Increased CSF-decorin predicts brain pathological changes driven by Alzheimer’s Aβ amyloidosis

Cerebrospinal fluid (CSF) biomarkers play an important role in diagnosing Alzheimer’s disease (AD) which is characterized by amyloid-β (Aβ) amyloidosis. Here, we used two App knock-in mouse models, AppNL-F/NL-F and AppNL-G-F/NL-G-F, exhibiting AD-like Aβ pathology to analyze how the brain pathologies translate to CSF proteomes by label-free mass spectrometry (MS). This identified several extracellular matrix (ECM) proteins as significantly altered in App knock-in mice. Next, we compared mouse CSF proteomes with previously reported human CSF MS results acquired from patients across the AD spectrum. Intriguingly, the ECM protein decorin was similarly and significantly increased in both AppNL-F/NL-F and AppNL-G-F/NL-G-F mice, strikingly already at three months of age in the AppNL-F/NL-F mice and preclinical AD subjects having abnormal CSF-Aβ42 but normal cognition. Notably, in this group of subjects, CSF-decorin levels positively correlated with CSF-Aβ42 levels indicating that the change in CSF-decorin is associated with early Aβ amyloidosis. Importantly, receiver operating characteristic analysis revealed that CSF-decorin can predict a specific AD subtype having innate immune activation and potential choroid plexus dysfunction in the brain. Consistently, in AppNL-F/NL-F mice, increased CSF-decorin correlated with both Aβ plaque load and with decorin levels in choroid plexus. In addition, a low concentration of human Aβ42 induces decorin secretion from mouse primary neurons. Interestingly, we finally identify decorin to activate neuronal autophagy through enhancing lysosomal function. Altogether, the increased CSF-decorin levels occurring at an early stage of Aβ amyloidosis in the brain may reflect pathological changes in choroid plexus, present in a subtype of AD subjects.

407 approved by Linköping animal ethical board except the CSF collection from 13 months old App NL-F/NL-F mice approved by RIKEN Center for Brain Science (W2021-2-020(2)). Two APP transgenic (Tg) mouse models were used, APP23 mice harboring only the Swedish mutation [13] (hemizygotes) and tg-ArcSwe mice containing both Swedish and Arctic mutations [14]. The CSF collection from APP23 mice were approved by Linköping animal ethical board (10925-2020). The CSF collection from tg-ArcSwe were approved by the Uppsala County animal ethics committee, the Swedish Board of Agriculture (5.8.18-20401/2020). Mice were kept on 12:12 light-dark cycle and with ad libitum access to food.

Mouse CSF collection and brain dissection
The mice were anesthetized by isoflurane and placed on the heating pad with fixed heads (the head angle was around 135° from the body) on the stereotaxic instrument. The skin was incised sagittally after removal of the fur. Under the dissection microscope, subcutaneous tissues and muscles were separated to expose the dura mater of cisterna magna. To reduce blood contamination, the exposed dura mater was cleaned by PBS-soaked cotton swabs and punctured with a 27-gauge needle avoiding blood vessels. CSF was sampled with a glass capillary and collected in low affinity Eppendorf tubes and directly frozen. The collected CSF was discarded if there was any blood contamination under microscope inspection. Thereafter, the mice were perfused with PBS through cardiac perfusion, and hippocampus and cortex were dissected.

