Multimarker synaptic protein cerebrospinal fluid panels reflect TDP-43 pathology and cognitive performance in a pathological cohort of frontotemporal lobar degeneration

Synapse degeneration is an early event in pathological frontotemporal lobar degeneration (FTLD). Consequently, a surrogate marker of synapse loss could be used to monitor early pathologic changes in patients with underlying FTLD. The aim of this study was to evaluate the relationship of antemortem cerebrospinal fluid (CSF) levels of 8 synaptic proteins with postmortem global tau and TDP-43 burden and cognitive performance and to assess their diagnostic capacity in a neuropathological FTLD cohort. We included patients with a neuropathological confirmation of FTLD-Tau (n = 24, mean age-at-CSF 67 years ± 11), FTLD-TDP (n = 25, 66 years ± 9) or AD (n = 25, 73 years ± 6) as well as cognitively normal controls (n = 35, 69 years ± 7) from the Penn FTD Center and ADRC. We used a semi-quantitative measure of tau and TDP-43 inclusions to quantify pathological burden across 16 brain regions. Statistical methods included Spearman rank correlations, one-way analysis of covariance, ordinal regression, step-wise multiple linear regression and receiver-operating characteristic curves. CSF calsyntenin-1 and neurexin-2a were correlated in all patient groups (rs = .55 to .88). In FTLD-TDP, we observed low antemortem CSF levels of calsyntenin-1 and neurexin-2a compared to AD (.72-fold, p = .001, .77-fold, p = .04, respectively) and controls (.80-fold, p = .02, .78-fold, p = .02, respectively), which were inversely associated with post-mortem global TDP-43 burden (regression r2 = .56, p = .007 and r2 = .57, p = .006, respectively). A multimarker panel including calsyntenin-1 was associated with TDP-43 burden (r2 = .69, p = .003) and MMSE score (r2 = .19, p = .03) in FTLD. A second multimarker synaptic panel, also including calsyntenin-1, was associated with MMSE score in FTLD-tau (r2 = .49, p = .04) and improved diagnostic performance to discriminate FTLD-Tau and FTLD-TDP neuropathologic subtypes (AUC = .83). These synaptic panels have potential in the differential diagnosis of FTLD neuropathologic subtypes and as surrogate markers of cognitive performance in future clinical trials targeting TDP-43 or tau.

Background A prominent feature of frontotemporal lobar degeneration (FTLD) is a region-specific degeneration and loss of synapses [1][2][3][4], which correlates with cognitive impairment [1]. Characterization of the changes in quantity and function of synaptic proteins and the relationship to pathologic abnormalities and clinical symptoms is essential to improve our understanding of early FTLD pathogenesis. An objective marker of synapse degeneration that is directly related to FTLD pathology would be invaluable for monitoring the progression of these early pathological changes, which could aid patient management, clinical trial recruitment and has potential as a measure of drug response in future clinical trials. Moreover, the clinical diagnosis of FTLD syndromes is challenged by the variable correspondence between the clinical syndrome and underlying neuropathological changes and there is a need for a biomarker that can distinguish neuropathologic subtypes. FTLD can be classified postmortem into four major neuropathological subtypes; FTLD-Tau (hyperphosphorylated tau inclusions), FTLD-TDP (transactive response DNA-binding protein-43; TDP-43), FTLD-FET (fused in sarcoma, Ewing's sarcoma, TATAbinding protein associated factor 15) and FTLD-UPS (ubiquitin and p62) [2][3][4]. Here, we have focused on FTLD-Tau and FTLD-TDP, which represent 90-95% of FTLD cases. The aim of this study is to a) compare the cerebrospinal fluid (CSF) profile of a panel of synaptic proteins in a neuropathological cohort of confirmed FTLD-Tau and FTLD-TDP, b) evaluate the relationship of antemortem CSF levels with postmortem Tau and TDP-43 burden and cognitive performance and c) determine their diagnostic accuracy. As reference groups we include cognitively normal controls where synapse degeneration is absent and Alzheimer's disease (AD) where synapse degeneration is widespread. The proteins included in the study (calsyntenin-1, glua2, glua4, Neurexin-2a, Neurexin-3a, neuroligin-2, syntaxin-1b, thy-1 and VAMP-2) were brought forward from our previous study in clinical AD cohorts [5]. These proteins are enriched at the synapse [9] and are involved in a range of synaptic functions including pre-synaptic differentiation [6], dendritic spine assembly [11][12][13][14], postsynaptic Ca2 + signalling [7], AMPA receptor trafficking [16,17], synaptic vesicle exocytosis [18][19][20][21] and synaptic plasticity [22][23][24][25].

