Human brain samples<\/p>\n
Post-mortem human brain and serum samples were obtained in accordance with institutional guidelines and with approval from the Harvard Medical School Institutional Review Board. All procedures complied with relevant ethical regulations. All post-mortem human brain and serum samples were fully deidentified before receipt, and no identifiable private donor information was accessible to the researchers. As such, informed consent was not applicable. Frozen post-mortem samples from the prefrontal cortex (BA9\/10\/47) were available for all cases included in the analysis. Cerebellar tissue and the most recently collected pre-mortem serum samples were available for a subset of individuals. The primary analysis was performed on tissue samples procured from the Rush Alzheimer\u2019s Disease Center, derived from participants in the Religious Orders Study (ROS) or Rush Memory and Aging Project (MAP) (referred to as ROSMAP). The ROSMAP is a longitudinal, clinical\u2013pathological study of ageing, cognitive decline and AD58<\/a>. Study participants agreed to comprehensive annual clinical and neuropsychological evaluation and brain donation at death. To assess cognitive function, 21 cognitive-function tests were used, 19 were in common and 11 were used to inform on clinical diagnoses, as previously described59<\/a>,60<\/a>. The follow-up rate exceeded 95% and the autopsy rate exceeded 90%. All individuals who underwent autopsy were subject to a uniform structured neuropathological evaluation of AD. Informed consent, an Anatomic Gift Act and a repository consent were obtained and the studies were approved by an Institutional Review Board of Rush University Medical Center. A second set of frozen frontal cortical brain samples was obtained from brain banks at the Massachusetts General Hospital, Duke University and Washington University, and is referred to as \u201ca second independent cohort\u201d. Brain tissue obtained from these sources had a confirmed pathological diagnosis of AD or NCI. Samples were randomly selected by the source institutions based on tissue availability and alignment with the requested diagnostic categories (NCI, MCI and AD). Within each diagnostic group, samples were matched for age and sex to ensure group comparability.<\/p>\n Absolute and relative metal levels were measured by ICP\u2013MS, with relative levels calculated as the ratio of cortical or cerebellar to serum concentrations from the same individual. Post-mortem interval had no significant effect on total or relative Li levels in this cohort. The study population comprised 40.2% male\u00a0individuals and 59.8% female\u00a0individuals. Within diagnostic subgroups, NCI cases comprised 40.8% male\u00a0individuals and 59.2% female\u00a0individuals; MCI cases, 42% male\u00a0individuals and 58% female\u00a0individuals; and AD cases, 36.4% male\u00a0individuals and 63.6% female\u00a0individuals. Individuals of both sexes were analysed, and those with MCI and AD, regardless of sex, exhibited significantly reduced cortical-to-serum Li ratios and lower total cortical Li levels. Donor sex was self-reported and provided by Rush Medical Center (ROSMAP study) and by further tissue sources, including Massachusetts General Hospital, Duke University and Washington University.<\/p>\n Isolation of plaque-enriched and non-plaque fractions<\/p>\n To fractionate brain parenchymal homogenates into amyloid plaque-enriched and non-plaque fractions, we modified a previously described protocol61<\/a>,62<\/a>. Frozen brain samples were weighed and then Dounce-homogenized (40 strokes per sample) in 5 volumes (v\/w) of ultrapure buffer containing 2% SDS (stock of ultrapure SDS 10%, ThermoFisher Scientific, 24730020) and 0.1\u2009M \u03b2-mercaptoethanol (VWR, 97064-878) in 50\u2009mM Tris HCl, pH 7.6 (ultrapure Tris-HCl, pH 7.5, Invitrogen, 15567-027) and water (Aristar Ultra, VWR 87003-236). The Li concentration of the complete buffer was below the detection threshold (\u22121). The homogenates were heated at 100\u2009\u00b0C for 10\u2009min and then transferred to a 15-ml Falcon tube fitted with a sieve consisting of woven mesh (polyethylene terephthalate) with a pore size of 100\u2009\u00b5m (pluriSelect, SKU 43-10100-60). The samples were passed through the sieve by gravity and the filtrate was then centrifuged (300g for 30\u2009min). The supernatant (soluble non-plaque fraction) was removed and stored at \u221280\u2009\u00b0C. The pellet was resuspended in water at a ratio of 5\u2009ml per gram of pellet mass and stored at \u201380\u2009\u00b0C (plaque-enriched fraction). To image subfractionated A\u03b2 and phospho-tau (Supplementary Fig. 1<\/a>), 10\u2009\u03bcl of the freshly collected plaque-enriched and non-plaque fractions was layered onto albumin-coated glass slides and allowed to dry overnight. They were then washed with ultrapure PBS (which we determined contained less than 25 parts per trillion (ppt) Li) and incubated with a rabbit monoclonal anti-A\u03b2 antibody (Cell Signaling, 8243) and a mouse monoclonal antibody to pSer202-tau (clone CP13) overnight in 2% BSA, 0.1% Triton X-100 in PBS, followed by labelling with secondary anti-rabbit IgG coupled to Alexa Fluor 594, or anti-mouse IgG coupled to Alexa Fluor 488 (1:300 in blocking buffer). The slides were then washed three times in ultrapure PBS and mounted.<\/p>\n ICP\u2013MS<\/p>\n For the analysis of metals in human and mouse biological samples, we modified previous protocols to optimize the detection of ultra-trace elements. We tested several protocols and found that the use of precleaned polyvinylidene difluoride (PVDF) vials fitted with perfluoroalkoxy alkane (PFA) caps, the use of ultra-trace grade reagents (nitric acid, hydrogen peroxide and water), combined with an extended sample digestion and homogenization, and a highly sensitive ICP\u2013MS instrument (PerkinElmer NexION 2000C), allowed the robust detection of ultra-trace metals in human and mouse samples. The commercial precleaned PVDF vials (Elemental Scientific, V-14-0712-C) and PFA caps (Elemental Scientific, V-14-0309-C) were further processed by fully immersing them in 10% trace-grade nitric acid (Fisher Chemical, A509-P212) for at least 48\u2009h, followed by abundant rinsing with double-distilled and deionized water and drying in a chemical hood for 48\u2009h. The chemical hood was thoroughly cleaned before the experiment and was used exclusively for ICP\u2013MS for the entire duration of the experiment to prevent contamination. We also used a protocol allowing for the simultaneous analysis of a large number of human brain samples (approximately 80\u2013120\u2009mg frozen brain material per region per case). First, we determined that the dry-to-wet ratio was unchanged in AD versus NCI. This was established in n\u2009=\u200945 NCI and n\u2009=\u200945 AD frozen cortical samples (100\u2013200\u2009mg per sample) that were weighed and then dried to a constant weight (48\u2009h in a dry oven at 60\u2009\u00b0C). The dry-to-wet ratios were 0.127\u2009\u00b1\u20090.048 for NCI and 0.123\u2009\u00b1\u20090.034 for AD and were not statistically different (P\u2009=\u20090.67), in agreement with previous work63<\/a>.<\/p>\n The frozen cortical and cerebellum samples\u00a0were first allowed to thaw, and were then weighed and digested in 5 volumes of nitric acid 67% (w\/m, relative to wet mass; BDH Aristar Ultra, VWR, 87003-226) for 72\u2009h with regular vortexing (20\u2009s per vial every 12\u2009h). The samples were fully digested after about 36\u2009h. The serum, the brain non-plaque fractions and the aqueous solutions were digested in an equal volume of nitric acid (67%) for 48\u2009h with regular vortexing (20\u2009s per vial every 12\u2009h). After digestion with nitric acid, hydrogen peroxide (30%; BDH Aristar Ultra, VWR, 87003-224) was added for 24\u2009h with regular vortexing (20\u2009s per vial every 12\u2009h). We added one volume of hydrogen peroxide (w\/m, relative to starting wet mass) to digested brain tissues and 0.75 volumes (relative to starting sample volume) to digested serum, non-plaque fractions and aqueous solutions. The samples were then diluted using a 2% nitric solution in ultrapure water (BDH Aristar, VWR, 87003-236). Indium was added to each solution as an internal standard (50 parts per billion; ppb). For all ICP\u2013MS runs, we also measured freshly made solutions of element standards (0, 10\u2009ppt, 50\u2009ppt, 100\u2009ppt, 1\u2009ppb, 10\u2009ppb and 50\u2009ppb) using a 30-element ICP standard (Aristar, VWR, 89800-580). Each run included n\u2009=\u200910 digestion blanks as well as n\u2009=\u200920\u201330 blank measurements to calculate the detection limits. The samples were injected into a PerkinElmer NexION 2000C ICP\u2013MS instrument fitted with a cross-flow nebulizer and peristaltic pump for sample introduction. The sample delay time was 30\u2009s with a pump speed of 24\u2009rpm. A wash solution of 2% nitric was used between analyses of samples. The human cortex, cerebellum and serum samples were each measured twice on two consecutive days (two technical replicates per sample) and the average value was obtained for each sample. The correlation coefficients between the lithium concentrations measured on day 1 and day 2 were r\u2009>\u20090.99 for frontal cortex, cerebellum and serum, showing that the ICP\u2013MS measurement was highly reproducible. After each run, ICP\u2013MS signal processing was done using GeoPro 2010 Software (Cetac Technologies). We derived the standard curves for each element, calculated the concentration of each element in the diluted solution, and used the dilution factors to derive elemental abundance in the original samples. Li levels in the cortex and cerebellum are reported per unit of wet weight (Fig. 1d,e<\/a>, Extended Data Fig. 1b<\/a> and Supplementary Table 1<\/a>). Limits of detection (LODs) and limits of quantification (LOQs) were calculated as follows: LOD\u2009=\u2009YB\u2009+\u20092tSB and LOQ\u2009=\u2009YB\u2009+\u200910SB, where YB is the average blank signal, t is the critical value of the one-tailed t-test (one-tailed, 95% confidence interval; for example, for 27 blank samples, df\u2009=\u200926 and t\u2009=\u20091.706) and SB is the standard deviation of a blank signal. LOD and LOQ values for all metals can be found in Supplementary Table 1<\/a>. All individual Li measurements in human samples (prefrontal cortex, cerebellum and serum) were above the LOQ. In recovery experiments, wet brain samples or fluids were spiked with lithium standard added at three levels (n\u2009=\u20097 replicates per spiking level). The recovery of Li from spiked samples ranged from 91% to 105%. All human sample measurements were double-blinded: one lab member not involved in the study relabelled the samples and kept a file with the old and new codes. After the ICP\u2013MS measurements, the samples were unblinded in the presence of the researchers involved in the study, as well as the lab member who was not involved in the study.