Network analysis of the progranulin-deficient mouse brain proteome reveals pathogenic mechanisms shared in human frontotemporal dementia caused by GRN mutations

Heterozygous, loss-of-function mutations in the granulin gene (GRN) encoding progranulin (PGRN) are a common cause of frontotemporal dementia (FTD). Homozygous GRN mutations cause neuronal ceroid lipofuscinosis-11 (CLN11), a lysosome storage disease. PGRN is a secreted glycoprotein that can be proteolytically cleaved into seven bioactive 6 kDa granulins. However, it is unclear how deficiency of PGRN and granulins causes neurodegeneration. To gain insight into the mechanisms of FTD pathogenesis, we utilized Tandem Mass Tag isobaric labeling mass spectrometry to perform an unbiased quantitative proteomic analysis of whole-brain tissue from wild type (Grn+/+) and Grn knockout (Grn−/−) mice at 3- and 19-months of age. At 3-months lysosomal proteins (i.e. Gns, Scarb2, Hexb) are selectively increased indicating lysosomal dysfunction is an early consequence of PGRN deficiency. Additionally, proteins involved in lipid metabolism (Acly, Apoc3, Asah1, Gpld1, Ppt1, and Naaa) are decreased; suggesting lysosomal degradation of lipids may be impaired in the Grn−/− brain. Systems biology using weighted correlation network analysis (WGCNA) of the Grn−/− brain proteome identified 26 modules of highly co-expressed proteins. Three modules strongly correlated to Grn deficiency and were enriched with lysosomal proteins (Gpnmb, CtsD, CtsZ, and Tpp1) and inflammatory proteins (Lgals3, GFAP, CD44, S100a, and C1qa). We find that lysosomal dysregulation is exacerbated with age in the Grn−/− mouse brain leading to neuroinflammation, synaptic loss, and decreased markers of oligodendrocytes, myelin, and neurons. In particular, GPNMB and LGALS3 (galectin-3) were upregulated by microglia and elevated in FTD-GRN brain samples, indicating common pathogenic pathways are dysregulated in human FTD cases and Grn−/− mice. GPNMB levels were significantly increased in the cerebrospinal fluid of FTD-GRN patients, but not in MAPT or C9orf72 carriers, suggesting GPNMB could be a biomarker specific to FTD-GRN to monitor disease onset, progression, and drug response. Our findings support the idea that insufficiency of PGRN and granulins in humans causes neurodegeneration through lysosomal dysfunction, defects in autophagy, and neuroinflammation, which could be targeted to develop effective therapies.

Background Frontotemporal lobar degeneration (FTLD) is the most common cause of dementia in people under the age of 60 [7]. Frontotemporal dementia (FTD) is the clinical manifestation of FTLD neuropathology and major clinical symptoms can be divided into either progressive deficits in executive function and behavior or language [26]. Importantly, ~ 30% of FTD patients have a familial history of FTD or related neurodegenerative disease, highlighting the important role of genetics in disease pathogenesis [26,39,43]. Taken together autosomal dominant mutations in three genes, progranulin (GRN), chromosome 9 open reading frame 72 (C9orf72), and microtubule associated protein tau (MAPT), account for the majority of FTD heritability [43]. Less common mutations in other genes encoding TAR DNA binding protein 43 (TDP-43; TARDBP) [19,32,95], sequestosome-1/p62 (SQSTM1) [67,121], charged multi-vesicular body protein 2b (CHMP2B) [104,120], valosin-containing protein (VCP) [45,126], TANK-binding kinase 1 (TBK1) [33,38,92], among other rare genes can cause FTD [44]. Despite these advances in understanding the genetic causes of FTD, the normal, physiologic function of many proteins encoded by genes mutated in FTD is still unclear. Understanding the function and dysfunction of proteins linked to FTD is critical to developing effective therapies.
In particular, the function of the progranulin (PGRN) protein, encoded by GRN, has remained an enigma. The PGRN protein family is over a billion years old and evolutionarily conserved across many species, suggesting PGRN has a critical function [85]. Heterozygous GRN mutations cause FTD by decreasing PGRN mRNA and protein by 50% or more [6,24,31,36,57,73]. Although FTD-GRN mutations lead to TDP-43 pathology and dysfunction [17,23,81], it is still unclear why deficiency of PGRN ultimately causes neurodegeneration.
PGRN is a pleiotropic, cysteine rich secreted protein composed of one half-length and seven full-length domain repeats called granulins that are connected by peptide linker regions [112]. PGRN is highly expressed in neuronal and microglial cells throughout the brain [88]. PGRN has been implicated in various physiological functions, ranging from extracellular signaling through membrane receptors [25,76,80,129], neuroprotection [35,65,117,130], to modulating inflammation [21,70,137]. PGRN can be cleaved by a variety of proteases to release individual ~ 6-kDa granulins, which are also bioactive [8,47,137]. However, the precise functions of PGRN and granulins are unclear, and the pathways that lead from deficiency of PGRN and granulins to FTD are still unknown, which has impeded progress towards developing therapies for FTD-GRN.
Another roadblock in the development of therapies for FTD is a lack of specific and sensitive biomarkers [40]. The identification of reliable biomarkers is important to distinguish FTD from Alzheimer's disease or other neurodegenerative diseases. Importantly, appropriate biomarkers are also useful to monitor disease progression and assess efficacy of potential drugs. Elevated levels of neurofilament light chain (Nfl) in CSF and plasma are one promising biomarker for symptomatic FTD patients that harbor mutations in GRN, C9orf72, or MAPT [58]. However, increased levels of Nfl in plasma or CSF are not specific to FTD and Nfl is increased in several other neurodegenerative diseases [124]. More research is needed to identify novel biomarkers to diagnose, discriminate, and ultimately treat the different sub-types of FTD.
To investigate the function of PGRN, provide insight into FTD pathogenesis, and identify potential biomarkers, we performed an unbiased quantitative proteomics analysis of whole-brain tissue from wild type (Grn +/+ ) and Grn knockout (Grn −/− ) mice at 3-and 19-months of age. We utilized the Grn −/− mouse model because they share many pathological features with human FTD-GRN, including microgliosis, lipofuscinosis, accumulation of ubiquitinated proteins, and behavioral impairment [1,82,96]. We identified a variety of proteins that were increased or decreased in the Grn −/− mouse brain proteome that were subsequently validated using biochemical and immunohistological approaches. In particular, lysosomal proteins were the most significantly altered in young Grn −/− mice, indicating that lysosomal dysfunction is an early event when PGRN expression is lost. In addition, we identified two proteins, transmembrane glycoprotein NMB (GPNMB) and galectin-3, which are predominantly expressed in microglia and are key hubs of a larger network of dysregulated proteins that change with age in Grn −/− mice. Moreover, GPNMB and galectin-3 were significantly elevated in the lysates of FTD-GRN brain samples compared to healthy controls. Together, these data suggest that common pathogenic pathways are dysregulated in both Grn −/− mice and human FTD cases caused by GRN haploinsufficiency.

