Control of mammalian brain ag e ing by the unfolded protein response transcription factor XBP1

: Brain ageing is the main risk factor to develop dementia and neurodegenerative diseases, associated with a decay in the buffering capacity of the proteostasis network. We investigated the 5 significance of the unfolded protein response (UPR), a major signaling pathway to cope with ER stress, to the functional deterioration of the brain during aging. Genetic disruption of the ER stress sensor IRE1 α accelerated cognitive and motor decline during ageing. Exogenous bolstering of the UPR by overexpressing an active form of the UPR transcription factor XBP1 restored synaptic and cognitive function, in addition to reducing cell senescence. Proteomic 10 profiling of hippocampal tissue indicated that XBP1s expression attenuated age-related alterations to synaptic function and pathways linked to neurodegenerative diseases. Overall, our results demonstrate that strategies to manipulate the UPR in mammals may sustain healthy brain ageing. incubated for Images were taken with Leica confocal with a 40x objective magnification. ImageJ and LAS X software were used to process the stacked images. The percentage of positive cells for p-H2AX was 5 graphed. Representative images are shown.


Abstract:
Brain ageing is the main risk factor to develop dementia and neurodegenerative diseases, associated with a decay in the buffering capacity of the proteostasis network. We investigated the 5 significance of the unfolded protein response (UPR), a major signaling pathway to cope with ER stress, to the functional deterioration of the brain during aging. Genetic disruption of the ER stress sensor IRE1α accelerated cognitive and motor decline during ageing. Exogenous bolstering of the UPR by overexpressing an active form of the UPR transcription factor XBP1 restored synaptic and cognitive function, in addition to reducing cell senescence. Proteomic 10 profiling of hippocampal tissue indicated that XBP1s expression attenuated age-related alterations to synaptic function and pathways linked to neurodegenerative diseases. Overall, our results demonstrate that strategies to manipulate the UPR in mammals may sustain healthy brain ageing.

Main text:
Normal ageing is associated with progressive cognitive impairment, representing the most prevalent risk factor for the development of dementia and neurodegenerative disorders.
Decades of research have defined the hallmarks of ageing, underscoring the biological processes that determine when and how organisms age, thus regulating healthspan and lifespan 1 . Subtle 20 structural and functional alterations in synapses are the main drivers of age-related cognitive decline 2 , but the molecular mechanisms dictating these perturbations are still elusive.
Proteostasis (homeostasis of proteins) is maintained by the dynamic integration of pathways that mediate the synthesis, folding, degradation, quality control, trafficking and targeting of proteins, and its disturbance has been posited as a pillar of the ageing process 3 . The complexity of 25 synaptic architecture and its dynamic regulation highlight the need to maintain the integrity of proteostasis at the level of the secretory pathway during the organismal lifespan to sustain normal brain function 4 . Studies in simple model organisms demonstrated that the activity of a central node of the proteostasis network, the unfolded protein response (UPR), declines with ageing 5 and that enhancing the expression of the UPR transcription factor X-BOX binding 30 protein-1 (XBP1s) in neurons extends health and lifespan in C. elegans 6 . XBP1 is a master regulator of the UPR, an adaptive pathway that mediates proteostatic recovery in cells suffering endoplasmic reticulum (ER) stress 7 . XBP1 is activated by the stress sensor inositol-requiring enzyme-1 alpha (IRE1), an ER-located RNase that catalyzes the unconventional splicing of the XBP1 mRNA to eliminate a 26-nucleotide intron 8 . This processing event shifts the coding 5 reading frame to generate an active and stable transcription factor termed XBP1s (for the spliced form) that controls the expression of various components of the proteostasis network [7][8] . In the brain, XBP1 has additional functions in the regulation of synaptic plasticity, neuronal differentiation, learning and memory 9 and global metabolic control 7 Studies in C. elegans indicate that the activity of the UPR in neurons operates as a central regulator of organismal 10 proteostasis during aging through a cell-nonautonomous mechanism that tunes proteostasis in the periphery to extend lifespan. [10][11][12][13] To date, this concept has not been validated in mammals, although correlative studies indicate altered UPR signaling in aged human tissue [14][15][16] .
To determine the possible contribution of the UPR to age-associated cognitive decline in humans, we analyzed RNA sequencing data from the hippocampus of elderly subjects diagnosed 15 or not with dementia in a unique aged population-based cohort from the Adult Changes in Thought (ACT) study (http://aging.brain-map.org/). Remarkably, unbiased functional enrichment analysis of most altered transcripts in demented patients highlighted unfolded protein binding and processing as the most enriched altered biological functions (Fig. 1a). Moreover, XBP1 was predicted as a possible transcription factors driving such alterations to the proteostasis network 20 ( Fig. 1a, Table S1). In addition, chaperones of the heat shock family accounted as the most altered genes in demented patients (Fig. 1b), while some classical UPR mediators were not altered (Extended Data Fig. 1a).
To determine the capacity of the ageing mammalian brain to engage the UPR, we intraperitoneally injected mice of different ages with tunicamycin, a well-established 25 pharmacological inducer of ER stress 17 . Next, we evaluated UPR transcriptional responses by measuring the mRNA levels of Xbp1s, Hspa5 (Bip/Grp78), Ddit3 (Chop) and Atf3 in the hippocampus, in addition to cerebral cortex and cerebellum. Remarkably, the capacity to induce Xbp1s was reduced in hippocampus in middle age and aged animals following experimental ER stress although no alteration in basal levels was found (Fig. 1c-