<\/p>\n The ICP\u2013MS findings from post-mortem human samples were replicated as follows. First, reduced Li content in the cortex of patients\u00a0with AD was observed using two independent methods, after measurement of total Li levels in frozen cortical material of cases from both ROSMAP (Fig. 1d<\/a>) and other sources (Fig. 1e<\/a>), as well as after fractionation and removal of amyloid plaques (Fig. 1g<\/a>). Second, decreased Li levels in the AD versus NCI prefrontal cortex (P\u2009=\u20092\u2009\u00d7\u200910\u22123) were also independently confirmed when n\u2009=\u200960 NCI and AD cases were processed and analysed by ICP\u2013MS in a different laboratory (the Spectroscopy Core Facility at the University of Nebraska, Lincoln). Third, decreased Li levels in the AD versus NCI prefrontal cortex (P\u2009=\u20093\u2009\u00d7\u200910\u22123) were also confirmed when n\u2009=\u200948 NCI and AD cases were processed using an alternative protocol. Frozen samples were thawed and dried to a constant weight by incubating in a dry oven at 60\u2009\u00b0C for 48\u2009h. The dried tissue was then digested in 1\u2009ml of 67% nitric acid using a heating block at 95\u2009\u00b0C for 3\u2009h. After digestion, 0.3\u2009ml of 30% hydrogen peroxide was added and the mixture was heated for a further 3\u2009h. Finally, the samples were diluted and analysed using ICP\u2013MS.<\/p>\n Li levels measured in the PFC of ageing NCI cases (ROSMAP cases: mean 2.36\u2009\u00b1\u20091.23\u2009ng per g, range 0.52\u20136.0\u2009ng per g; non-ROSMAP cases: mean 3.50\u2009\u00b1\u20092.27\u2009ng per g, range 0.89\u20139.94\u2009ng per g; Fig. 1d,e<\/a>) were similar to those measured in a previous study64<\/a> (4.1\u2009\u00b1\u20091.7\u2009ng per g in the prefrontal cortex of aged non-diseased cases; age, 71\u2009\u00b1\u200912\u2009years). Similarly, Li levels measured in the cerebellum (ROSMAP cases: mean 2.90\u2009\u00b1\u20091.69\u2009ng per g, range 0.58\u20138.40\u2009ng per g; Extended Data Fig. 1b<\/a>) were similar to those measured in the previous study64<\/a> (2.9\u2009\u00b1\u20091.3\u2009ng per g). Finally, consistent with previous studies, we observed significantly elevated levels of sodium25<\/a> and zinc26<\/a>, along with reduced copper12<\/a>,65<\/a> levels, in the AD cortex (Fig. 1a,b<\/a> and Supplementary Table 1<\/a>).<\/p>\n LA-ICP\u2013MS<\/p>\n The Li composition in the human and mouse brains in situ was analysed using LA-ICP\u2013MS. Frozen human and mouse brains were first embedded in OCT medium and then sectioned using a cryostat, and the resulting sections (80\u2009\u03bcm thick) were mounted onto glass slides. Before data acquisition, the samples were placed vertically in a rack and air-dried for 1\u2009h. The LA-ICP\u2013MS spectrometer consisted of a laser ablation system (213\u2009nm Nd:YAG, Cetac Technologies) connected to a Perkin Elmer NexION 2000C ICP\u2013MS (Perkin Elmer). Using the line tool, we manually selected the area to be ablated. For human samples, we ablated a region of the prefrontal cortex. For mouse samples, we processed coronal sections where the cortex and hippocampus were readily identifiable. The analyte signal was collected using multiple parallel line scans along the entire selected area, progressing in the direction of ablation cell gas flow, using a spot size of 200\u2009\u00b5m at 75\u2009\u00b5m\u2009s\u22121. The laser operated at an energy level of 70% and a pulse repetition rate of 20\u2009Hz. The typical run time for one sample was about 4\u20135\u2009h. We also ablated parts of each section that did include brain tissue but contained embedding medium (OCT) and subtracted this background signal from the total signal. Levels of 7Li were normalized to carbon (12C) to correct for any variations in the amount of tissue ablated. Similar conclusions were reached when the analysis did not include normalization to 12C. Matrix-matched standards were obtained by spiking homogenized samples of human or mouse tissue with three different concentrations of metal standard solution containing the analytes of interest. After homogenization, the mixtures were frozen and 80-\u03bcm sections were cut using the cryostat. The final concentrations of these standards were validated by ICP\u2013MS. After LA-ICP\u2013MS data acquisition, signal processing was done using Iolite Software 2018 (Iolite). A Li distribution matrix was generated computationally, using the multiple parallel line rasters. To identify the regions occupied by amyloid plaques, the section immediately adjacent to the section analysed by LA-ICP\u2013MS was processed for A\u03b2 immunofluorescence. In brief, the adjacent section was first fixed with 4% PFA for 2\u2009h then washed three times with PBS. The section was then blocked for 1\u2009h with 2% BSA, 2% fetal bovine serum, 0.1% Triton X-100 in PBS. The anti-A\u03b2 primary antibody (Cell Signaling, 8243), diluted 1:250 in blocking buffer, was then added and the section was incubated overnight at 4\u2009\u00b0C. The next day, the section was washed three times with PBS (in total, 30\u2009min), and a secondary anti-rabbit Alexa 594 antibody (diluted 1:300 in blocking buffer) was added for 3\u2009h. The section was finally washed three times with PBS (for 30\u2009min) and mounted. We acquired multiple pictures of A\u03b2 immunofluorescence spanning the entire section using an Olympus FV3000 confocal microscope. The images were then stitched together and imported into Iolite, where the distribution of A\u03b2 immunofluorescence was computationally superimposed on the LA-ICP\u2013MS Li distribution matrix. For each human or mouse sample, we manually selected multiple regions containing A\u03b2 plaques (plaque or \u2018P regions\u2019) as well as neighbouring regions devoid of plaques (non-plaque or \u2018NP regions\u2019).<\/p>\n The mean Li levels in P and NP regions were determined, and after correcting for background and normalizing to 12C, the P:NP ratios were calculated. Three other isotopes were also assessed: 57Fe, 63Cu and 66Zn. All measurements in P and NP regions exceeded the LOQ, which was 0.82\u2009ng per g for 7Li, 0.23\u2009\u00b5g per g for 57Fe, 0.44\u2009\u00b5g per g for 63Cu and 55.1\u2009ng per g for 66Zn). As positive controls, 57Fe, 63Cu and 66Zn were all enriched in plaques relative to non-plaque regions in the AD brain, consistent with previous observations66<\/a>.<\/p>\n Lithium salts<\/p>\n LiO was obtained from Innophos Nutrition and LiC was from Rockwood Lithium. The purity and Li content were verified by mass spectrometry and ICP\u2013MS, respectively. Sources for other Li salts used in conductivity assays are provided in Supplementary Table 16<\/a>.<\/p>\n Li salts were dissolved in distilled, deionized drinking water and administered ad libitum to mice. The low Li dose corresponded to 4.3\u2009\u03bcEq\u2009l\u22121 (equivalent to 0.03\u2009mg (30\u2009\u2009\u00b5g) of elemental Li per litre). The background Li concentration in the water was minimal (0.109\u2009\u00b5g\u2009l\u22121). Solutions of 4.3\u2009\u00b5M LiO and 2.15\u2009\u00b5M LiC were prepared to deliver equivalent amounts of elemental Li, accounting for the two Li atoms per molecule of LiC (Li2CO3). A 4.3\u2009\u00b5M sodium orotate (NaO) solution was also prepared to assess the effects of the orotate anion in the absence of Li. Two more Li doses were also tested: 43\u2009\u00b5Eq\u2009l\u22121 (delivered as 43\u2009\u00b5M LiO) and 430\u2009\u00b5Eq\u2009l\u22121 (delivered as 430\u2009\u00b5M LiO or 215\u2009\u00b5M Li2CO3). To control for the orotate anion at the high dose, a 430\u2009\u00b5M NaO solution was also tested. Average daily water consumption was comparable across all treatment groups and the vehicle (water-only) group. To evaluate Li uptake and its biological effects in the brain, mice received the Li-containing water for defined periods. Animals were randomly assigned to treatment and control groups, with control mice receiving plain drinking water.<\/p>\n Conductivity measurements<\/p>\n To measure the conductivity of Li salts, the salts were dissolved in water to achieve Li concentrations of 4.3\u2009mEq\u2009l\u22121, 430\u2009\u03bcEq\u2009l\u22121, 43\u2009\u03bcEq\u2009l\u22121 or 21.5\u2009\u03bcEq\u2009l\u22121 in each case. Conductivity was measured using an ST300C conductivity meter (OHAUS, 83033964) equipped with a STCON7 electrode (OHAUS, 30080693) calibrated with potassium chloride conductivity standards. For each lithium salt, three independent solution replicates (n\u2009=\u20093) were prepared. Conductivity values are reported in \u03bcS per cm at 25\u2009\u00b0C.