Autophagy is impaired in AD brains
p62 is a cargo receptor that recognizes and transports ubiquitinated proteins to autophagosomes to be degraded by lysosomes. p62 is normally degraded along with the cargo during functional autophagy and therefore, p62 accumulation and aggregation indicate impaired autophagy. To investigate autophagy status in AD brains, we immunostained brains from healthy controls (Fig. 1a-e) and AD subjects (Fig. 1f-j) for p62 and quantified the p62-positive signals. The results showed a pronounced p62 accumulation in both cortex (Fig. 1f) and hippocampus including dentate gyrus (DG) (Fig. 1g), CA1 (Fig. 1h), and CA3 (Fig. 1i) of AD brains (semi-quantified in Fig 1k,l), whereas no p62 accumulation was found in brains of non-demented individuals (Fig. 1a-d). p62 accumulated as large intracellular aggregations in neurons (Fig. 1f and 1g, arrow in depicted area 1, Fig. 1h, arrow in depicted area) and was also identified as small round dots or outstretched structures (Fig. 1f and 1g, arrows in depicted area 2). This dot-shaped staining may represent axonal beadings, which are the series of swellings along the axons located in the molecular layer of DG. In addition, we found p62 in apoptotic bodies (Fig. 1i, arrows in depicted area) and corpora amylacea (Fig. 1j, arrows in depicted area). Interestingly, tunica intima of the vessels also contains substantial amount of p62-postive staining (Fig. 1j). In summary, p62 accumulation was frequently observed in AD brains, which indicates a disturbed autophagy-lysosomal system potentially caused by an inhibition of autophagy in both neurons and cells in the tunica intima of vessels.
To further characterize alterations in the autophagy system in AD brains, we analyzed another autophagy key protein in the autophagy pathway, Atg5, in AD postmortem brains. The results showed that Atg5 was significantly upregulated in neurons in cortex (Extended Data Fig. 1f) and in the hippocampus (granular cell layer of DG (Extended Data Fig. 1g), CA1 (Extended Data Fig. 1h), and CA3 (Extended Data Fig. 1i)) of AD brains as compared to healthy control brains (Extended Data Fig. 1a-e) (semi-quantified in Extended Data Fig 1k,l). Interestingly, we found that Atg5 was not only increased in the neurons (Extended Data Fig. 1f,g,h,i, arrows in depicted area) but also in glial cells (Extended Data Fig. 1j, arrow in depicted area), potentially as an attempt to increase autophagy in those cells.
Autophagy is impaired in AppNL-G-F/NL-G-F mice
We next asked the question if autophagy is altered in two App knock-in AD mouse models, AppNL-F/NL-F and AppNL-G-F/NL-G-F mice. The two models exhibit different degree of Aβ pathology, neuroinflammation, synaptic alteration and cognitive impairment all of which being significantly stronger in the AppNL-G-F/NL-G-F mice due to the arctic mutation which enhances Aβ oligomerization (10) (Fig. 2a,b). To characterize the autophagy status in the App knock-in mice, we performed western blot and immunohistochemical analysis for autophagy markers LC3 and p62 using 12 months old homozygous Appwt/wt, AppNL-F/NL-F and AppNL-G-F/NL-G-F mouse brains. We found that both p62 and LC3 II were specifically increased in the cortex of AppNL-G-F/NL-G-F mice but not in AppNL-F/NL-F mice as compared with Appwt/wt mice (Fig. 2c,d). In addition, the LC3 II levels were also significantly increased in the hippocampus of AppNL-G-F/NL-G-F mice whereas it was decreased in the AppNL-F/NL-F mice as compared to Appwt/wt mice (Fig. 2e,f). Consistently, immunostaining of p62 showed a significant accumulation of p62 in AppNL-G-F/NL-G-F mice as compared to Appwt/wt (Fig. 2g,h), whereas the p62 mRNA expression level was unaltered (Fig. 2i). These data together show that autophagy is inhibited in the brains of AppNL-G-F/NL-G-F mice.