e). UPR mediators Ddit3 and 30
Hspa5 showed the same trend in this brain region (Fig. 1c,). Similar findings were obtained in the brain cortex of the same animals (Extended Data Fig. 1b), whereas no differences were detected in the cerebellum (Extended Data Fig. 1c). Increased generation of reactive oxygen and nitrogen species, such as nitric oxide (NO), has been observed during ageing 18 . Previous reports indicated that S-nitrosylation of IRE1α, resulting from posttranslational modification of cysteine thiol (or more properly thiolate anion) groups at Cys931 and Cys951 by NO-related species, 5 inhibits its ribonuclease activity 19 . Accordingly, we evaluated the levels of S-nitrosylation of IRE1α (abbreviated SNO-IRE1α) during normal brain ageing. We measured the ratio of SNO-IRE1α to input IRE1α by biotin-switch assay in the brains of young and old animals and found a significant increase in this ratio in aged samples (Fig. 1f), potentially accounting, at least in part, for the decrease in IRE1α activity and thus compromised ability to adapt to ER stress with age. 10 Notably, when we also performed standard immunoblots of IRE1α to carefully quantify by levels by densitometry in young and aged animals, we found that total IRE1α also decreased with ageing ( Fig. 1g), as suggested previously based on mRNA levels 20 . Thus, while the absolute level of S-nitrosylated IRE1α may not be that different in young vs. aged brain, the proportion of SNO-IRE1α is significantly greater (Fig. 1f), reflecting overall inhibition of enzyme activity. 15 These results suggest that the occurrence of specific molecular alterations in the ageing hippocampus interfere with the capacity to adapt to ER stress.
In order to evaluate and quantify cognitive and motor decline associated with normal ageing in rodents, we implemented a battery of tasks to assess the behavior of young, middleaged, and aged wild type mice (Extended Data Fig.2a,e,h,i,j,k). Additionally, we measured 20 compound muscular action potentials (CMAPs) in three muscles during ageing (Extended Data Fig.1). We detected spontaneous decline in the performance of animals starting at middle age and progressing thereafter using distinct cognitive and motor evaluations (Extended Data Fig.1), some of which were previously reported 21 . Since most of the cognitive tasks implemented here are dependent on normal hippocampal function, we also evaluated the electrophysiology of 25 hippocampal slices derived of young or aged animals (Extended Data Fig.1m). Firing rates of CA1 neurons were recorded during spontaneous activity or following picrotoxin (PTX) treatment, an antagonist of GABAergic inhibitory interneurons that fosters excitatory activity in hippocampal circuits (Extended Data Fig. 2m). Despite no significant alterations at basal level, a decline in the firing rates of CA1 neurons of aged mice following PTX treatment was detected, 30 indicative of functional ablation. In line with those results, density of dendritic spines in CA1 pyramidal cells was also significantly diminished in aged animals when compared to middle aged or young mice (Extended Data Fig. 2n).
To assess the significance of the UPR to brain health span, we conditionally ablated the RNase domain of IRE1α in the nervous system using CRE transgenic lines driven by the Nestin promoter (IRE1 cKO ) for general deletion in the brain 22 or by the Camk2a promoter (IRE1 cKO/CaK ) 5 (Extended Data Fig. 2a), restricting the deletion to specific neuronal populations. Disruption of the IRE1α pathway in the central nervous system resulted in reduced performance in agesensitive cognitive tests, including new object recognition (NOR) and contextual fear conditioning (CFC) (Fig. 2a, 2b). IRE1α deletion using the Camk2a-CRE line also resulted in similar effects (Extended Data Fig.2c). Importantly, aged mice interacted with objects for the Remarkably, only aged IRE1 cKO animals presented a significant decay in spatial memory acquisition when tested in the Barnes maze, reflected in a higher percentage of errors (Fig. 2c), although the latency to find the targets did not differ between genotypes (Extended Data Fig.3b). 15 Notably, genetic disruption of IRE1α function did not alter the cognitive performance of young animals, indicating the occurrence of age-dependent phenotypes ( Fig. 2a- 6 , we further confirmed our results in XBP1 conditional knockout animals, which have normal IRE1α expression, but lack XBP1 expression in the hippocampus, and observed reduced 20 performance in the NOR assay in middle aged but not in young animals (Fig. 2d).
We next targeted the UPR in the brain of adult animals via local delivery of CRE recombinase into the hippocampus of IRE1α floxed animals using adeno-associated viruses (AAVs) (Fig. 2e and Extended Data Fig. 3d). Middle-aged mice were tested in the NOR assay prior to AAV-CRE injection and then monitored again following 4-weeks of brain surgery. 25 Remarkably, targeting IRE1α in the hippocampus of middle-aged mice impaired the capacity to discriminate novel objects when compared with empty AAV (Fig. 2f), correlating with reduced density of dendritic spines in the CA1 region (Fig. 2g). Importantly, injection of AAV-CRE into the brain of young IRE1 flox/flox animals did not alter NOR performance or the distribution of dendritic spines (Fig. 2f-g), confirming the occurrence of age-related impairment. We previously 30 reported that XBP1 regulates the expression of brain-derived neurotrophic factor (BDNF), an important factor regulating synaptic plasticity thus facilitating learning and memory processess 10 .
