S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E NEUROSCIENCE Copyright © 2023 The Authors, some Adipose tissue coregulates cognitive function rights reserved; exclusive licensee American Association Núria Oliveras-Cañellas1,2,3,4†, Anna Castells-Nobau1,2,3,4†, Lisset de la Vega-Correa1,2,3,4†, for the Advancement Jessica Latorre-Luque1,2,3,4, Anna Motger-Albertí1,2,3, Maria Arnoriaga-Rodriguez1,2,3, of Science. No claim to Josep Garre-Olmo5, Cristina Zapata-Tona1,2, Clàudia Coll-Martínez6, Lluís Ramió-Torrentà4,6,7, original U.S. Government José Maria Moreno-Navarrete1,2,3, Josep Puig8, Francesc Villarroya3,9, Rafel Ramos4,10, Works. Distributed Verònica Casadó-Anguera11,12, Elena Martín-García11,12, Rafael Maldonado11,12, under a Creative Commons Attribution Jordi Mayneris-Perxachs1,2,3‡*, José Manuel Fernández-Real1,2,3,4‡* NonCommercial License 4.0 (CC BY-NC). Obesity is associated with cognitive decline. Recent observations in mice propose an adipose tissue (AT)–brain axis. We identified 188 genes from RNA sequencing of AT in three cohorts that were associated with perfor- mance in different cognitive domains. These genes were mostly involved in synaptic function, phosphatidyli- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 nositol metabolism, the complement cascade, anti-inflammatory signaling, and vitamin metabolism. These findings were translated into the plasma metabolome. The circulating blood expression levels of most of these genes were also associated with several cognitive domains in a cohort of 816 participants. Targeted mis- expression of candidate gene ortholog in the Drosophila fat body significantly altered flies memory and learn- ing. Among them, down-regulation of the neurotransmitter release cycle–associated gene SLC18A2 improved cognitive abilities in Drosophila and in mice. Up-regulation of RIMS1 in Drosophila fat body enhanced cognitive abilities. Current results show previously unidentified connections between AT transcriptome and brain func- tion in humans, providing unprecedented diagnostic/therapeutic targets in AT. INTRODUCTION The brain regulates adipose tissue performance. For instance, There is an increased awareness that peripheral tissues modulate sympathetic neuroadipose connections mediate lipolysis (9) and cy- brain function shaping different cognitive domains. Recent tokine production (10), while neuroimmune interactions are in- studies in rodents revealed that restoring glucose metabolism of volved in thermoregulation (11). Recent observations in rodents myeloid cells reversed cognitive decline in aging (1), whereas selec- show that adipose tissue also regulates brain function. Genetically tive gastrointestinal vagal afferent ablation impaired hippocampus- modified mice engineered to release the peptide NaKtide in adipo- dependent episodic and spatial memory in rats (2). The liver has cytes led to improved hippocampal memory through the inhibition also been proposed to play a major role in regulating feeding behav- of Na- and K-dependent adenosine triphosphatase (Na,K-ATPase) ior in mice (3). signaling in adipocytes (12). Visceral adipose NLR family Pyrin Obesity is associated with increased risk of cognitive impairment Domain Containing 3 (NLRP3) was linked to impaired memory (4, 5). An increased fatness in middle-aged adults (<65 years) pre- in obese mice via interleukin-1R1 (IL-1R1) on CX3CR1+ cells, dicts decreased cognitive abilities in late life, mainly memory, exec- leading to microglial activation (13). An increased abundance of utive functioning, and learning (6–8). However, the precise beige adipocytes in subcutaneous fat restored hippocampal synaptic mechanisms involved remain largely unknown. plasticity in mice through induction of anti-inflammatory cytokine IL-4, protecting from obesity-induced cognitive impairment (14). We have investigated in humans and animal models the mech- anisms involved in the bidirectional interactions between adipose 1 Department of Diabetes, Endocrinology and Nutrition, Dr. Josep Trueta University tissue gene expression and cognitive abilities. An RNA sequencing Hospital, Girona, Spain. 2Nutrition, Eumetabolism and Health Group, Girona Bio- (RNA-seq) analysis of adipose tissue identified genes associated medical Research Institute (IdibGi), Girona, Spain. 3CIBER Fisiopatología de la Obe- sidad y Nutrición (CIBERobn), Instituto de Salud Carlos III, Madrid, Spain. with different cognitive domains in discovery and different valida- 4 Department of Medical Sciences, School of Medicine, University of Girona, tion human cohorts. The gene expression of those genes most con- Girona, Spain. 5Department of Nursing (Serra-Hunter Professor), University of sistently associated with a battery of neuropsychological tests was Girona, Girona, Spain. 6Neuroimmunology and Multiple Sclerosis Unit, Department then targeted in the fat body of Drosophila and in adipose tissue of Neurology, Dr. Josep Trueta University Hospital, Girona, Spain. 7Girona Neuro- degeneration and Neuroinflammation Group, Girona Biomedical Research Insti- of mice to evaluate the possible functional implications. Last, we tute (IdibGi), Girona, Spain. 8Department of Radiology (IDI), Girona Biomedical studied the expression of several of these genes in peripheral Research Institute (IdIBGi), Dr. Josep Trueta University Hospital, Girona, Spain. blood mononuclear cells (PBMCs) in association with cognitive 9 Department of Biology, University of Barcelona, Barcelona. Spain. 10Vascular Health Research Group of Girona (ISV-Girona), Jordi Gol Institute for Primary traits in humans. Care Research (Institut Universitari per a la Recerca en Atenció Primària Jordi Gol I Gorina -IDIAPJGol), Girona, Spain. 11Laboratory of Neuropharmacology-Neuro- phar, Department of Medicine and Life Sciences, Universitat Pompeu Fabra (UPF), Barcelona, Spain. 12Hospital del Mar Medical Research Institute (IMIM), Bar- RESULTS celona, Catalonia, Spain. Adipose tissue gene expression is linked to several *Corresponding author. Email: jmayneris@idibgi.org (J.M.-P.); jmfreal@idibgi.org (J. cognitive domains M.F.