The unfolded protein response transcription factor XBP1s ameliorates Alzheimer’s disease by improving synaptic function and proteostasis

Overexpression of XBP1s in the brain reduces amyloid β load in AD mice

To determine the consequences of enforcing adaptive UPR responses in the AD brain, we first used a transgenic mouse model generated by our laboratory that overexpresses XBP1s through the control of the Prion promoter (Tg^XBP1s).³⁷ These animals were crossed with the 5xFAD model, which expresses a combination of five human mutations in APP and presenilin-1 (PSEN1) genes³⁹ (Figure S1A). We confirmed the overexpression of the XBP1s transgene using real-time PCR of cDNA obtained from hippocampus and brain cortex of Tg^XBP1s mice crossed with 5xFAD animals (Figure S1B).

We next determined the effects of overexpressing XBP1s in the progression of key AD features. We evaluated amyloid β deposition in the brain of Tg^XBP1s and 5xFAD animals (termed Tg^XBP1s/5xFAD) of experimental mice at 6 and 8 months of age (Figure 1). Immunohistochemical analysis of amyloid β content using the 4G8 antibody showed a significant reduction in the cerebral cortex of Tg^XBP1s/5xFAD mice compared with 5xFAD littermate control animals (Figure 1A). Similar results were obtained when hippocampal tissue was analyzed (Figure 1D). Overall, Tg^XBP1s/5xFAD mice showed a 30%–40% reduction in amyloid β load in both brain regions compared with 5xFAD at the same age (Figures 1B and 1E). These results were also confirmed by measuring the number of amyloid β plaques per area in both cortical and hippocampal regions. We observed a decrease in the number of amyloid deposits when XBP1s was overexpressed (Figures 1C and 1F). Then, we stained brain tissue with thioflavin S (ThS) to evaluate fibrillar amyloid deposits. Expression of XBP1s also reduced ThS-reactive deposits in the cortex and hippocampus of 5xFAD mice at 8 months old (Figure 1G). We also analyzed the distribution of small and large places after amyloid β staining using the 4G8 antibody followed by image quantification. Analysis of animals at 6 months of age revealed minor changes in the proportion of small and large plaques when XBP1s is overexpressed in the brain (Figure S2B), suggesting a general reduction of amyloid β plaques independent of the size.

Because amyloid β oligomers are proposed as a causative agent of synaptic dysfunction and behavioral impairment in AD,⁴⁰ we monitored the relative levels of soluble and insoluble amyloid β species. We performed serial extractions with detergents followed by amyloid β quantification in cortical and hippocampal brain homogenates using a specific human ELISA to detect the amyloid β42 isoform.²⁵ Tg^XBP1s/5xFAD animals contained lower aggregated/fibrillar (formic acid insoluble) amyloid β42 species in the cortex (Figure 1H). The soluble species showed similar levels following serial extraction in 5xFAD animals (Figure S2).

We then characterized other typical neuropathological alterations of AD in our experimental groups by monitoring astrocyte activation in the brain. Immunofluorescence analysis of GFAP staining (total burden signal by area) indicated that astrogliosis was significantly induced in the cortex and hippocampus of 5xFAD mice at 8 months (Figure 1I). Overexpression of XBP1s attenuated the content of reactive astrocytes in both brain regions analyzed (Figure 1I). These results indicate that XBP1s overexpression reduced two central neuropathological features of experimental AD, amyloid β deposition and gliosis.

XBP1s overexpression in the brain ameliorates cognitive impairment in AD mice

We and others previously reported that XBP1s expression in the brain enhances synaptic function and improves learning and memory.^35,36,37 To monitor the possible effects of overexpressing XBP1s on the behavioral performance of 5xFAD, we evaluated spatial memory acquisition using the Barnes maze (BM) test by measuring the total latency during the training phase (Figure 2A). In line with previous studies,^41,42 5xFAD animals spent significantly less time than non-transgenic mice in the target quadrant (Figure 2B). Remarkably, Tg^XBP1s/5xFAD mice increased total time spent in the target quadrant compared with 5xFAD control animals, showing significant prevention of the memory impairment (Figure 2B). The 5xFAD model may show differential effects depending on the gender.⁴³ As our experimental cohorts included male and female animals, we reanalyzed data to determine if XBP1s overexpression has preferential beneficial effects. We assessed a selected pool of experiments of animals at 8 months of age, including histological analysis of amyloid beta burden (Figure S2C), in addition to cognitive assessment using Barnes maze (Figure S2D). Our results indicate that gender did not influence the modification of AD features when XBP1s was overexpressed.

