Distinct cell type-specific protein signatures in GRN and MAPT genetic subtypes of frontotemporal dementia

Frontotemporal dementia is characterized by progressive atrophy of frontal and/or temporal cortices at an early age of onset. The disorder shows considerable clinical, pathological, and genetic heterogeneity. Here we investigated the proteomic signatures of frontal and temporal cortex from brains with frontotemporal dementia due to GRN and MAPT mutations to identify the key cell types and molecular pathways in their pathophysiology. We compared patients with mutations in the GRN gene (n = 9) or with mutations in the MAPT gene (n = 13) with non-demented controls (n = 11). Using quantitative proteomic analysis on laser-dissected tissues we identified brain region-specific protein signatures for both genetic subtypes. Using published single cell RNA expression data resources we deduced the involvement of major brain cell types in driving these different protein signatures. Subsequent gene ontology analysis identified distinct genetic subtype- and cell type-specific biological processes. For the GRN subtype, we observed a distinct role for immune processes related to endothelial cells and for mitochondrial dysregulation in neurons. For the MAPT subtype, we observed distinct involvement of dysregulated RNA processing, oligodendrocyte dysfunction, and axonal impairments. Comparison with an in-house protein signature of Alzheimer’s disease brains indicated that the observed alterations in RNA processing and oligodendrocyte function are distinct for the frontotemporal dementia MAPT subtype. Taken together, our results indicate the involvement of different brain cell types and biological mechanisms in genetic subtypes of frontotemporal dementia. Furthermore, we demonstrate that comparison of proteomic profiles of different disease entities can separate general neurodegenerative processes from disease-specific pathways, which may aid the development of disease subtype-specific treatment strategies.

(FTD-C9), or mutations in the progranulin gene (FTD-GRN) or the microtubule-associated protein tau gene (FTD-MAPT) [27,61,64]. The neuropathological hallmark of FTD is specific proteinopathy, with 50% of the cases showing TDP-43 aggregates, 40% tau aggregates, and 5 ~ 10% showing FET protein family aggregates [5,42]. While TDP-43 aggregates are classically linked to FTD-C9 and FTD-GRN, and tau-aggregates to FTD-MAPT [5], neuropathological features transcend specific subtypes, and combinations of neuropathological and clinical features are seen in sporadic FTD. This heterogeneity, combined with a relatively rare occurrence (3-26 in 100.000 people worldwide [71]), makes it challenging to study the disease in humans on a large scale and in a stratified manner. Cell and animal models of FTD have revealed potential disease mechanisms, however, demonstrating their involvement in distinct subtypes has proven difficult and attempts to translate findings into therapeutic strategies have failed so far [2,23,60].
Key to the development of treatment strategies is to identify cell type-specific pathways that drive disease initiation and progression. To date, several genomewide association and transcriptomic studies have identified susceptibility genes, implicating impairments in lysosomal autophagy and the immune system across the FTD spectrum [6,8,18,19,63,68]. A few studies have focussed on proteome changes in neuropathological subtypes, including FTD associated with TDP-43 pathology (FTD-TDP) [25,26,31,38,48,70], FUS pathology [43], and in the genetic subtype FTD-C9 [3]. However, a systematic proteomic analysis of dysregulated proteins and pathways in affected cell types in genetic FTD is lacking.
Here, we performed a stratified analysis of two genetic subtypes, FTD-GRN and FTD-MAPT, to enable identification of disease mechanisms that are either shared or distinct for these subtypes. Data-independent quantitative proteomic analysis (DIA) of frontal and temporal cortical tissues from FTD patients with genetically-confirmed GRN or MAPT mutations, and non-demented controls (NDCs) was performed. Brain region-specific protein expression profiles for both subtypes were identified. Expression-weighted cell type enrichment (EWCE) analysis was performed to reveal cell types contributing to the FTD subtype-specific disease processes. Subsequent gene ontology (GO) analysis uncovered biological processes that are distinct for the different FTD subtypes and the cell types involved. Finally, by comparing FTD-MAPT with Alzheimer's disease (AD), both general neurodegenerative and FTD-specific processes were revealed.

Selection of FTD cases
A schematic overview of the workflow is presented in Fig. 1. Post-mortem brain tissues were obtained from the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam, and from the Queen Square Brain Bank for Neurological Disorders, UCL Institute of Neurology, London. All materials have been collected from donors from whom written informed consent for brain autopsy and the use of the material and clinical information for research purposes has been obtained. Approval was granted for the Netherlands Brain Bank by the Ethics Committee of the Vrije Universiteit Medical Center (April 30,2009), and for the Queen Square Brain Bank for Neurological Disorders by the London Central Research Ethics Committee (August 6, 2013). A total of 22 brains from patients with familial FTD were available and eligible based on clinical and neuropathological reports, and genetic origin was validated by genetic screening. Cortical tissues from the middle frontal gyrus and middle temporal gyrus were obtained from nine cases with FTD-GRN and 13 cases with FTD-MAPT. In addition, middle frontal and temporal cortical tissues were obtained from 11 sex-matched non-demented controls. Extended information on all cases and controls in the RiMOD-FTD cohort is listed in Table 1.