Mouse brain sample processing for proteomics and biochemical analysis
Grn −/− mice used in this study were originally generated in Dr. Aihao Ding's laboratory [131,132] and purchased from the Jackson Laboratory (B6(Cg)-Grntm1.1Aidi/, IMSR Cat# JAX:013175, RRID:IMSR_JAX:013175). Mice were housed in the Department of Animal Resources at Emory University and all work was approved by the Institutional Animal Care and Use Committee (IACUC) and performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Mice were sacrificed at various ages and whole brains were dissected from the skull and frozen immediately in liquid nitrogen. Brain tissue was ground to a fine powder under liquid nitrogen using a mortar and pestle and stored at − 80 °C as previously described [48]. Frozen cerebral cortex brain samples from 3-month old Grn +/+ wild type (n = 4) and Grn −/− knock out (n = 4) mice and 19-month old Grn +/+ wild type (n = 4) and Grn −/− knock out (n = 4) mice were collected. For proteomic analysis, approximately 100 mg of mouse brain tissue powder was homogenized in 500 μL of urea lysis buffer (8 M urea, 100 mM NaH 2 PO 4 , pH 8.5), supplemented with 5 μL (100 × stock) HALT protease and phosphatase inhibitor cocktail (Pierce) using a Bullet Blender (Next Advance) and 750 mg of steel beads (Next Advance). Protein supernatants were then transferred to a new 1.5 mL Eppendorf tube and sonicated (Sonic Dismembrator, Fisher Scientific) 3 times for 5 s with 15 s intervals of rest at 30% amplitude. Protein concentration was measured with the bicinchoninic acid (BCA) assay, and samples were frozen in aliquots at − 80 °C. Protein integrity was checked by one-dimensional SDS-PAGE. For subsequent immunoblots or ELISAs, an equal weight of brain powder was homogenized with 6 × volume per weight in cytoplasmic extraction buffer (CEB), membrane extraction buffer (MEB) from a Subcellular Fractionation kit (Pierce), or with 10 × volume per weight in RIPA buffer (150 mM NaCl, 0.1% SDS, 1% Triton-X 100, 0.5% sodium deoxycholate, 50 mM Tris, pH 8.0), supplemented with 1 × HALT protease and phosphatase inhibitor cocktail (UK286007, Thermo scientific) as previously described [47]. Brain powder suspensions were sonicated (Sonic Dismembrator, Fisher Scientific) 5 times for 2 s with 8 s intervals of rest at 30% amplitude. Brain lysates were obtained by centrifugation at 500 × rcf for 10 min at 4 °C. Protein supernatants were transferred to a new 1.5 mL Eppendorf tube. Protein concentration was measured with the bicinchoninic acid (BCA) assay and saved samples were saved at − 80 °C.

Tandem mass tag (TMT) peptide labeling and electrostatic repulsion-hydrophilic interaction chromatography (ERLIC) fractionation
Proteolytic digestion of protein samples and cleanup was performed as previously described [14]. Briefly, protein samples were reduced with 1 mM dithiothreitol (DTT) for 30 min, alkylated with 5 mM iodoacetamide (IAA) in the dark for an additional 30 min and then diluted 8-fold with 50 mM triethylammonium bicarbonate (TEAB). Overnight digestion was performed with 1:100 (w/w) Lysyl endopeptidase (Wako) followed by an additional 12-h digestion with Trypsin at 1:50 (w/w). Peptide solutions were acidified and desalted with a C18 Sep-Pak column (Waters). A 2 μg equivalent of each sample elution was pooled and used to create a global internal standard (GIS) and all samples were dried under vacuum. Tandem mass tag (TMT) peptide labeling was performed according to manufacturer's instructions and as previously described [14]. One batch of 10-plex TMT kits (Thermo Fisher) was used to label 8 samples and two GIS mixtures per batch. Electrostatic repulsion-hydrophilic interaction chromatography (ERLIC) offline fractionation was performed as previously described [14,15]. Briefly, dried samples were re-suspended in 100 μL of ERLIC buffer A (90% acetonitrile with 0.1% acetic acid) and separated on a PolyWAX LP column (20 cm by 3.2 mm packed with 300 Å pore, 5 μm beads (PolyLC Inc.) and elution fractions were recovered over a 45-min gradient from 0 to 50% ERLIC buffer B (30% ACN with 0.1% FA).