In agreement with the observed behavioral phenotypes in IRE1 cKO mice, Bdnf expression was reduced in the aged hippocampus of these animals (Fig. 2h).
In order to determine whether IRE1α ablation in the brain exacerbates age-associated 5 motor decay, we evaluated animals at behavioral, electrophysiological and morphological levels.
Aged IRE1 cKO mice manifested impaired performance in the wire hanging and rotarod tests when compared to littermate control animals ( Fig. 2i and 2j). However, muscle electrophysiological properties did not show significant differences between genotypes in three distinct muscles tested (Extended Data Fig. 3e-g). Similarly, analysis of neuromuscular junction (NMJ) morphology did 10 not reveal any alterations between aged IRE1 cKO and control mice (Extended Data Fig. 3h-i). We then determined if the disruption of the IRE1α pathway in the brain results in altered markers of ageing. Thus, we evaluated the accumulation of senescent cells using β-galactosidase and γ-H2AX staining in hippocampal tissue, as increased senescent cells in brain tissue correlate with age-dependent cognitive decline 23 . Remarkably, an increase in the content of senescent cells was 15 observed in the brain of middle-aged IRE1 cKO animals but not in young animals (Fig. 2k, Extended Data Fig. 3j). Overall, our results indicate that genetic disruption of the IRE1α pathway in the nervous system accelerates the natural emergence of age-associated behavioral and neuromorphological alterations in the central nervous system.
In order to test the consequences of artificially bolstering an adaptive UPR during ageing, 20 we developed strategies to increase the levels of the spliced and active form of XBP1 in the brain. For this purpose, we initially used transgenic mice that overexpress XBP1s under the control of the PrP promoter 9 (referred to here as Tg XBP1s ; Extended Data Fig. 4a) and evaluated their cognitive performance during ageing. Remarkably, XBP1s overexpression prevented the development of age-related deterioration in brain function, as evaluated in the NOR, NOL and 25 Barnes maze tests; in fact, these mice performed comparably to non-transgenic young animals ( Fig. 3a-c). Additionally, Tg XBP1s mice showed reduced age-dependent coordination and motor decay when compared to littermate controls in the wire hanging and rotarod tests (Fig. 3d-e). We then investigated possible molecular pathways that may explain the protective effects of XBP1s overexpression during normal brain ageing. Unexpectedly, we did not observe the upregulation 30 of Bdnf in the brain of middle age and aged Tg XBP1s mice (Extended Data Fig. 4b). To define global changes driven by the expression of XBP1s in the brain, we performed an unbiased proteomic analysis of hippocampal tissue derived from animals at different ages. Remarkably, comparison between young and aged wild-type animals indicated alterations in a cluster of cell signaling proteins involved in long-term potentiation, calcium signaling and metabolic control 5 (Extended Data Fig. 4c, left; Table S2). Evaluation of middle aged and aged animals (Extended Data Fig.3C, right) uncovered alterations in a cluster of proteins related to synaptic vesicle recycling, endocytosis, cytoskeletal dynamics, among other processes (Extended Data Fig. 4c, Table S2). Comparison of Tg XBP1s animals with age-matched litter-mates (Extended Data  Table S3), consistent with the functional improvements detected in these animals at the cognitive level. Interestingly, enriched terms were associated to several age-related neurodegenerative diseases (amyotrophic lateral sclerosis, prion diseases, Alzheimer's disease), synaptic physiology (long term-potentiation, neurofilament, glutamatergic synapse and exocytosis) and myelin sheet (  Table S3).
Finally, we evaluated whether artificial activation of XBP1s-dependent responses could potentially reverse the natural decay in normal brain function observed during ageing. For this purpose, we performed bilateral injections of AAVs to express XBP1s in the hippocampi of ageing mice that already manifested cognitive decline (Extended Data Fig. 5a-b). Remarkably, 20 the administration of AAV-XBP1s to middle-aged and aged animals resulted in improved performance in the various cognitive tests compared to age-matched animals injected with control virus (Fig. 4a-d). Administration of AAV-XBP1s also had beneficial effects in electrophysiological properties of hippocampal slices (Fig. 4e-f). In fact, we observed that basal firing rates and bursting activity in hippocampal CA1 neurons were decreased after treating aged 25 animals with