ECM and autophagy related proteins are altered in CSF of App knock-in mice
In a translational approach to identify potential AD-related biomarkers including alterations in proteins associated with or regulating autophagy, we next investigated CSF of the 12 months old AppNL-F/NL-F, AppNL-G-F/NL-G-F, and Appwt/wt mice. Taking the limited volume of mouse CSF into account, a label-free MS approach was used for the detection and quantification of the mouse CSF proteomes. This led to the identification of 427-703 proteins (a complete list of all identified proteins is presented in Supplementary Table 1). A qualitative analysis of proteins that were detected in all 12 samples, revealed that 246 proteins were identified in all three groups while some proteins were identified only in two of the groups and some proteins were uniquely detected in one group (Fig. 3a). Principal component analysis (PCA) of the CSF proteome using the 246 proteins identified in all the samples indicated a separation between the three groups (Fig. 3b). Among the proteins that were significantly changed (p < 0.05) in AppNL-F/NL-F and AppNL-G-F/NL-G-F mice as compared to Appwt/wt mice, 13 proteins were commonly changed whereas 25 and 23 proteins were specifically changed in either AppNL-F/NL-F or AppNL-G-F/NL-G-F mice respectively (Fig. 3c). PCA of the significantly altered proteins (p < 0.05) resulted in a tighter clustering of the three groups (Fig. 3d). Volcano plots visualized a similar number of up and downregulated proteins in both AppNL-F/NL-F and AppNL-G-F/NL-G-F mouse CSF as compared to Appwt/wt mice (Fig. 3e,f, protein list in Supplementary Table 2). A detailed list of significantly (p < 0.05) altered proteins are shown in the heatmap (Fig. 3g,h). Among those significantly altered proteins, two proteins (Cathepsin B and Alpha-mannosidase) are associated with autophagy and seven proteins (SPARC-like protein 1, fibronectin, ecm1 protein, collagen alpha-1(I), basement membrane-specific heparan sulfate proteoglycan core protein, fibulin-1, vitronectin) are ECM proteins (Table 1). Interestingly, two proteins, decorin and lumican, are related to both autophagy and ECM and significantly increased in AppNL-F/NL-F mice as compared to Appwt/wt (Table 1). Comparing the two App knock-in mouse models directly with each other revealed significant (p < 0.05) changes in additional autophagy and ECM related proteins (Extended Data Fig. 2a,b, Table 1). Taking these data together indicate that AppNL-F/NL-F and AppNL-G-F/NL-G-F mice are, at least to some extent, different mouse models of AD, which may also include the autophagy status reflected in their CSF proteomes.
Several studies have previously found that the dysfunction of BBB and BCSFB, including changes in ECM proteins, increase upon aging and in AD (30-32). Therefore, to identify additional brain barrier changes in the CSF from the App knock-in mice not detected by the MS analysis, we also analyzed the CSF of 18 months old App knock-in mice (n = 5) using a mouse proximity extension assay (PEA) panel containing several autophagy and ECM associated proteins (but not including decorin and lumican). From a panel of 92 proteins, one autophagy associated protein (Tripeptidyl-peptidase 1) and two ECM associated proteins (Matrilin-2 and CCN family member 4) were significantly (p < 0.05) altered in App knock-in mice (Table 1, Supplementary Table 3). In addition, indispensable BBB associated protein platelet-derived growth factor subunit B, was significantly decreased in AppNL-G-F/NL-G-F mice as compared to AppNL-F/NL-F mice (Supplementary Table 3). Altogether, the analysis of CSF from App knock-in mice by MS and PEA showed alterations in both autophagy and ECM-related proteins as well as BBB-associated proteins indicating changes in the BBB/BCFCB. Interestingly, decorin and lumican which are related to both autophagy and ECM are significantly altered in the CSF of AppNL-F/NL-F mice but not in that of AppNL-G-F/NL-G-F mice which indicates that changes in the CSF of these two proteins could reflect the early events in the development of the pathologies in the App knock-in mice.
Decorin is similarly increased in CSF of AppNL-F/NL-F mice and in CSF of preclinical AD subjects having abnormal amyloid
In an attempt to further understand how changes in the mouse CSF proteomes of the AD mouse models reflect those observed in patients, we next compared them to MS-characterized CSF proteomes of a large human cohort European Medical Information Framework for Alzheimer’s Disease Multimodal Biomarker Discovery (EMIF-AD MBD) (33). According to the specific criteria, 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 (34). The CSF proteome alterations in NC, MCI and AD were based on the comparison to the control subjects having normal CSF Aβ42 and tau, and normal cognition.