LC-MS/MS and TMT data acquisition
Assuming equal distribution of peptide concentration across all ERLIC fractions, 10 μL of loading buffer (0.1% TFA) was added to each of the fractions and 2 μL was separated on a 25 cm long 75 μm internal diameter fused silica column (New Objective, Woburn, MA) packed inhouse with 1.9 μm Reprosil-Pur C 18 -AQ resin. The LC-MS/MS platform consisted of a Dionex RSLCnano UPLC coupled to an Orbitrap Fusion mass spectrometer with a Flex nano-electrospray ion source (Thermo Fisher). Sample elution was performed over a gradient of 3 to 30% Buffer B (0.1% formic acid in ACN) over 105 min (flow rate started at 300 nL/min and ended at 350 nL/ min), from 30 to 60% B over 20 min at 350 nL/min, and from 60 to 99% B over 5 min at 350 nL/min. The column was equilibrated with 1% B for 10 min at a flow rate that increased from 350 nL/min to 400 nL/min. The MS was operated in positive ion mode and utilized the synchronous precursor selection (SPS)-MS3 method for reporter ion quantitation as described [90]. The full scan range was 380-1500 m/z at a nominal resolution of 120,000 at 200 m/z and automatic gain control (AGC) set to 2 × 10 5 . Collision-induced dissociation (CID)-Tandem MS/MS at 35% normalized collision energy (CE) and higher energy collision dissociation (HCD) SPS-MS3 at 65% normalized collision energy (CE) were collected at top speed with 3 s cycles. For SPS, the top 10 product ions were notched and fragmented.

Protein identification and quantification
Raw data files from the Orbitrap Fusion were processed using Proteome Discover (version 2.1). Collected MS/ MS spectra were searched against the UniProt mouse proteome database (54,489 total sequences). SEQUEST parameters were specified as: trypsin enzyme, two missed cleavages allowed, minimum peptide length of 6, TMT tags on lysine residues and peptide N-termini (+ 229.162932 Da) and carbamidomethylation of cysteine residues (+ 57.02146 Da) as fixed modifications, oxidation of methionine residues (+ 15.99492 Da) and deamidation of asparagine and glutamine (+ 0.984016 Da) as a variable modification, precursor mass tolerance of 20 ppm, and a fragment mass tolerance of 0.6 Da. Peptide spectral match (PSM) error rates were determined using the target-decoy strategy coupled to Percolator [16] modeling of true and false matches. Reporter ions were quantified from MS3 scans using an integration tolerance of 20 ppm with the most confident centroid setting. An MS2 spectral assignment false discovery rate (FDR) of less than 1% was achieved by applying the target-decoy strategy. Following spectral assignment, peptides were assembled into proteins and were further filtered based on the combined probabilities of their constituent peptides to a final FDR of 1%. In cases of redundancy, shared peptides were assigned to the protein sequence with the most matching peptides, thus adhering to principles of parsimony. In total, 8695 proteins were identified by tandem mass spectrometry. The search results and TMT quantification as well as raw LC-MS/MS files are included in the ProteomeXchange online repository with identifier (to be uploaded and assigned). Prior to data analysis, preliminary network connectivity outliers were determined as samples with connectivity beyond 3 standard deviations from the mean using Oldham's "Sample-Networks" v1.06 R script as previously published, but no cases were identified for removal [84,102].

Data preparation
Each TMT batch consisted of a single age group, either 3 month or 19 month, of Grn +/+ and Grn −/− mice which led to some proteins being measured in one batch and not in the other. As a result, batch wise missing values for these proteins across the two TMT multiplexes led to exactly 50% missing values. To limit the effects of batch wise differences on data analysis methods, initial analysis was conducted on the 6566 unique proteins which were measured across all batches. Proteins with 50% of quantified values were later mapped into the existing network and volcano plots (2129 proteins).

Differential expression analysis and MetaScape
Differentially enriched or depleted proteins (p ≤ 0.05) were identified by one-way ANOVA with post-hoc Tukey HSD test comparing four groups: 3-month-old Grn +/+ , 3-month-old Grn −/− , 19-month-old Grn +/+ mice and 19-month-old Grn −/− mice. Differential expression of proteins were visualized with volcano plots generated using the ggplot2 package in Microsoft R Open v3.4.2. Significantly differentially expressed proteins were determined by both having a p ≤ 0.05 and a fold change difference of greater than log 2 (1.25) or less than − log 2 (1.25) (a minimum 25% fold change).
Proteins that were significantly differentially expressed in Grn +/+ and Grn −/− mouse brain proteomes were analyzed using MetaScape as described [136]. Briefly, differentially expressed genes were analyzed using the MetaScape web portal (https ://metas cape.org/) to identify enriched ontology clusters in the data set. Statistically enriched terms (i.e. GO/KEGG terms), accumulative hypergeometric p-values, and enrichment factors were calculated and used for filtering. The significant terms were hierarchically clustered into a tree based on Kappastatistical similarities among their gene memberships, then 0.3 kappa score applied as threshold to cast the tree into term clusters.

Weighed co-expression network analysis
Following previously described procedures of WGCNA [102], a weighted protein co-expression network was generated using the protein abundance network of 6566 unique proteins. WGCNA::blockwiseModules() function was used with the following settings: soft threshold power beta = 29, deepSplit = 4, minimum module size of 25, TOMdenom = "mean", merge cut height of 0.07, pam-Stage = TRUE and a reassignment threshold of p < 0.05. Hierarchical protein correlation clustering analysis was conducted using 1-TOM, and initial module identifications were established using dynamic tree cutting as implemented in the WGCNA::blockwiseModules() function [66]. Module eigenproteins were defined as the first principal component of coexpression module protein log 2 (abundances) [74]. The module membership measure is defined as k ME . K ME is the pearson correlation between the expression pattern of the protein and the module eigenprotein. Bicor correlation was used for pairwise complete correlation of non-missing measurements with the cognate samples' levels in module eigenproteins. The top correlated eigenprotein was used to assign the proteins with 50% missing values to a module, albeit with diminished confidence due to N = 8 instead of 16.