Primary neuron culture
Twenty-four well plates were coated with Poly-D-lysine (Sigma-Aldrich, cat. P6407) for one hour at room temperature and washed with Milli-Q water followed by drying overnight. Embryos were separated from App wt/wt mice E16-E18 and the heads were kept in HBSS (Thermo Fisher Scientific, Cat. 14175095) on ice for brain dissections. Brains were dissected under the dissection microscope to separate the cortex/hippocampus. Cortex/ hippocampus were chopped and transferred to the falcon tube together with HBSS. HBSS was removed until only tissues were left at the bottom. Neurobasal medium (Thermo Fisher Scientific, Cat. 21103049) 97% + B-27 (Thermo Fisher Scientific, Cat. 17504044) 2% + Glutamax (Thermo Fisher Scientific, Cat. 35050038) 1% were added and tissues were separated by pipetting up and down 20-30 times. Cells were counted by a hematocytometer with trypan blue staining. 1.5 × 10 5 cells for 24 well plates, 3 × 10 5 cells for 12 well plates and 5 × 10 6 cells for 10 cm petri dishes were seeded. After one week, 50% of the medium was replaced by a fresh medium.  10:96 For decorin treatment study, after 18 DIV, cells were treated with 200 nM recombinant mouse decorin (R&D Systems, Cat. 1060-DE-100) and/or 100 nM bafilomycin A1 (Sigma-Aldrich, Cat. B1793) for six hours and thereafter the cells were collected in PBS. Cell solutions were centrifuged at 1500 rpm for five min at 4 °C to pellet the cells, which were lysed in RIPA buffer (Thermo Fisher Scientific, Cat. 89901). The lysate was sonicated for one min and centrifuged at 15,000 rpm for 20 min at 4 °C. The supernatant was transferred to new tubes for western blot. Each condition was triplicated and the whole set of experiments repeated twice.
For human Aβ42 treatment study, after 12 DIV, cells were treated for 24 h with 50 nM recombinant Aβ42 (Met-Aβ42) which was prepared as described previously [15]. Then the cells were collected in RIPA buffer, sonicated for one min and centrifuged at 15,000 rpm for 20 min at 4 °C. The supernatant was transferred to new tubes for western blot. Each condition was repeated eight times and four out of eight were used for western blot analysis. Protein concentration was measured by BCA Protein Assay Kit (Thermo Fisher Scientific, Cat. 23225). SDS sample buffer containing 0.1 M dithiothreitol was loaded to the samples for western blot analysis.

Immunohistochemistry
The paraffin-embedded mouse brains (12 months old) were sectioned into four µm thick sections for most of immunofluorescence staining except for vessel staining for which 10 µm thick sections was used. For decorin staining, slides were treated with 0.5 units of chondroitinase ABC (Sigma-Aldrich, Cat. C2905-10UN) in 0.1 M Tris-acetate (pH: 7.3) for four h in 37 °C before blocking [16]. For other antibodies, antigen retrieval was performed, whereas the combination of autoclaving and chondroitinase ABC treatment was performed for double staining. After blocking, slides were incubated with anti-Aβ antibody (82E1) (Immuno-Biological Laboratories, Cat.

Phospho-explorer antibody array and analysis of biological processes
The phosphorylation profiling (Additional file 1: Table S1) of decorin-treated and non-treated mouse primary neurons was determined by Phospho Explorer Antibody Array (Full Moon BioSystems, Cat. PEX100) following the manufacturer's protocol. The alterations of phospho-protein levels in decorin-treated/non-treated with ratios > 1.5 or < 0.6 were subjected to gene ontology enrichment analysis [17][18][19]. The biological processes with false discovery rate < 0.01 were considered as significantly enriched.

Decorin enzyme-linked immunosorbent assay (ELISA) of mouse CSF
Mouse CSF decorin levels were measured by a decorin mouse ELISA kit (Abcam, Cat. ab155454) following the manufacturer's protocol.

Proximity extension assay
One μL of mouse CSF samples were applied to Mouse Exploratory Panel, Olink according to manufacturer's protocol. The tryptic peptides were adsorbed to the media using two repeated cycles of 40 μL sample loading and the column was washed using 3 × 200 μL of 0.5% trifluoroacetic acid. Finally, the peptides were eluted in 3 × 50 μL of 70% acetonitrile and dried. Dried peptides were resolved in 21 μL of 0.1% formic acid prior to nano-LC-MS/MS. The nano-LC-MS/MS experiments were performed using a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) equipped with a nano electrospray ion source. The peptides were separated by C18 reversed phase liquid chromatography using an EASY-nLC 1000 system (Thermo Fisher Scientific). A set-up of pre-column and analytical column was used. The precolumn was a 2 cm PepMap Acclaim (C18 100 µm, 5 µm particles) (Thermo Fisher Scientific) while the analytical column was a 10 cm EASY-Spray column (C18 75 µm, 3 µm particles, Thermo Fisher Scientific). Peptides were eluted with a 150 min linear gradient from 4 to 100% acetonitrile at 250 nL min-1. The mass spectrometer was operated in positive ion mode, acquiring a survey mass spectrum with resolving power 70,000 (full width half maximum), m/z 400-1750 using an automatic gain control target of 3 × 10 6 . The 10 most intense ions were selected for higher-energy collisional dissociation fragmentation (25% normalized collision energy) and MS/MS spectra were generated with an automatic gain control target of 5 × 10 5 at a resolution of 17,500. The mass spectrometer was operated in data-dependent mode. Acquired raw files were processed by MaxQuant (version 1.5.1.2) (the software is available at http:// www. maxqu ant. org). Tandem mass spectra were searched with Andromeda against the UniProt Mus musculus database (release January 2017). The search settings were set as: maximum 10 ppm and 0.02 Da error tolerance for the survey scan and MS/MS analysis respectively; enzyme specificity was trypsin/Lys-C; maximum two missed cleavage sites were allowed; cysteine carbamidomethylation was set as static modification, and Oxidation (M) was set as dynamic modification. The search criteria for protein identification were set to at least two matching peptides. No proteins were identified and quantified using only one peptide. A maximum false discovery rate of 1% for peptides and proteins was selected. Both razor and unique peptides were used for quantification. For decorin, at least five peptides per sample were used for the quantification (Additional file 2: Table S2). A decoy sequence database was built by reversing the target sequence database. A list of known contamination was also included in the identification. The protein intensity values were used for further data analysis.