Study design
This is a single-center cross-sectional study. Antemortem CSF samples obtained between 1992 and 2015 were retrospectively selected from samples at the Penn FTD Center and Alzheimer's Disease Research Center (ADRC) at the University of Pennsylvania (Philadelphia, USA) [8,9]. The study includes patients with FTLD-related syndromes followed to autopsy with a neuropathological diagnosis of FTLD-Tau (n = 24) or FTLD-TDP (n = 25) and patients with a clinical and neuropathological diagnosis of AD (n = 25). Patients included for study had prospective CSF collection as part of ongoing observational research programs at The Penn FTD Center aimed at developing biological markers that reflect pathophysiological changes in FTLD proteinopathies to improve antemortem diagnosis and prognostication. Patients underwent a Mini-Mental State Examination (MMSE) by neurologists masked to synaptic biomarker data within 5 months of the lumbar puncture. We also analyzed CSF samples from cognitively normal controls (n = 35) that were recruited from the community and screened by revising their medical and medication history. Controls self-reported a negative neurological and psychiatric history and were within the normal range of cognitive performance (Mini-Mental State Examination > 27). CSF samples from patients and controls were processed in the same way. Pre-analytical processing details for these CSF samples can be found elsewhere [8].
Peak areas for the endogenous (light) and isotopicallylabeled (heavy) transitions were extracted and used as input for processing. Using the dataProcess function of MSstats v3.5 package in R [14], light and heavy transitions were log-2 transformed and normalized using the EqualizeMedians function. The light/heavy intensity was calculated for all transitions and summarized using the Tukey's median polish function. We removed transitions with between-run interference (between RunInterferenc-eScore < 0.8). All remaining transitions passed the minimum log2 intensity cut-off designated by the MSstats package (7.724). We did not impute missing transition data. One of the three calsyntenin-1 peptides was not detected in any samples and was excluded from further analyses. The final number of peptides corresponding to each protein was as follows:

Statistical analysis
Statistical analyses were performed in R version 4.1.2 [15]. Outliers were excluded using the 3 × IQR rule. Where residuals deviated from a Gaussian distribution (Shapiro-Wilk p < 0.05), we used square root transformed values, which did not deviate from a Gaussian distribution. Group differences were compared using χ2 test for categorical variables or one-way analysis of covariance (ANCOVA) for continuous variables including sex and age-at-CSF as covariates with post hoc Tukey method. A Kruskal-Wallis test was used to compare Global Tau and TDP-43 burden across male and female participants. We used Spearman rank correlations to determine the relationship between antemortem synaptic protein levels in the CSF and demographic variables. We performed ordinal regression to test for association of synaptic proteins with Tau and TDP-43 scores in individual brain regions, including age-at-death and time from CSF to autopsy as covariates. We performed a mixed entry backward stepwise linear regression (stepAIC function) to identify the best possible predictors of postmortem Global tau and Global TDP-43 (sum of scores across all regions). Age-atdeath and time from CSF to death were forced into the model and test variables included all 8 synaptic proteins, biological sex, and AD comorbidity. Akaike Information criteria and variant inflation factors were used to set a limit on the total number of variables included in the final model. The same approach was taken for regressions using MMSE as the outcome measure, with the modification that age-at-CSF and years of education were forced into the model. All reported r 2 values were adjusted for the number of predictors in the model. All ANCOVA, Spearman and regression p-values were adjusted for multiple testing of 8 synaptic proteins using the Benjamini-Hochberg method. We assessed the diagnostic utility (area-under-the-curve; AUC) of CSF levels of the synaptic proteins and multimarker panels using receiver operating characteristic (ROC) curves in the pROC package implemented in R. We used the 'DeLong' method to compare ROC curves [16]. Table 1 shows the demographic, clinical and neuropathological data for the samples included in the study grouped according to their final neuropathological diagnosis. Forty-five percent of the study participants were female. The FTLD-Tau group had a lower proportion of females (25%) compared to controls (57%, p = 0.03), whereas the proportion was comparable across all other groups (p > 0.05). There was no difference in years of education across groups (p = 0.40). The mean age-at-disease onset (6 year difference, post-hoc p = 0.02) and mean age-at-CSF analysis (7 year difference, post-hoc p = 0.04) were higher in the AD group compared to FTLD-TDP. The mean age-at-death was higher in AD compared to FTLD-Tau (8 year difference, post-hoc p = 0.03) and FTLD-TDP (9 year difference, post-hoc p = 0.006). There was no difference in mean disease duration (p = 0.07) or time interval between disease-onset and CSF analysis (p = 0.79), whereas the mean time interval between CSF analysis and death was longer in the AD group compared to FTLD- Tau (1.4 year difference, post-hoc p = 0.04) and FTLD-TDP (2.4 year difference, post-hoc p = 0.01). Based on these differences, we included sex and age-at-CSF collection as a covariate in statistical analysis for group comparisons. Eighty-three percent of the FTLD-Tau and 76% of FTLD-TDP showed little to no AD pathology and 72% of the AD patients showed no TDP-43 pathology beyond the amygdala, a proportion that was comparable across groups (p = 0.63). The global pathological burden of tau (p = 0.66) and TDP-43 (p = 0.31) pathology was comparable between male and female participants. Global pathological burden was not associated with the time interval from CSF collection to death (tau; p = 0.12, TDP-43; p = 0.21) or disease duration (tau; p = 0.08, TDP-43; p = 0.19) but was inversely related to age-at-death (tau; p = 0.04, TDP-43; p = 0.02). The mean MMSE score and time interval between MMSE and CSF collection were comparable between FTLD-Tau and FTLD-TDP (both p > 0.29).

Correlation between antemortem CSF levels of the synaptic proteins
We first sought to compare the pair-wise correlation between antemortem CSF levels of the synaptic proteins in each neuropathologic group (Fig. 1).

Association of antemortem CSF levels of the synaptic proteins with post-mortem pathological burden in FTLD
In FTLD-Tau, we observed no association between antemortem CSF levels of the synaptic proteins with age-at-death (all p > 0.11). In FTLD-TDP, neurexin-3a was inversely associated with age-at-death (r s = -0.55, p = 0.04). We observed no association between antemortem CSF levels of the 8 synaptic proteins with postmortem global tau burden (Additional File 2; all p > 0.17). Calsyntenin-1, neurexin-2a, thy-1 and neuroligin-2 were inversely associated with postmortem global TDP-43 burden (Fig. 3). With the exception of neuroligin-2 and thy-1, biological sex also significantly contributed to the associations (all p = 0.03). Moreover, the inclusion of covariates explained global postmortem TDP-43 burden better than simple linear regression models for each synaptic protein alone (all ANOVA p < 0.003). In a secondary analysis, we observed a stronger association of calsyntenin-1 when the outcome measure was restricted to regions with low to medium TDP-43 burden (brainstem + cerebellum + thalamus + basal ganglia; r 2 = 0.61; p = 0.03). We also observed an association for neurexin-3a in regions with low to medium TDP-43 burden (r 2 = 0.69; p = 0.01). This may be due to a ceiling effect in regions with high burden.
To determine whether a combination of synaptic proteins could better explain postmortem tau and TDP-43 burden, we performed stepwise mixed entry linear regression starting with the synaptic proteins