Gene ontology (GO) and cell-type enrichment analysis
To characterize groups of differentially expressed proteins and co-expressed proteins, we used GO Elite v1.2.5 as previously published [102] with pruned output visualized using an in-house R script. Overrepresentation of ontologies in each module was determined by Z-score value. Enrichment of cell type across co-expression modules was investigated by intersecting module proteins with lists of proteins known to be expressed by each cell marker [103] and assessing significance of overlap using a one-tailed Fisher exact hypergeometric overlap test. After assessing significance, the p-values were corrected by the Benjamini-Hochberg method. Cell type-specific gene lists are provided in Additional file 1: table S1.

Mouse plasma samples
Mouse blood samples were collected by cheek vein puncture into EDTA tubes and chilled on ice for 1 h. Plasma samples were separated by centrifugation at 500 rcf for 10 min at 4 °C and stored at − 80 °C for ELISA.

Human samples processing Human brain samples
Human brain tissue was provided by the Emory Brain Tissue Bank (Emory University Goizueta Alzheimer's Disease Research Center, Atlanta, Georgia, USA) and the Mayo Clinic Brain Bank (Jacksonville, Florida). A summary of the neuropathological and clinical descriptors of human post-mortem samples used for immunostaining and ELISA is provided as a supplementary table (Additional File 2: table S2). This included a total of frozen frontal cortex samples from FTD-GRN patients (n = 21, 11 males and 10 females, mean age 66.81 ± 1.66 years), cognitively normal controls (n = 23, 13 males and 10 females, mean age 62.17 ± 2.51 years), among which 5 corresponding paraffin-embedded brain sections in each group were collected. Human brain tissue was processed as described for mouse brains described above. Briefly, an equal weight of brain powder was homogenized with 10 × volume per weight in RIPA buffer and saved for protein analysis (BCA), immunoblot, and ELISA.

Human CSF samples
Human cerebrospinal fluid (CSF) samples were obtained from the Advancing Research and Treatment for Frontotemporal Lobar Degeneration (ARTFL) and the Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects (LEFFTDS) studies (ARTFL-LEFFTDS), which were housed at the National Centralized Repository for Alzheimer Disease and Related Dementias (NCRAD). CSF was obtained by lumbar puncture from individuals with genetic mutations associated with FTD (FTD-GRN n = 13, FTD-C9orf72 n = 13, or FTD-MAPT n = 12) or cognitively normal controls (n = 14) and stored at − 80 °C before analysis. NCRAD receives government support under a cooperative agreement grant (U24 AG21886) awarded by the National Institute on Aging (NIA).

ELISA
General ELISA protocols were performed as previously described [55,63,64]. Levels of GPNMB and galectin-3 in tissue were quantified using ELISAs Duosets according to the manufacturer's protocol: human GPNMB (AF2550, R&D Systems), mouse GPNMB (AF2330, R&D Systems), human Galectin-3 (DY1154, R&D Systems), or mouse Galectin-3 (DY1197, R&D Systems). Additional ELISA reagents were from ELISA reagents kit 2 (DY008, R&D Systems). Briefly, ELISA plates were coated with 100 µL/ well coating antibody diluted in coating buffer overnight at room temperature. After washing with 1 × wash buffer 3 times, plates were blocked with 1X reagent diluent 300 µL/well at least 1 h. Human brain RIPA lysates, CSF samples, mouse brain RIPA lysates or mouse plasma were diluted at 1:20, 1:4, 1:50, or 1:10 respectively, in 1X reagent diluent and added to plates with standard at 100 µL/ well and incubated for 2 h. Next, 100 µl of diluted detection antibody was added per well. 100 µL of the working dilution of Streptavidin-HRP was added to each well and plates were incubated for 20 min in the dark. Then, 100 µL of substrate solution was added to each well and incubated for 15 min while still avoiding light. To stop the reaction, 50 µL of 2 N sulfuric acid was added to each well. Finally, ELISA plate sample absorbance (450 nm signal and 570 nm for background correction reading) was measured on an Epoch plate reader (BioTek) and processed using Gen5 software (BioTek). All samples were run in duplicate and values fell within the standard curve generated with recombinant human or mouse GPNMB or galectin-3 protein provided in DuoSet ELISA kits. Protein measurements in human and mouse brain samples were normalized to the amount of total protein added per well.

Statistical analysis
Standard curve generation and statistical analyses were performed by GraphPad Prism 8.0. An unpaired student's t-test for two groups and one-way or two-way ANOVA for more than two groups were used to generate p values.
In 19-month-old Grn −/− mice, 119 proteins were increased and 20 proteins were decreased compared to Grn +/+ mice. In the brains of 19-month-old Grn −/− mice we found that an even larger number of lysosomal proteins were increased, including glycoprotein NMB (GPNMB), which is the most upregulated protein in aged Grn −/− mice (Fig. 1c). In addition to lysosome alterations, GO analysis revealed novel upregulated pathways in 19 month old Grn −/− mice including proteins involved in gliogenesis (C1qa, Dbi, Gfap, P2rx4, Stat3, Bin1, Zfp365) as well as inflammation, complement, and coagulation cascades (C1qa, C1qb, C1qc, C4b, Itgb2, and A2m) [136]. Thus, our initial analysis of the mouse brain proteome reveals an early dysregulation of proteins involved in lysosomal function and lipid metabolism in 3-month-old Grn −/− mice, which is exacerbated with age, leading to upregulation of proteins involved in the innate immune response and inflammation.
Co-expression protein analysis with Grn +/+ and Grn −/− mouse Next, we performed weighted co-expression network analysis (WGCNA) on the Grn +/+ and Grn −/− mouse brain proteome datasets to determine the relationship between the changes in protein abundance we observed and biological pathways and cell types [66]. Network analysis identified 26 modules of strongly co-expressed groups of proteins ( Fig. 1d; Additional file 1: table S4,  table S5). Each protein co-expression module is defined by its first principal component, an eigenprotein, which is also the most representative weighted protein expression pattern across samples for a group of co-expressed proteins [102]. Modules were clustered based on relatedness defined by the correlation distance between eigenproteins (Fig. 1d). We calculated the relationship between modules and the biweight midcorrelation (bicor) [106] of eigenproteins to genotype and age of Grn mice and identified a number of modules that significantly correlated (Fig. 1e). Comparison of modules in the Grn +/+ 19-month versus Grn +/+ 3-month data set or Grn −/− 19-month versus 3-month Grn −/− data set identified many of the same modules, suggesting this analysis identifies common or shared protein modules that are generally altered during aging (Additional file 3: Fig. S1, online resource).