Mass spectrometry analysis of human CSF
CSF proteomic results reported in a previous study were included in the present research [12]. Briefly, we selected individuals with AD pathology defined as abnormal CSF-Aβ42, and we subdivided this group into abnormal (a + t + , n = 151) and normal CSF-t-tau groups (a + t −, n = 77). Based on their cognitive performance, AD individuals were classified in three clinical stages as preclinical AD (normal cognition, i.e., NC), prodromal AD (mild cognitive impairment, i.e., MCI) and mild to moderate AD-type dementia according to study specific criteria [20]. MS was performed using the tandem mass tag (TMT) technique with 10 + 1 multiplexing as previously described, and high-pH reverse phase HPLC for peptide prefractionation [12,21]. The EMIF-AD MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier https:// doi. org/ 10. 6019/ PXD01 9910 [22]. Fold-changes were computed for a + t − and a + t + groups according to a control group (n = 82) with intact cognition and normal CSF amyloid and tau markers.

Comparison of mouse and human CSF proteomes and analysis of biological processes
The same protein identifiers were used to compare mouse and human CSF proteomes. We performed two types of comparisons. In the first exploratory comparison, proteins with fold change > 1.0 was considered to be upregulated whereas < 1.0 was considered downregulated in the mouse and human CSF proteomes. In the second comparison only signifcantly changed proteins in the mouse and human CSF samples (p < 0.05) were compared.

Statistical analysis
For the analysis of mouse CSF proteomics data, protein intensities were first transformed to log2 scale and a twotailed Student's t-test was performed on proteins which were identifed in all samples in order to identify proteins that were significantly altered in App knock-in mouse models, using Qlucore software. A p < 0.05 was considered as statistically significant. Venn diagrams were generated by the Interactive Venn tool [23] or Venny 2.1 [24].
To investigate the variation among individual mouse CSF proteomes, a principal component analysis (PCA) was performed in Qlucore. Volcano plots were generated by GraphPad Prism 8 and significantly altered proteins in App knock-in models, p < 0.05, were indicated. In order to visualize the significantly altered proteins in App knockin mice, heat maps were generated by Morpheus software from the Broad Institute [Morpheus, https:// softw are. broad insti tute. org/ morph eus]. For the analysis of biochemical data, one-way ANOVA followed by Dunnett's multiple comparisons test was performed for the three group comparisons or more while two tailed student's t-test was performed for the two-group comparisons in GraphPad Prism 8 when the datasets were normally distrubuted. If the dataset was not normally distributed, non-parametric Kruskal-Wallis tests with Dunn's post hoc analysis was used. Decorin distribution in different cell types in mouse hippocampus were analyzed by twoway ANOVA followed by Tukey's post hoc test. CSF-decorin levels across the AD spectrum were analyzed by one-way ANOVA followed by Tukey's multiple comparisons test. Associations between human CSFdecorin levels (outcome) and CSF Aβ42, t-tau and p-tau levels (predictors) were determined using linear regression across the total group, and differences between subgroups (i.e., a − t −, a + t −) in NC group were tested with the 'emtrends' function from the emmeans R package (v1.5.2). ROC curve analyses were performed for CSFdecorin levels and AUC with 95% CI were calculated with R package pROC (v1.18.0).
The correlation analysis between parameters of Aβ pathology and decorin expression in the mice were made by Spearman's rank correlation coefficient since some of the parameters were not normally distributed. The data were grouped from App wt/wt and App NL-F/NL-F mice and App wt/wt and App NL-G-F/NL-G-F mice, respectively.

