Neurotoxic astrocytes secreted glypican-4 drives Alzheimer’s tau pathology

Apolipoprotein E4 (APOE4) is the most crucial genetic risk factor of late-onset Alzheimer’s disease (AD). However, the mechanism through which APOE4 induces AD risk remains unknown. Here, we report the astrocyte-secreted protein glypican-4 (GPC-4), as a novel binding partner of APOE4, drives tau pathology. APOE4-carrying AD patients display more tau accumulation compared to APOE4-noncarring AD patients. GPC-4 is highly expressed in APOE4 AD patients, and is regulated by microglial factors via NF-κB signaling pathway. The astrocyte-secreted GPC-4 induced both tau accumulation and spreading in vitro and in vivo . Further, GPC-4 is required for APOE4-mediated surface trafficking of low-density lipoprotein receptor-related protein 1 (LRP1) and tau propagation. GPC-4 activates unfolded protein response (UPR) pathway IRE1α, and pharmacological inhibition of IRE1α with KIRA6 blocks GPC-4 induced tau propagation. Together, our data comprehensively demonstrate that the APOE4-induced AD risk is directly mediated by GPC-4, and that perturbing GPC-4 induced IRE1α pathway has therapeutic opportunities.


INTRODUCTION
Apolipoprotein (APOE) plays a major role in the circulation of high-density and very-low-density lipoproteins and mediates the transport of fats between the cells 1 . In the brain, APOE is secreted by glial cells, primary astrocytes 2 . Cholesterol released by astrocytes in the form of APOEcontaining high-density lipoprotein-like particles are vital for neuronal survival 3 . and acetylated tau antibody (Lys174) on postmortem tissues of APOE2/3 (control), APOE2/2 (AD), APOE3/3 (AD) and APOE4/4(AD) individuals. There is no significant tau accumulation in APOE2/2 AD and APOE2/3 control (Fig. 1A). We observed presence of neurofibrillary tangles, neuropil threads and neuritic plaque both in APOE3/3 AD and APOE4/4 AD patients (Fig. 1A).
The neurofibrillary tangles, neuropil threads and neuritic plaque are different forms of tauassociated neurofibrillary changes observed in the AD brain 26 . Interestingly, this triad of tau pathologies is significantly increased in APOE4/4 AD compared to APOE3/3 AD (Fig. 1B). Our data shows an accumulation of more tau protein in APOE4/4 AD patients compared to other APOE variants.
Interestingly, immunoprecipitation from APOE44 human brains revealed that the majority of APOE4 protein exists as a monomeric protein or very small APOE4-containing particles (Fig.   S1A), suggesting the presence of abundant lipid-free APOE4 proteins/very small particles in human brains. Further, tau seeding biosensor assay showed the APOE4-containing particles and purified human APOE4 protein induced tau seeding in similar ways ( Fig. S1B and S1C). It is consistent with previous studies 30,31 ; suggesting that APOE4 protein, not lipids, possess pathogenic activity.
By using a primary neuronal culture from the human tauopathy mouse model P301S, we first investigated which APOE variant is crucial for tau protein phosphorylation (Fig. S2A). We found that APOE2 significantly reduced tau phosphorylation compared to control and APOE4 treatments ( Fig. S2C), while the total tau (tau-5) remained unchanged (Fig. S2B). Further, addition of both APOE2 and APOE4 also reduced AT8 levels (Fig. S2C). To address APOE2-mediated neuronal tau phosphorylation, as tau accumulation is also observed in glial cells 32 , we performed IHC in the presence of APOE2, APOE4, and both APOE2 and APOE4 (Fig. S2D). We observed the APOE2 significantly reduced AT8 levels in neurons (Fig. S2E).
We next determined whether APOE2 and APOE4 differentially influence tau spreading. For this purpose, we cocultured neurons from tau KO mice that express GFP while tau is absent and from P301S mice. With the cocultured neurons from these two mice, we expect that because the GFP+ neurons (tau KO) do not express endogenous tau proteins, the tau+ signal in GFP+ neurons can be considered to be the tau protein from P301S neurons. As expected, we observed the tau+ signal from the GFP+ neurons (tau KO neurons), suggesting the tau proteins are released by P301S neurons and were taken up by GFP+ neurons. (Fig. S2F). When the neuronal cultures were treated with APOE isoforms, we found that APOE4 robustly enhanced tau spreading (Fig. 1C, D). Upon adding both APOE2 and APOE4 to the neuronal cultures, we observed no additional increase in tau spreading because APOE2 hampers both tau phosphorylation and spread. It suggests that APOE2 induces a protective response by decreasing the AT8 levels and APOE4-induced tau spreading.