Aged Grn −/− mouse brains accumulate lysosomal proteins and markers of neuroinflammation
Because GO and network analysis revealed that dysregulation of the lysosomal pathway was an early and highly significant event in the Grn −/− mouse brain proteome, we focused on validating and examining these changes using biochemical and immunological orthogonal approaches. First, we examined the expression levels and neuroanatomical location of two lysosomal proteins, cathepsin Z (protein abbreviation, Cat Z; gene, Ctsz) and cathepsin D (protein abbreviation, Cat D; gene Ctsd), enriched in modules M7 and M6, respectively. We performed immunoblotting of whole brain lysates from 3-month (n = 8) and 18-month-old (n = 8) Grn +/+ and Grn −/− mice. First, we examined Cat Z, a unique cysteine cathepsin, previously implicated in neurodegenerative diseases, but not examined in the context of FTD or PGRN biology [3,13,110]. Levels of Cat Z increased in Grn +/+ (1.5-fold) and Grn −/− (2.3-fold) in whole brain lysates of 18-month mouse brain as measured by quantitative immunoblotting (Fig. 3a, b). Next, we examined the levels of Cat D, a key lysosomal aspartyl protease that has been suggested to play an important role in PGRN function [9, 115]. We observed a significant increase in both the pro-(~ 46 kDa) and heavy chain (~ 33 kDa) isoforms of Cat D (Fig. 3a, c, d). There were no significant differences in Cat Z and Cat D levels between Grn +/+ and Grn −/− mouse brain at 3 months, suggesting PGRN deficiency leads to an age-dependent increase of both Cat Z and Cat D.
Next, we performed immunostaining on 19-month-old (n = 14) Grn −/− brain coronal sections to determine the regions of Cat Z and Cat D expression and upregulation. We found strong immunostaining of Cat D in the thalamus, corpus callosum, striatum, and hippocampus in Grn −/− mouse brains (Fig. 3e). Immunostaining of Cat Z was also increased in the thalamus, corpus callosum, and striatum of Grn −/− mice, but to a lesser extent than Cat D (Fig. 3f ). Taken together, immunoblotting and immunostaining of Grn −/− mouse brains confirm our proteomic results and demonstrate that PGRN deficiency leads to an age-dependent increase in the lysosomal proteins Cat Z and Cat D.
Next, we performed immunohistochemical staining for GPNMB on sagittal brain sections from 3-, 12-, and 24-month-old Grn +/+ , Grn +/− , and Grn −/− mice to determine where GPNMB expression is most upregulated in the brain (Fig. 4e). We noted elevated GPNMB staining in the thalamus of Grn −/− mice at 12 and 24 months of age compared to both Grn +/+ and Grn +/− brains, in agreement with our ELISA results. Further, there was no appreciable difference in GPNMB staining between Grn +/+ and Grn +/− mice at any age. Finally, we completed a more extensive survey of GPNMB staining on coronal brain sections of 19-month old Grn +/+ and Grn −/− mice. We observed strong increases in GPNMB immunoreactivity in the corpus callosum and the thalamus and lower levels of increase in the hippocampus, cortex, and striatum (Fig. 4f ).

Cellular localization of GPNMB and galectin-3 in Grn −/− mice
Next, we aimed to determine which cells express GPNMB and galectin-3 in 19-month-old Grn −/− mice brains. We performed double immunofluorescent staining for GPNMB or galectin-3 with antibody markers for microglia (Iba-1), astrocytes (GFAP), and neurons (NeuN) (Fig. 6a, b). We detected strong co-localization between GPNMB, galectin-3, and Iba-1, indicating microglia express both proteins in aged Grn −/− mouse brain. This further supports our network-based proteomic analysis that detected a strong overlap between module M7, which contains GPNMB, galectin-3, and microglia cell markers (Fig. 1f ). In contrast, we did not detect co-localization between GFAP-positive astrocytes and GPNMB or galectin-3. Similarly, we did not detect robust co-localization of GPNMB or galectin-3 signal in NeuN-positive neurons. On occasion, we observed GPNMB co-staining in some cells that were weakly positive for NeuN. In summary, double immunofluorescent suggests that microglia are the major source of GPNMB and galectin-3 expression in Grn −/− mouse brains.