Results
To investigate how AD pathologies driven by Aβ translate to CSF, we used label-free MS to analyze the CSF of App knock-in mice which exhibit robust Aβ amyloidosis. To stratify the data, the mouse CSF proteomes were next compared with the proteomes of human CSF from a previously published large AD cohort [12]. This resulted in identification of protein hits related to ECM and autophagy which were followed up and investigated by biochemical means. The workflow of the whole project is presented in Fig. 1.

ECM proteins are altered in CSF of App knock-in mice
In a translational approach to identify potential ADrelated CSF biomarkers induced by Aβ and its downstream effects, we investigated the protein changes in CSF of 12 months old App NL-F/NL-F and App NL-G-F/NL-G-F mice which exhibit different degree of Aβ pathology (Fig. 2a, b). Taking the limited volume of mouse CSF into account, a label-free MS approach was used which identified 427-703 proteins of which 246 were detected in all samples and their levels quantified ( Fig. 2c; protein list in Additional file 2: Table S2). Principal component analysis (PCA) of these 246 proteins indicated a separation between the three groups ( Fig. 2d) whereas PCA of the significantly altered proteins (p < 0.05) resulted in a tighter clustering (Fig. 2e, f ). Interestingly, ECM proteins decorin and lumican were upregulated in App NL-F/NL-F mice while SPARC-like protein 1 and fibronectin were downregulated (Fig. 2g, i; Additional file 3: Table S3). However, in App NL-G-F/NL-G-F mice, most of ECM proteins were downregulated including collagen alpha-1(I) chain, basement membrane-specific heparan sulfate proteoglycan core protein, fibronectin, SPARC-like protein 1, fibulin-1, vitronectin and ecm1 protein (Fig. 2h, j; Additional file 3: Table S3). Comparing the CSF proteomes of the two App knock-in mouse models directly with each other revealed significant (p < 0.05) changes in additional ECM related proteins (Additional file 4: Fig. S1a, b; Additional file 3: Table S3). Several studies have previously found that the dysfunction of BBB and BCSFB, including changes in ECM proteins, increase upon aging and in AD [25][26][27] and we therefore also analyzed the CSF of 18 months old App knock-in mice (n = 5) using a mouse proximity extension assay (PEA) panel. Indeed, two ECM proteins, matrilin-2 and CCN family member 4, were significantly (p < 0.05) altered (Additional file 3: Table S3, Additional file 5: Table S4). Altogether, several ECM proteins were significantly altered in CSF of the App knockin mice, and they are differently altered depending on the severity of the brain Aβ pathology.

Decorin is similarly increased in CSF of both App NL-F/NL-F mice and preclinical AD subjects with abnormal CSF-Aβ42
In an attempt to further understand how changes in the CSF proteomes of App knock-in mouse models reflect those observed in human patients, we next compared them to MS-characterized CSF proteomes of a large human European Medical Information Framework for Alzheimer's Disease Multimodal Biomarker Discovery (EMIF-AD MBD) cohort (n = 310) [12]. In this cohort (clinical data for the patients is summarized in Additional file 6: Table S5), individuals with abnormal CSF-Aβ42 were classified into three clinical stages; preclinical AD (normal cognition, i.e., NC), prodromal AD (mild cognitive impairment, i.e., MCI) and mild to moderate AD-type dementia, based on their cognitive performance [20]. The groups were further classified according to CSF-t-tau levels, being normal (a + t −) or abnormal (a + t +). The CSF proteome alterations in the NC, MCI and AD groups were based on the comparison to the healthy control subjects having normal CSF-Aβ42 and CSF-tau, and normal cognition. Interestingly, the number of proteins with a relative expression level below one in both the App knock-in mice and a + t − human subjects were higher as compared to a + t + subjects, indicating that the alterations in the CSF proteomes of App knock-in mice are, at least to some extent, more similar to the CSF alterations observed in a + t − human subjects (Table 1, Additional file 7: Table S6, Additional file 8: Table S7). This may reflect that App knock-in mice have a strong Aβ pathology whereas the tau pathology is less pronounced [28]. Further mouse and human CSF proteome comparisons were therefore performed with a + t − human subjects. A qualitative direct comparison of CSF MS data from a + t − NC subjects and App NL-F/NL-F mice showed that 33 proteins exhibited relative expression levels above one (Fig. 3a). Notably, among those, only decorin (DCN) was significantly upregulated in both App NL-F/NL-F mice and a + t − NC subjects (p < 0.05) (Fig. 3b). The comparison of App NL-F/NL-F mice and a + t − MCI and AD subjects revealed two significantly downregulated ECM proteins SPARC-like protein 1 (SPARCL1) and fibronectin (FN1) (Fig. 3c). Comparing App NL-G-F/NL-G-F mouse CSF proteome with the human CSF proteome revealed significant downregulation of ECM proteins SPARCL1, FN1 and ecm1 protein (ECM1) (Fig. 3d-f ). Taken together, several ECM proteins were significantly and commonly altered in mouse and human CSF proteomes and among those, decorin is increased already at an early stage of the Aβ pathology in both App NL-F/NL-F mice and a + t − NC subjects.

