. L. Vandevrede is supported by K23AG073514 and grants from the Alzheimer's Association and American Academy of Neurology, site PI for clinical trials sponsored by Biogen, consulted for Retrotope. M. Grossman is deceased, to the best of our knowledge, he had no relevant disclosures. A. Pantelyat is supported by NIH/NIA K23AG059891, NIH/National Institute of Neurological Disorders and Stroke U01NS102035; NIH/NIA R01AG038791; Scientific Advisory Board, MedRhythms, Inc. J.C. Rojas NIA/NIH K23AG59888 is site principal investigator for Clinical trials sponsored by Eisai and Eli-Lilly. Go to Neurology.org/N for full disclosures.