Immunohistochemistry for neuropathological characterization
Routine immunohistochemistry was carried out by the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam. In addition, we performed extended staining on multiple brain regions, including all cortical areas, hippocampus, caudate nucleus, and putamen, using AT8 (MN1020, Thermo Fisher Scientific, 1:400) and pTDP-43 (CAC-TIP-PTD-M01, Cosmo Bio, 1:1000) antibodies. The pattern of TDP-43 pathology was classified according to the morphology and distribution of neuronal inclusions as proposed by Neumann et al. [50].

Brain tissue preparation and laser microdissection
Sections (10 µm) of fresh frozen tissue were mounted on polyethylene naphthalate-membrane slides (Leica, Herborn, DE), fixed in 100% ethanol for 1 min and stained using 1% (wt/vol) Toluidine Blue in H 2 O (Fluka Analytical, Buchs, Switzerland) for 1 min. Laser microdissection was performed using a Leica AS LMD system. A volume of 1.2 mm 3 of grey matter tissue from the frontal and temporal cortical regions was collected in Eppendorf tubes containing 30-µL M-PER lysis buffer (Thermo Scientific, Rockford, IL, USA) supplemented with reducing sodium dodecyl sulphate sample buffer (Thermo Scientific). Microdissected tissue was stored at −80 °C until further use.

Protein separation by electrophoresis and in-gel digestion
Microdissected tissue lysates were incubated at 95 °C for 5 min, followed by incubation with 50-mM iodoacetamide for 30 min at room temperature in the dark. Proteins were size separated on a NuPAGE 4-12% Bis-Tris acrylamide gel (Invitrogen, Carlsbad, CA, USA) using MOPS sodium dodecyl sulphate running buffer (Invitrogen) according to the manufacturer's protocol. Gels were fixed and stained with colloidal Coomassie Blue G-250 overnight while shaking. After destaining in ultrapure H 2 O, each gel lane was sliced into four equal-sized parts and each part was cut into blocks of approximately 1 mm 3 and collected in a 96-wells plate. Destaining, trypsin digestion, and peptide extraction were done as described previously [11].

Micro LC and data-dependent acquisition mass spectrometry of strong cation-exchange fractions for library preparation
For library preparation, pooled protein extracts from a mix of FTD-MAPT and NDC samples were used. Extracted peptides were analysed by micro liquid chromatography with tandem mass spectrometry (LC-MS/ MS) using an Ultimate 3000 LC system (Dionex, Thermo Scientific) coupled to the TripleTOF 5600 mass spectrometer (Sciex). Peptides were trapped on a 5 mm Pepmap 100 C18 column (300 μm i.d., 5 μm particle size, Dionex) and fractionated on a 200 mm Alltima C18 column (300 μm i.d., 3 μm particle size). The acetonitrile concentration in the mobile phase was increased from 5 to 18% in 88 min, to 25% at 98 min, 40% at 108 min and to 90% in 2 min, at a flow rate of 5 μL/min. The eluted peptides were electro-sprayed into the TripleTOF MS with a micro-spray needle voltage of 5,500 V. The mass spectrometer was operated in a data-dependent acquisition (DDA) mode with a single MS full scan (m/z 350-1250, 150 ms) followed by a top 25 MS/MS (m/z 200-1800, Fig. 1 RiMOD-FTD study workflow. After tissue preparation of all 66 brain cortical tissue samples, we performed DIA mass spectrometry (SWATH), followed by differential protein expression analysis. The identified proteins with differential expression in the FTD subtypes were further subjected to downstream bioinformatics analyses. These included (1) expression-weighted cell type enrichment based on existing single cell RNAseq (scRNAseq) data resources, (2) several gene ontology analyses, and (3) an extensive comparison with an in-house sporadic AD proteomic data set. The results of these analyses were integrated and led to the identification of distinct cell type-specific protein signatures in GRN and MAPT genetic FTD subtypes 150 ms) at high sensitivity mode in UNIT resolution, precursor ion > 150 counts/s, charge state from + 2 to + 5, with an exclusion time of 16 s once the peptide was fragmented. Ions were fragmented in the collision cell using rolling collision energy, and a spread energy of 5 eV.