Next, to determine whether the expression of XBP1s could prevent synaptic dysfunction in the context of experimental AD, we measured long-term potentiation (LTP) in the hippocampus, a long-lasting form of synaptic plasticity. We recorded glutamatergic transmission evoked by Schaffer’s collaterals stimulation to monitor field excitatory postsynaptic potentials (fEPSPs) in the CA1 region.⁴⁴ We confirmed a drastic decrease in the slope of fEPSP in brain tissue of 5xFAD (Figure 2C). Strikingly, overexpression of XBP1s in the nervous system prevented LTP impairment in 5xFAD mice and significantly improved it compared with wild-type (WT) (Figure 2D). Thus, the beneficial effects provided by XBP1s to the cognitive function of AD mice correlated with prevention of synaptic dysfunction at the electrophysiological level.

To validate the possible protective effects of XBP1s expression on AD, we used an alternative disease model based on intracerebroventricular infusion of amyloid β oligomers (AβOs).^45,46 AβOs were prepared weekly from synthetic amyloid β_1–42 and revealed a mixture of low- and high-molecular weight amyloid β species, characterized by size-exclusion chromatography (SEC-HPLC) (Figure 2E). Seven days after the injection of AβOs, we assessed cognitive alterations using the novel object recognition test, a task previously shown to be impaired in wild-type mice injected with AβO preparations.^45,47 We assessed cognitive impairments in wild-type and Tg^XBP1s animals after injection of AβOs compared with control mice injected with a saline solution (Figure 2F). Strikingly, mice overexpressing XBP1s in the brain were fully protected against the adverse effects of AβO exposure, showing a behavioral performance similar to vehicle-injected control animals (Figure 2F). Overall, these results indicate that XBP1s overexpression significantly restores synaptic function and spatial memory acquisition in mouse models of AD.

Gene therapy to deliver XBP1s into the hippocampus of AD mice restores cognitive function

To evaluate the therapeutic potential of XBP1s overexpression with a strategy that can be translated into the clinic, we locally targeted the hippocampus using brain stereotaxis to deliver adenovirus-associated viruses (AAV; serotype 2) expressing XBP1s.³⁷ These viral particles also expressed EGFP for detection. As control we used a vector containing a stuffer sequence instead of the XBP1 cDNA, and the EGFP cassette for detection (AAV-Mock). 5xFAD mice were injected with AAV-XBP1s or AAV-Mock at 6 weeks and then evaluated for behavioral performance 7–9 months later. Remarkably, local administration of AAV-XBP1s into the hippocampus of 5xFAD mice improved the performance in learning and memory tasks as assessed using the Morris water maze (MWM) (Figure 3A). Similar results were obtained when memory was measured in the Barnes maze test (Figure 3B). Importantly, analysis of the latency to the target hole during the test day also indicated that the performance of 5xFAD mice injected with AAV-XBP1s was equivalent to non-transgenic animals (Figure 3C). We confirmed an improvement of cognition by measuring spatial working memory in the Y maze, observing that the bilateral injection of AAV-XBP1s into the hippocampus improved the performance of 5xFAD mice (Figure 3D). Thus, the local delivery of XBP1s into the hippocampus prevents the progression of AD-related cognitive decline.