Association of antemortem CSF levels of the synaptic proteins with cognitive performance in FTLD
We first tested the association of the synaptic proteins from Panel A[CLSTN1-VAMP2] with MMSE score in FTLD patients including sex, age-at-CSF, and years of education as covariates. The panel was associated with MMSE in the total FTLD group (Table 2b;  To determine whether other combinations of synaptic proteins could improve the association of Panel A, we performed stepwise linear regression starting with the synaptic proteins and covariates. Compared to Panel A, a 6 variable model including calsyntenin-1, gluA4, syntaxin-1b, VAMP-2, age-at-CSF and education showed a stronger association with MMSE in FTLD-Tau (Table 2b; r 2 = 0.49, p = 0.04). Age-at-CSF, calsyntenin-1, VAMP-2 and gluA4 were the significant contributors to the model (henceforth termed Panel B[CLSTN1-VAMP2-GLUA4]). When controlling for the other variables, elevated age (p = 0.03), elevated CSF calsyntenin-1 (p = 0.005), lower CSF VAMP-2 (p = 0.01) and lower CSF gluA4 (p = 0.03) were associated with worse MMSE score. Additional file 5 plots the relative regression line for each variable. An alternative 6 variable model, including gluA4, neuroligin-2, neurexin-3a, syntaxin-1b, age-at-CSF and education, showed a stronger association with MMSE in FTLD-TDP (Table 2b; r 2 = 0.38, p = 0.02) and in all FTLD (Table 2b; r 2 = 0.30, p = 0.002). GluA4, neurexin-3a and syntaxin-1b, age-at-CSF and years of education were the significant contributors to this model (henceforth termed Panel C[GLUA4-NRX3A-STX1B]). When controlling for the other variables, less education (p = 0.02), elevated age (p = 0.005), lower CSF gluA4 (p = 0.002), elevated neurexin-3a (p = 0.003) and elevated CSF syntaxin-1b (p = 0.03) were associated with worse MMSE score. Additional file 6 plots the relative regression line for each variable.
Thus, several multimarker panels of synaptic proteins are associated with cognitive performance in FTLD. Panels B and C were not associated with postmortem global TDP-43 or global Tau burden (all p > 0.1).

Table 1 Demographic, clinical and neuropathological data for the samples included in the study
Unless otherwise specified, the results are expressed as mean (SD). Global tau and TDP pathology refers to sum of ordinal values across 16 brain regions. Comorbidity refers to FTLD without incidental AD pathology (NIA-AA stage ≤ B1) and AD without TDP-43 pathology beyond the amygdala.   (Table 2c; AUC = 0.81) compared to calsyntenin-1 (p = 0.56) and was significantly worse than the pTau/ Aβ42 ratio (AUC = 0.96, p < 0.03 vs Panel B).