Micro LC and data-independent acquisition mass spectrometry for experimental samples
The conditions used for micro liquid chromatography of experimental samples were the same as those for the library preparation. The mass spectrometer was operated in a data-independent acquisition (DIA) mode, where experiments consisted of a parent ion scan of 150 ms followed by a window of 8 Da with scan time of 80 ms, and stepped through the mass range between 450 and 770 m/z. The total cycle time was about 3.2 s, which yielded in general 9-10 measurement points across a typical peptide with an elution time of 30 s. The collision energy for each window was determined based on the appropriate collision energy for a 2 + ion, centered upon the window with a spread of 15 eV.

DIA data extraction and analysis
We first analysed the DDA data from our pooled library samples and identified 3,422 protein groups by Max-Quant search (version 1.5.2.8) [13] against the human proteome using the UniProt FASTA (release February 2015) and Biognosys iRT FASTA databases. The proteins and their corresponding fragment ions were then converted into a spectral library with Spectronaut, version 11 [9], for which the Q-value threshold for peptides imported from the MaxQuant msms.txt output table was set to 0.01, and all other settings were left to default. Analysis of DIA data from the experimental samples was done in Spectronaut using our DDA spectral library and the default settings. Across-run normalization based on total peak areas was performed by Spectronaut. Peptide abundances were exported as a Spectronaut report and further processed using the R language for statistical computation, version: 3.4.4 [56]. For a total overview of the studied samples, peptides were selected using a quality value cut-off condition of ≤ 10 -3 in 50% of samples (this filter was applied per sample fraction). For further pairwise statistical comparison between conditions, peptides in each sample fraction were selected using a quality value cut-off condition of ≤ 10 -3 in 50% of the samples of one condition. Peptide abundances were computed by summation of the peak area of the top two abundant fractions, preceded by peptide normalization using the normalizeCyclicLoess function from the limma R package, which was set to 'fast' and iterations were set to 3. Protein abundances were computed by summation of the normalized abundancies of the top five most abundant peptides for a respective protein.

Statistical analysis of differential protein expression
Differential expression analysis between conditions was performed on log-transformed protein abundances. Permutation-based modified t-statistics with multiple testing correction by False Discovery Rate (FDR) was applied using the SAM function from the siggenes R package, using the method 'd.stat' , running 1000 permutations. An FDR adjusted q-value threshold of 0.05 was used to discriminate proteins of interest after differential expression analysis.

Cell type enrichment analysis
Cell type enrichment analysis can help to stratify data from mixed cell populations, without the need for physical cell sorting [4,45]. This is frequently applied on bulk RNA expression data, but to our knowledge has not been used for protein expression data yet. As single cell proteomic data sets are for now unattainable, we set out to identify cell type enrichment in our FTD protein signatures based on published transcriptomic profiles -assuming that proteins are detected in cells that express the corresponding genes. Cell type enrichment analysis of differentially expressed proteins was based on several relevant and extensive human brain single cell transcriptome data resources. For analysis of our frontal cortical protein signatures, single nuclei RNAseq data of 10,319 cells from post-mortem frontal cortical tissue of four adult controls was used [39]. For analysis of temporal cortical protein signatures, a combination of single cell RNAseq data of 466 cells from eight adult control donors [14] and single nuclei RNAseq data of 15,928 cells from eight adult control donors [30] from temporal cortical tissue of either surgical procedures or post-mortem, was used.
Pre-processing and analysis of single cell and single nuclei RNAseq data sets was performed using the Python package Scanpy (version 1.5.1) [74] as described previously [46]. In short, cell-gene matrices were filtered for outliers and gene expression was normalized per cell. All cells were clustered using Louvain clustering implementation [69] on the top 1000 highly variable genes. To identify cell types, marker genes and expected cell types were inferred from the original publications of the data sets. Briefly, angiotensinogen (AGT ), electrogenic sodium bicarbonate cotransporter 1 (SLC4A4), and excitatory amino acid transporter 2 (SLC1A2) were taken as markers for astrocytes, vascular endothelial growth factor receptor 1 (FLT1), dual specificity protein phosphatase 1 (DUSP1), and nostrin (NOSTRIN) for endothelial cells, vesicular glutamate transporter 1 (SLC17A7) for excitatory neurons, glutamate decarboxylase 1 (GAD1) for inhibitory neurons, amyloid beta A4 precursor proteinbinding family B member 1-interacting protein (APB-B1IP) and TYRO protein tyrosine kinase-binding protein (TYROBP) for microglia, myelin-associated oligodendrocyte basic protein (MOBP) for oligodendrocytes, and protocadherin-15 (PCDH15) and platelet-derived growth factor receptor alpha (PDGFRA) as markers for oligodendrocyte precursor cells (OPCs). Clusters that could not be clearly identified with one cell type were labelled 'unknown' .
Normalized gene expression data and cell type label matrices were subsequently used for expressionweighted cell type enrichment analysis using the EWCE package, version 1.2.0 [65] in R. The total set of DIA quantified proteins for both FTD subtypes was used as the background set, from which 20,000 random lists were generated for bootstrapped analysis of the probability distribution of cell type expression. In addition, proteins that showed specificity values ≥ 0.5 for a certain cell type were considered as highly enriched for that cell type. Furthermore, to look into synapse enriched proteins, we used the knowledgebase SynGO (version: 20180731) [35].
Proteins that were annotated within SynGO were considered enriched for the synapse, with a division into preand postsynaptic based on ontology structure.