GPNMB and Galectin-3 are increased in FTD-GRN brain tissue
Finally, we asked if the pathologic changes we observed in aged Grn −/− mice also occur in the brains of FTD-GRN patients. We generated detergent lysates of the frontal lobe from 21 FTD-GRN samples and 23 cognitively normal controls. Initial immunoblotting of a subset of these cases suggested an increase in both GPNMB and galectin-3 in FTD-GRN brain homogenate compared to controls. Next, in order to more accurately quantify all samples, we measured human GPNMB and galectin-3 using sandwich ELISAs. Levels of GPNMB (p < 0.0001) and galectin-3 (p < 0.001) were both significantly increased in FTD-GRN brain homogenates compared to controls (Fig. 7a, b).
Subsequently, we examined the distribution of GPNMB in more detail in FTD-GRN brains. First, we examined GPNMB expression in the frontal lobe of five FTD-GRN cases compared to age-matched, cognitively normal controls. Specificity of the anti-GPNMB antibody for immunohistochemistry was validated using recombinant GPNMB protein to block staining (Additional file 3: Fig.  S6, online resource). Interestingly, the strongest signal for GPNMB was detected in the white matter of the frontal lobe in FTD-GRN brains (Fig. 7e, f ). Intensity of GPNMB immunoreactivity was 6.5-fold higher (p < 0.01) in the frontal lobes of FTD-GRN brains compared to matched regions from the brains of cognitively normal controls (Fig. 7e, f, Fig. S7a-d). Adjacent sections of FTD-GRN brains, but not controls, were immunopositive for phosphorylated TDP-43, confirming the presence of FTLD pathology in these cases [54,81] (Fig. 7g, h, Fig. S7e-h). Then, to address cell-type specificity, we performed double immunofluorescent staining of GPNMB and antibody markers for microglia (Iba-1), astrocytes (GFAP), and neurons (NeuN) (Fig. 7j-u). We detected strong co-localization between GPNMB and Iba-1 in human FTD-GRN brains. No co-localization between GPNMB and GFAP or NeuN signal was detected. These data suggest microglia are the predominant source of GPNMB expression in human FTD-GRN brains. Huang et al. acta neuropathol commun (2020) 8:163 Next, we asked if GPNMB levels were increased in FTD bio-fluids. We obtained CSF from patients with pathogenic genetic FTD mutations collected from the Advancing Research and Treatment for Frontotemporal Lobar Degeneration (ARTFL) and Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects (LEFFTDS) studies that were stored in the National Centralized Repository for Alzheimer's Disease and Related Dementias (NCRAD) [95,97]. GPNMB protein levels were quantified with sandwich ELISAs in CSF samples from individuals that were cognitively normal (controls; n = 14) and individuals with known mutations that cause FTD (FTD-GRN n = 13, FTD-C9orf72 n = 13, or FTD-MAPT n = 12). GPNMB levels were significantly up regulated (p < 0.05) in the FTD-GRN group (3.07 ± 0.35 ng/ mL CSF) compared to controls (1.92 ± 0.31 ng/mL CSF) (Fig. 7d). In contrast, there was no significant difference in GPNMB levels between controls and FTD-C9orf72 or FTD-MAPT CSF samples.