AAV-XBP1s. Such alterations facilitated both increased firing rates and busting activity in CA1 neurons following PTX treatment ( Fig. 4e-f). Furthermore, induction of longterm potentiation (LTP), widely thought to represent an electrical correlate of learning and memory, was significantly improved in hippocampal slices derived from aged mice treated with AAV-XBP1s compared to controls ( Fig. 4g; see fiber volley amplitudes in Extended Data Fig.  30 5c). Importantly, these findings were associated with a significant increase in dendritic spine density in CA1 neurons compared to age-matched control animals (Fig. 4h). Interestingly, aged mice also exhibited decreased accumulation of senescent cells in the hippocampus following XBP1s overexpression (Fig. 4i, Extended Data Fig. 5d). Lastly, we evaluated proteomic changes in the hippocampus of aged animals treated with AAV-XBP1s ( Fig. 4j- Table S4), similar to the findings obtained in Tg XBP1s animals. We also detected important changes in proteins related to extracellular matrix, suggesting an impact to secretory pathway proteostasis (Table S4). Overall, our proteomic profiling suggests that the enforcement of XBP1s expression influences age-associated 10 alterations in synaptic function, consistent with the positive effects of XBP1s administration in prolonging brain health span.
Our findings underline for the first time a protective function of the IRE1α-XBP1s axis in sustaining mammalian brain healthspan, grounded on its function at synaptic physiology maintenance. Surprisingly, proteomic studies could not detect evident changes in canonical 15 XBP1s-target genes involved in proteostasis control described in other organs, but rather highlighted altered expression of a cluster of proteins related to synaptic function and neurodegenerative diseases. Many of the identified hits are cargoes of the secretory pathway, suggesting that XBP1s overexpression modulates neuronal proteostasis at some extents.
Importantly, accumulating evidence suggest that UPR mediators, and more specifically XBP1, 20 have alternative functions in the nervous system by controlling synaptic plasticity 24 and dendritogenesis 25,26 . Importantly, the appearance of senescent cells in brain function was prevented by XBP1s overexpression, whereas IRE1α deficiency exacerbated the accumulation of senescent cells, consistent with a role of XBP1 as an ageing modifier. Overall, our results suggest that the UPR exerts global effects in sustaining proper brain function during ageing. 25 Prior reports have demonstrated the involvement of the IRE1α-XBP1s pathway in a variety of age-related neurodegenerative conditions, including Parkinson´s disease, Alzheimer´s disease, ALS and frontotemporal dementia 24 and retinal degeneration in diabetes 27 . However, whether the activity of the UPR can restore neuronal function in the context of normal mammalian ageing in the absence of disease was unknown, and the mechanisms driving brain 30 ageing versus neurodegenerative diseases are predicted to be different. Unexpectedly, our proteomic profiling suggested that XBP1s overexpression modulates the expression of a variety of proteins related to several neurodegenerative diseases (Table S2, Table S3 and Table S4).
These findings support the concept that targeting central components of the proteostasis network, such as the UPR, may influence the risk to develop brain diseases, where XBP1 operates as an 5 intersection between the biology of ageing and the emergence of neurodegenerative conditions.
In this line, a previous study suggested that XBP1s directly controls the expression of several Alzheimer's disease-related genes 28 and a polymorphism in the XBP1 promoter operates as risk factor to develop the disease 29 . Studies in invertebrate models (yeast, C. elegans and D. melanogaster) have uncovered a central role of ER proteostasis and the UPR in ageing (see 10 examples in 6,[10][11][12][13][30][31][32][33]. Interestingly, the beneficial effects of caloric restriction, a major intervention that extend lifespan, were recently linked to modulatory effects on ER proteostasis [32][33] . Importantly, the positive consequences of activating neuronal and glial UPR on lifespan of worms involves the global control of organismal proteostasis through a cellnonautonomous mechanism 6,12,31 . It remains to be determined if an increase in XBP1s expression 15 in the aged brain can be translated into the propagation of adaptive signals that improve the function of other organs, thus mitigating their deterioration during the course of natural ageing. Because recent studies suggested a correlation between UPR alterations in elderly human tissue [14][15][16] , strategies to improve ER proteostasis or boost the adaptive activity the UPR may extend brain healthspan, reducing the risk of developing dementia and other age-associated 20 neurodegenerative diseases. 10 animals/group. One-way ANOVA followed by Tukey's post-test compared tunicamycin treated groups, **: P < 0.01). (f, g) Relative levels of S-nitrosylated IRE1α (d) and total IRE1α (e) in brains from young and aged mice.