AAV-XBP1s administration into the hippocampus of AD mice restores synaptic plasticity

As most cognitive paradigms employed here are dependent on the normal function of the hippocampus, we evaluated the distribution of dendritic spines in CA1 pyramidal cells because their density correlates with the learning capacity.⁴⁸ We injected animals with either AAV-XBP1/EFGP or AAV-Mock/EGFP to assess the morphology and content of dendritic spines using EFGP fluorescence and confocal microscopy analysis. 5xFAD mice showed reduced dendritic spines per length (Figure 4A, left panel). Consistent with our behavioral data, delivery of AAV-XBP1s into the hippocampus restored the content of dendritic spines to numbers almost comparable with littermate wild-type animals (Figure 4A, right panel). Next, we performed electrophysiological studies in isolated hippocampal tissue from 5xFAD mice treated with AAV-XBP1s. Administration of AAV-XBP1 into the hippocampus of 5xFAD mice significantly improved fEPSP slopes (Figure 4B). Virtually identical results were obtained when the magnitude of LTP was quantified (Figure 4C), suggesting that the local expression of XBP1s in the hippocampus strongly improves synaptic plasticity in 5xFAD animals.

To further evaluate the consequences of ectopically expressing XBP1s on AD mice, we measured amyloid deposition after 9 months of injection of the AAV particles. Although the local injection of AAVs only transduced a portion of the hippocampus, we could detect a slight but significant reduction in amyloid β deposition and plaque number in animals receiving the AAV-XBP1s vector (Figure 4D), suggesting improved proteostasis.

Finally, we injected a cohort of wild-type animals to assess the possible toxic effects of overexpressing XBP1s in the hippocampus after a long-term administration of AAV particles. Remarkably, we could confirm the expression of the Xbp1s transgene 12 months after injection, with no overall neurotoxicity (Figures S3A and S3B). In addition, no signs of ER stress were observed in these brain samples when we measured the mRNA levels of the PERK/ATF4 target genes Bip and Chop by real-time PCR (XBP1-independent targets) in wild-type animals (Figure S3C).

Altogether, these results indicate that local administration of AAV-XBP1s into the hippocampus is safe and effective in restoring synaptic plasticity and cognitive function in an animal model of AD.

XBP1s corrects the proteomic alterations observed in the brain of 5xFAD mice

Because XBP1s-target genes are functionally diverse, this UPR transcription factor might attenuate the progression of experimental AD by multiple mechanisms. For example, XBP1s may improve ER proteostasis by regulating canonical UPR targets, by regulating synaptic function, or by modifying other biological functions previously linked to XBP1s (i.e., energy and lipid metabolism, cell differentiation, inflammation, etc.^12,14,49). To determine the major gene expression alterations in the hippocampus of AD mice and how XBP1s may modify them, we performed quantitative proteomics of dissected hippocampus to compare tissues derived from Tg^XBP1s, 5xFAD, Tg^XBP1s/5xFAD, and littermate control animals (WT) (Figure 5A). Interestingly, overexpression of XBP1s induced different proteomic changes in WT mice and the AD model (r = −0.02, p = 0.34), indicating that the disease context influences the consequences of XBP1s overexpression on gene expression. Notably, 76% of the alterations observed in the 5xFAD model were corrected when XBP1s was overexpressed in the disease model (r = −0.71, p = 1.61e-217), suggesting global benefits at the proteomic level in the hippocampus (Figure 5A).

To identify processes targeted by XBP1 in the AD model, we performed gene set enrichment analysis on proteins altered by 5xFAD mice and corrected by XBP1s overexpression. Thus, we searched for biological processes significantly down-regulated in the disease model and up-regulated after XBP1s overexpression or vice versa (Figure 5B). In both cases, we found a statistically significant overlap between the pathophysiological processes altered in AD mice that are modified in opposite directions by XBP1s overexpression in the disease context. Pathways associated with ion transport and cell homeostasis showed upregulation in the AD model and were reversed upon XBP1s overexpression. At the same time, components involved in actin and the cytoskeleton regulation were down-regulated in the disease model and reversed by XBP1s (Figures 5B, S4A, and S4B).