<\/p>\n In vitro binding of lithium to A\u03b2<\/p>\n To assess the in vitro binding of Li to A\u03b2, both oligomeric and fibrillar forms of A\u03b242 were prepared. Human A\u03b21\u201342 peptide (1\u2009mg) was initially dissolved in 80\u2009\u03bcl of 1% NH4OH then diluted with PBS to a final concentration of 1\u2009mg\u2009ml\u22121 (stock solution) and stored at \u221280\u2009\u00b0C. Oligomeric A\u03b242 was generated by resuspending the stock solution in PBS followed by overnight incubation at 4\u2009\u00b0C. Fibrillar A\u03b242 was obtained by incubating the same stock at 37\u2009\u00b0C for 72\u2009h. For Li binding assays, 10\u2009\u00b5g of either oligomeric or fibrillar A\u03b242 (10\u2009\u00b5l of the 1\u2009mg\u2009ml\u22121 stock) was added to 90\u2009\u00b5l of Li-containing solutions. These solutions included varying concentrations of either LiO or LiC, matched for Li content and dissolved in ultrapure water (BDH Aristar, VWR, 87003-236). Ultrapure water alone served as the negative control. Samples were incubated at 37\u2009\u00b0C for 16\u2009h. After incubation, the mixtures were transferred to dialysis membranes for 24\u2009h to remove unbound Li (Pur-A-Lyzer mini dialysis kits were used: 6\u20138\u2009kDa cut-off for oligomers and 25\u2009kDa for fibrils). After dialysis, the samples were transferred into precleaned PVDF vials and digested by adding an equal volume of 67% nitric acid (final volume, 200\u2009\u00b5l), followed by a 24-h digestion period. The digested samples were then diluted to 800\u2009\u00b5l with 2% nitric acid prepared in ultrapure water. Li content was quantified using a PerkinElmer NexION 2000C ICP\u2013MS instrument. Elemental Li standards were prepared and standard curves showed excellent linearity (r\u2009>\u20090.99). The bound 7Li levels were calculated across a range of Li salt concentrations, and binding curves were plotted using GraphPad Prism (v.9.4.1). Binding affinities (EC50) and 95% confidence intervals for LiO and LiC were determined using nonlinear regression analysis ([agonist] versus response\u2013variable slope, four-parameter model; Supplementary Table 13<\/a>). Binding to A\u03b242 oligomers and fibrils was modelled across the full concentration range (0\u2013500\u2009\u03bcEq\u2009l\u22121) as well as within the higher-affinity subranges (0\u201350\u2009\u03bcEq\u2009l\u22121 for oligomers and 0\u201330\u2009\u03bcEq\u2009l\u22121 for fibrils; Supplementary Table 13<\/a>).<\/p>\n Mice<\/p>\n Animal housing and experimental procedures were approved by the Institutional Animal Care and Use Committee of Harvard Medical School. All mice were housed socially (2\u20134 animals per cage) in a room with a 12\u2009h:12\u2009h\u2009 light:dark cycle (lights on at 06:00), controlled for temperature (18\u201322\u2009\u00b0C) and humidity (40\u201360%). Sentinel mice housed in each rack were tested quarterly and confirmed to be free of pathogens. All cages were individually ventilated. The standard diet 5053, as well as the chemically defined control and Li-deficient diets, were irradiated. Reverse osmosis deionized water and deionized water containing LiO, LiC or NaO was provided ad libitum in bottles that were changed at least weekly.<\/p>\n Wild-type mice were on a C57BL\/6J background. We analysed both adult (3\u20136 months old) and aged (up to 26 months old) wild-type mice, treated for varying durations. The 3xTg mice14<\/a> carried APPSwe and tauP301L mutant transgenes, as well as a PS1 knock-in mutation, and were in a hybrid C57BL\/6J and 129Sv\/Ev background. The J20 mice13<\/a> transgenic mice expressed a mutant form of the human amyloid protein precursor bearing both the Swedish (K670N\/M671L) and the Indiana (V717F) mutations (APPSwInd) in a C57BL\/6J background. For breeding, 10\u201320 females (all litter-mates derived from the same cross) were typically mated with 8\u201312 males (all litter-mates derived from the same cross). Mice were identified by numbered ear tags and were randomly selected for behavioural and histological analyses.<\/p>\n To assess treatment effects on disease onset and progression, animals were treated either before pathology emerged (5\u20136 months old for 3xTg; 3 months old for J20) or after pathology was established (starting at 9 months old for 3xTg and 17 months old for J20). To investigate age-related effects in wild-type mice, chronic treatments were initiated in adulthood (10\u201312 months old) and continued for 10\u201314 months during ageing. Experiments included both sexes, and results were consistent between males and females. The number and sex of animals used in each experimental group can be found in the Source Data file. Investigators remained blinded to genotypes and treatment conditions throughout data collection and analysis. No prior sample-size calculations were done, but the number of animals used was consistent with similar studies in the field.<\/p>\n Mouse diet<\/p>\n Li levels in the cortex were comparable between human NCI cases (RUSH cohort: range 0.52\u20136.0\u2009ng per g; non-RUSH cohort: range 0.89\u20139.94\u2009ng per g) and mice (wild type and J20: range 1.61\u20134.59\u2009ng per g). Similarly, serum Li levels in human NCI cases (range 1.53\u201310.41\u2009ng\u2009ml\u22121) overlapped with those in mice (wild type, J20, 3xTg: range 0.75\u20134.50\u2009ng\u2009ml\u22121), supporting the relevance of mouse models for studying the biological effects of lithium.<\/p>\n The regular mouse 5053 is a grain-based diet that does not allow Li levels to be manipulated experimentally. To obtain a Li-deficient diet, we used a standard, chemically defined mouse AIN-93M diet that is calorically and nutritionally equivalent to the 5053 diet and was formulated as a standard diet for laboratory rodents by the American Institute of Nutrition in 1993. We tested 5 samples of the regular mouse 5053 diet and 5 samples of the AIN-93M diet and found that the average Li content was 104.8\u2009ng per g in the 5053 diet and 103\u2009ng per g in the AIN-93M diet. The AIN-93M chemically defined diet was modified to exclude Li. The Li-deficient and control AIN-93M diets were formulated by Dyets. We measured Li levels in the Li-deficient diet and confirmed that Li was depleted by 92.0% relative to the chemically defined control diet. The abundances of the other 26 metals that we measured by ICP\u2013MS were identical (data not shown). The solid diets were irradiated before administration to animals. The diets were stored in closed plastic bags that were placed inside cardboard boxes (devoid of light) at \u221220\u2009\u00b0C for up to 4 months before administration to animals.<\/p>\n DNA extraction and genotyping by PCR<\/p>\n We collected about 0.5\u20131.0\u2009cm of mouse tails in clean Eppendorf tubes; 500\u2009\u03bcl of tail lysis buffer (10\u2009mM Tris pH 8, 100\u2009mM NaCl, 10\u2009mM EDTA, 0.5% SDS) containing 0.4\u2009mg\u2009ml\u22121 Proteinase K was added to each tube, and the tubes were incubated overnight in a 56\u2009\u00b0C water bath. The next day, 500\u2009\u03bcl of isopropanol was added to precipitate the DNA, and the tubes were shaken vigorously for 20\u2009s. Tubes were centrifuged for 10\u2009min at 18,000g and the isopropanol was carefully removed, avoiding the DNA pellet. We then added 70% ethanol and shook the tubes to wash the DNA pellet. We next centrifuged the tubes for 10\u2009min at 18,000g. We removed the ethanol and air-dried the DNA pellet for 2\u201316\u2009h. The DNA was resuspended in 100\u2009\u03bcl acetate-EDTA buffer and placed in a 56\u2009\u00b0C water bath overnight. To amplify DNA regions by PCR, we mixed 3\u2009\u03bcl of DNA sample with corresponding amounts of forward and reverse PCR primers, PCR master mix and nuclease-free water, and ran the reactions in a thermocycler. Sample loading dye was added to the PCR products and the samples were run on 1\u20133% agarose gels (prepared by dissolving agarose in TAE buffer, to which Gel Red was added to allow DNA visualization). We also loaded a 100-bp DNA ladder. Gels were visualized using a UV transilluminator.<\/p>\n Quantitative RT\u2013PCR<\/p>\n Total RNA was extracted from cells and tissues using TRIzol reagent (Invitrogen) followed by DNase treatment to remove genomic DNA contamination. Primers were obtained from Harvard\u2019s PrimerBank: for mouse Gsk3b, forward 5\u2032-TGGCAGCAAGGTAACCACAG-3\u2032 and reverse 5\u2032-CGGTTCTTAAATCGCTTGTCCTG-3\u2032; for mouse Gapdh, forward 5\u2032-CTTTGTCAAGCTCATTTCCTGG-3\u2032 and reverse 5\u2032-TCTTGCTCAGTGTCCTTGC-3\u2032. Real-time PCR was performed for 40 cycles. The specificity and purity of PCR and RT\u2013PCR products were confirmed by the presence of single-peak melting curves.<\/p>\n GSK3\u03b2 inhibitor treatment<\/p>\n Li-deficient and control 3xTg mice 12 months old, maintained on their respective diets for three months, were treated with the GSK3\u03b2 inhibitor CHIR-99021 or a vehicle control. A stock solution of CHIR-99021 was prepared in DMSO and diluted in 0.9% saline to a final concentration of 10\u2009mg\u2009ml\u22121, containing 2% DMSO. The solution was warmed to 70\u2009\u00b0C to ensure dissolution of the compound. Mice received intraperitoneal injections of CHIR99021 at a dose of 50\u2009mg per kg body weight, once daily for 14 consecutive days. Control animals received equivalent volumes of vehicle (2% DMSO in saline). All animals tolerated the treatment without visible abnormalities and were included in the analysis.<\/p>\n Blood chemistry<\/p>\n BUN and creatinine measurements were done by IDEXX Laboratories, using mouse serum samples. TSH levels in the mouse serum were assessed by ELISA (Elabscience, E-EL-M1153).<\/p>\n Behavioural testingOpen field<\/p>\n Mice were placed in an open field box (75\u2009cm\u2009\u00d7\u200975\u2009cm) and movements were tracked in real-time using TopScan Lite software (CleverSys) coupled to a camera. Each mouse was recorded for 10\u2009min, and the average speed and distance travelled were automatically recorded. Mice had no prior exposure to the open-field arena (spontaneous test). All behavioural experiments were performed by researchers who were blinded to the experimental conditions.