Overrepresentation analysis
Gene ontology (GO) enrichment was performed using g:Profiler web server (version: rev 1760 e93 eg40) [58], with all settings on default, g:Profiler-based multiple testing correction (g:SCS method), and with the total of DIA quantified proteins as background for both FTD subtypes. Classical GO terms, i.e. biological process (BP), cellular component (CC), and molecular function (MF) were examined, taking only terms containing five or more proteins into account. When possible, GO terms were further organized into GO groups in keeping with shared functions and proteins. For visualization, only 'Best Per Parent' GO terms are shown. These were selected by hierarchical filtering (moderate), where for every parent GO term its sibling term with the strongest p-value was chosen. For detailed analysis regarding affected synaptic processes we used SynGO (version: 20180731) [35], with FDR-based multiple testing correction and the total of DIA quantified proteins as background for both FTD subtypes. For detailed analysis on affected mitochondrial processes we used the human MitoCarta inventory (version 3.0) [57] in combination with the PANTHER Classification System (version 14.0) [47]. Differentially expressed proteins that were annotated to the mitochondrion according to MitoCarta were analysed for GO enrichment using the PANTHER Overrepresentation Test, with a Fisher's Exact test and FDR-based multiple testing correction, and with the total set of MitoCarta annotated proteins detected within the DIA quantified proteins as background.

Immunoblotting for validation of differential protein expression
Post-mortem middle frontal gyrus and middle temporal gyrus cortical tissues from a random subset of patients and controls (n = 8/group) were selected from the original study cohort. Additionally, frontal and temporal cortex tissues from newly confirmed genetic FTD-GRN (n = 2) and FTD-MAPT (n = 3) cases were requested from the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam, to use as an independent validation cohort ( Table 2).
Protein extracts for immunoblotting were prepared by lysis of whole cortical tissue in Laemmli reducing SDS sample buffer using a 1:20 tissue weight to lysis buffer ratio. Proteins were denatured at 98 °C for 5 min, with the exception of samples used for immunoblotting of mitochondrial proteins, which were denatured at 50 °C for 5 min. Proteins were separated by SDS-PAGE using Criterion ™ TGX stain-free ™ precast gels (Bio-Rad, Hercules, CA, USA) and transferred (40 V o/n at 4 °C) onto a 0.45 µm PVDF membrane (Merck Millipore), which was pre-incubated in 100% methanol. Membranes were blocked with 5% non-fat milk (Sigma-Aldrich, St. Louis, MO, USA), incubated with primary antibody at RT for 2 h and then with matching HRP-conjugated secondary antibodies at RT for 1 h (Agilent Dako, Santa Clara, CA, USA). After washing, the membranes were scanned Table 2 Demographic, clinical, and post-mortem characteristics for cortical frontal and temporal samples from the independent validation cohort ApoE profile information was unknown for these cases. n/a; not available

Alzheimer's disease proteomics
The cortical temporal FTD-MAPT protein profile was compared with a cortical temporal Alzheimer's disease (AD) protein profile. Proteomic data was taken from a subset of samples originating from the post-mortem brain cohort of the 100-plus Study, a research initiative focussing on the mechanisms of healthy aging. From this cohort, containing NDCs, AD patients, and healthy centenarians, we selected post-mortem middle temporal gyrus tissue from 10 sporadic AD cases with Braak tau score ≥ 5 and 10 non-demented controls. All tissues came from the Netherlands Brain Bank, Netherlands Institute for Neuroscience, Amsterdam, and were age and sex matched with the temporal FTD-MAPT samples (Table 3). Tissue was processed as described above, with a few alterations; an equal volume of 0.5 × 10 9 µm 3 of grey matter tissue was collected for each sample. Proteins were size separated on 10% Bis-Tris acrylamide gels using 1.5 M Tris/Glycine SDS running buffer pH 8.3. Gels were fixed overnight and shortly stained with colloidal Coomassie Blue G-250 the next morning. After destaining, trypsin digestion, and peptide extraction, samples were dried and dissolved again with 100 µl Mobile phase A (2% acetonitrile/0.1% formic acid) to be cleaned using the OASIS filter plate according to protocol (Waters Chromatography Europe BV, Etten-Leur, The Netherlands). DIA proteomics data was obtained using an identical approach, with the exception of gel fractionation. A spectral library was prepared using peptides from pooled protein extracts from a mix of NDC, AD, and centenarian samples, which identified 4,948 protein groups by MaxQuant search (version 1.6.3.4) against the human proteome using the UniProt FASTA (release May 2018) and Biognosys iRT FASTA databases. Data extraction and analysis, and statistical comparison of NDC vs AD cases was performed using the same methods.