Discussion
Here we report a comprehensive evaluation of the antemortem CSF levels of 8 synaptic proteins in a neuropathological cohort of FTLD (FTLD-Tau and FTLD-TDP), AD and cognitively normal controls. CSF levels of calsyntenin-1 and neurexin-2a were correlated in all patient groups and were lower in the FTLD-TDP neuropathological subtype compared to neuropathologically confirmed AD and cognitively normal controls. Furthermore, antemortem CSF levels of calsyntenin-1 and neurexin-2a inversely correlated with global TDP-43 burden post-mortem. Combining calsyntenin-1 with VAMP-2 and neurexin-3a improved this association with TDP-43 burden and was also correlated with cognitive performance in FTLD. Combining calsyntenin-1 with VAMP-2, gluA4 and syntaxin-1b showed high accuracy in the differential diagnosis of FTLD-TDP and FTLD-Tau neuropathologic subtypes and was associated with cognitive performance in FTLD-Tau. Thus, when measured in CSF, combinations of multiple synaptic proteins were better predictors of global TDP burden and cognitive performance and showed better diagnostic accuracy compared to individual synaptic proteins. To our knowledge, this is the first study to relate CSF levels of synaptic proteins to FTLD neuropathological subtypes.
Calsyntenin-1 is abundant in most neurons of the CNS [36] where it has been shown to modulate postsynaptic calcium signaling and promote dendritic spine assembly [15]. VAMP-2 and syntaxin-1b are synaptic vesicle recycling proteins predominantly found at glutamatergic synapses [17] as part of the synaptic exocytosis core vesicular complex [18] where they regulate the releasable pool of glutamate at the pre-synapse [19]. GluA4 is the predominant AMPA receptor subunit in PV interneurons [44]. The neurexins are a complex family of proteins that are generated from three different genes (NRXN1, NRXN2, NRXN3) with alternative promoters (α, β, γ) and extensive alternative splicing resulting in over 1000 distinct neurexin isoforms, each associated with specific neuronal cell types [20]. The formation of a transsynaptic complex with their post-synaptic partners, the neuroligins, allows the structural assembly of excitatory synapses, by triggering the recruitment of postsynaptic NMDA and AMPA glutamate receptors [21,22]. The composition of these panels thus support the growing evidence for involvement of the glutamatergic system in FTLD syndromes [23].
A previous study has reported an association between low CSF calsyntenin-1 levels in patients with behavioral variant FTD (bvFTD) compared to cognitively normal controls and compared to presymptomatic carriers of C9orf72, GRN or VCP mutations [24]. As the pathology underlying clinically defined FTLD syndromes is heterogeneous, the results of the previous study cannot be directly compared to the current study; clinical diagnoses of bvFTD is split between FTLD-TDP and FTLD-Tau subtypes and the small number of cases with a bvFTD diagnosis in the current study (n = 10) prohibits meaningful comparison across syndromes. Nevertheless, both studies point towards altered abundance of calsyntenin-1 in CSF from patients clinically diagnosed with FTLD syndromes and those with neuropathologically confirmed FTLD compared to cognitively normal controls. Neurexins have also been shown to facilitate the calciumdependent release of synaptic vesicles [25,27,28].
Another important finding of this study is that while AD neurofibrillary tangle pathology (Braak score < 2 [31]) was present in 17% of the FTLD-Tau and 24% of the FTLD-TDP patients, stepwise regression showed that AD comorbidity was not a good predictor of TDP-43 burden and was therefore excluded from the final model. On the other hand, AD comorbidity was a predictor of MMSE and was included in Panel C. Thus, AD comorbidity may influence cognitive performance in FTLD. This is consistent with a recent study showing that neurodegenerative comorbidities contribute to transition from mild cognitive impairment to dementia [29].

Study limitations
A limitation of this work is the relatively small study size compared to other CSF biomarker studies. However, CSF samples from well-annotated autopsy cases of these uncommon conditions are scarce and the neuropathological confirmation reduces the substantial heterogeneity inherent in clinical FTD cohorts. It should also be noted that the mean time from CSF to death was 5 years in FTLD-Tau and 4 years in the FTLD-TDP subtypes, reaching up to 10 years in some cases. This long interval and variability could influence the relationship between participants' CSF biochemical signature and their final neuropathological findings. Finally, our control participants lack neuropathological confirmation. However, complete clinical and neuropsychological evaluations were performed to exclude significant medical (and specifically neurological) conditions in these participants.

Conclusions
A biomarker that can distinguish FTLD neuropathologic subtypes in clinical cohorts would be a valuable addition to the biomarker arsenal. In this regard, Panel B[CLSTN1-VAMP2-GLUA4] identified in this study provides proof-of-concept that these synaptic proteins are differentially altered and can distinguish between the two most common neuropathologic subtypes. Moreover, all panels described here were associated with MMSE score and therefore show proof-of-concept as surrogate measures of cognitive performance. These panels may be relevant for future clinical trials targeting TDP-43 or tau pathology, where there will be a need for a surrogate measure of cognitive performance, independent of the drug target. In addition, the postsynaptic modulator, calsyntenin-1 both alone and in combination with other synaptic proteins, has potential as an objective measure of TDP-43-mediated degeneration in FTLD. Validation of these findings in independent cohorts is needed to fully determine their clinical potential.