<\/p>\n Morris water maze<\/p>\n To assess spatial learning and memory, we trained and tested mice in a large circular pool (1.1\u2009m in diameter) filled with 21\u2009\u00b0C water, which was rendered opaque by the addition of non-toxic white paint. We placed four distinct visual cues (representing different geometric shapes, patterns and colours) on each wall, to facilitate spatial orientation and the acquisition of spatial memory. Mice were given four training trials a day for 5\u20137 consecutive days. Each training trial lasted for 1\u2009min. Mice were trained to remember the location of a hidden platform that was submerged 2.5\u2009cm below the water surface. The location of the hidden platform (south-east) remained the same during the 5\u20137-day training period. If, after a 60-second trial, the animal failed to locate the platform, it was placed on the platform and allowed to remain on the platform for 15\u2009s. Mice were trained four times a day and entered the pool in a randomized order of rotating entrance points (compass directions N, S, E, W, NE and SW). During each training trial, the latency to find the hidden platform was recorded. Then, 24\u2009h after the last training trial, a probe trial was conducted. The platform was removed and mice entered the arena from the NW location (opposite from the platform). The number of entries in the target area (representing the area where the platform had been located during the training trials), the total time spent in the target area, as well as the time spent in all quadrants, and the swimming speed were recorded during the 60-s probe trial. We also conducted separate trials in which a visible platform (platform elevated above the water level, on which a small red flag had been placed) was presented. Mice were given several training sessions and the time (latency) to reach the visible platform was recorded. Mouse movements, as well as average speed, distance travelled, latency to reach a quadrant or target area and number of entries in the target area, were tracked in real time using TopScan Lite software (Clever Sys) and the different measures were automatically recorded. For measurements of learning (latency to reach the platform during the training trials), mice underwent repeated measurements (four measurements a day for 6\u20137 consecutive days).<\/p>\n Novel-object recognition<\/p>\n Mice were placed in the same open-field box with two novel identical objects for 10\u2009min and allowed to freely explore the identical objects. The next day, mice were reintroduced in the open-field box and presented with a novel object, as well as one of the two objects they explored the previous day. The mice were allowed to explore the objects for 10\u2009min and their movements were tracked in real time using TopScan Lite software (Clever Sys) coupled to a camera. The box and items were cleaned with 70% ethanol between mice. We automatically recorded the time each mouse spent exploring each object, on both day 1 (two identical objects) or day 2 (one novel object and one familiar object), and derived a novelty (discrimination) index, defined as the ratio of time spent exploring the novel object relative to the familiar one.<\/p>\n Y maze<\/p>\n Spontaneous alternation, which is a measure of spatial working memory, was assessed by allowing the mice to freely explore a Y-shaped maze for 8\u2009min. The Y maze consisted of 3 arms (each 40\u2009cm\u2009\u00d7\u20098\u2009cm\u2009x\u200915\u2009cm) at an angle of 120\u00b0 from each other. Mice typically preferred to investigate a new arm of the maze, rather than returning to one that was previously visited. Using TopScanLite software, we recorded each entry in one of the three arms (A, B and C) and then derived the percentage of total correct alternations over the 8-min duration of the trial. A correct alternation (triad) is a succession of entries into three different arms (A\u2013B\u2013C, A\u2013C\u2013B, B\u2013A\u2013C, B\u2013C\u2013A, C\u2013A\u2013B or C\u2013B\u2013A).<\/p>\n Mouse neuropathology<\/p>\n Mice were anaesthetized with isoflurane and carbon dioxide and then perfused with PBS at 4\u2009\u00b0C for 20\u2009min. Brains were rapidly removed and the two hemispheres were separated. One hemisphere was dissected into subregions (frontal cortex, temporal cortex, occipital cortex, hippocampus and cerebellum). Each subregion was placed in a separate Eppendorf tube, snap-frozen in liquid nitrogen and then stored in a freezer at \u221280\u2009\u00b0C. The second hemisphere was placed in 4% paraformaldehyde for 48\u2009h. The fixed brain was then processed for paraffin embedding, using standard procedures. Paraffin-embedded blocks were sectioned and 6-\u03bcm sections were mounted on glass slides and used for histological analyses.<\/p>\n Paraffin-embedded mouse brain blocks were sectioned and the sections were mounted on glass slides. We deparaffinized the sections by immersion in two xylene baths for a total of 10\u2009min, followed by a 5-min immersion in a 50% xylene:50% ethanol solution. The sections were then rehydrated by immersion in solutions of decreasing concentrations of ethanol (95%, 90%, 70% and 50%) and then placed in water. Sections then underwent antigen retrieval using the Diva decloaker (BioCare). Sections were blocked with 3% BSA, 3% fetal bovine serum (FBS) and 0.1% Triton X-100 in PBS for 45\u2009min at room temperature. Primary antibodies (Supplementary Table 16<\/a> has a list of antibodies used for immunolabelling) were diluted in 3% BSA, 3% FBS and 0.1% Triton X-100 in PBS. After overnight incubation at 4\u2009\u00b0C, sections were washed three times with PBS. Secondary antibodies, diluted in 3% BSA, 3% FBS and 0.1% Triton X-100 in PBS were either biotin-coupled (1:200; Vector Labs) or coupled to Alexa fluorophores (1:300, Invitrogen). After three 10-min washes with PBS, sections were mounted with Pro-Long anti-fade mounting medium with DAPI (Invitrogen) and then imaged using confocal microscopy. For the A\u03b2 labelling shown in Extended Data Fig. 1e<\/a>, we incubated sections with an anti-rabbit biotinylated IgG secondary antibody (VectorLabs) for 1\u2009h, followed by three washes in PBS (1\u2009min each) and the addition of avidin-streptavidin-HRP-coupled complex (1:200 in 2% BSA and 0.1% Triton X-100 in PBS; VectorLabs). After three washes with PBS, we added diaminobenzidine (DAB) substrate (prepared by dissolving DAB and peroxide tablets in PBS; Sigma-Aldrich) and incubated for several minutes, until a brown precipitate formed. Sections were then washed with PBS, dehydrated with increasing ethanol concentrations (50%, 70%, 90%, 95% and 100%), followed by incubation with a 50% ethanol:50% xylene solution and two immersions in 100% xylene (5\u2009min each). Sections were mounted with a hydrophobic mounting medium (Permount). For Thioflavin S staining, after deparaffinization the brains were incubated with filtered 1% aqueous Thioflavin-S for 8\u2009min at room temperature, then washed twice (3\u2009min each) in 80% ethanol, once in 95% ethanol (3\u2009min), three times in distilled water and finally mounted. For sections labelled by immunofluorescence, multiple confocal images were acquired using an Olympus Fluoview Confocal Microscope FV3000. For DAB-stained sections, we acquired pictures using a bright-field microscope coupled with a camera.<\/p>\n For analysis of the A\u03b2 plaque burden, pictures of A\u03b2 immunoreactivity (using the rabbit anti-A\u03b2 monoclonal antibody, clone D54D2, Cell Signaling, 8243, dilution 1:250) in the hippocampus were processed using a macro developed for use with Fiji\/ImageJ 2.9.0. In brief, confocal pictures were all saved in the same folder and were all automatically opened in Fiji and processed serially. For each picture, the background was subtracted (rolling ball radius was set for 25). Pictures then underwent de-noising, using a Gaussian blur filter (radius of one pixel). The images were then thresholded using the Default Fiji threshold set at 120. Particles with a minimal size of 5\u2009\u03bcm2 were retained and their number, average size and mean fluorescence intensity were automatically recorded for each picture in an Excel file. To calculate the A\u03b2 plaque burden, the total area occupied by A\u03b2 plaques was divided by the area of the selection. Three coronal sections (6\u2009\u03bcm thick) were sampled for each animal, in the rostral, intermediate and ventral hippocampus. Two 20\u00d7 images were acquired per section, using an Olympus FluoView LV1000 confocal microscope. The average A\u03b2 burden was obtained by averaging the A\u03b2 plaque density (area occupied by A\u03b2 plaques divided by the total area analysed) in all pictures acquired for each animal.<\/p>\n For analysis of tau pathology, pictures of p-Ser202 tau (CP13, dilution 1:150) or p-Ser396\/Ser404 tau (PHF1, dilution 1:200) immunoreactivity in the hippocampus CA1 were processed using a macro developed for use with Fiji\/ImageJ 2.9.0. In brief, confocal pictures were all saved in the same folder and were all automatically opened in Fiji and processed serially. Pictures underwent de-noising, using a Gaussian blur filter (radius of one pixel). The images were then thresholded using the Default Fiji threshold set at 150. The number of tau-positive neurons in the selected CA1 area was then manually counted for each thresholded picture and the area was measured. For each picture, we calculated the average density of tau-positive neurons (the total number of tau-positive neurons divided by the area of the region). The average tau-positive neuron densities were calculated for each animal by averaging all the pictures acquired.