Cohort description
Cortical tissues from the middle frontal gyrus and middle temporal gyrus were collected from nine FTD-GRN brains, 13 FTD-MAPT brains, and 11 NDC brains. Both patient groups had significantly lower ages than NDCs and their post-mortem brain weights and CSF pH values were lower (Table 1). Neuropathological examination (see images in Additional File 1) revealed that all FTD-GRN cases exhibited TDP-43 immunoreactivity in both cortical areas, consistent with TDP-subtype A. All FTD-MAPT cases were characterized by tau-positive neuronal inclusions, neuropil threads, and tangles in both cortical areas, with MAPT-variant specific features.

Proteomic analysis of FTD-GRN and FTD-MAPT shows brain region-specific protein expression
Using DIA LC-MS/MS we measured abundances of 22,995 unique peptides. Applying our quality value cut-off on all samples together, yielded 9,545 unique peptides, mapping to 2,040 unique proteins measured. Analysis of technical replicates showed a median coefficient of variation of 0.13 in protein abundances, indicating high reproducibility between samples (Additional File 2). For further statistical analysis, quality value cut-off selection of peptides was performed for single disease comparisons (FTD vs NDC) and frontal and temporal samples separately (Table 4). Significant differential protein expression (Fig. 2, Table 5, and Additional File 3) for FTD-GRN vs NDC was mainly found for frontal cortex (579 proteins), and almost absent in temporal cortex (one protein) (q < 0.05). In contrast, for FTD-MAPT vs NDC, significant differential protein expression was found in temporal cortex (488 proteins), and not in frontal cortex (q < 0.05). The top 50 significantly differentially expressed proteins with the largest fold change are listed for both subtypes in Table 6. The differential expression of several well-known neurodegeneration-related proteins (e.g. glial fibrillary acidic protein (GFAP) and MAPT) is highlighted for the most-affected brain region of both subtypes in Additional File 3.

Cell type enrichment analysis reveals cell-specific involvement in FTD-GRN and FTD-MAPT
EWCE analysis of significantly differentially expressed proteins demonstrated distinct cell type involvement for FTD-GRN and FTD-MAPT (Fig. 3). In both, higher expressed proteins showed enrichment for astrocytes and endothelial cells, and lower expressed proteins showed enrichment for neurons. In FTD-MAPT, lower expressed proteins showed additional enrichment for oligodendrocytes.
As these cell type-specific protein expression patterns might be due to changes in cell numbers, protein abundances associated with a particular cell type could also indicate cell loss or gain instead of specific protein regulation. Therefore, we analysed the fold changes between FTD and NDC of all highly-enriched cell type-specific proteins. This showed that the majority of these proteins is normally distributed within the range of NDC protein variation, with values ranging from negative (lower expressed) to positive (higher expressed), and that significantly differentially expressed proteins form a separate population within this distribution (see analysis in Additional File 4). This demonstrates that cell typespecific changes in protein expression are unlikely to be merely the result of cell ratio changes. Taken together, EWCE analysis shows that differential protein expression in genetic FTD subtypes is partly linked to distinct cell types.