CSF-decorin positively correlates with CSF-Aβ42 in a + t − preclinical AD subjects and predicts a subtype of AD
Having found that decorin is specifically increased in CSF of a + t − NC subjects but not in a + t− MCI and AD subjects suggested an association of decorin to early AD pathologies and hence be of interest as a future potential biomarker (Fig. 4a). Therefore, we next investigated how decorin levels were related to the AD CSF biomarkers Aβ42, t-tau and p-tau in human CSF. Within the a − t − healthy subjects CSF-decorin levels followed a nonlinear trajectory with CSF-Aβ42 (beta(se) = -0.16(0.18), p = 0.36). In contrast, within a + t − NC subjects CSFdecorin levels significantly correlated with CSF-Aβ42 levels (beta(se) = 0.61(0.25), p = 0.01), a correlation that was significantly (p = 0.016) different from that of a-thealthy subjects (Fig. 4b). In other words, in healthy subjects with normal levels of CSF-Aβ42, CSF-decorin levels tend to increase with decreasing CSF-Aβ42 levels until CSF-Aβ42 reaches pathological levels. When CSF-Aβ42 levels reach abnormal levels, the correlation switches and becomes positive. That is, within the group of a + t − preclinical AD subjects that have abnormal CSF-Aβ42 levels, the CSF-decorin levels start to decrease (Fig. 4e). In addition, plotting CSF-decorin against CSFt-tau and CSF-p-tau within the a-t− healthy group revealed non-linear associations between CSF-decorin and CSF-t-tau or CSF-p-tau. In contrast, the correlations became significantly negative in a + t − NC subjects for CSF-decorin and CSF-t-tau (beta(se) = − 0.89(0.32), p = 0.0053) (Fig. 4c) and CSF-decorin and CSF-p-tau (beta(se) = − 1.00(0.21), p < 0.0001) (Fig. 4d). Furthermore, across the whole disease spectrum including NC, MCI and AD subjects (n = 310), lower CSF-decorin levels correlated to higher CSF-t-tau (beta(se) = − 0.11(0.04), p = 0.0014) (Additional file 9: Fig. S2b) and CSF-p-tau levels (beta(se) = − 0.16(0.04), p < 0.0001) (Additional file 9: Fig. S2c) but no association with CSF-Aβ42 levels was observed (beta(se) = 0.09(0.06); p = 0.11) (Additional file 9: Fig. S2a) which is expected considering that decorin increases early in the a + t − preclinical AD subjects and then subsequently decreases upon disease progression (Fig. 4e). We next explored whether increase of CSFdecorin levels might have use as a biomarker for early Aβ amyloidosis by performing receiver operating characteristic (ROC) analysis in a + t − preclinical AD subjects vs healthy control. The area under the curve (AUC) was 0.61 (95% confidence interval (CI) = 0.50, 0.73) and not significant within this rather limited cohort (p = 0.058) (Fig. 4f ). However, we previously observed distinct AD subtypes showing non-linear alterations in CSF proteomics, which may weaken decorin relationships with AD if not taken into account [12]. One subtype (including all the subjects in preclinical AD, MCI and AD) was characterized by innate immune activation, specific increases in proteins associated with ECM and increased transthyretin, which may indicate ChP dysfunction. Interestingly, CSFdecorin alterations could very well classify these subjects from healthy control with an AUC (95% CI) = 0.70 (0.62, 0.79) (p = 1.73 × 10 -6 ) (Fig. 4g). Taken these data together indicate that the increase of CSF-decorin levels may reflect the innate immune activation and potentially ChP dysfunction in the brain.