Since the present study is an exploratory study, we chose to perform two kinds of comparisons of the mouse and human proteomes. Firstly, we compared how many proteins with relative expression levels below one and above one as compared to healthy controls, were common in the mouse and human CSF proteomes. Secondly, we analyzed the number of proteins with significantly changed expression levels that were commonly altered in both mouse and human CSF proteomes. Because of App knock-in mice exhibit robust Aβ pathology but less pronounced tau pathology (AppNL-G-F/NL-G-F mice have higher tau phosphorylation at Ser-396/Ser-404, and Ser-422 as compared to Appwt/wt mice, but lack neurofibrillary tangles) (35), their CSF proteomes were compared with human proteomes stratified for CSF t-tau status i.e., abnormal-amyloid/abnormal-tau (a+t+) (Supplementary Table 4) and abnormal-amyloid/normal-tau (a+t-) (Supplementary Table 5). Though a similar number of commonly upregulated proteins comparing the mouse and human cohorts was found, the comparison revealed that 76-90 proteins were commonly downregulated in mouse and a+t- human subjects whereas much fewer, 24-29 proteins, were commonly downregulated in mouse and a+t+ human subjects (Fig. 4a). This indicated that alterations in CSF proteome of App knock-in mice are, at least to some extent, more similar to the CSF alterations observed in a+t- human subjects which may reflect that App knock-in mice have a strong Aβ pathology whereas the tau pathology is less pronounced. Further analyses were therefore performed in a+t- human subjects.
A qualitative direct comparison of CSF MS data from patients and AppNL-F/NL-F mice, as presented in the Venn diagrams, showed that 33, 46 and 35 proteins exhibited a relative expression level above one whereas 76, 84 and 81 proteins were below one in NC, MCI and AD, respectively, and AppNL-F/NL-F mice (Fig. 4b, protein list in Supplementary Table 5). Notably, restricting the comparisons to the proteins with significantly altered levels (p < 0.05), only one protein decorin (DCN), was found to be significantly upregulated in both NC subjects and AppNL-F/NL-F mice (Fig. 4c). In the MCI vs AppNL-F/NL-F mouse comparison six proteins, including one ECM protein, SPARC-like protein 1 (SPARCL1), as well as two BBB-associated proteins, dickkopf-3 (DKK3) and neurotrimin (NTM), were significantly and commonly downregulated. In the AD vs AppNL-F/NL-F mouse comparison the same proteins that were altered in the MCI vs AppNL-F/NL-F mouse comparison were found to be altered. In addition, one ECM protein, fibronectin (FN1), and one BBB-associated protein, contactin-1 (CNTN1), were commonly and significantly altered in the AD vs AppNL-F/NL-F mouse comparison (Fig. 4d). Comparing AppNL-G-F/NL-G-F mouse CSF proteome with the human CSF proteome revealed that 38, 50 and 37 proteins exhibited expression levels above one whereas 84, 90 and 86 proteins exhibited expression levels below one in NC, MCI, AD, respectively (Fig. 4e, protein list in Supplementary Table 5). Restricting the comparisons to only the significantly (p < 0.05) changed proteins (Fig. 4f), apolipoprotein A1 (APOA1) and apolipoprotein A2 (APOA2) which are related to cognitive status and late-onset AD (36, 37), were significantly upregulated in both AD subjects and AppNL-G-F/NL-G-F mice whereas nine proteins including two ECM proteins, SPARC-like protein 1 (SPARCL1) and ecm1 protein (ECM1), and two BBB-associated proteins, limbic system–associated membrane protein (LSAMP) and natriuretic peptide precursor C (NPPC), were significantly and commonly downregulated in MCI. In AD, one additional ECM protein fibronectin (FN1) was similarly and significantly altered, as well as one additional BBB-associated proteins contactin-1 (CNTN1) was significantly and commonly downregulated (Fig. 4g). Taken together, the comparison of mouse and human CSF proteomes revealed that several ECM and BBB/BCSFB-associated proteins were significantly and commonly altered, indicating that the App knock-in mice recapitulate to some extent the changes observed in human CSF. Among the ECM proteins, decorin was found to be the earliest altered protein as shown by a significant increase in both AppNL-F/NL-F mice having a mild and limited Aβ pathology, and preclinical a+t- subjects, whereas the cohorts representing the later stages in the AD spectrum i.e. MCI and AD did not exhibit a significant increase of decorin.