Astrocyte secreted glypican-4 induces tau pathology
How do the APOE variants differentially induce tau pathology? Alboleda-Velasquez et al recently reported that, compared to APOE2, APOE4 showed a strong interaction with heparin 14 . Therefore, we hypothesized that Heparan Sulfate Proteoglycans (HSPGs) may be involved in APOE4mediated tau pathology. To determine which HSPGs are crucial for tau pathology, we screened a list of HSPGs in vitro and found that glypican-4 (GPC-4) robustly induced AT8 levels in neuronal cultures from P301S animals ( Fig. 2A-C). The IHC experiment yielded essentially the same result ( Fig. 2D, E). To test the effect of GPC-4 on tau spreading, we cocultured neurons of P301S and tau KO animals and found that GPC-4 increased tau spreading from P301S neurons to tau KO neurons (Fig. 2F, G).
The GPC-4 is primarily expressed in astrocytes 33 . To validate whether astrocyte-secreted GPC-4 is sufficient to induce tau pathology, we treated P301S neuronal culture with astrocyte-conditioned medium (ACM) as described in Figure 2H. ACM-treated neurons expressed significantly higher levels of AT8 ( Fig. 2I-K). We next treated the astrocytes with GPC-4 shRNA, and collected GPC-4 deprived ACM (Figs. 2L). Interestingly, GPC-4 deprived ACM failed to induce tau pathology ( Fig. 2M-O). It suggests that astrocyte-derived GPC-4 is sufficient to induce tau pathology.

Neurotoxic astrocytes secrete glypican-4
Given that APOE4 carrying AD patients displayed an accumulation of more tau proteins, we next investigated whether GPC-4 is differentially expressed in APOE variants. We found that APOE4-carrying AD patients expressed more GPC-4 protein compared to APOE4-noncarrying AD patients (Fig. 3A). Immunostaining from human postmortem brain tissues revealed that indeed astrocytes express GPC-4 protein (Fig. 3B). Single-cell RNA-sequencing (ScRNAseq) revealed the presence of AD-associated genes and astrocytes in humans 34-36 . We therefore wondered which subtype of astrocytes express GPC-4. We generated a heatmap with those genes to investigate which subtype of astrocytes express GPC-4 (Fig. 3C). AD-genes are mainly expressed in astrocyte subtype 2 and 3, suggesting that these are the clusters which are associated with AD. Interestingly, GPC-4 is expressed within AD associated astrocyte subcluster 3 (Fig. 3C).

Glypican -4 drives APOE4-induced LRP1 trafficking and tau propagation
The LRP1 (low-density lipoprotein receptor-related protein 1) is involved in tau uptake and spreading 38 . We next determined whether the APOE variants have differential effects on the LRP1 receptor. For this purpose, we monitored the effects of APOE2 and APOE4 on the total and surface LRP1 levels. We found that the addition of APOE2 had no effect on both the total and surface LRP1 levels ( Fig. 4A-C). By contrast, APOE4 enhanced the surface LRP1 levels but not the total LRP1 levels ( Fig. 4A-C). Further, active exocytic and endocytic pathways are required for APOE4-induced surface LRP1 and APOE2 mediated downregulation of APOE4-induced surface LRP1, respectively (Fig. S3).
We next reasoned that GPC-4 being an astrocyte-secretory factor would interact with neuronal LRP1 to induce tau pathology in neurons. To test this notion, we targeted the likelihood of GPC-4-LRP1 interaction and assessed the effect of GPC-4 on the key signal transduction protein. First, we validated the GPC4-LRP1 interaction from human postmortem brain tissue by coimmunoprecipitation (Fig. 4D). In the neuronal culture, the addition of GPC-4 unaltered the total LRP1 (Fig. 4E, F), whereas the surface LRP1 levels increased greatly (Fig. 4E, G). This suggests that GPC-4 interacts with LRP1 and regulates trafficking of LRP1.
We next investigated whether APOE4 is dependent on GPC-4 to induce surface LRP1 levels. To investigate this notion, we treated neurons with either APOE4 alone or APOE4 with GPC-4 shRNA (Fig. 4H). The total LRP1 was unaltered (Fig. 4I). However, APOE4 in the presence of GPC-4 shRNA showed a significant reduction of surface LRP1 compared to APOE4 alone (Fig.   4J). Altogether, these data indicate that APOE4 increases surface LRP1 through GPC-4.