Increased CSF-decorin in three different mouse models of Aβ pathology
Having found that decorin was increased in CSF proteome of App NL-F/NL-F mice, we were interested in the potential decorin alterations in the same mouse brains as used for CSF analysis. Interestingly, decorin expression was significantly higher in ChP in App NL-F/NL-F mice (Fig. 5a, b). Decorin were also localized in neurons in both cortex and hippocampus wherein decorin was highly expressed in CA2 pyramidal neurons and parvalbumin (PV)-positive interneurons, but little expressed in somatotropin release-inhibiting factor (SRIF)-positive interneurons (Fig. 5c-e). Interestingly, the neurites of PV-interneurons in App knock-in mice, especially in the App NL-G-F/NL-G-F mice, contained significantly less decorin (Fig. 5f, g). In addition, decorin was highly expressed in the tunica externa of arteries (Additional file 10: Fig.  S3a) and veins (Additional file 10: Fig. S3b) on the surface of the brain but not present in the vessels of brain parenchyma (Additional file 10: Fig. S3c). However, no significant differences in decorin levels were observed in the vessels (Additional file 10: Fig. S3d, e). Summarizing the decorin alterations in CSF and in the brains of App knock-in mice by Spearman's correlation analysis, revealed that the increase of CSF-decorin positively correlated both with decorin levels in ChP and Aβ plaque load in App NL-F/NL-F mice (Additional file 11: Fig. S4a) and that the reduced decorin expression in the interneurons was associated with the severe Aβ amyloidosis in App NL-G-F/NL-G-F mice (Additional file 11: Fig. S4b).
To deepen the understanding of the relationship of decorin and the progression of Aβ amyloidosis, we measured the CSF-decorin levels in App knock-in mice of different ages as well as in two additional APP tg mouse models exhibiting Aβ pathology; APP23 and tg-ArcSwe, which have the same genetic background (C57BL/6J) as the App knock-in mice, by enzyme-linked immunosorbent assay (ELISA). APP23 mice contain only the Swedish mutation and form Aβ plaques from around 12 months of age, while tg-ArcSwe mice having both the Swedish and the Arctic mutation start to form Aβ plaques at around six months of age [29]. Notably, a significant increase of CSF-decorin levels were detected already in three months old App NL-F/NL-F mice (Fig. 5h). At 13 months of age, the CSF-decorin levels were significantly increased in both App NL-G-F/NL-G-F and APP23 mice while there was a nonsignificant trend towards increased levels in App NL-F/NL-F and tg-ArcSwe mice (Fig. 5i). The CSF-decorin levels in 18 months old App NL-F/NL-F and tg-ArcSwe mice were not altered (Fig. 5j). Since we found increased CSF-decorin levels in the early preclinical stage of AD, we were interested in the CSF-decorin changes in the App NL-F/NL-F mice exhibiting a mild Aβ pathology and how it is changed upon development of the Aβ amyloidosis. Therefore, we deepened the analysis of the App NL-F/NL-F mice and measured the CSF-decorin levels in 3, 7, 13, 18, 22 months of age. Interestingly, we found a clear decrease in the aged mice after the initial increase similar to the human CSFdecorin alterations (Fig. 5k). To investigate the causal relationship between decorin and Aβ42, we treated App wt/wt mouse primary neurons with 50 nM of recombinant human Aβ42. This led to significantly decreased levels of decorin in the neurons (Fig. 5l, m), whereas the decorin levels in the conditioned media were significantly increased (Fig. 5n). This indicated that a low concentration of Aβ42 increased the neuronal secretion of decorin. Taken together, CSF-decorin levels are increased in both App knock-in and APP23 mice exhibiting Aβ amyloidosis, and the levels go down upon increased Aβ42 pathology in the brain, similar to the CSF-decorin alterations in a + t − human subjects.