Because the major protective effects of XBP1s overexpression in 5xFAD model mice involved improvements in synaptic plasticity and memory performance, we decided to focus our proteomic validation analysis on assessing the impact on actin cytoskeleton function. The regulation of actin cytoskeleton dynamics is essential to sustain neuronal connectivity, synaptic plasticity, dendritic spine formation, and general brain function.^41,42 We identified 23 genes enriched in more than half of the top 10 biological processes detected (Figures S4A and S4B), among which we identified cofilin-1 (Cfl1) as a strong hit, a protein previously associated with neurodegenerative diseases.⁵⁰ We confirmed the downregulation of cofilin-1 in hippocampal brain extracts derived from in 5xFAD animals using western blot analysis, a phenomenon restored in Tg^XBP1s/5xFAD animals (Figure 5C), validating a major hit of our proteomic analysis. We also identified 7 key proteins involved in processes linked with ion transport that increased abundance in the disease model but decreased their levels after XBP1s overexpression (Figures S4C and S4D). In addition, we analyzed the promoters of 13 of the genes identified in our analysis that are linked to cytoskeleton function and searched for DNA regulatory elements in the proximal promoter regions that may be controlled by XBP1 including ERSE, ERSEII, and UPRE sequences (Figure S5A). All the genes contained different number of putative XBP1 binding sites, and the cofilin-1 gene was among the top 3 candidates with more DNA binding sites (Figure S5B).

To increase the relevance of our findings to human aging and AD, we assessed the levels of the identified genes in publicly available proteomics datasets (Figure 5D; Table S1). We selected 30 proteins that are altered in our AD model and that are corrected by XBP1 overexpression. On average, proteins with decreased abundance in the disease model and restored by XBP1s overexpression also decreased during normal human aging and also in tissues derived from AD patients (Figure 5D). Similarly, increased abundance was observed in AD patients and people aged 70 years or older in genes up-regulated in 5xFAD animals and down-regulated in Tg^XBP1s/5xFAD mice. A similar consistency was observed with a previous proteomics dataset of the 5xFAD disease model.⁵¹ To calculate the global significance of these trends, we pooled the fold change values of these proteins in all the external datasets analyzed (Figure 5D). Overall, we observed that during human aging and AD, the proteins changed in the same direction as in our AD model (p for deviation from zero < 0.05), suggesting that proteins altered in the disease model and restored by XBP1s are also affected during human aging and in the brain of AD patients.

In addition, we assessed the possible correlation between the levels of XBP1s and cofilin-1 in gene expression datasets of AD subjects. Unfortunately, in any of the proteomic datasets available the abundance of Xbp1s was measured, possibly because it is a low abundant protein. Alternatively, we analyzed gene expression datasets of AD in humans from Noori et al.⁵² From the 43 datasets analyzed, 39 measured the mRNA levels of XBP1. We were able to detect a slight positive correlation between XBP1 expression and cofilin-1 (Figure S6). Overall, our proteomic analysis indicates that XBP1s has important effects at the global gene expression level, and these changes might be relevant for AD in humans.

Generation of a transgenic mouse model of AD overexpressing XBP1s in the nervous system

XBP1s overexpressing mice (Tg^XBP1s) were described previously.³⁷ These animals were crossed with the AD model 5xFAD animals (formerly JAX stock no. 008730). 5xFAD mice overexpress two transgenes, which have three mutations in the human APP gene (Swedish, Florida, and London familial Alzheimer’s disease), and two in the human presenilin-1 gene (M146L+L286V). These animals begin to accumulate amyloid β aggregates at 2 months. Increased senile plaques, synaptic degeneration, gliosis, and cognitive impairment, are observed between 4 and 5 months of age.^39,71 Animals were housed in groups of a maximum of five in individually ventilated cages under standard conditions (22°C, 12 h light-dark cycle), receiving food and water ad libitum. All animal manipulations were carried out according to standard regulations and approved by the Animal Welfare Committee of the Faculty of Medicine of the Universidad de Chile. Mice were euthanized by CO₂ inhalation, and all efforts were made to minimize suffering. The left hemispheres were frozen at −80°C for biochemical analyses, whereas the right hemispheres were stored for histological studies.