<\/p>\n Fluorescent image analysis was also performed using MetaMorph NX 2.5 (Meta Series, Molecular Devices). Mean fluorescence intensity for specific markers was quantified in each animal either in the nucleus (\u03b2-catenin) or across the entire cell body (for example, GSK3\u03b2, pSer9-GSK3\u03b2, pTyr216-GSK3\u03b2, GPNMB and LPL), based on co-labelling with cell-type-specific markers (such as MAP2, aspartoacylase and Iba1). Between 50 and 300 cells per mouse were analysed, and background signal was subtracted. Synaptophysin and PSD-95 fluorescence intensities were quantified in the CA1 region of the hippocampus, and FluoroMyelin, MBP and SMI-312 intensities were measured in the corpus callosum, with background subtraction applied in all cases. Cell densities of Iba1+, CD68+, aspartoacylase+, PDGFR\u03b1+ and GFAP+ populations were also determined in relevant brain regions by quantifying 50\u2013500 cells per mouse. For each measurement, multiple images were acquired at 4\u00d7, 10\u00d7, 20\u00d7 or 40\u00d7 magnification per animal, spanning the region of interest. Values were averaged for each animal before statistical analysis. The following primary antibodies were also used: anti-aspartoacylase [N1C3-2] (GeneTex, GTX113389; rabbit polyclonal, dilution 1:200), anti-aspartoacylase (clone D-11; Santa Cruz Biotechnology, sc-377308, mouse monoclonal, dilution 1:50), anti-\u03b2-catenin (clone E247; Abcam, ab32572; rabbit recombinant monoclonal, dilution 1:250), anti-\u03b2-catenin (clone 1B8A1; PTGlab, 66379-1-Ig, mouse monoclonal, dilution 1:200), anti-CD68 (clone KP1; Abcam, ab955; mouse monoclonal, dilution 1:200), anti-GFAP (Sigma-Aldrich, G9269; rabbit polyclonal, dilution 1:200), anti-GSK3\u03b2 (clone 3D10; Novus Bio, NBP1-47470SS; mouse monoclonal, dilution 1:200), anti-pTyr216-GSK3\u03b2 (Millipore Sigma, SAB4300237; rabbit polyclonal, dilution 1:100), anti-pSer9-GSK3\u03b2 (Abcam, ab131097; rabbit polyclonal, dilution 1:100), anti-Iba1 (clone EPR16588; Abcam ab178846; rabbit recombinant monoclonal, dilution 1:2,000), anti-PSD-95 (clone K28\/43; Biolegend, 810401; mouse monoclonal, dilution 1:250), anti-synaptophysin (clone SY38; Millipore Sigma, mouse monoclonal, MAB5258-I; dilution 1:200), anti-neurofilament marker (pan axonal marker; clone SMI-312; Biolegend, 837904; mouse monoclonal, dilution 1:200), anti-GPNMB (clone 2B10B8; PTGlab, 66926-1-Ig; mouse monoclonal, dilution 1:200), anti-LPL (Novus Bio, AF7197-SP; goat polyclonal, dilution 1:200), anti-MAP2 (Phosphosolutions, 1099-MAP2; goat polyclonal, dilution 1:500), anti-MBP (clone D8X4Q; Cell Signaling, 78896; rabbit monoclonal, dilution 1:200) and anti-PDGFR\u03b1 (R&D Systems, AF1062; goat polyclonal, dilution 1:200). Secondary antibodies were used at a 1:300 dilution: donkey anti-goat Alexa 594 (Invitrogen, A-11058), donkey anti-rabbit IgG (H+L) Highly Cross-Adsorbed antibody, Alexa 488 (ThermoFisher Scientific, A21206<\/a>), donkey anti-mouse IgG (H+L) Highly Cross-Adsorbed antibody, Alexa 594 (ThermoFisher Scientific, A21203<\/a>), donkey anti-mouse Alexa 647 (Invitrogen, A-31571), donkey anti-rabbit Alexa 647 (Invitrogen, A-31573) and donkey anti-goat Alexa 488 (Invitrogen, A-11055).<\/p>\n Golgi labelling<\/p>\n The brains were processed and stained using the FD Rapid Golgistain Kit (FD Neurotechnologies, PK401) following the manufacturer\u2019s protocol with minor modifications. Immediately after dissection, the brains were fixed overnight in 4% PFA. After cryosectioning, free-floating sections of 100\u2009\u03bcm were shortly (10\u2009min) fixed in 4% PFA, then stained using the kit\u2019s reagents and mounted using a glycerin-containing medium. Then 12 dendrites per mouse were imaged in the hippocampus or the cortex using a confocal microscope. Dendritic spine density was quantified using Fiji software v.2.9.0.<\/p>\n Transmission electron microscopy<\/p>\n The 3xTg mice were fed either a Li-deficient diet (n\u2009=\u20098) or a control diet (n\u2009=\u20098) from 6 to 12 months of age. At the end point, mice were perfused with a fixative containing 2.5% glutaraldehyde and 2.5% paraformaldehyde in 0.1\u2009M sodium cacodylate buffer, pH 7.4 (Electron Microscopy Sciences, 15949). After perfusion, 1\u20132-mm3 brain sections were generated and post-fixed overnight at 4\u2009\u00b0C in fresh fixative. The corpus callosum was subsequently dissected and processed for embedding in TAAB Epon resin at the Harvard Electron Microscopy Core Facility. In brief, tissue was washed in 0.1\u2009M cacodylate buffer, post-fixed in 1% osmium tetroxide and 1.5% potassium ferrocyanide for 1\u2009h, rinsed in distilled water and incubated in 1% aqueous uranyl acetate for 1\u2009h. After two further water rinses, samples were dehydrated through graded ethanol (50%, 70%, 90% and twice in 100%, for 10\u2009min each) followed by 1\u2009h in propylene oxide. Samples were then infiltrated overnight in a 1:1 mixture of propylene oxide and TAAB Epon (Marivac), embedded in pure TAAB Epon the next day and polymerized at 60\u2009\u00b0C for 48\u2009h. Ultrathin sections (approximately 80\u2009nm thick) were cut on a Reichert Ultracut-S microtome, mounted on copper grids, stained with lead citrate and imaged using either a JEOL 1200EX or a Tecnai G2 Spirit BioTWIN transmission electron microscope. Images were captured using an AMT 2k CCD camera and saved in TIFF format. Quantification of myelin sheath thickness, axon diameter and g-ratio was performed using MetaMorph NX 2.5 software (Meta Series, Molecular Devices). A total of 1,376 axons (control group) and 1,396 axons (Li-deficient group) were analysed from eight randomly selected fields per animal (\u00d74,800 magnification) spanning the corpus callosum.<\/p>\n A\u03b2 detection by ELISA<\/p>\n Mouse endogenous A\u03b2x\u201340 and A\u03b2x\u201342 levels were measured using a previously established protocol67<\/a>. In brief, hippocampi or cortices were homogenized in 20 volumes (v\/w) of tissue lysis buffer consisting of 20\u2009mM Tris-HCl (pH\u20097.4), 1\u2009mM EDTA, 1\u2009mM EGTA and 250\u2009mM sucrose, supplemented with protease inhibitors (Roche) and 100\u2009\u03bcM AEBSF. Soluble A\u03b2 species were extracted from tissue homogenates by diethanolamine treatment. Mouse A\u03b2(x\u201340) and A\u03b2x\u201342 were quantified using Wako ELISA kits (292-64501 and 294-62501, respectively). The LOQs were 7.44\u2009pmol\u2009l\u22121 for A\u03b2(x\u201340) and 4.75\u2009pmol\u2009l\u22121 for A\u03b2x\u201342. Sample concentrations ranged from 26.21 to 52.98\u2009pmol\u2009l\u22121 for A\u03b2(x\u201340) and from 11.72 to 22.40\u2009pmol\u2009l\u22121 for A\u03b2x\u201342, all above the respective LOQs.<\/p>\n Assessment of microglial function in vitroMicroglial purification for cell-culture assays<\/p>\n Wild-type Li-deficient and control mice were transcardially perfused with 1\u00d7 cold PBS. The cortex and hippocampus were dissected and minced using a scalpel before transferring them to 5\u2009ml digestion buffer (2\u2009U\u2009ml\u22121 of Dispase II, 20\u2009U\u2009ml\u22121 DNase I, 10\u2009\u03bcM HEPES in HBSS without calcium or magnesium). Samples were incubated for 30\u2009min at 37\u2009\u00b0C on an orbital shaker. The tissue was then homogenized by successive trituration with a Pasteur pipette followed by a 1\u2009ml pipette. An equal volume of 1\u00d7 HBSS was added to the homogenate and the resulting mix was passed through a 70-\u00b5m cell strainer, then centrifuged at 300g for 10\u2009min at 4\u2009\u00b0C. Samples were resuspended with 40% Percoll (GE Healthcare, 17-0891-02) to remove myelin, and microglia were enriched using CD11b beads, as described above. Microglia were resuspended in pre-warmed media (DMEM\/F12 with 2% FBS, 100\u2009ng\u2009ml\u22121 IL-34, 50\u2009ng\u2009ml\u22121 TGF\u03b21, 25\u2009ng\u2009ml\u22121 M-CSF) and counted using a haemocytometer. Microglia were seeded into 96-well glass-bottom plates precoated with poly-l-ornithine (Sigma, P4957) at 5,000 cells per well using 100\u2009\u03bcl of medium. A half-medium change was performed on day 2 and the downstream assay was done on day 3.<\/p>\n Primary microglia were also purified using an alternative protocol. After perfusion, the cortex and hippocampus were dissected, placed in 3\u2009ml buffer (0.9% Hepes, 50\u2009mM NaCl, pH 7.4) and minced with small scissors for 4\u2009min. Then, 7\u2009ml Dispase buffer (2\u2009U\u2009ml\u22121 Dispase II in 0.9% Hepes, 50\u2009mM NaCl, pH 7.4) was added and the tissue was incubated for 1\u2009h at 37\u2009\u00b0C on an orbital shaker. The tissue was then homogenized by gently triturating with a 10\u2009ml pipette with a wide bore, to prevent cell shearing. The enzyme activity was halted by the 1:1 addition of 10% fetal bovine serum in PBS (10\u2009ml) at 4\u2009\u00b0C. The homogenate was passed through a 70-\u00b5m cell strainer to remove meninges and clumped cells. The homogenate was then spun for 10\u2009min at 1,000g and 4\u2009\u00b0C, and the supernatant was discarded. The pellet was resuspended in 6\u2009ml of 75% isotonic percoll in PBS (high percoll; GE Healthcare, 17-0891-02). Then 5\u2009ml of 35% isotonic percoll in PBS (low percoll) was added, followed by 4\u2009ml of PBS. The resulting discontinuous gradient was allowed to settle for 15\u2009min at 4\u2009\u00b0C. We then centrifuged the tubes at 800g for 45\u2009min at 4\u2009\u00b0C. We then aspirated the top (PBS-containing) layer and part of the upper percoll layer. The band containing microglia (approximately 1.5\u2009ml), situated at the interface between the 35% percoll and 75% percoll layers, was gently collected. Then 50\u2009ml of PBS was gently added and the tube was inverted 20 times. The microglia were then centrifuged at 1,000g for 10\u2009min. The supernatant was discarded and the pellet was resuspended in a pre-warmed (at 37\u2009\u00b0C) buffer containing 2% fetal bovine serum, 50\u2009U\u2009ml\u22121 penicillin and 50\u2009\u00b5g\u2009ml\u22121 streptomycin in RPMI medium.