The correlation of CSF-decorin with CSF-amyloid switches from negative to positive upon onset of abnormal CSF-amyloid
We next studied in human CSF how decorin levels were related to the CSF AD biomarkers Aβ amyloid, t-tau and p-tau. Within the NC group, CSF-decorin levels follow a non-linear trajectory, with higher levels of decorin corresponding to lower CSF-amyloid in a-t- subjects, albeit not significantly (beta(se) = -0.16(0.18), p = 0.36) (Fig. 5a). In contrast, lower CSF-decorin levels corresponded to lower CSF-amyloid levels in a+t- subjects (beta(se) = 0.61(0.25), p = 0.01), a correlation that is significantly (p = 0.016) different from that of a-t- subjects (Fig. 5a). In other words, in subjects with pathological levels of CSF-amyloid, CSF-decorin levels tend to increase with decreased CSF-amyloid levels towards reaching abnormal CSF-amyloid. When CSF-amyloid levels are abnormal, the correlation switches and becomes positive. In addition, plotting CSF-decorin against CSF-t-tau and CSF-p-tau within the NC group reveals non-linear associations between decorin and t-tau or p-tau in a-t- subjects, whereas the correlations become significantly negative in a+t- subjects for both decorin and t-tau (beta(se) = -0.89(0.32), p = 0.0053) (Fig. 5b) and decorin and p-tau (beta(se) = -1.00(0.21), p < 0.0001) (Fig. 5c). Furthermore, across the total group including NC, MCI and AD subjects (n=310), lower CSF-decorin levels were related to higher CSF-t-tau (beta(se) = -0.11(0.04), p = 0.0014) (Fig. 5e) and CSF-p-tau levels (beta(se) = -0.16(0.04), p < 0.0001) (Fig. 5f), but no association with CSF-amyloid levels was observed (beta(se) = 0.09(0.06); p = 0.11) (Fig. 5d). Taken together, a switch from negative to positive correlation of CSF-decorin and CSF-amyloid occurs in NC a+t- subjects when CSF-amyloid reaches pathological levels, and at this stage, decorin additionally negatively correlates with both CSF-t-tau and CSF-p-tau. In addition, CSF-decorin levels continuously decreases with higher CSF-t-tau and CSF-p-tau levels in the whole AD spectrum. This indicates early change in CSF-decorin is associated with Aβ amyloidosis.
ECM associated biological processes were similarly changed both in App knock-in mouse models and human subjects with a+t-
Considering that a similar number of proteins exhibit levels either below one or above one in both AppNL-F/NL-F and AppNL-G-F/NL-G-F mice as compared to the a+t- human subjects (Fig. 4a), we decided to investigate whether those proteins are the same. This comparison also allowed us to detect changes that occur throughout the course of AD both in human and mouse as well as to pinpoint the ones that were specifically altered in one of the App knock-in mouse models. The Venn diagrams show that nine to 18 proteins were exclusively changed in one of the App knock-in mouse models and similarly changed in NC, MCI and AD human subjects, indicating that unique proteins are changed in the two App knock-in mouse models (Extended Data Fig. 3a). To get a better understanding of which biological processes are enriched from those exclusively changed proteins, we performed gene ontology enrichment analyses (protein list used for gene ontology is in Supplementary Table 6). Interestingly, the enrichment analyses revealed several pathways related to the ECM; collagen fibril organization, collagen catabolic process, extracellular matrix organization and extracellular matrix disassembly, as well as one pathway related to glycosaminoglycan metabolism; mucopolysaccharide metabolic process (Extended Data Fig. 3b,c). This data indicated that changes in BBB/BCSFB composition may be present in the App knock-in mice which similarly takes place in human subjects with a+t- in the CSF.
Decorin is increased in the ChP of AppNL-F/NL-F mice and decreased in parvalbumin (PV) positive interneuron neurites of AppNL-G-F/NL-G-F mice
Having found that decorin was increased in CSF of AppNL-F/NL-F mice, we were interested in the potential decorin alterations in the mouse brains since the pathological changes in the brain can translate to CSF. Therefore, we investigated the brains of the 12 months old mice used for CSF collection and performed immunohistochemistry of decorin to characterize decorin expression and localization. Interestingly, by establishing an immunostaining protocol based on the removal of glycosaminoglycan sidechains, we were able to successfully localize decorin in ChP, blood vessels and neurons. Indeed, decorin expression was significantly higher in ChP in AppNL-F/NL-F mice as compared to Appwt/wt mice but unaltered in AppNL-G-F/NL-G-F mice (Fig. 6a,b). In addition to that, we found that decorin was highly expressed in the tunica externa of arteries (Extended Data Fig. 4a) and veins (Extended Data Fig. 4b) on the surface of the brain but not present in the vessels of brain parenchyma (Extended Data Fig. 4c). However no significant differences in decorin levels were observed in the vessels (Extended Data Fig. 4d,e).