Materials and Methods
Human Study Population 5 RNA-seq human data was extracted from The Aging, Dementia and Traumatic Brain Injury Study (https://aging.brain-map.org/overview/home) which is a detailed neuropathologic,

Stereotaxic injections
Young (3 month-old), middle-aged (12 month-old), and aged (18 month-old) mice received bilateral stereotaxic injections of AAV (1 × 10 9 viral genomes (VGs)/µL) in the hippocampi using the coordinates AP: -1.9, DV: +1.7, and ML: ±1.0. Male mice were deeply anesthetized 25 using isoflurane (4%) and a stereotaxic apparatus coupled to a Hamilton microsyringe was used for the procedure. Cranium was exposed though a skin incision, and bilaterally symmetrical holes were opened using a dental drill. Injections were performed at approximately 0.5 µL/min in a total of 1 µl. Mice were returned to their home cages and kept under close monitoring until they were awake. Mice were injected with either control AAV serotype 2 vector (Mock), AAV2-CRE (Addgene, #105545-AAV2) or AAV2-XBP1s (produced at Genzyme) under the control of the CMV promoter, in addition to an eGFP cassette to monitor transduction efficiency, as we previously reported 34 .

Behavioral tests 5
Behavioral experiments were performed in a masked fashion, both for genetically-modified animals and AAV-injected mice, using groups of age-matched controls. Injected mice underwent behavioral evaluation 1 month following injections. Cognitive and motor tests were performed in the following order: New object recognition, new object location, Barnes maze, wire hanging test, rotarod, and contextual fear conditioning. Not all animals were exposed to all tests in order 10 to avoid possible secondary effects mediated by interaction of the various tests.

Novel object recognition
Novel object recognition (NOR) was performed as previously described [35][36]  were cleaned with 70% ethanol before each session to avoid olfactory cues. Exploration time was defined as the amount of time mice had their noses oriented toward an object with its nose within 3 cm or less. Other behaviors such as rearing near the object or resting against the object were not considered as exploration. Exploration time spent during the training phase was also recorded. Animals that failed to interact with one or both objects, or that showed signs of stress were excluded from the analysis. All analysis was performed in a masked fashion.

Novel object location
The novel object location task (NOL) was performed as previously described [35][36] . Mice were placed in the same arena used for the NOR assay. The arena was divided into quadrants. For the 5 training phase, mice were introduced to two identical objects (Lego blocks, 5 x 4 x 4 cm) placed in two randomized quadrants of the arena. Objects were placed equidistant from one other. At 3 or 24 h after the test phase, mice were again placed in the arena facing the wall but now with one of the objects placed in a distinct quadrant. Trials were recorded on a digital camera coupled to a computer for subsequent evaluation of time spent exploring objects. Both the objects and the 10 arena were cleaned with 70% ethanol before each session to avoid olfactory cues. Exploration time was defined as detailed above by a blinded researcher. As per standard protocol, animals that failed to interact with one or both objects or showed signals of stress were excluded from the analysis.