Decorin activates autophagy-lysosomal pathway by modulating lysosomal degradation
Previous study has shown that decorin activates autophagy in endothelial cells [30] which prompted us to investigate the relationship between decorin and autophagy in a neuronal setting. Interestingly, treating App wt/wt mouse primary neurons with decorin decreased the level of p62, microtubule associated protein 1 light chain 3 (LC3)-II and LC3-II/LC3-I ratio (Fig. 6a, b). However, co-treatment with bafilomycin A1 (an inhibitor of lysosomal proteolysis) and decorin led to no change of LC3-II and p62 levels but to a decrease of LC3-II/LC3-I levels as compared to bafilomycin A1 treatment alone (Fig. 6a, c). This indicates that decorin may stimulate autophagy flux by enhancing downstream lysosomal degradation rather than by increasing initiation of autophagy. The levels of lysosomal-associated membrane protein 1 (LAMP1), a marker for the amount and integrity of lysosomes, were significantly increased in decorin-treated neurons whereas the levels of a master regulator of the lysosome biogenesis, transcription factor EB (TFEB), were not changed (Fig. 6d, e). Interestingly, the LAMP1 levels were decreased by co-treatment of decorin and bafilomycin A1 as compared to the single bafilomycin A1 treatment which indicates that decorin may ameliorate the lysosomal inhibition (Fig. 6d, f ), which is additionally supported by a decrease in the levels of autophagy receptor NBR1 (Fig. 6d, f ). These data demonstrate that decorin influences the lysosomal degradation and potentially rescues partially the lysosomal inhibition by bafilomycin. The analysis of the decorin signaling pathways revealed no significant changes of phospho-proteins associated with autophagy upstream pathways (Additional file 12: Fig. S5). However, several proteins (apoptosis signal-regulating kinase 1 (ASK1), SAPK/Erk kinase (MKK4/SEK1), focal adhesion kinase (FAK), mothers against decapentaplegic homolog 3 (Smad3)) related to c-Jun N-terminal kinases (JNK)/mitogen-activated protein kinase (MAPK) signaling pathways exhibited significantly altered phosphorylation status (Fig. 6g, h) which play a role in autophagy regulation [31].