Glypican-4 drives APOE4-mediated tau propagation
Given that both APOE4 and GPC-4 enhanced tau pathology, we next examined whether the differential efficacy of APOE2 and APOE4 is dependent on the GPC-4 interaction.
Immunoprecipitation analysis from a human postmortem brain showed that APOE directly interacts with GPC-4 ( Fig. S4A). Glycosaminoglycan chains in proteoglycans including in HSPGs play an important role in binding with other proteins. To test whether APOE2 and APOE4 differentially interact with glycosaminoglycans of GPC-4, we incubated a GPC-4-coated Sepharose column with either APOE2 or APOE4 for 1 h and eluted it with NaCl gradients. We observed that neither isoforms showed preferential binding with GPC-4 (data not shown).
However, we cannot rule out that protein segments rather than the glycan part may bind to the APOE variants differently. To test this, we incubated GPC-4 either with APOE2 or APOE 4 at room temperature for 1 h as described in Figure S4B and analyzed with a native gel. APOE2+GPC-4 combination did not show any major shifts, whereas APOE4+GPC-4 mix showed a robust shift both with GPC-4 and APOE antibodies. Further, treatment with 2-mercaptoethanol disturbed the shift, suggesting that core proteins of GPC-4 and APOE4 are in direct interaction.
Given that GPC_4 strongly interacts with APOE4 and is involved in APOE4-mediated LRP1 trafficking, we next reasoned that GPC-4 would play an important role in APOE4-mediated tau pathology. We treated the primary neuronal culture with APOE4 or APOE4 with GPC-4 shRNA for 24 h and then incubated it with 1 µg/ml of human tau protein for 1 h. Immunohistochemistry with human tau antibody HT-7 showed that APOE4 increased tau uptake but this was blocked in the presence of GPC-4 shRNA. This result highlights the role of GPC-4 in regulating the APOE4-induced tau uptake (Fig. 4K, L). We next investigated the role of APOE4 and GPC-4 in tau propagation using P301S animals (Fig. 4M). We observed APOE4-mediated tau accumulation in the ipsilateral hippocampus (Fig. S5); further, APOE4 induced tau spreading to contralateral hippocampus and cortical regions (Fig. 4N). Interestingly, APOE4-mediated tau pathology was dramatically reduced in the absence of GPC-4 ( Fig. 4N, O). This demonstrates that APOE4 induces both tau accumulation and spreading through GPC-4 (Fig. S4C).

GPC-4 drives tau propagation via IRE1α pathway
To further investigate the role of GPC-4 in tau pathology in vivo, we injected GPC-4 protein stereotaxically in the hippocampal CA1 region of P301S animals (Fig. 5A). After one week of incubation period, we observed a tremendous accumulation of phosphorylated tau in the CA1 region ( Fig. 5B-D). Additionally, injection of GPC-4 in cortical regions as well induced tau accumulation (Fig. S5D).
We next examined a possible molecular mechanism that underlies GPC-4 induced tau pathology.
Unfolded protein response (UPR) pathways play a major role in protein misfolding disorders such as AD 39,40 . IRE1α is one of the major UPR pathways involve in protein quality control 41 .
Phosphorylated IRE1α (pIRE1α) is detected within degenerating pyramidal neurons in AD patients 42 . We therefore tested whether GPC-4 induced tau pathology is associated with IRE1α pathway.
Addition of GPC-4 protein in neuronal culture enhanced phosphorylation of IRE1α protein (Fig.   5E, F). We next blocked the IRE1α pathway with a pharmacological compound, KIRA6, to test whether GPC-4 induced tau pathology can be reversed 43 . We injected GPC-4 protein in P301S animals, i.p injected KIRA6 daily for 3 weeks (5mg/kg), and analyzed on contralateral hippocampus for tau propagation (Fig. 5G). Interestingly, GPC-4 induced tau propagation is blocked in KIRA6 treated animals (Fig. 5H, I). It suggests that GPC-4 induces tau pathology via the IRE1α pathway.