GO analysis identifies distinct biological processes involved in FTD-GRN and FTD-MAPT
To determine which biological processes are affected in FTD, we used GO analysis on the differentially higher and lower expressed proteins separately (see extensive results in Additional File 5). 'Best Per Parent' GO terms, further categorized into GO groups, are shown for FTD-GRN in Fig. 4-IA and for FTD-MAPT in Fig. 4-IIA.
FTD-GRN and FTD-MAPT showed overlap in GO groups reflecting processes that may be generally affected in FTD. Higher expressed proteins overlapped for 'Tissue development' , 'Cell adhesion' , and 'Extracellular space' , and lower expressed proteins overlapped for 'Neuron' and 'Synapse' , pointing towards a shared impairment of neuronal function and connectivity. In-depth SynGO analysis shows that differentially expressed synaptic proteins in FTD-GRN and FTD-MAPT are localized in presynaptic and postsynaptic compartments, and are implicated in a wide range of functions (see visualization in Additional File 6 and extensive results in Additional File 7), likely reflecting the overall impact of neurodegeneration on the synapse. Table 4 Number of unique peptides and proteins measured within the FTD cohort Peptides and proteins are selected using quality value filtering on peptide level (see methods for details). Quality value peptide selection separated on brain area and FTD subtype gave the opportunity to analyse the highest number of proteins per statistical comparison In FTD-GRN only, higher expressed proteins are enriched for GO terms related to metabolism and the immune system ( Fig. 4-IA). EWCE analysis of 'Metabolism' proteins shows enrichment for astrocytes ( Fig. 4-IB), and 'Immune' proteins are enriched for endothelial cells (Fig. 4-IC). Furthermore, distinct for FTD-GRN, prominent enrichment of mitochondria-related GO terms for lower expressed proteins is observed. A total of 85 proteins point towards affected mitochondria. Detailed dissection of the mitochondrial regulation shows specific enrichment for the oxidoreductase complex (see Additional File 8), suggesting specific functional alterations rather than overall (structural) downregulation or loss of mitochondria in FTD-GRN. Especially respiratory chain complex I (RCCI) seems to be affected, with 12 of in total 38 RCCI proteins Fig. 2 Differential protein expression in most-affected areas for genetic FTD subtypes shows brain region-specific protein signatures. A Differential protein expression (at q < 0.05) in frontal cortical tissue for FTD-GRN vs NDC. B Differential protein expression (at q < 0.05) in frontal cortical tissue for FTD-MAPT vs NDC. C Differential protein expression (at q < 0.05) in temporal cortical tissue for FTD-GRN vs NDC. D Differential protein expression (at q < 0.05) in temporal cortical tissue for FTD-MAPT vs NDC. Differential expression analysis was done using permutation-based modified t-statistics. An FDR-adjusted q-value threshold of 0.05 was used to discriminate proteins of interest. The number of differentially expressed proteins is extensive in both FTD-GRN (n = 580) and FTD-MAPT (n = 488) cases Table 5 Number of significantly differentially expressed proteins within the FTD cohort Results of differential expression analysis are shown both at non-corrected (p < 0.05) and multiple comparison corrected (q < 0.05) statistical cut-offs     Fig. 4-ID), indicating that these cells might be particularly affected in FTD-GRN.
In FTD-MAPT only, higher expressed proteins are enriched for GO terms related to RNA processing. EWCE analysis of 'RNA processing' proteins showed no specific involvement of cell types (Fig. 4-IIB). Lower expressed proteins in FTD-MAPT showed distinct enrichment for ' Axon' , 'Ion transport' , and 'Plasma membrane' . EWCE analysis on ' Axon' (Fig. 4-IIC) and 'Plasma membrane' proteins ( Fig. 4-IIE) alluded to the involvement of oligodendrocytes. A comparison of statistical effect sizes of all GO group proteins distinct for FTD-GRN and FTD-MAPT confirms that these biological processes are indeed strongly biased towards their respective FTD subtype (see visualizations in Additional File 9 and Additional File 10).
In the process of selecting targets for validation study, proteins were chosen for their biological relevance and distinct regulation in only one FTD subtype. Immunoblotting of selected target proteins for FTD-GRN and FTD-MAPT was carried out on a random subset of samples from our cohort as well as on samples from an independent cohort (Table 2). Results confirm lower expression of mitochondrial RCC proteins for subunits I-IV in FTD-GRN only, though detected differences are not statistically significant per protein. PLP1 expression, which was lower in FTD-MAPT specifically according to our mass spectrometry data, indeed shows a strong trend (p = 0.0595) for decreased expression in FTD-MAPT (see Additional File 11 for all immunoblotting results). Raw images of gels and immunoblots can be found in Additional File 12.
Taken together, GO analysis illustrates both the presence of general neurodegenerative processes and subtype-distinct biological processes for FTD-GRN and FTD-MAPT, with a cell type-specific involvement in many of these processes.

Comparing FTD-MAPT with AD confirms both distinct and general neurodegenerative protein signatures
Tau pathology is a shared disease hallmark between FTD-MAPT and AD. Comparison of differential protein expression in the same brain area between the two might help identify proteins implicated in shared neurodegenerative mechanisms, as well as proteins representing distinct aspects of pathological mechanisms for both diseases. For this, we used a temporal cortical AD vs NDC proteomic data set in which we quantified 3,332 unique proteins, of which 962 were significantly differentially expressed compared to NDC (q < 0.05) (see Additional File 13). When comparing the 1,847 proteins that were quantified in both the FTD-MAPT and AD datasets (also Additional File 13), 195 differentially expressed proteins were shared, and 259 differentially expressed proteins were distinct for FTD-MAPT. The majority of shared proteins have the same direction of differential expression (Fig. 5B).
When comparing EWCE results between AD (Fig. 5A) and FTD-MAPT (Fig. 3C), a shared association of astrocytes with higher expressed proteins, and of neurons with lower expressed proteins was revealed, indicating more common neurodegenerative processes leading to astrogliosis and neurodegeneration as shared mechanisms. Significant endothelial and microglial enrichment is seen in AD, whereas oligodendrocyte involvement is specific for FTD-MAPT, suggesting that these cell types specifically contribute to AD and FTD-MAPT pathology, respectively.  GO analysis of the 195 proteins shared between FTD-MAPT and AD ( Fig. 5C and Additional File 14) highlighted GO groups 'Neuron' and 'Synapse' linked to lower expressed proteins. SynGO analysis showed that the 55 shared differentially expressed synaptic proteins are implicated in a wide range of functions (extensive results seen in Additional File 15). In addition, EWCE analysis demonstrated that shared processes are indeed enriched for astrocytes and neurons (Fig. 5D).
GO analysis of the 259 distinct FTD-MAPT proteins revealed distinct enrichment of 'RNA processing' GO terms for higher expressed proteins, and ' Axon' GO terms, specifically indicating the myelin sheath, for lower expressed proteins (Fig. 5E). EWCE analysis confirmed the subtype-specific involvement of oligodendrocytes in FTD-MAPT and additionally identified oligodendrocyte precursor cells (OPC) as a cell type involved in FTD-MAPT pathology (Fig. 5F). OPCs were not significantly enriched previously (Fig. 3C), demonstrating the power of excluding general neurodegeneration-associated proteins through filtering with the AD protein set.
Thus, comparison of FTD-MAPT with AD aided in delineating common neurodegenerative mechanisms as well as FTD subtype-specific changes, including the specific involvement of certain brain cell types.