<\/p>\n The BV2 microglial cell line has been maintained in the Yankner laboratory for more than 20 years and stored long-term in liquid nitrogen at \u2013180\u2009\u00b0C. After revival, the BV2 cells were authenticated based on their characteristic microglial morphology (small, round to slightly elongated shape, clear cytoplasm and occasional short processes) as well as positive immunolabelling for the microglial markers CD11b and Iba1. Mycoplasma contamination testing was not done.<\/p>\n Microglial A\u03b2 uptake and degradation assays<\/p>\n Microglial A\u03b2 uptake and degradation assays were done as previously described68<\/a>. Human amyloid A\u03b21\u201342 was purchased from Anaspec (AS-20276). Next, 1\u2009mg of A\u03b21\u201342 peptide was dissolved in 80\u2009\u03bcl 1% NH4OH, followed by dilution with PBS to 1\u2009mg\u2009ml\u22121 (stock) and storage at \u221280\u2009\u00b0C. Oligomeric A\u03b21\u201342 was prepared by resuspending the stock solution in DMEM\/F12 to 500\u2009\u03bcg\u2009ml\u22121 (100\u2009\u03bcM) and overnight incubation at 4\u2009\u00b0C. On day 3, the medium was replaced with DMEM\/F12 containing 2% FBS and A\u03b242 oligomers diluted to a final concentration of 2\u2009\u03bcg\u2009ml\u22121 (0.4\u2009\u00b5M). To assess A\u03b242 uptake, cells were incubated for 3\u2009h at 37\u2009\u00b0C, followed by three washes with 1\u00d7 PBS and fixation with 4% PFA for 15\u2009min. To assess microglial A\u03b242 degradation, cells incubated with A\u03b242 oligomers for 3\u2009h were first washed three times with warm medium. The cells were then incubated with warm medium devoid of A\u03b242 for an extra 3\u2009h. They were then washed with 1\u00d7 PBS and fixed with 4% PFA for 15\u2009min. The fixed cells were washed twice with PBS and blocked with 2% BSA, 2% FBS, 0.1% Triton X-100 in PBS for 1\u2009h. Anti-Iba1 and anti-A\u03b2 (6E10) antibodies, diluted 1:500 in the blocking buffer, were then added and incubated overnight at 4\u2009\u00b0C. The next day, the cells were washed three times with PBS 1\u00d7 and incubated with secondary antibodies (1:300 in blocking buffer) at room temperature for 2\u2009h. Cells were washed three times with 1\u00d7 PBS, then mounted and analysed by confocal microscopy. We also assessed A\u03b242 uptake and clearance by microglia using fluorescently labelled (HiLyte Fluor 555) human A\u03b21\u201342 (AnaSpec, AS-60480). The fluorescently labelled A\u03b242 was directly added to the medium containing 2% FBS to reach a concentration of 2\u2009ng\u2009\u00b5l\u22121, and the uptake and degradation assays were conducted as detailed above.<\/p>\n Microglial stimulation and treatment with GSK3\u03b2 inhibitors<\/p>\n To assess cytokine release, primary microglia isolated from control and Li-deficient mice were treated with 50\u2009ng\u2009ml\u22121 LPS on day 2 for 16\u2009h followed by supernatant collection. Inflammatory cytokines were detected and measured using a mouse cytokine array kit (R&D Systems, ARY006). To assess the effects of GSK3\u03b2 inhibitors on microglial function, microglia were pretreated with 3\u2009\u03bcM CHIR99021 or 1\u2009\u03bcM of PF-04802367 on day 2 for 24\u2009h before A\u03b242 uptake and clearance or cytokine-detection assays.<\/p>\n snRNA-seqSample preparation<\/p>\n We performed snRNA-seq on the hippocampus of 12-month-old 3xTg mice that were fed a Li-deficient (n\u2009=\u20095 mice) or chemically\u2009defined control (n\u2009=\u20094 mice) diet for five weeks. Mice were transcardially perfused with ice-cold PBS at a speed of 6\u2009ml\u2009min\u22121 for 8\u2009min to repress the transcriptional response during the brain dissection and sample preparation69<\/a>,70<\/a>. Hippocampal tissue was dissected on ice and flash frozen in liquid nitrogen. Both frozen hippocampal tissues were subsequently thawed together in 500\u2009\u00b5l HB buffer (0.25\u2009M sucrose, 25\u2009mM KCl, 5\u2009mM MgCl2, 20\u2009mM Tricine-KOH, pH\u20097.8, 1\u2009mM DTT, 0.15\u2009mM spermine and 0.5\u2009mM spermidine) and homogenized with the tight pestle of a dounce homogenizer in the same HB buffer with the addition of 0.32% of IGEPAL (Sigma) (average of 25\u201330 times per sample) on ice. Subsequently, single nuclei were diluted to 9\u2009ml in HB buffer, passed through a 40-\u03bcm filter and separated from debris and multinuclei by iodixanol gradient centrifugation. Specifically, we prepared a 50% iodixanol solution by diluting 60% iodixanol (Optiprep density gradient medium, Sigma D1556) with diluent (150\u2009mM KCl, 30\u2009mM MgCl2, 120\u2009mM Tricine-KOH, pH7.8), and subsequently diluted them with HB buffer and supplemented with 0.04% BSA and 64\u2009U\u2009ml\u22121 RNasin Plus ribonuclease inhibitor (Promega, N2611) to prepare 40% iodixanol and 30% iodixanol. We layered 1\u2009ml of 40% iodixanol in the bottom, 1\u2009ml of 30% iodixanol in the middle and gently layered 9\u2009ml of the diluted nuclei suspension on top of the 30% iodixanol layer. The three layers were visually confirmed to be distinct and were subjected to 18\u2009min of 10,000g centrifugation. Single nuclei were carefully recovered from the 30% iodixanol layer in between the 30% and 40% interface. An aliquot was taken for trypan blue staining and visual inspection of nucleic morphology under a microscope, which showed a homogeneous size distribution and absence of major debris or doublets. The numbers of nuclei were determined initially by haemocytometer and subsequently confirmed with an automated counter. The remainder of the nucleic suspension was diluted for nuclei encapsulation and sequencing library preparation at the Harvard Single Cell Core, according to the 10X Genomics manual. The size and quality of the prepared libraries were confirmed on Agilent high-sensitivity TapeStation and the library was independently quantified by qPCR. The prepared libraries were sequenced by Nova-Seq S4 at the Harvard Biopolymers Facility, at an average coverage of 32,897 reads per nucleus (Supplementary Table 3<\/a>). Sequencing data and individual animal metadata have been deposited at the NCBI Gene Expression Omnibus GSE272344<\/a> and linked to BioProject PRJNA1136488.<\/p>\n Single-nucleus RNA-seq quality control<\/p>\n We aligned the demultiplexed raw sequencing reads to the mouse genome (mm10 from 10X Genomics) using Cell Ranger (v.6.1.2)71<\/a>, with the \u2013include-introns option, to account for nuclear pre-mRNAs. The generated counts table was loaded to Seurat (v.4)72<\/a> to generate Seurat objects. Cells with more than 10% of reads being attributed to mitochondrial transcripts were filtered out. Cells that expressed fewer than 200 features (low-quality cells) or more than 8,000 features (apparent doublets) were also filtered out. These thresholds were determined by visual inspection of the distribution of features among cells (Seurat, VlnPlot) and are generally consistent with previous reports57<\/a>,73<\/a>. The cells that passed quality controls were log-normalized using the NormalizeData function from Seurat with a scale factor of 10,000. Variable features were identified using the FindVariableFeatures function from Seurat with the vst selection method and 2,000 features. Data were scaled using ScaleData and principal component analysis (PCA) was performed using RunPCA with the identified variable features using Seurat. Nearest neighbours were found using FindNeighbors with dimension 1:30, which was determined by ElbowPlot following the Seurat manual. The number of clusters was determined by the FindClusters function using Seurat v.4. UMAP and TSNE were performed using RunUMAP and RunTSNE using Seurat with dimensions 1:30 and do.fast=TRUE parameter. Potential doublets were removed using DoubletFinder (v.3)74<\/a> following the default parameters. After filtration, cluster-specific markers were determined using the FindAllMarkers function (Seurat v.4), with parameters only.pos = F, min.pct = 0.25 and max.cell.per.ident = 500.<\/p>\n Cell-type-specific annotation and differential gene expression analyses<\/p>\n Cell types were identified by cross-referencing the transcriptome of each individual cell to the Mouse Cell Atlas75<\/a> and independently validated by confirming the expression of established cell-type-specific markers on a cluster-to-cluster basis. Violin and heat scatter plots to demonstrate the expression distribution of selected established markers were plotted using the VlnPlot and FeaturePlot functions of Seurat. Examples of cell-type markers include Slc17a7 (excitatory neurons), Prox1 (granule cells), Gad1 and Gad2 (inhibitory neurons), Mbp (oligodendrocytes), Aldoc, Aqp4 (astrocytes), Pdgfra (OPCs), Vtn (pericytes), Cx3cr1 and Tgfbr1 (microglia), Cldn5 and Flt1 (endothelial cells), Prlr and Folr1 (choroid plexus cells; Supplementary Fig. 4<\/a>). Cell type-specific abundance and differential gene expression analyses were performed on the main cell types (excitatory neurons, inhibitory neurons, granule cells, microglia, astrocytes, oligodendrocytes, OPCs and endothelial cells). Clusters of the same cell types were combined to increase the statistical power, as described previously69<\/a>,70<\/a>,76<\/a>, and clusters with mixed cell types (less than 80% homogenous) were removed for cell-type-specific analyses. The relative abundance of each cell type was computed by dividing the number of cells of the particular cell type by the total number of cells. Two-tailed unpaired Student\u2019s t-tests were done to determine whether there were significant differences in the relative abundance of each cell type between the control and Li-deficient conditions. DEGs were computed by the MAST77<\/a> test in Seurat. Genes expressed in fewer than 1% of the cells in each cell type were filtered out. All the DEGs (FDR\u20094. DEGs with FDR\u20092fold change|\u2009>\u20090.1 were used for Gene Ontology enrichment analyses using Metascape78<\/a> v.3.5.20240101. The heat scatter plot for DEGs (Fig. 3d<\/a>) was generated using the FeaturePlot function of Seurat.