Discussion
In this study we aimed at determining how AD brain pathologies are reflected in CSF and their translational potential toward clinical use, using data from both AD mouse models, and human subjects. Here, we used two App knock-in mouse models exhibiting different degree of Aβ pathology, and analyzed their CSF by label-free MS. This revealed alterations in several ECM related proteins, including decorin and lumican that were specifically increased in the CSF of App NL-F/NL-F mice. Interestingly, decorin and lumican are highly expressed in the recently identified vascular cell type fibroblast-like cells which reside within the perivascular Virchow-Robin space in the mouse brain [32]. The perivascular pathway facilitates CSF flow and enhances the clearance of Aβ   [33]. Most importantly, decorin was similarly and significantly increased in CSF from both App NL-F/NL-F mice and a + t − NC human subjects. In addition, an increase of CSF-decorin is also found in APP Tg mice exhibiting Aβ amyloidosis, further generalizing the correlation of Aβ amyloidosis and CSF-decorin. Notably, among a + t − NC a   Quantitative western blot of LAMP1, p-TFEB S142, T-TFEB and NBR1 levels in mouse primary neurons treated with decorin and/or bafilomycin A1. Brain homogenate from Atg7 conditional knockout mouse was loaded in the right most lane. g Phosphorylation profiling of proteins with ratios greater than 1.5 and lower than 0.6 comparing decorin-treated to non-treated. Scale bars, log10 of the ratio. h The 34 phospho-proteins identified in (g) were separately uploaded to GO and top 10 significantly enriched GO terms of biological processes are shown. (False discovery rate < 0.01). Data in (b, c, e, f) were analyzed by student's t-test. Data are represented as mean ± SEM; *p < 0.05, ***p < 0.001; ns: not significant. Baf bafilomycin A1, p-TFEB S142 phosphorylated TFEB Ser142, T-TFEB total TFEB subjects, a significant positive correlation of CSF-decorin and the well-established biomarker CSF-Aβ42 was found, and the correlation was significantly switched from the non-linear negative correlation in the a-t− healthy subjects. In other words, among a − t − healthy subjects that are on the trajectory towards preclinical AD, the CSF-decorin levels transiently increases as CSF-Aβ42 decreases. After reaching a peak in CSF-decorin levels in a + t − preclinical AD subjects, CSF-decorin levels start to decrease along with a continuous decrease in CSF-Aβ42 levels, and in a + t − MCI subjects the CSFdecorin levels are back to that of a-t-healthy subjects. Hence, a change in CSF-decorin level is associated with an early Aβ pathology or even increased levels of soluble Aβ species. This transient increase in CSF-decorin followed by a subsequent decrease may explain that CSFdecorin and CSF-Aβ42 do not correlate over the whole disease spectrum. In addition, CSF-Aβ42 levels are low already at an early stage of AD and continue to be stable during the AD dementia stage [34]. The ROC analysis of the a + t − preclinical AD subjects (n = 36) showed that CSF-decorin levels could not classify this group with high accuracy (AUC = 0.061, p = 0.058). However, decorin could indeed classify with high sensitivity and specificity a subtype of CSF-amyloid positive subjects including NC, MCI, AD and whose CSF proteome was characterized by innate immune activation, increased ECM proteins and increased transthyretin indicating a potential ChP dysfunction (AUC = 0.70, p = 1.73 × 10 -6 ). Further analysis of ChP pathology in AD brains of this AD subtype and its correlation with CSF-decorin is warranted. Moreover, the CSF-decorin levels significantly negatively correlated with both CSF-t-tau and CSF-p-tau in a + t − NC subjects as well as over the whole AD spectrum. This additionally supports the link of decorin and the tau pathology in AD brain which needs to be addressed in mouse models of tau pathology. A growing number of studies have shown that BBB breakdown occurs in early AD and is closely related to cognitive dysfunction [35][36][37]. Hence, finding a potential CSF biomarker, such as decorin, for the early changes of BBB would be suitable for predicting the early changes of cognition. In our study we additionally found other ECM proteins including SPARC-like protein 1, fibronectin and ecm1 protein to be significantly downregulated in a similar manner in App knock-in mouse models and a + t − MCI and AD human subjects showing that such changes are translational between mice and human. Considering that ECM is an important constituent of BBB/BCSFB, which are most likely affected in AD [2,4], our proteomic findings suggest that the App knock-in mice could also have changes in the BBB/BCSFB. Decorin has been identified in Aβ depositions in AD brains [16] whereas a disturbed decorin expression is observed in fibroblasts from sporadic AD patients [38]. Our finding that Aβ42 induces secretion of decorin from neurons may explain the decreased level of decorin in PV-positive interneurons in hippocampus. On the other hand, it cannot be ruled out that decorin alterations in other cell types e.g., glia cells, induced by Aβ amyloidosis may eventually lead to CSF-decorin changes. This needs to be further investigated. Furthermore, we showed that decorin expression in ChP was increased in App NL-F/ NL-F mice which correlated both with Aβ pathology and elevated CSF-decorin level. Taking these data together indicates that early Aβ amyloidosis, potentially involving soluble Aβ species, in the brain may increase the decorin expression in ChP which may also lead to elevated decorin levels in the CSF since CSF is produced in ChP. A time course analysis of CSF-decorin in App NL-F/ NL-F mice revealed an increase of CSF-decorin levels at three months of age even before the establishment of Aβ plaque deposition followed by a continuous decrease upon Aβ pathology development, which is similar to the CSF-decorin alterations in a + t − human subjects. An increase in CSF-decorin was also found in the App NL-G-F/ NL-G-F mice by ELISA. The discrepancy between MS and decorin ELISA data of the 12-13 months old App NL-G-F/NL-G-F mice may be due to differences in detection of different proteolytic fragments of decorin by the two methods (decorin was identified and quantified based on five to 11 different peptides, one of which contains a glycosylation site, in the MS-based proteomic approach, peptide list for decorin in Additional file 2: Table S2). Additionally, decorin is a proteoglycan which has several glycosylation sites, and the glycosylation levels may differ between the two App knock-in models. It is likely that the glycosylated and non-glycosylated peptides of decorin may not all be identified in the MS analysis since they have different mass-to-charge ratio.

App wt/wt App NL-F/NL-F App NL-G-F/NL-G-F App wt/wt App NL-F/NL-F App NL-G-F/NL-G-F App wt/wt App NL-F/NL-F App NL-G-F/NL-G-
Decorin has previously been shown to activate autophagy in endothelial cells, through VEGFR2/ AMPKα activation and mTOR inhibition [39,40], and in breast carcinoma cells [41] and activates autophagy flux in glioma cells [42] and intestinal cells [43]. We here extend this to include also neuronal autophagy, where autophagy-lysosomal pathway could be stimulated in vitro by decorin mainly via enhancing lysosomal degradation. Such activation may have implications for Aβ metabolism since Aβ is degraded through the autophagylysosomal system. Additionally, this autophagic effect by decorin may also involve JNK/MAPK signaling pathways which is involved in autophagy-mediated cell death [44]. This is interesting since autophagy is altered in AD [45,