DISCUSSION
Our study addresses at the molecular level why APOE4 carrying individuals are at risk of developing AD pathophysiology. We found that APOE4 AD patients have more tau accumulation both inside and outside of neurons, and APOE4 protein induces tau propagation. A subtype of neurotoxic astrocyte secreted GPC-4 triggers tau accumulation and propagation. We further demonstrated that APOE4-induced tau uptake and propagation is dependent on GPC-4 protein.
We finally showed that GPC-4 induces tau pathology via the IRE1α pathway.
Human IPSCs-derived APOE4 neurons expressed more pTau 44 . However, human tau PET studies tend to contradict each other 10,45,46 . To our knowledge there are no studies carried out to compare tau accumulation in APOE homozygous variants of human postmortem brain tissues. Our immunohistochemical studies clearly demonstrated the presence of more neurofibrillary tangles, neuropil threads and neuritic plaque in APOE4/4 AD patients. Further insights into how these tau pathologies would function differently in APOE variants is required to understand the disease progression.
An association of APOE4 with tau is independent of amyloid beta levels 22,23 . Our data demonstrating APOE4 mediated tau propagation are in agreement with these studies. Importantly, why APOE4 behaves as an AD risk factor 5 , while APOE2 is protective against AD 8 is still unknown. With differential actions of APOE2 and APOE4 in tau pathology, our in vitro studies showed that APOE4-induced tau propagation is reversed by APOE2 protein. Furthermore, APOE4 enhanced the surface expression of LRP1, but the same effect was reversed in the presence of APOE2. Interestingly, compared to APOE2, APOE4 showed a strong interaction with GPC-4, suggesting that the molecular architecture of APOE4 carrying individuals possibly determine the onset of AD pathologies. However, it is still possible the geographic regions, race or environmental factors might play an important role in APOE4-AD risks. For example, APOE4-carrying European descents are at higher risk of developing AD compared to African-Americans or Hispanic populations 2 . Therefore, to further understand an association of APOE4 with AD, molecular studies at the level of geographic region or race would be required.
Activated A1 astrocytes are considered as neurotoxic astrocytes whose secretory molecules may be involved in worsening of AD pathology 37,47 . Interestingly, we found that GPC-4 is mainly expressed within a neurotoxic astrocytic population and astrocytes-derived GPC-4 induces tau pathology. Proteostasis is vital for cellular protein quality control. Altered proteostasis leads to cellular stress, aging, increased protein half-life, protein misfolding and more deleterious effects 39,48 . Activation of GPC-4 mediated IRE1α pathway suggests that GPC-4 alters the normal proteostasis process. Further, in the absence of GPC-4, APOE4 failed to induce tau accumulation and spreading. It supports the idea that APOE4 is neither necessary nor sufficient in development of AD, but potentially other cofactors such as GPC-4 would be needed 49 .
In conclusion, our results reveal that astrocyte secreted GPC-4 drives tau pathology. Moreover, GPC-4 mediates APOE-4 dependent LRP1 trafficking and tau pathology. Further, IRE1α pathway is needed for GPC-4 induced tau pathology. Thereby, our study provides a mechanism which explains why APOE4 carriers are at risk in developing AD pathogenesis and a way to intervene the tau spread.     Immunohistochemistry from the human brain shows that GPC-4 is expressed in GFAP astrocytes.

Protein binding assays
To investigate whether GPC-4 differentially interacts with APOE2 and APOE4, the protein binding assay was performed. The concentration of any indicated protein was at 1 g/l. The final volume was made to five microliters with Tris buffer. The mixture was incubated at a shaker for one hour. The samples were boiled with or without 2-mercaptoethanol and separated by a native gel.