<\/p>\n Purified microglia RNA-seqPurification of microglia<\/p>\n Li-deficient or control wild-type and 3xTg mice were perfused transcardially using PBS at 4\u2009\u00b0C. The cortex and hippocampus were quickly dissected and pooled, then minced using a scalpel, before transferring to 5\u2009ml dissection buffer (HBSS without calcium and magnesium) and Protector RNase inhibitor (Sigma) at 4\u2009\u00b0C in a dounce homogenizer. Brain samples were dounced 20 times with a loose pestle and 10 times with a tight pestle. The cell suspension was passed through a pre-wetted 70-\u03bcm cell strainer into a pre-chilled 15\u2009ml tube. Cells were then spun down at 300g for 10\u2009min at 4\u2009\u00b0C. Cell pellets were resuspended in 10\u2009ml ice-cold 40% Percoll and centrifuged at 800g for 20\u2009min at 4\u2009\u00b0C. Myelin debris was removed by vacuum suction and the cell pellet was washed with 5\u2009ml ice-cold HBSS and spun again for 5\u2009min at 300g at 4\u2009\u00b0C. The pellets were resuspended in 180\u2009\u03bcl ice-cold MACs buffer (0.5% BSA, 2\u2009mM EDTA, Protector RNase inhibitor in PBS) with 20\u2009\u03bcl of CD11b microbeads (Miltenyi Biotec, 130-049-601) and incubated on ice for 15\u2009min. After incubation, 1\u2009ml of MACs buffer was added to samples and cells were centrifuged for 5\u2009min at 300g and 4\u2009\u00b0C. Microglia were then isolated using LS columns with QuadroMACS Separator following the manufacturer\u2019s instructions. In brief, LS columns were pre-washed three times with 3\u2009ml MACs buffer. Samples were resuspended in 500\u2009\u00b5l MACs buffer and transferred to LS columns, followed by three more washes with 3\u2009ml MACs buffer. Finally, microglia were released with 5\u2009ml MACs buffer (without EDTA), then used for RNA extraction.<\/p>\n Microglial RNA sequencing quality control and analysis<\/p>\n Total microglial RNA was extracted from MACs-purified microglia using RNAzol RT (Sigma, R4533). RNA extracted from each microglial preparation was quantified using an Agilent Tapestation 4200 instrument, with a corresponding Agilent Tapestation High Sensitivity RNA assay (5067-5579). The samples were normalized to 4\u2009ng of input in 9.5\u2009\u03bcl, and the polyadenylated mRNA was selected for using 3\u2032 SMART-Seq CDS Primer II A as part of the Takara SMART-Seq v.4 Ultra Low Input RNA (634894) workflow, which generated cDNA. From there, an Agilent Bioanalyzer High Sensitivity DNA assay (5067-4626) was used to quantify the cDNA concentration. Libraries were obtained using the Illumina NexteraXT kit (FC-131-1096). Adapter ligation, indexing and amplification were done subsequently as part of the same workflow. After amplification, residual primers were eluted away using KAPA Pure Beads (07983298001) in a 0.6\u00d7 SPRI-based clean-up. The resulting purified libraries were run on an Agilent 4200 Tapestation instrument, with a corresponding Agilent D5000 ScreenTape assay (5067-5588 and 5067-5589) to visualize the libraries and check the size and concentration of each library. Molarity values obtained from this assay were used to normalize all samples in equimolar ratio for one final pool. The pooled library was denatured and loaded onto a single lane of an Illumina NovaSeq 6000 S4 flow cell to generate 100-bp paired-end reads. The pool was loaded at 200\u2009pM (normalized to 1\u2009nM pre-denaturation), with 1% PhiX spiked in as a sequencing control. The base-call files were demultiplexed through the Harvard BPF Genomics Core pipeline and the resulting fastq files were used in subsequent analysis. Raw RNA-sequencing data in FASTQ format were subjected to quality assessment using FastQC (v.0.11.9) and sequencing reads were aligned to mouse genome (mm10) using a STAR aligner79<\/a> with the following options: –outFilterMismatchNmax 999 –outFilterMismatchNoverLmax 0.04 –alignSJDBoverhangMin 1 –alignSJoverhangMin 8 –outFilterMultimapNmax 20 –outFilterType BySJout –alignIntronMin 20 –alignIntronMax 1000000 –alignMatesGapMax 1000000. Microglia RNA-seq yielded an average of 100 million uniquely mapped reads for each sample, and gene expression levels were quantified using htseq-count80<\/a>. To reduce the computational burden and focus on biologically relevant genes, we initially prefiltered the count data. Genes were retained if they had at least five counts in at least three samples. To validate the purity of the isolated microglia, we determined that microglial markers (Csf1r, P2ry12 and Tmem119) were strongly enriched, whereas neuronal (Map2 and Nsg2), astrocytic (Gfap and Aldh1l1) and oligodendrocyte (Olig2 and Mog) marker genes were negligibly expressed (Supplementary Fig. 7b,c<\/a>). We also verified that markers of ex vivo microglia activation81<\/a> (Fos, Jun, Hspa1a and Zfp36) were minimally expressed in our samples (Supplementary Fig. 7b,c<\/a>). Differential gene expression analysis was done using DESeq2 (ref. 82<\/a>) to identify DEGs between Li-deficient and control microglia, with an adjusted P value cut-off of 0.05 (Supplementary Tables 10<\/a> and 11<\/a>). Upregulated and downregulated DEGs from Li-deficient wild-type and 3xTg microglia were further analysed for overlapping DEGs, and the overlapping DEGs were subjected to Gene Ontology enrichment analysis using Metascape78<\/a> v.3.5.20240101. Results are summarized in Fig. 4a<\/a> and provided in Supplementary Table 12<\/a>.<\/p>\n Ingenuity Pathway Analysis<\/p>\n Signalling pathway and molecular network analyses were done on DEGs identified from the snRNA-seq and microglia RNA-seq datasets (FDR\u200983<\/a>. Significantly enriched pathways and disease or function annotations were identified and ranked based on the FDR, calculated using a one-sided Fisher\u2019s exact test followed by a Benjamini\u2013Hochberg correction for multiple comparisons. To visualize the results, the top pathway-enriched DEGs were integrated into a signalling network using IPA\u2019s build and overlay function (Extended Data Fig. 4b,c<\/a>).<\/p>\n RNA-seq of hippocampus from 3xTg mice treated with LiORNA extraction<\/p>\n Twelve-month-old 3xTg mice that were treated with 4.3\u2009\u00b5M LiO (n\u2009=\u20099 females) or vehicle (water; n\u2009=\u20099 females) from 6 to 12 months of age were transcardially perfused with cold PBS 1\u00d7 and the hippocampi were rapidly dissected and snap frozen. The total hippocampal RNA was extracted using Trizol reagent (Ambion, 15596018) and purified using a Direct-zol RNA Mini Prep kit (Zymo Research, R2050) according to the manufacturer\u2019s instructions. RNA integrity and concentration were assayed using an Agilent 2100 Bioanalyzer instrument. All RNA samples had an RNA integrity number of more than 8.2.<\/p>\n RNA library preparation and sequencing<\/p>\n Libraries were prepared using Illumina TruSeq Stranded mRNA sample-preparation kits from 500\u2009ng of purified total RNA according to the manufacturer\u2019s protocol. The finished dsDNA libraries were quantified using a Qubit fluorometer, Agilent TapeStation 2200, and RT\u2013qPCR using a Kapa Biosystems library quantification kit according to the manufacturer\u2019s protocols. Uniquely indexed libraries were pooled and sequenced on an Illumina NextSeq 500 instrument with paired-end 75-bp reads by the Dana-Farber Cancer Institute Molecular Biology Core Facilities. Samples were pooled with multiple samples per lane and sequenced. There were two sequencing batches (batch 1, n\u2009=\u20095 mice per group; batch 2, n\u2009=\u20094 mice per group).<\/p>\n RNA sequencing quality control and quantification of gene expression<\/p>\n Quality control of sequencing reads (Supplementary Table 14<\/a>) was done using FastQC v.0.11.5 (https:\/\/www.bioinformatics.babraham.ac.uk\/projects\/fastqc\/<\/a>). Reads were aligned to the Mouse GRCm38 genome with GENCODE M21 gene models using STAR79<\/a> v.2.7.0f with options –outSAMunmapped Within –alignSJDBoverhangMin 1 –alignSJoverhangMin 8 –outFilterMultimapNmax 20 –outFilterType BySJout –alignIntronMin 20 –alignIntronMax 5000000 –alignMatesGapMax 5000000 –twopassMode Basic. The expression of genes was quantified as gene counts using STAR at the same time as alignment with option –quantMode GeneCounts.<\/p>\n Gene-expression normalization and covariate adjustment<\/p>\n Gene counts were input to edgeR. Genes were deemed expressed if at least n\u2009=\u20099 samples (where n is the group size) had more than one count per million (CPM). Genes not satisfying these criteria were removed, keeping the original library sizes. This filtering retained n\u2009=\u200914,862 expressed genes out of 55,536 annotated genes for subsequent analyses. Counts were then normalized using the TMM method in edgeR. Finally, log(CPM) values were calculated for analyses other than differential expression.<\/p>\n To adjust gene expression for covariates, we fit the linear regression model for each gene and cohort separately using lm() in R: gene expression\u2009~\u2009group\u2009+\u2009covariates, where gene expression is log(CPM), and using the group and covariates: factor, two levels: LiO and water (reference level), covariates (sequencing batch (factor, two levels)) and one RUV with residuals (RUVr) covariate (continuous). The final normalized and adjusted gene-expression values were derived from adding the regression residuals to the estimated effect of the group level to preserve the effect of the group on expression. These normalized and adjusted gene-expression values were used to perform gene\u2013gene regression analysis and gene\u2013gene group regression analysis, and to visualize gene expression.