Barnes maze 15
The Barnes maze task was performed on a white circular surface (0.9 m in diameter) with 20 holes equally spaced around the perimeter, as previously described 37 . A dark escape box (10 × 20 × 7.8 cm) was located under one of the holes, which was the target. A ramp was placed under the target hole so that mice could reach the escape tunnel easily. The circular open field was elevated 75 cm above the floor. Distal spatial cues (with different colors and shapes) were placed outside 20 of the maze. The maze was rotated daily, with the spatial location of the target unchanged with respect to the distal visual room cues. A cylindrical start chamber was placed in the center of the maze and removed after 10 s. Training sessions consisted of four trials per day conducted for 4 days with a maximum time of 3 minutes each. If animals were unable to find the target after 3 minutes, they were gently placed in the right hole by the tail. A stopwatch sound was used as an 25 aversive stimulus in order to induce exploration. Once a mouse reached the target, the noise was immediately stopped, and the mouse was left inside the box for 1 minute and then returned to its home cage. A maximal period of 15 minutes was given for the inter-trial interval. The maze and the escape box were cleaned with 70% ethanol between each trial to avoid olfactory cues. The numbers of primary pokes and the primary latency to reach the target were manually counted, as 30 defined by the number of pokes and time to reach the target for the first time, as some animals did not promptly enter the escape box during training sessions. Every trial was recorded on a video camera placed in the ceiling and fed to a computer. On the fifth day, the escape box was removed, and one single trial was performed as a probe trial to evaluate spatial memory acquisition with a maximum time of 90 s. Then, following 7 days without training, mice were 5 again exposed to a test phase to evaluate long-term memory formation.

Contextual fear conditioning
On the first day, mice were placed in the contextual fear-conditioning chamber (Med Associates). During the first 2 min, no stimulus was applied, following which 80 db of white noise (the conditioned stimulus; CS) was generated for 30 s. Two seconds later, mice were 10 exposed to a 0.5 mA electric shock (the unconditioned stimulus) for 2 s. After the shock, mice were let in the cage for 2 more minutes and freezing episodes, represented by the percentage of freezing response to total activity, were measured by an automated system. On the next day, mice were again placed in the same chamber and exposed to the same protocol but this time without any electroshock. Freezing events and freezing percentage time were recorded for 5 min 15 using an automated system.

Wire hanging test
The wire hanging test is a well-established protocol for measuring muscular strength in mice. We generated an arbitrary score to take into consideration the number of times one animal fell from the wire during the course of the test (total of 180 s). Every time an animal fell from the wire, its 20 arbitrary score on the test was decreased by 1 (starting at 10 and decreasing to unity). If an animal fell more than 9 times, the test was stopped. Every time an animal reached one side of the wire, the stopwatch was paused, and the animal was placed in the middle of the wire before the stopwatch was started again. A soft cotton sheet was placed under the wire to avoid an harm to the animal during the falls. The test was scored as follows: 25 Final Score = (10 -number of falls) x time remaining on the wire (s).

Rotarod test
Mice were placed on a rotating cylinder and trained to learn how to walk on it at constant speed.
Mice that could not learn the task after successive trials were excluded. Mice that learned the task were then analyzed in 5 trials with the apparatus configured to accelerate from 4-40 RPM (rotations per minute) over 10 minutes during the course of 3 days. The mean time until the mouse fell off of the cylinder for each trial was computed as the score.

Tissue collection and processing
Briefly, mice were deeply anesthetized with ketamine/xylazine and perfused with ice cold saline. 5 The brain was removed from the skull and separated into two hemispheres. The hippocampus, cerebral cortex and cerebellum of the left hemisphere were dissected out and stored frozen at -80 °C until analysis. The right hemisphere was postfixed in 4% paraformaldehyde in PBS overnight at 4 °C, followed by cryopreservation in 30% sucrose in PBS and freezing medium (OCT, TissueTek). Subsequently, 40 µm-thick sagittal sections were obtained free-floating on a Leica 10 cryostat for immunostaining using anti-NeuN (Millipore, MAB377, 1:100) to label neurons. Five serial sections every 200 µm were stained per animal. Fluorescence images were acquired using a confocal microscope (Nikon C2+).