Examining the brain parenchyma, we found that decorin was localized in neurons in both hippocampus and cortex. In hippocampus, decorin was not only expressed in CA2 pyramidal neurons but also in some other neurons of CA layers. To identify the decorin-positive neuronal subtype we performed double staining of decorin with two interneuronal markers, PV and somatotropin release-inhibiting factor (SRIF) (38). The results showed that a majority of decorin positive neurons co-localized with PV positive interneurons (Fig. 6c) whereas there was only a minor co-localization with SRIF positive interneurons (Fig. 6d). The quantification showed that the distributions of decorin positive cells in Appwt/wt, AppNL-F/NL-F, and AppNL-G-F/NL-G-F mice were: more than 50% of decorin was found in CA2 pyramidal neurons, PV positive interneurons account for 30-40%, whereas approximately 10% were SRIF positive interneurons or other cell types (Extended Data Fig. 5). Although the distribution of decorin expressing cells were not different between App knock-in and Appwt/wt mice, the decorin positive interneuron neurites of App knock-in mice, especially in the AppNL-G-F/NL-G-F mice, contained significantly less decorin (Fig. 6e,f) as compared to Appwt/wt mice.
Decorin levels in CSF correlates with Aβ pathology in the brains of AppNL-F/NL-F mice
To better understand how Aβ amyloidosis is associated with both autophagy status and decorin expression in App knock-in mice as well as how these changes can be mirrored in the CSF, we performed Spearman’s correlation analysis of Aβ plaque load, p62 and LC3 II levels, decorin expression in the brains and CSF-decorin levels. The analysis of the data from Appwt/wt and AppNL-F/NL-F mice (Fig. 7a) showed that the CSF-decorin levels positively correlated with both decorin expression in ChP (r = 0.79, p = 0.028) and Aβ plaque load in cortex (r = 0.76, p = 0.04). In addition, the LC3 II levels in hippocampus negatively correlated with Aβ plaque load in hippocampus (r = -0.84, p = 0.017), CSF-decorin levels (r = -0.86, p = 0.011) and decorin expression in Chp (r = -0.93, p = 0.002). Moreover, the decorin expression in the ChP negatively correlated with decorin in interneuron neurites (r = -0.74, p = 0.046). All these together indicated that autophagy is altered during early Aβ amyloidosis and the increase of CSF-decorin correlates with the increase of decorin in ChP possibly driven by the mild Aβ pathology in AppNL-F/NL-F mice.
The correlation analysis of the data obtained from Appwt/wt and AppNL-G-F/NL-G-F mice (Fig. 7b) showed that both p62 (r = 0.86, p = 0.012) and LC3 II (r = 0.89, p = 0.006) positively correlated with Aβ plaque load in cortex whereas only LC3 II positively correlated with Aβ plaque load in hippocampus (r = 0.77, p = 0.033). The decorin expression in interneuron neurites negatively correlated with Aβ plaque load (r = -0.91, p = 0.005) and LC3 II (r = -0.93, p = 0.002) in hippocampus. In sharp contrast to the Appwt/wt and AppNL-F/NL-F dataset, the CSF-decorin levels in Appwt/wt and AppNL-G-F/NL-G-F dataset did not correlate with any of the measured parameters. The correlation findings indicate that the autophagy impairment and the reduced decorin expression in the interneurons were caused by the severe Aβ amyloidosis in AppNL-G-F/NL-G-F mice whereas the correlation of decorin levels in CSF and in Chp with Aβ is lost in the AppNL-G-F/NL-G-F mice.
Decorin activates autophagy-related lysosomal degradation in primary neurons
Having found that autophagy was inhibited in AppNL-G-F/NL-G-F mouse brains and that less decorin was expressed in the neurites of hippocampal interneurons in AppNL-G-F/NL-G-F mice together with the previous findings that decorin activates autophagy in endothelial cells (39) prompted us to investigate the relationship between decorin and autophagy in a neuronal setting. We, therefore, treated primary cortical/hippocampal neuron culture derived from Appwt/wt mice with decorin and measured the effect of decorin on autophagy flux as determined by p62 and LC3 western blot analysis. The single decorin treatment data showed that decorin significantly decreased LC3 II and LC3 II/LC3 I ratio as compared to control and also a tendency towards decreased p62 level (p = 0.0891) (Fig. 8a,b). Furthermore, co-treatment with bafilomycin A1 (an inhibitor of lysosomal proteolysis) and decorin led to no change of LC3 II and p62 levels but an increase of LC3 II/LC3 I level as compared to Bafilomycin A1 treatment alone (Fig. 8c,d). However, the reason for the increase of LC3 II/LC3 I was from LC3 I reduction (Fig. 8d). All these together indicated that decorin activates autophagy-associated lysosomal degradation rather than causing an upstream induction of autophagy in Appwt/wt primary neurons.