Behavioral studies

Cognitive impairment was measured using different behavioral tests, including the Barnes maze, Morris water maze, and Y maze (YM). Briefly, the BM test consists of a circular surface containing 21 holes. Among them, only one is equipped with a hiding chamber. Mice were placed on the maze surface and were allowed to explore it for 1.5 min. Mice were stimulated with sound to stress them in looking for the hiding chamber. Room environments had spatial cues for mouse orientation. This training procedure was performed four times a day for four consecutive days per animal. On the fifth day, memory was evaluated by measuring the time each animal took to find the chamber, the time spent in the target quadrant, and the number of errors made by each animal before arriving at the objective. Errors and time to the objective were measured when the animal’s nose was entirely inside one hole. On the 5th day, short-term memory (STM) was measured. On day 12, long-term memory (LTM) was assessed. The MWM was performed as previously described.²⁵ In this assessment, animals learn to swim underwater to a hidden platform. Mice at 6 or 9–10 months of age were placed in the pool and allowed to explore it for 1 min. The testing room was equipped with spatial cues for orientation. This training procedure was performed six times a daily for four consecutive days per animal. On day 5, the platform was removed from the pool to measure the animals’ time spent in the target quadrant until the animal found the platform. Additionally, each animal’s time spent finding the platform was measured during the training phase. Learning performance was measured as total latency on the fifth day. The spatial working memory was analyzed using a Y maze protocol. The Y maze array consists of three arms (A, B, and C) that radiate from the center in the shape of Y. The behavior task was initiated by placing the mouse in the center of the Y, which allows free access to three arms. The animal behavior was recorded for 8 min, and the number of entries in each arm and the sequence of entrances were measured. A “spontaneous alternation” is defined as a set of three consecutive arm choices without a repeated entry (e.g., ABC, BCA, CBA). A spontaneous alternation score was calculated using the formula number of alternances/(number of total entries − 2) × 100. All behavior analysis were performed using commercial software for video tracking Any-Maze.

Electrophysiological analysis

Hippocampal slices were prepared as previously reported.^25,37 Six- to 9-month-old mice were deeply anesthetized with isoflurane, and their brains were quickly removed. Slices (350 μm) were dissected in an ice-cold dissection buffer using a vibratome (Vibratome 1000 plus; Ted Pella Inc.). Synaptic responses were evoked by stimulating the Schaffer collaterals with 0.2 ms pulses delivered through concentric bipolar stimulating electrodes and recorded extracellularly in the stratum radiatum of the CA1 subfield as described.^25,37 Long-term potentiation was induced by four-theta burst stimulation (10 trains of four pulses at 100 Hz; 5 Hz inter-burst interval) delivered at 0.1 Hz. LTP magnitude was calculated as the average (normalized to baseline) of the responses recorded 50–60 min after conditioning stimulation.

Histological analysis

Fixed brains were collected in serial coronal sections either on a freezing cryostat or embedded in paraffin and processed on a microtome at 25- or 12-μm-thick 12 serial slices (10 sections/stain/animal) from lambda 0 to lambda −4 mm, respectively. For the paraffin sections, a dehydration process was done after the staining. After formic acid-induced epitope retrieval, primary antibody 4G8 was incubated overnight at a 1:1,000 dilution at room temperature (RT) (BioLegend, San Diego, CA). Horseradish peroxidase (HRP)-linked secondary goat anti-mouse antibody at a 1:1,000 dilution (Invitrogen, Carlsbad, CA) was incubated for 2 h at RT. The peroxidase reaction was visualized using DAB Kit (Vector) following the manufacturer’s instructions. Finally, sections were dehydrated in ascendant ethanol, cleared in xylene, and coverslipped with DPX mounting medium (Innogenex, San Ramon, CA). For fibrillar Aβ quantification, sections were incubated in thioflavin S solution (0.025% in 50% ethanol) for 10 min after defrosting. Sections were coverslipped with DPX mounting medium (Innogenex, San Ramon, CA). Astrogliosis was visualized after staining with rabbit monoclonal anti-glial fibrillary acidic protein (GFAP) antibody at a 1:1,000 dilution (Abcam, Cambridge, MA). All samples were analyzed on an inverted epifluorescent microscope (Olympus IX71), and quantification analysis was performed using the ImageJ software and quantifying the area of fluorescent signals per area (three pictures per animal covering brain cortex and two pictures for hippocampus), subtracting background signals.