Dendritic spine imaging
Brain slices were cut at 40-µm thickness on a cryostat. AAV2-GFP fluorescence was previously 15 confirmed in injected animals to validate viral transduction using an inverted epifluorescence microscope and then imaged on a confocal microscope, Nikon Eclipse T1, at 60x magnification with additional digital zoom of 3x. Similar regions were compared among animals (CA1 region, spines in primary and secondary dendrites between the stratum radiata and the pyramidal layer, AP: -1.9 to -2.1 from the bregma). Z-stacks were acquired in 0.5-µm slices, laser intensity at 0.5-20 1% and 12.5us/pixel at 1024x1024 resolution. Z-Stacks were then summed using ImageJ software for total maximum intensity to generate one single stacked 8-bit image. The number of spines was manually quantified in scaled images and divided by the length of the dendrite analyzed; 5-10 dendrites per animal were used for analysis.

Immunofluorescence 25
Brain slices were incubated in citrate buffer at 96 °C for 30 min for epitope exposition and washed in PBS. After this, slices were washed in TBS and incubated in blocking solution (3% BSA and 0.05% Triton X-100) for 60 min at room temperature. Slices were then incubated overnight at 4 °C with primary antibody anti-phospho-Histone H2AX Ser139 (Sigma-Aldrich, 05-636) diluted 1:1000 in 3% BSA in TBS. Secondary antibody was Alexa Fluor 568 (Invitrogen A-11031) diluted 1:2000 in 3% BSA in TBS. The samples were washed 3 times using TBS, and in the final wash, DAPI was added and incubated for 5 minutes. Images were taken with Leica TCS SP8 confocal microscope with a 40x objective magnification. ImageJ and LAS X software were used to process the stacked images. The percentage of positive cells for p-H2AX was 5 graphed. Representative images are shown.

Senescence-associated beta-galactosidase (SA-βgal) staining
Histochemical detection of SA-βgal activity was performed according to Debacq-Chainiaux et al. 38 . Briefly, slices were incubated in a 1 mg/mL of solution of 5-bromo-4-chloro-3-indolyl β-d- incubation, slices were washed with TBS and mounted on superfrost microscope slides (ThermoFisher 6776214) using Fluoromount-G (ThermoFisher, 00-4958-02). Images were acquired on a Leica DM500 binocular microscope equipped with a Leica ICC50 W camera using 4x and 10x objective magnifications. ImageJ software was used to process the images. Positive 15 area for SA-βgal activity was measured and representative images are shown.

Multielectrode array (MEA) recording
The same brain hippocampal slices were used to record both multielectrode arrays (MEA, 252 electrodes, Multichannel Systems) and local field potentials. Hippocampal slices were mounted 10 on a MEA matrix bathed in an ACSF medium (in mM: NaCl 125, KCl 2.5, glucose 25, NaHCO 3 25, NaH 2 PO4 1. 25 traces, whitening them (to remove correlated noise), clustering action potential waveforms, and fitting them on the whitened traces.

Firing and burst rate recordings
Firing rate was computed as the number of spikes divided by the recording length using Neuroexplore (https://www.neuroexplorer.com/) software. Burst rate analysis was computed by 20 an inter-spike interval routine implemented by a custom Matlab procedure 41 . Spontaneous Activity (SA) was recorded for 10 min, following a 30-min application of picrotoxin (PTX, 100 µM).

Compound muscle action potential (CMAP) recordings 25
CMAPs were measured in isoflurane-anesthetized mice using an electromyographic apparatus (Keypoint, Dantec, Les Ulis, Francer AD Instruments, Oxford, UK). Briefly, the sciatic nerve was stimulated by two electrodes placed over the lumbar vertebral column. CMAP was recorded by two electrodes placed in the belly and in the tendon of the right gastrocnemius, tibialis anterior, and triceps muscles. A reference earth electrode was placed in the left gastrocnemius 30 and connected to the electromyography apparatus 42 .