Discussion
This study describes the differential proteins expression in cortical frontal and temporal regions of the brain in FTD-GRN and FTD-MAPT. Region-specific protein signatures indicated the involvement of cell type-specific distinct biological processes in these FTD subtypes. Importantly, comparing FTD-MAPT to AD revealed overlapping neurodegenerative processes as well as the existence of FTD-MAPT-specific disease mechanisms.
The occurrence and extent of differential protein expression for both subtypes reflect known brain atrophy patterns. In FTD-GRN, atrophy is usually found in a diffuse hemispheric manner, including in frontal and temporal lobes [10,62]. Six FTD-GRN patients, for which we had neuropathological reports, presented with severe frontal and little to no temporal atrophy, explaining the frontal focus of their differential protein expression. FTD-MAPT patients are characterized by predominant temporal lobe atrophy [10,62], which is reflected by the extensive differential protein expression specifically in the temporal cortex of our cases.
A clear indication that disease-specific involvement of cell types and biological processes in neurodegeneration exists, even in end-stages, came from comparison with a protein signature of AD. Although both FTD subtypes and AD share involvement of astrocytes and neurons, there was clear discrepancy in microglial enrichment. Although microglia have been implicated in FTD [24], significant microglial enrichment was absent in this study. A recent investigation in FTD demonstrated variable involvement of microglia according to brain region and subtype, with a higher burden in white versus grey matter [75]. As our study specifically analysed grey matter tissue, this could explain the lack of microglial enrichment. Further comparison between FTD-MAPT and AD confirmed the existence of both shared general neurodegenerative as well as FTD-MAPT-specific processes.
In FTD-GRN, we found distinct involvement of immune-related processes, primarily linked to endothelial cells. Involvement of endothelial cells in FTD has only been described incidentally, mainly in relation to bloodbrain-barrier (BBB) pathology [32,33,52]. Their possible active involvement in immune processes is a new finding. The increased expression of intercellular adhesion molecule 1 (ICAM1; 6,80x) hints towards a role for the Fig. 5 Comparison of profiles between FTD-MAPT and AD demonstrates presence of FTD subtype-specific and general neurodegenerative protein signatures. A EWCE analysis of lower (n = 406) and higher (n = 556) expressed proteins in temporal cortical AD vs NDC shows astrocyte, endothelial cell, and microglial cell enrichment (all p = 0) for higher expressed proteins, and excitatory and inhibitory neuron enrichment (both p = 0) for lower expressed proteins. Comparison with EWCE results for FTD-MAPT confirms distinct involvement of oligodendrocytes in FTD-MAPT. B Overlay of differentially expressed proteins shared between FTD-MAPT and AD (n = 195) on the protein profile for FTD-MAPT shows that the majority of shared proteins have the same direction of differential expression. C GO analysis on shared proteins. D EWCE analysis on shared proteins. Significant enrichment is seen for astrocytes (p = 0) for higher expressed proteins, and for excitatory (p = 0) and inhibitory (p = 0.002) neurons for lower expressed proteins. E GO analysis on proteins that are only differentially expressed in FTD-MAPT (n = 259). 'RNA processing' , 'Ion transport' , 'Axon' and 'Neuron' GO groups are confirmed to be distinct for the FTD-MAPT subtype when compared with AD. F EWCE analysis on proteins only differentially expressed in FTD-MAPT. Significant enrichment is seen for astrocytes (p = 0) and OPCs (p = 0.0074) for higher expressed proteins, and for excitatory neurons and oligodendrocytes (both p = 0) for lower expressed proteins. EWCE analysis is performed using bootstrapped t-test statistics with multiple testing correction by the Benjamini-Hochberg method. traversing of leukocytes across the BBB, which appears to be affected more by ICAM1 expression than inflammatory molecules or BBB integrity [1,41]. This finding is strengthened by multiple GO terms pointing towards the presence of activated peripheral immune cells and an ongoing immune response. Discernment of the specific response type is difficult, as higher expression of proinflammatory (e.g. PPIA), anti-inflammatory (e.g. ANXA1 and ASAH1), and ambiguous inflammatory proteins (e.g. CD44 and PRDX1/6) was detected. Further research is needed to characterize the precise role of the immune response in FTD-GRN. Neuroinflammation is increasingly implicated in FTD disease progression [7,8]. A recent proteomic study of frontal