Immunoprecipitation
The total protein was isolated from frozen human brain samples using 1% Triton in Tris buffer.
Briefly, Pierce Direct IP method was used to isolate proteins of interest. The AminoLink Plus Coupling Resin was washed with a coupling buffer, and the resin was incubated with 10g of LRP1 or APOE antibodies for 2 h in a rotator. The unbound antibodies were washed away, and 500 g of isolated proteins was added to resin and incubated overnight, 4 0 C, in a rotator. It was washed three times to remove unbound proteins, and antibody-coupled proteins were eluted with an elution buffer.

Ion exchange chromatography
The method was adapted from previous publication 14 . The DEAE Sepharose was equilibrated with a wash buffer. Ten micrograms of GPC-4 protein were added to the column and incubated for 5 h at 4 0 C. After incubation, the unbound GPC-4 was washed away, 5 times, with a washing buffer.
And then, either APOE2 or APOE4 protein was added and incubated for 1 h. After washing, 5 times, the bound proteins were eluted with NaCl gradients. The excess salt was removed by using 10 kD Amicon Ultra Centrifugal filter tubes.

Electrophoresis and western blot
Samples were lysed with SDS-lysis buffer (Laemmli Sample Buffer, Biorad), boiled at 95 0 C for 5 mins with 50 mM DTT, and separated based on their molecular weights using TGX Stain-Free Gels (Biorad). The separated proteins were transferred to PVDF membrane, blocked with 2% BSA for 1 h, incubated with primary antibodies overnight at 4 0 C. Membranes were washed with PBST 6 x10 mins, incubated with HRP-conjugated secondary antibodies for 1 h, again washed with PBST 6 x10 mins and then developed using ECL solution (Clarity Western ECL substrate, Biorad).

Tau uptake assay
On day 11 of neuronal culture, the cells were treated either with GPC-4 shRNA, LRP1 shRNA or scrambled shRNA. After 24 h of treatment, it was incubated with 5 g/ml APOE4 protein for 24 h as described in the manuscript. Finally, the neurons were incubated with 1 g/ml of human tau protein for 1h and washed with PBS 3 times before proceeding to immunohistochemistry.

Tau seeding assay
The Tau Biosensor (ATCC CRL-3275) cells stably expressing the repeat domain of Tau conjugated yellow fluorescent protein (YFP) were cultured in 24well-coverslips at 37 °C, 5% CO2 in a complete medium. Brain extracts from Alzheimer's disease patients were used. APOE4contaning particles were isolated from APOE44 postmortem human brain. 3% (final volume) Lipofectamine 2000 were mixed with brain extracts alone, brain extracts + isolated APOE4 particles and brain extracts + purified APOE4 protein for one hour at room temperature. It was added to the CRL cells, incubated for two days, fixed with 4%PFA for one hour and YFP signals were imaged at SP5 Leica microscope. The tau seeded cells were blindly counted.

Immunohistochemistry and imaging
The cultured cells containing coverslips were fixed for an hour with 4% paraformaldehyde. After a brief wash with PBS, incubated with 10% Donkey serum in PBST for 30 min. Primary antibodies were added and incubated for 48 h at 4 0 C, washed 3 x 10 mins with PBST, probed with corresponding secondary antibodies for 2 h, then washed again with PBST before counterstaining with DAPI. The tissues were mounted with Aqua-Poly/Mount and imaged using Leica TCS SP5 confocal laser scanning microscope. The following primary antibodies were used: TAU-5 (Thermo Scientific), AT8 (Thermo Scientific), MAP2 (Millipore), LRP1(Abcam), HT-7 (Thermo Scientific) and GFP (Abcam).

Data and statistical analysis
Quantification of western blots and immunocytochemical staining were performed by Image J software. For western blot quantification, the intensity of each immunoreactive band was normalized to the corresponding beta-actin immunoreactive band. Normalized values were grouped and comparted for statistical analysis. To analyze tau uptake (HT-7 staining), we imaged the neurons with 40x objective. Before staining, the coverslips were washed thoroughly with PBS to remove any extracellular tau. For different places/coverslip were imaged. The whole field of imaged area was quantified. .scRNAseq data was analyzed using a pipeline provided by the