Biochemical and molecular 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

Detection of S-nitrosylated (SNO-)IRE1α and total IRE1α levels
We performed biotin-switch assays using whole-brain tissue samples as previously described 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), with each step followed by a gradient from buffer A (95% water, 5% acetonitrile, 0.1% formic acid) to buffer B. Electrospray was 25 performed directly from the analytical column by applying a voltage of 2.2 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 higher energy collisional dissociation (HCD) using a normalized collision energy of 38%, isolation window of 30 0.7 m/z, a resolution of 45,000, AGC target 1e5, maximum IT 60 ms, and scanned from 100 to was performed using Proteome Discoverer 2.4 (ThermoFisher). Spectra were searched using SEQUEST against a 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 5 filtering and the following settings: 10 ppm peptide precursor tolerance, 6 amino acid minimum

Neuromuscular junction (NMJ) staining and analyses
The levator auris longus (LAL) muscles were dissected and whole-mount fixed in 0. were post-fixed with 1% FA in 1X PBS for 10 min at 22 °C, flat mounted and imaged. Z-stack images were collected at 1-µm intervals on a Zeiss LSM 700 confocal microscope. Maximal intensity projection images were reconstructed in 3D using ImageJ software. The morphology of >50 NMJs per mouse was manually determined and expressed as a percentage of the total. The area of >50 acetylcholine receptor (AChR) densities per mouse was determined for each 5 postsynaptic structure using ImageJ software.

Statistical analysis
Statistical significance respective to age was queried for single comparisons with a Student's t test and for multiple comparisons with a one-way ANOVA followed by Tukey's post-hoc test.
When analyzing both age and genotype in the same comparison, a two-way ANOVA followed 10 by Tukey's post-hoc test was implemented. P values were considered significant when they were < 0.05. The number of animals in each group varied from 4-15 depending on the experiment based on a Power Analysis of prior data.

Data availability
The mass spectrometry proteomics data have been deposited to the ProteomeXchange 15 Consortium via the PRIDE 49 partner repository with the dataset identifier PXD020539. All data generated or analyzed during this study are included in this published article (and its 20 supplementary information files).

Supplementary Information is available for this paper.
Correspondence and requests for materials should be addressed to: chetz@med.uchile.cl Reprints and permissions information is available at www.nature.com/reprints.

Delta voltage (mv)
Young

Middle aged Aged
Animals were evaluated in the wire hanging test with a score based on total time of hanging and falls off of the wire in a 3 min period. Histogram shows mean and SEM of score in arbitrary units (au; n = 9, 11, 15 for each age group, respectively). **P < 0.01 by one-way ANOVA followed by Tukey's post-hoc test. (k) Animals were evaluated in the rotarod test to analyze muscular function and coordination. Histogram shows mean and SEM of latencies to fall from the rod under constant acceleration in 4 trials split over 4 consecutive days (n = 16, 8, 17 for each age group, respectively). **: P < 0.01 by one-way ANOVA followed by Tukey's post-hoc 5 test. (l) Animals were deeply anesthetized and had their compound muscle amplitude potentials (CMAPs) measured in the gastrocnemius, tibialis anterior, and triceps muscles. Histogram shows mean and SEM of differences between maximum and minimum amplitudes (n = 8, 8, 8 for each age group, respectively). **: P < 0.01, ***P < 0.001 by one-way ANOVA followed by Tukey's post-hoc test. (m) Brain slice electrophysiological analysis assessing firing rates in CA1 pyramidal neurons (n = 4 animals/group; n = 832-1584 neurons/animal). Mean and SEM of firing rates were measured during spontaneous activity or

Potential relation between XBP1s, calcium signaling and ageing
XBP1s was reported to modulate intracellular calcium dynamics in neurons, correlating with increased neurotoxicity in an AD model 3 and calcium homeostasis unbalance has long been 5 proposed to be a central player in neuronal ageing 4 . Our findings indicate that many proteins associated with calcium signaling in neurons; namely calcineurin, calmodulin, calcium transporter ATP2B4 and other signaling components such as adenylyl cyclase, are altered following XBP1s overexpression in aged mice (Tables S3 and S4). Likely, increased long-term potentiation, dendritic spike stabilization and increments in memory formation in aged mice 10 reported here are also dependent on calcium homeostasis.

Extension to human ageing
One central query in our hypothesis is whether our findings can be extended to human ageing, as we point that unfolded protein machinery account as the main changes observed in the 15 hippocampus of demented patients (Fig. 1A-B; Table S1) . Interestingly, a recent large-scale proteomic human study has pinpointed promising protein targets that sustain cognitive stability during ageing 5 showing that myelin-associated proteins NEFL, NEFM and MBP are associated with cognitive resilience. Those findings overlap with ours, both in the Tg XBP1s and AAV-XBP1s approaches, as those proteins were altered following XBP1s overexpression (Table S3, Table  20 S4). Also, altered expression of extracellular matrix component laminin-A and glial fibrillar associated protein (GFAP), two proteomic hits highlighted in our analysis, were also linked to cognitive resilience in two human cohorts studies 5 .