<\/p>\n To adjust for technical variation, we used the RUVr method84<\/a> implemented in the RUVSeq v.1.18.0 Bioconductor package. We performed a first pass edgeR analysis, up to and including the glmFit() step with the covariates listed above, excluding the RUVr covariates. Then we used residuals() with argument type\u2009=\u2009\u2018deviance\u2019 to obtain a matrix of deviance residuals. The specified number of unwanted factors (RUVr covariates) used in final analyses were then estimated by the RUVr function using log(CPM) expression values and the residuals. The number of unwanted factors was selected based on separation of groups in PC plots using normalized and adjusted gene-expression values and checking that histograms of differential expression P\u00a0values showed a uniform or anti-conservative pattern.<\/p>\n Differential expression and gene set enrichment analysis<\/p>\n Differential expression analysis between groups with covariate adjustment using the covariates listed above was performed for expressed genes using edgeR (estimateDisp, glmFit and glmLRT with default arguments) in R. Genes were considered differentially expressed if FDR\u200915 and Fig. 5f<\/a>.<\/p>\n GWAS-DEG enrichment analysis<\/p>\n Before doing the GWAS-DEG enrichment analysis, we converted the mouse gene symbols to their human orthologues, using a two-step process. First, we used the alias2SymbolTable function in the Limma R package85<\/a> v.3.58.1 to map any gene aliases to their corresponding main symbols. Subsequently, the resulting gene symbols were converted to human orthologues using the MGI orthologue table86<\/a>. If there were multiple mapping candidates, all possible conversions were applied. For example, if a mouse gene had multiple human orthologues, records with all the relevant human gene symbols were generated. This standardized gene nomenclature enabled cross-species comparisons in subsequent analyses.<\/p>\n To do the GWAS-DEG enrichment analysis for microglia isolated from Li-deficient mice, we used MAGMA87<\/a> v.1.10. The gene set of DEGs identified by microglia bulk RNA-seq analysis was used and the summary statistics from the GWAS catalogue AD (accession ID: MONDO_0004975)88<\/a> were obtained and formatted for MAGMA input. The GWAS catalogue AD contains GWAS records from multiple studies. If multiple records were found for the same variant, we retained the entry with the lowest P\u00a0value. Gene-set analysis was conducted using the default MAGMA settings, with multiple testing correction applied to account for the number of gene sets tested. Enrichment results were considered significant at a false discovery rate (FDR) of 0.05.<\/p>\n Overlap of mouse and human DEGs<\/p>\n To assess the overlap between DEGs derived from our transcriptomic analyses and DEGs derived from the analysis of human brain samples with varying degrees of AD pathology15<\/a> (Fig. 3d<\/a>), we first converted mouse gene symbols to their human orthologues, as described above. We matched the cell types analysed in our mouse studies with those analysed in humans15<\/a>. To assess the statistical significance of the overlap between the two DEG sets, we did a Fisher\u2019s exact test. To control for multiple comparisons arising from the analysis of different cell types and DEG directions (upregulated and downregulated), we adjusted the P\u00a0values using the Benjamini\u2013Hochberg procedure. Two sets of adjusted P\u00a0values were calculated (Fig. 3d<\/a> and Supplementary Table 8<\/a>): one set for the overlap of genes upregulated in both datasets (indicated in red), and another for the overlap of genes downregulated in both datasets (indicated in blue).<\/p>\n Proteome analysis by mass spectrometry<\/p>\n The proteomic analysis was done at the Harvard Center for Mass Spectrometry. Hippocampal homogenates from Li-deficient and control 3xTg mice containing an equal amount (100\u2009\u00b5g) of protein were reduced with 200\u2009mM tris[2-carboxyethyl] phosphine (TCEP) at 55\u2009\u00b0C for 1\u2009h, then alkylated with 375\u2009mM iodoacetamide at room temperature for 30\u2009min in the dark. Proteins were precipitated using the methanol\/chloroform\/water precipitation method and then digested with trypsin overnight at 37\u2009\u00b0C. TMT labelling of digested samples was done according to the manufacturer\u2019s instructions (ThermoFisher). In brief, TMT labelling reagents were dissolved with 41\u2009\u00b5l anhydrous acetonitrile, and an equal volume of TMT reagent mix was added to each sample. After incubation for 1\u2009h at room temperature, the reaction was quenched with 8\u2009\u00b5l of 5% hydroxylamine. Equal amounts of peptides from each sample were combined and dried in a SpeedVac. The peptides were then separated using an Agilent 1200 HPLC system with a PolyWAX LP column (PolyLC), 200\u2009\u00d7\u20092.1\u2009mm, 5\u2009\u03bcm and 300\u2009A running under electrostatic repulsion\u2013hydrophilic interaction chromatography (ERLIC) mode conditions. Peptides were separated across a 90-minute gradient from 0% buffer A (90% acetonitrile, 0.1% acetic acid) to 75% buffer B (30% acetonitrile, 0.1% formic acid) with 20 fractions collected by time. Each fraction was dried in a SpeedVac and resuspended in 0.1% formic acid solution before analysis by mass spectrometry. Each ERLIC fraction was submitted for a single liquid chromatography\u2013tandem mass spectrometry (LC-MS\/MS) experiment that was done on a Q Exactive HF-X High Resolution Orbitrap (Thermo Fisher) coupled with an Ultimate 3000 nanoLC (Thermo Fisher) at the Harvard Center for Mass Spectrometry. Peptides were first isolated on a trapping cartridge (300\u2009\u00b5m\u2009\u00d7\u20095\u2009mm PepMap Neo C18 trap cartridge, Thermo Scientific) before separation on an analytical column (\u00b5PAC, C18 pillar surface, 50-cm bed, Thermo Scientific). The LC gradient was as follows: 2\u201327% in mobile phase B (0.1% formic acid in acetonitrile) for 70\u2009min and increased to 98% mobile phase B for 15\u2009min at a flow rate of 300\u2009nl\u2009min\u22121. The mass spectrometer operated in data-dependent mode for all analyses. Electrospray-positive ionization was enabled with a voltage of 2.1\u2009kV. A full scan ranging from 400 to 1,600\u2009m\/z was done with a mass resolution of 12\u2009\u00d7\u2009104 and an automated gain control (AGC) target set to 1\u2009\u00d7\u2009106.<\/p>\n Proteomics quality control<\/p>\n The top three most intensive precursor ions from each scan were used for MS2 fragmentation (normalized collision energy of 32) at a mass resolution of 3.0\u2009\u00d7\u2009104 and an AGC of 1\u2009\u00d7\u2009105. The dynamic exclusion was set at 50\u2009s with a precursor isolation window of 1.2\u2009m\/z. Raw data were submitted for analysis in Proteome Discoverer 3.0 software (Thermo Scientific). The MS\/MS data were searched against the UniProt reviewed Mus musculus (mouse) database along with known contaminants, such as human keratins and common lab contaminants. Sequest HT searches were performed using the following guidelines: a 10-ppm MS tolerance and 0.02-Da MS\/MS tolerance; trypsin digestion with up to two missed cleavages; carbamidomethylation (+57.021\u2009Da) on cysteine, TMT 6-plex tags on peptide amino termini and lysine residue (+229.163\u2009Da) were set as static modification; oxidation (+15.995\u2009Da) of methionine was set as variable modification; minimum required peptide length was set to \u22656 amino acids. At least one unique peptide per protein group was required to identify proteins. Of 13,404 proteins identified, only n\u2009=\u20093,392 proteins were identified with high confidence (MS2 spectra assignment, FDR\u200989<\/a>,90<\/a>,91<\/a>. The sample labels were control (samples 1, 3, 5 and 7) and Li-deficient (samples 2, 4, 6 and 8) and were all 3xTg homozygous females, aged 15 months (treatment from 6 to 15 months of age). An ANOVA followed by Tukey\u2019s post-hoc test was used to assess differences in protein abundance between Li-deficient and control samples. P\u00a0values were adjusted for multiple comparisons using the Benjamini\u2013Hochberg method to control the FDR. Proteins with an adjusted P\u20097). The proteins identified with high confidence and included in the statistical analysis are listed in Supplementary Table 7<\/a>. All other proteins, identified with lower confidence, can be accessed from the files deposited at the ProteomeXchange Consortium through the PRIDE partner repository with the dataset identifier PXD063039<\/a>. This represents 21 files, including one .msf file (containing all search results: peptide-spectrum matches, peptide groups, protein groups, modifications, scores, FDR and metadata) and 20 .raw files (containing MS1\/2 spectra and metadata), one for each of the 20 fractions analysed by mass spectrometry.<\/p>\n Statistics and reproducibility<\/p>\n Statistical analysis was done using GraphPad software v.10.3.0 (507). The statistical tests used are noted in the figure legends. Throughout the paper, all tests are two sided and unpaired unless stated otherwise. A significance level of 0.05 was used to reject the null hypothesis. The sample size, age and sex of experimental animals, as well as the summary of each statistical test (including degrees of freedom, confidence intervals and P\u00a0values) can be found in the Source Data file. All animal experiments were done once per condition using biologically independent samples (individual animals), with group sizes indicated in the corresponding figure legends. Representative immunolabelling images shown in the figures are from one animal per group, selected from multiple animals that consistently showed similar results.<\/p>\n Reporting summary<\/p>\n