Small and large amyloid β plaques were analyzed using the ImageJ analyze particles tool. First, the quantification area was manually selected from multiple images of the same size. Once the area was selected, a mask was applied, and particles between a minimum of 0.0005 in² and a maximum of infinite, to avoid default pixels and be able to count the total plaques. Plaques larger than 0.004 in² were considered large and counted following the same procedure. With this information, the small plaques are obtained by subtracting the large plaques from the total. To verify that all images were quantified properly, a mask picture was saved, and the selection of small/large plaques was visually inspected at the time of quantification.

ELISA quantification of amyloid β species

The dissected brain was separated into cortical and hippocampal areas to generate the brain homogenates (BH). To measure Aβ₄₂ levels, the BH was processed using a previously described serial extraction protocol.^25,72,73 10% BH (cortex and hippocampus) were centrifuged in L100K ultracentrifuge tubes (Beckman-Coulter, Brea, CA) at 32,600 rpm for 1 h at 4°C in a 42.2 Ti rotor. Supernatants were collected, and pellets were resuspended in 70% formic acid (Thermo Fisher Scientific, Waltham, MA). Then, samples were centrifuged for 30 min, and supernatants were collected. To neutralize the samples, formic acid fractions were diluted on 1 M Tris Buffer pH 11 (Sigma-Aldrich, St. Louis, MO) 20 folds to neutralize the samples. ELISA was used to measure the levels of Aβ₄₂ in the brain (kit KHB3442; Invitrogen) following the manufacturer’s instructions. Finally, samples were measured using an ELISA reader (EL800 BIO-TEK; BioTek, Winooski, VT) at 450 nm.

Amyloid β oligomer preparations

Amyloid β oligomers (Aβo) were prepared weekly from synthetic Aβ_1–42 (California Peptide, Salt Lake City, UT), and were routinely characterized by size-exclusion chromatography under non-denaturing conditions and, occasionally, by western immunoblots and transmission electron microscopy, as previously described.^74,75 Briefly, the peptide was dissolved in HFIP to 1 mM and stored as a dried film at −80°C after solvent evaporation. The film was resuspended in DMSO to a final concentration of 5 mM and thoroughly vortexed. The solution was then diluted in ice-cold PBS to 100 mM and left at 4°C overnight. The solution was centrifuged at 14,000 × g for 10 min at 4°C to remove insoluble aggregates (protofibrils and fibrils), and the supernatant containing Aβo was collected. Protein concentration was determined using the BCA assay (Thermo-Pierce). Intracerebroventricular (i.c.v.) infusion of Aβo in mice was performed as described.^45,47,76

AAVs and stereotaxic injections

The generation of AAV-XBP1s and AAV-Mock was described before^77,78,79 In brief, the whole murine Xbp1s expression cassette was excised from pcDNA3-XBP1s as a MfeI/SphI fragment and inserted into a previral plasmid pAAVsp70 containing AAV inverted terminal repeats (ITRs). The Xbp1s cDNA is controlled by the CMV promoter. The vector carries an EGFP expression cassette controlled by the EF1a promoter that serves as a fluorescent marker to report transduced cells. This vector is referred to as AAV-XBP1s, whereas the control vector contained a stuffer DNA sequence, in addition to the EGFP cassette (AAV-Mock). Recombinant AAV-XBP1s (serotype 2/2) was produced by triple transfection of HEK293 cells using a rep/cap plasmid and pHelper (Stratagene, La Jolla CA) and purified by column affinity chromatography, as previously described⁷⁷ To obtain pure and concentrated AAV particles, cell lysates of infected HEK293T cells were treated with trypsin and nuclease followed by ion-exchange chromatography using ceramic hydroxyapatite and DEAE-Sepharose in combination with Cellufine sulfate affinity chromatography. Viral titers were determined by real-time TaqMan PCR assay with specific primers for the bovine growth hormone (BGH) poly-adenylation sequence, which is present in both plasmids.