Discussion
Although PGRN haploinsufficiency is a well-established cause of FTD, the function of PGRN and the pathogenic cascade caused by PGRN deficiency that ultimately leads to neurodegeneration is still unclear [22,86]. In this study, we performed deep proteomic analysis of whole brains from Grn −/− mice, which recapitulate many features of FTD, including behavioral impairments and neuroinflammation [5,10,60,82].
The first major finding is that the levels of lysosomal proteins are altered in young 3-month old c mouse brains. Our proteomics data is the first to find lysosomal dysregulation at such an early time point. This is likely due to the sensitivity of TMT-based proteomics and the known discrepancy between RNA transcript levels and protein expression [102]. In particular, we detected significant increases in well-established proteins that localize to the lumen or membrane of the lysosome including glucosamine (N-Acetyl)-6-Sulfatase (GNS), prosaposin (Psap), lysosomal integral membrane protein-2 (Limp-2/Scarb2), cathepsin A (CtsA), and hexosaminidases A and B (Hexa; Hexb). Importantly, the genes encoding all of these proteins harbor pathogenic mutations that cause lysosome storage disorders [91]. Furthermore, expression of these and other lysosomal genes are under control of the transcription factors TFEB and TFE3 and are up-regulated under conditions of lysosomal stress or dysfunction [93,98,101]. Thus, early lysosome dysfunction in Grn −/− mice may manifest by increased expression of many lysosomal proteins. In agreement with our findings, the mRNA transcripts of many of the lysosomal genes we have identified have been reported to be increased in older Grn −/− mice [42,135]. Taken together, our data indicate that lysosomal dysfunction occurs early in Grn −/− brains and likely initiates pathogenesis and eventual neurodegeneration.
Our findings also provide insight into the function of PGRN. Recently, a potential role of PGRN in lysosome homeostasis has emerged based on the discovery that multiple cases of homozygous GRN mutation carriers develop neuronal ceroid lipofuscinosis-11 (CNL11), a lysosomal storage disease [4,18,50,105]. We, and other labs, have found that PGRN is trafficked to the lysosome and processed by cathepsins into granulins, which may be bioactive [47,68,133]. However, the function of PGRN and granulins within the lumen of the lysosome is still unclear. Because PGRN deficiency causes impairment of lysosomal protease activity and accumulation of lipofuscin [125], one possibility is that loss of PGRN/granulins directly or indirectly decrease the levels and/or activity of a lysosomal hydrolase.
With this idea in mind, our observation that a subset of the most significantly downregulated proteins (Acly, Apoc3, Asah1, Gpld1, Ppt1, Naaa) in the 3 month Grn −/− brain proteome are involved in lysosomal lipid catabolic process is important. Of these proteins two, ATP-citrate lyase (Acly) and apolipoprotein (apo) C-III (See figure on next page.) Fig. 7 GPNMB and galectin-3 levels are elevated in FTD-GRN brains. a, b GPNMB and galectin-3 levels (ng/mg protein) were measured in frontal lobe tissue lysates generated from cognitively normal controls (CTL; n = 27) and FTD-GRN patients (n = 25). Data analyzed using unpaired t-test. c Representative immunoblots for GPNMB and galectin-3 in frontal lobe lysates from cognitively normal controls (n = 8) and FTD-GRN (n = 8) patients. d GPNMB levels (ng/mL) in CSF samples form cognitively normal controls (n = 14), FTD-GRN (n = 9), FTD-C9orf72 (n = 12) and FTD-MAPT (n = 12) samples quantified by ELISA. Data analyzed using one-way ANOVA. e, f GPNMB immunostaining was performed on frontal lobe tissue sections from cognitively normal controls (n = 5) (e) and FTD-GRN (n = 5) (f) patients. g, h Immunostaining for p-TDP 43 was stained on adjacent sections from identical samples in e, f as marker of FTLD pathology. i GPNMB staining intensity in human brain sections (e, f) were measured and presented as fold change. Representative immunofluorescence staining for cell markers (green) (j, n, r), GPNMB (red) (k, o, s), DAPI (blue) (i, p, t) in paraffin sections of brains from FTD-GRN cases. Iba-1, GFAP, NeuN used for markers of human microglia, astrocytes, and neurons respectively. GPNMB and Iba-1 signals overlap (arrow) (m) whereas, no overlapping signal was observed in co-staining with GFAP or NeuN (q, u). Scale bars were labeled in the images. Data analyzed by unpaired t-test. Scale bars (20 µm) labeled in images and quantitative data are shown as mean ± SEM, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 Huang et al. acta neuropathol commun (2020) 8:163 (Apoc3), are involved in lipogenesis and lipid homeostasis [30,94]. Intriguingly, the remaining proteins are all involved in lipid degradation pathways. N-acylsphingosine amidohydrolase 1 (Asah1) hydrolyzes sphingolipid ceramides into sphingosine and free fatty acids in the lysosome [87]. N-Acylethanolamine Acid Amidase (Naaa) degrades bioactive fatty acid amides, such as N-palmitoylethanolamine [114]. Palmitoyl-protein thioesterase 1 (Ppt1) removes thioester-linked fatty acyl groups like palmitate from modified cysteine residues during lysosomal degradation of proteins [59]. Finally, glycosylphosphatidylinositol specific phospholipase D1 (Gpld1) hydrolyzes the inositol phosphate linkage in proteins anchored by phosphatidylinositol glycans (GPI-anchor) to release proteins from the membrane and has recently been linked to the cognitive benefits of exercise for the aged brain [49]. Of note, earlier work found that loss of PGRN leads to an accumulation of polyunsaturated triacylglycerides and reduced diacylglycerides and phosphatidylserines in Grn −/− fibroblasts and mouse brains [29]. Our data provide additional evidence that PGRN deficiency leads to impaired degradation and recycling of lipids in the lysosome. Further work is necessary to determine if PGRN or granulins have a direct or indirect role in lysosomal lipid metabolism and homeostasis. Our analysis of differentially expressed proteins in 19-month-old Grn −/− brains uncovered an even greater number of lysosomal proteins that are increased including Hexa, Hexb, Tpp1, and Fuca2. Intriguingly, proteins involved in inflammation and immune response (i.e. complement genes C1qa, C1qb, and C1qc) are significantly increased. Our data agrees with earlier work demonstrating that multiple Grn −/− mouse models have age-dependent microgliosis and astrogliosis throughout the brain including the cortex, hippocampus, and thalamus [1,37,82,127,132]. The upregulation of C1qa, C1qb, and C1qc we observe is consistent with previous transcriptomic data from Grn −/− mice, which reported upregulation in C1qa, C1qb, and C1qc before the onset of neurodegenerative features [29,70]. Further, microglia isolated from 5.5-month old Grn −/− mice are activated and upregulate expression of genes associated with a microglial neurodegenerative phenotype (MGnD) [41]. Thus, gliosis and a robust inflammatory response is a consistent observation in multiple Grn −/− mouse models that increases with age, suggesting this is may be a downstream consequence initially triggered by lysosome dysfunction.
We also identified novel proteins that increase with age in the brains of Grn −/− mice. In particular, two of the most strongly up-regulated proteins were GPNMB (also known as osteoactivin) and galectin-3, suggesting they may play an important role in disease progression. Our co-localization studies demonstrate that GPNMB and galectin-3 are strongly expressed by microglia in aged Grn −/− mouse brain. GPNMB is a widely expressed transmembrane type I protein that has been implicated in many cellular functions including cell adhesion, cell migration, cell proliferation, and cell differentiation [113]. Intriguingly, variants in GPNMB are associated with Parkinson's disease, highlighting the potential importance of GPNMB broadly in neurodegenerative diseases [52,89]. Galectin-3 is a member of the lectin family, contains a carbohydrate-recognition-binding domain that mediates binding of β-galactosides, and plays an important role modulating inflammation [28]. Galectin-3 has also been implicated in brain innate immunity associated with neurodegeneration [16]. High levels of both GPNMB and galectin-3 levels have been found in the brain of 5xFAD mice [51] as well also the brains of PD and AD patients [11,61,75,100]. Additionally, levels of GPNMB were elevated in grey and white matter of spinal cord of ALS patients [79].
We observed the strongest increases in GPNMB and galectin-3 expression in the thalamus of Grn −/− mouse, providing additional evidence that this region of the brain is particular vulnerable to loss of PGRN [70]. We also found high levels of GPNMB and galectin-3 in the corpus callosum, which is highly myelinated. Abundant galectin-3 staining was found along white matter tracts in the cortex and striatum, suggesting that the expression and upregulation of GPNMB and galectin-3 in microglia is related to myelin and white matter.
Importantly, myelin basic protein (MBP), a major component of white matter, is decreased in Grn −/− mouse brain. Furthermore, modules enriched for oligodendrocyte proteins (M19, p = 0.033) and neuronal and synaptic proteins (M5, p = 0.002) are not significantly decreased until 19-months of age in Grn −/− mouse brain, indicating that synaptic and neuronal loss and demyelination are a late-stage consequence of Grn deficiency. Our proteomics data is supported by a previous report that found defective myelination in the cerebral cortex of Grn −/− mice [108]. Our discovery that GPNMB and galectin-3 are elevated in the white matter of frontal lobe of FTD-GRN cases provides additional clinical relevance suggesting alterations of these proteins may contribute to neurodegeneration.
Intriguingly, demyelination that is observed as white matter hyperintensities (WMH) on MRI brain scans is a frequent and specific occurrence in FTD cases caused by GRN mutations [20,107,128]. Our data provide additional support to the idea that FTD-GRN cases are especially vulnerable to demyelination and loss of white matter, although the exact mechanism is unclear. One possibility is that aberrant microglial activation due to PGRN deficiency, caused by lysosome dysfunction and protein aggregation, leads to phagocytosis, synaptic pruning, and de-myelination that eventually manifests as white matter damage and WMH. Intriguingly, microglia expressing either GPNMB or galectin-3 have been implicated in the process of myelin phagocytosis and demyelination [46,69,109]. Taken together, we find that increases in microglial GPNMB and galectin-3 levels are correlated with demyelination. Currently it is unclear if GPNMB and galectin-3 upregulation is a cause or consequence of this process.
Lack of specific and sensitive biomarkers are another roadblock in the development of therapies for FTD [40]. Importantly, appropriate biomarkers are useful for monitoring disease progression as well as assessing the efficacy of potential drugs. Elevated levels of neurofilament light chain (Nfl) in CSF and plasma are one promising biomarker for symptomatic FTD patients that harbor mutations in GRN, C9orf72, or MAPT [118]. However, Nfl is not specific to FTD-GRN and is increased in several other neurodegenerative diseases [124]. In this study, we report our finding of elevated levels of GPNMB in aged Grn −/− mouse plasma as well as CSF from human FTD patients with GRN mutations. Previous reports found that GPNMB was elevated in the serum of type I Gaucher disease [138], but not significantly different in AD patients [51], compared to healthy controls. Because GPNMB levels were only increased in CSF from GRN mutation carriers, but not C9 or MAPT cases, GPNMB levels could serve as a biomarker to differentiate between FTD caused by GRN mutantions or other FTD genes.
It is currently unlcear why GPNMB is elevated in the microglia of Grn −/− mice and human FTD-GRN patients. One intriguing possibility is that upregulation of GPNMB expression is driven by lysosome dysfunction. This idea is supported by previous reports that lysosomal stress, induced by chemical inhibition of lysosome acidification or function, causes upregulation of GPNMB in macrophages [34,111]. GPNMB is also elevated in the substantia nigra of patients with Parkinson's disease [75], a neurodegenerative disease increasingly linked to lysosome dysfcuntion [56,83,123]. Moreover, chemical [75] or genetic inhibition [61] of β-glucocerebrosidase (GBA; GCase) activity leading to the accumulation of glucosylceramides also causes increased expression of GPNMB. Progranulin deficiency reduces glucocerebrosidase activity [116,134] suggesting that the accumulation of glucosylceramide or other sphingolipids could be a proximal cause of GPNMB upregulation in the Grn −/− mouse brain, an idea that needs further investigation. Although the precise mechanism that causes GPNMB upregulation in progranulin deficiency is unclear, our data suggest that measurement of GPNMB levels in the CSF could be used to monitor changes in microglial activation and response to therapies in FTD-GRN patients, similar to substrate reduction therapy in lysosome storage disorders [71,78,119]. One limitation of our data is a small sample size and lack of longitundal testing. Thus, further studies are necessary to investigate the utility, specificity, and sensitivity of GPNMB as a biomaker in FTD and related neurodegenerative diseases. Moreover, it will be important to determine if GPNMB and galectin-3 expression in microglia is deleterious, which would open a new therapeutic target for FTD and other diseases with PGRN deficiency.