cortex tissue from FTD-TDP patients identified an 'inflammatory' protein module enriched for astrocytes and microglia [70]. The fact that we identify endothelial cells at the forefront of immune-related processes in our FTD-GRN cohort might be specifically related to the genetic cause of FTD in these cases. This is illustrative of the strength of our approach to specifically select genetic cases, which limits possible interference of biological processes related to sporadic FTD. We also identified a striking pattern of mitochondrial dysregulation, suggesting specific functional alterations of mitochondria in FTD-GRN. This finding is consistent with Umoh et al., in which decreased expression of a mitochondrial module in frontal cortex tissue from FTD-TDP patients was reported [70]. Although mitochondrial dysfunction seems to be a common denominator of neurodegeneration [20,49], the differential expression of RCCI proteins in our cohort was most prominent in FTD-GRN when compared to FTD-MAPT. EWCE analysis showed enrichment for neurons, suggesting they might be the focal site of mitochondrial dysregulation.
In FTD-MAPT, protein signatures indicated affected processing and trafficking of RNA. Disturbances of RNA processing and nucleocytoplasmic transport (NCT) have repeatedly been reported in FTD in relation to TDP-43 [12,22,40,51], FUS [40], and C9ORF72 repeat expansion [21,67]. Recently, for the first time, FTD-related MAPT mutations were linked to microtubule-mediated nuclear deformation and disruption of NCT in human iPSCderived neurons [53]. By specifically selecting genetic cases in our cohort we were able to more prominently highlight the involvement of RNA processing and transport in FTD-MAPT. Comparison with AD supported the view that these processes are specifically associated with FTD-MAPT. Our results also illustrate that FTD-MAPT cases and FTD cases harbouring TDP-43 pathology (FTD-TDP) regardless of the presence of genetic mutations, might be more alike than previously expected.
In addition, ' Axon' and 'Plasma membrane' proteins showed enrichment for neurons and oligodendrocytes in FTD-MAPT. Interestingly, tau inclusions have been described in oligodendrocytes of patients with FTD-MAPT [16,34], and studies with tau transgenic mice indicate that these inclusions disrupt axonal transport, leading to impairments in myelin and axon integrity [29]. Alternatively, tau accumulation in axons may indirectly impair oligodendrocyte function due to their functional interaction. Comparison of FTD-MAPT to AD highlighted involvement of the myelin sheath as highly distinctive, hinting at impaired axon-myelin interactions in FTD-MAPT. Perhaps the distinguishing factor in FTD-MAPT is indeed the presence of tau pathology of genetic origin in all cell types, versus the neuronal and extracellular tau pathology in (nongenetic) AD.
A potential limitation of proteomic studies using whole-tissue is the fact that protein abundances are dependent on expression changes in multiple cell types, and are consequently affected by changes in cell type ratios within the studied tissue. As neuronal loss is characteristic of neurodegeneration, changes in cell type ratios within the diseased brain are expected. However, our analysis suggested that detected protein expression differences are not simply caused by disease-induced cell loss or gain. Furthermore, to improve stratification of brain cell types involved in the FTD subtypes, we applied cell type enrichment analysis. This approach seems to adequately address the issue of mixed cell type populations, as we were able to demonstrate the involvement of different cell types within our whole-tissue data. Nonetheless, the results should be carefully interpreted, as our method of using scR-NAseq data sets for the inference of cell types on a protein level has limitations as well. For instance, it is well known from literature that mRNA-to-protein correlations are only ~ 40% [17,28,36,37,55,66,72] and so mRNA specificity across cell types might not always reflect the cell type specificity at the protein level. In addition, a recent study has demonstrated a cell-type specific aging effect on the transcriptomic level [59], suggesting that age-related regulation of mRNA and protein levels might influence the cell type specificity ratios we infer from scRNAseq data sets. A benefit from our 'enrichment'-based approach is that cell type inference is done using protein groups and not individual proteins, which reduces the possible effect of individual poorly-correlated or aging-sensitive proteins. Future (single) cell type-specific proteomics approaches [15,44,73] may further help disentangle the different causal or consequential processes for brain cell types in FTD.