Forty-five-day-old 5xFAD mice were anesthetized using isoflurane and fixed to a mouse stereotaxic frame (David Kopf Instruments). Bilateral injections of 2.5 μL of AAV-XBP1s or AAV-Mock virus were performed at a single point in the hippocampal region using a 5 μL Hamilton syringe (Hamilton) using the following coordinates: AP, −1.8 mm; ML, 1.8 mm y; DV, −1.8 mm. The injection was conducted at a rate of 0.5 μL/min, and the needle was left in place for 5 min before retraction.

Dendritic spine imaging

Brain slices were cut at 25 mm thickness in a cryostat. AAV-GFP fluorescence was previously confirmed in injected animals to validate viral transduction in an inverted fluorescence microscope and then imaged in a confocal microscope Nikon Eclipse T1 at 60x magnification with an additional digital zoom of 3×. Similar regions were compared within each animal (CA1 region, spines in primary and secondary dendrites between the stratum radiata and the pyramidal layer; AP, 1.9–2.1 from the bregma). Z stacks were acquired in 0.5 mm slices, laser intensity at 0.5–1%, and 12.5us/pixel at 1,024 × 1,024 resolution. Z stacks were then summed using ImageJ software with total maximum intensity to generate one stacked 8-bit image. The number of spines was manually quantified in scaled images and divided by the dendrite length analyzed.

Real-time PCR analysis

Total tissue RNA was extracted using Trizol Reagent (Invitrogen) according to the manufacturer’s instruction. cDNA was synthesized with random primers using a High Capacity cDNA Reverse Transcription KIT (Applied Biosystems) and subsequently subjected to quantitative PCR analysis using HOT FIREPol EvaGreen qPCR Mix plus (ROX) (Solis Bio Dyne) on a Stratagene Mx3000P machine (Agilent Technologies). Actin mRNA expression was used to normalize all samples. The sequences of primers used were as follows:

Quantitative proteomic analysis

Hippocampal tissue of WT, Tg^XBP1s, 5xFAD, and Tg^XBP1s/5xFAD mice were homogenized in TEN buffers as described above. 20 μg of lysate was precipitated with chloroform/methanol for each sample. Samples for mass spectrometry analysis were prepared as described.⁸⁰ Air-dried pellets were resuspended in 1% RapiGest SF (Waters) and diluted to final volume in 100 mM HEPES (pH 8.0). Proteins were reduced with 5 mM Tris (2-carboxyethyl) phosphine hydrochloride (Thermo Fisher Scientific) for 30 min and alkylated with 10 mM iodoacetamide (Sigma-Aldrich) for 30 min at room temperature in the dark. Proteins were digested for 18 h at 37°C with 0.5 μg trypsin (Promega). After digestion, the peptides from each sample were reacted for 1 h with the appropriate tandem mass tag (TMT) isobaric reagent (Thermo Fisher Scientific) in 40% (v/v) anhydrous acetonitrile and quenched with 0.4% ammonium bicarbonate for 1 h. Samples with different TMT labels were pooled and acidified with 5% formic acid. Acetonitrile was evaporated on a SpeedVac, and debris was removed by centrifugation for 30 min at 18,000g. MudPIT microcolumns were prepared as described.⁸¹ LC-MS/MS analysis was performed using a Q-Exactive HF mass spectrometer equipped with an Ultimate 3000 nLC 1000 (Thermo Fisher). MudPIT experiments were performed by 10 μL sequential injections of 0%, 10%, 20%, 30%,…, 100% buffer C (500 mM ammonium acetate in buffer A) and a final step of 90% buffer C/10% buffer B (100% acetonitrile, 0.1% formic acid, v/v/v) and each step followed by a gradient from buffer A (95% water, 5% acetonitrile, 0.1% formic acid) to buffer B. Electrospray was performed directly from the analytical column by applying a voltage of 2 kV with an inlet capillary temperature of 275°C. Data-dependent acquisition of MS/MS spectra was performed with the following settings: eluted peptides were scanned from 300 to 1800 m/z with a resolution of 120,000. The top 15 peaks for each full scan were fragmented by HCD using a normalized collision energy of 30%, isolation window of 2.0 m/z, a resolution of 30,000, ACG target 1e5, maximum IT 60 ms, and scanned from 100 to 1800 m/z. Dynamic exclusion was set to 10 s. Proteome Discoverer 2.2 (Thermo Fisher Scientific) performed peptide identification and protein quantification. Spectra were searched using SEQUEST against an UniProt mouse proteome database. The database was curated to remove redundant protein and splice-isoforms, and common contaminants were added. Searches were carried out using a decoy database of reversed peptide sequences using Percolator node for filtering and the following settings: 50 ppm peptide precursor tolerance, 6 amino acid minimum peptide length, trypsin cleavage (unlimited missed cleavage events), static Cys modification of 57.0215 (carbamidomethylation), and static N-terminal and Lys modification of 229.1629 (TMT sixplex), FDR 0.01, 2 peptide IDs per protein. TMT reporter ion intensities were normalized based on total peptide abundance in each channel. Subsequently, a common pooled sample calculated TMT ratios for each identified protein. Finally, the reference-normalized TMT intensities were compared between WT (n = 5), Tg^XBP1s (n = 5), 5xFAD (n = 5) and Tg^XBP1s/5xFAD (n = 5). Statistical significance was assessed by a two-tailed unpaired t test using the FDR approach ⁵⁰ and Q = 1% in GraphPad Prism. Enrichment for Gene Ontology (GO) terms was tested in EnrichR.^82,83