Conclusion
Our results demonstrate the utility of a systems biology approach in understanding a complex disease like FTD. We were able to relate changes in the Grn −/− mouse brain proteome to known phenotypic signatures of FTD. Further analysis of these proteomic changes across age also provided insight on the mechanism(s) in which neurodegeneration occurs as result of PGRN deficiency. We identified novel proteins in the Grn −/− mouse proteome that are decreased at 3-months, which suggest an impairment of lysosomal metabolism of lipids, including sphingolipids, which are particularly important for neuronal survival [2,14]. Lysosomal dysregulation is exacerbated with age in the Grn −/− mouse brain leading to neuroinflammation, synaptic loss, and decreased markers of oligodendrocytes, myelin, and neurons. For the first time, we identified increased levels of two proteins, galectin-3 and GPNMB, which have not been linked to FTD or PGRN deficiency previously and may serve as novel biomarkers or drug targets. Previous data demonstrate that upregulation of GPNMB and galectin-3 in microglia can be beneficial or harmful depending on the context, timing, and disease [12,15,16,62,109]. Further studies are necessary to understand the contribution of GPNMB and galectin-3 to FTD and related neurodegenerative disease. In summary, our findings support the idea that insufficiency of PGRN and granulins in humans cause FTD through lysosomal dysfunction and neuroinflammation and suggest novel therapeutic approaches.
Additional file 1. Compilation of multiple supplementary tables. Table S1 is a list of cell type-specific protein markers used to identify enrichment of cell types in WGCNA modules. Table S3 is a summary of differential protein expression in Grn −/− (KO) compared to Grn +/+ (WT) mouse brain. Table s4 is a compilation of proteins identified in mouse brain proteome and their correlation to WGCNA modules. Table S5 is a summary of the enrichment of Gene Ontology (GO) terms for each module in WGCNA network. Table S2 summarizing the neuropathological, clinical diagnosis, age, sex, and other characterizes of human post-mortem samples used for immunostaining and ELISA.

Additional file 2.
Additional file 3. Compilation of multiple supplementary figures (S1-S7). Fig. S1 is a two-color heatmap showing the relationship between WGCNA modules and the bicor correlation of age.  Fig. S6 is an immunoadsorption validation experiment demonstrating an anti-human GPNMB antibody binds antigen specifically on human brain sections. Fig. S7 GPNMB immunostaining was performed on frontal lobe tissue sections from four FTD-GRN patients.