To identify changes in the proteome we fitted a linear model of proteomic levels versus genotype and the samples run for each protein and calculated the summary statistics using a moderated t test implemented in the R package limma.⁸² Correlations between results of the comparative proteomics analysis were performed using Spearman’s method on significance scores calculated as the sign of the fold changes (1 for positive values and −1 for negative values) multiplied by the logarithm of the p values. The gene set enrichment analysis was performed using proteins ranked by significance scores as input for the function gseGO from the R package ClusterProfiler.⁸⁴ p values for the enriched GO terms were adjusted for multiple testing using the Benjamini-Hochberg method.⁸⁵ The statistical significance of the overlaps was calculated using a Fisher’s Exact Test.⁸⁶ To determine the genes leading the top enriched processes, we performed a leading-edge analysis using the R package fgsea.^57,87 To search proteomics datasets of the 5xFAD mouse model, AD, and human aging, we used the PRIDE database (see raw values in Table S1).^58,88 We normalized the log₂ fold changes using quantile normalization to compare the effect sizes across datasets.

Bioinformatic analysis

Mouse genes were converted into human orthologues using WORMHOLE (software in: https://wormhole.jax.org/). The comparison between the levels of XBP1 and the proteins was performed using a Pearson correlation between the log₂ fold changes of our experiment and the observed between AD patients and controls. Pre-processed gene expression data were obtained from Noori et al. Data base in: https://www.sciencedirect.com/science/article/pii/S0969996120305003).

To study the presence of putative XBP1s binding sites on the proximal promoter regions of the protein hits identified in our proteomic analysis, we selected the 10 actin-related biological processes obtaining 13 main hits. Each gene was then analyzed in the Eucaryotic Promoter Database (link: https://epd.epfl.ch//index.php) in a range of −1,600 to +200 nt. It should be noted that the specific sequences of XBP1s in the cis-regulatory elements ERSE and ERSEII are CCACG and UPRE CGTGG⁸⁹ (Sung-Min 2021) and at the same time alternative sequences mentioned in previous articles were analyzed^19,89 (Figure S5A).

Statistical analyses

Data are expressed as mean ± SEM. After confirming normal distribution with the skewness/kurtosis statistic test, Student’s t test was used to analyze differences in histological and biochemical analysis of Aβ and APP. For behavioral studies, one-way ANOVA followed by Tukey’s multiple post-test or Newman-Keuls multiple post-test measured significant differences. Two-way ANOVA followed by Bonferroni post-test was used in electrophysiological and behavioral experiments. Statistical analyses were performed using Graph Pad Prism 5.0 software. Statistical differences were considered significant for values of p <0.05.