-R.) To gain insights into the fat-brain axis, we performed a transcrip- †These authors contributed equally to this work. ‡These authors contributed equally to this work. tomics analysis (RNA-seq) of visceral adipose tissue (VAT) from 17 Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 1 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E subjects with severe obesity [body mass index (BMI) > 35 mg/kg2] were also overrepresented in the inhibitory control (Fig. 1F) and at- aged 28 to 60 years (table S1). Ten different cognitive tests assessed tention domains (fig. S2C). Semaphorin signaling and interactions the functional performance of three cognitive domains: memory, (Fig. 1F), which play an important role in adult neuronal plasticity executive function, and attention. The tests and their interpretation (25), were associated with inhibitory control, while alterations in are shown in Materials and Methods. complement activation were linked to executive function. We also Following normalization by trimmed mean of M value and cal- found an overrepresentation of pathway involved in inflammation culating weights to account for variations in precision between ob- associated with the different cognitive domains. Hence, alterations servations, we fitted robust linear regression models with empirical in both notch signaling and complement activation were signifi- Bayes moderation of the standard errors to identify gene transcripts cantly associated with working memory (Fig. 1D). associated with each cognitive test score controlling for age, sex, BMI, and years of education, all confounding factors know to Adipose tissue gene expression was associated with several affect cognitive function. There are other lifestyle-related factors cognitive functions later in life known to cause or contribute to cognitive function such as physical We next sought to validate these results performing an RNA-seq of activity, diet, or alcohol use (15). However, we found no significant both VAT and subcutaneous adipose tissue (SAT) in an indepen- associations of any of the 10 cognitive tests with physical activity dent longitudinal cohort of 22 subjects with severe obesity aged Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 (table S2), macronutrients, alcohol intake, glycemia, or hemoglobin 23 to 57 years that underwent the same battery of cognitive tests A1c (fig. S1). Note that all subjects had morbid obesity (table S1). (table S1). Patients from this cohort underwent the same battery We observed significant associations with gene transcripts in all of cognitive tests that patients in the discovery cohort. However, cognitive tests at Benjamini-Hochberg procedure for false discovery in this case, the cognitive testing was performed 2 to 3 years after rate (pFDR) of <0.05 (Fig. 1; fig. S2, A and B; and tables S3 to S12). the collection of adipose tissue samples. Once more, we did not find Among these genes, NUDT2, AMPH, UNC5B, and OAT were those any significant association of physical activity, diet, alcohol intake, with most significant associations across the different cognitive glucose, or hemoglobin A1c levels (fig. S1). Thus, models were ad- domains (significant in seven, six, six, and eight cognitive tests at justed for age, sex, BMI, and education years. We again found sig- pFDR of <0.05, respectively). The attributed action of the proteins nificant genes associated with all cognitive test at a pFDR of <0.05 encoded by these genes was concordant with their function in the both in VAT (Fig. 2, A to C, and tables S18 to S28) and in SAT (fig. central nervous system (CNS), but their expression in adipose tissue S3 and tables S29 to S37). Consistent with the previous results, en- in association with cognition is unprecedented. Variants in the richment analyses highlighted an overrepresentation of pathways NUDT2 gene have been recently associated with intellectual disabil- with important roles in the CNS (tables S38 to S40). Axon guidance, ity (16, 17). AMPH encodes for a protein highly concentrated in nervous system development, neuronal system, and tryptophan presynaptic terminals (18) and is one of the six proteins altered in (Fig. 2, D and E) were replicated as the pathways most significantly the cerebrospinal fluid of patients with mild cognitive impairment associated with attention (Fig. 2, D and E), while semaphorin sig- and Alzheimer’s disease (19). Amph in Drosophila is an important naling was associated with inhibitory control (Fig. 2F). The activa- postsynaptic protein that regulates the location of key components tion of the complement system in VAT was again strongly associated of glutamatergic signaling and dendrite dynamics, maturation, and with the performance in several cognitive domains (Fig. 2, D to F). synaptogenesis (20, 21). OAT is the gene that encodes for the In SAT (tables S41 to S43), pathways related to synaptic transmis- enzyme ornithine aminotransferase (OAT), which plays a key role sion, neuronal system, and one‑carbon metabolism were associated in transforming ornithine and arginine into glutamate and γ-ami- with both long-term (fig. S3, D and E) and short-term memory (fig. nobutyric acid as well as major excitatory and inhibitory neuro- S3, F and G). Complement activation in SAT was again among the transmitters, respectively (22). In addition, genes involved in pathways most significantly linked to attention (fig. S3, H and I). tryptophan metabolism such as KYNU, QPRT, and INMT were The gene encoding for Ezrin (EZR) was one with most associa- among the most significant genes associated with inhibitory tions across the different cognitive domains in VAT (significant in control (Fig. 1A), attention (fig. S2A), and executive function four tests at a pFDR of <0.05), being the most significantly associ- (Fig. 1B). Consistently, subjects with obesity had impaired ated gene with the digit span forward score (Fig. 2C and table S22). memory and inhibitory control through alterations in tryptophan In line with previous findings, complement-related genes were par- metabolism (23, 24). ticularly and strongly associated with inhibitory control (Fig. 2A). To facilitate interpretation and identify relevant adipose tissue This cognitive function was also associated with early immediate pathways associated with cognitive function, we next performed genes from the orphan nuclear receptor 4A family (NR4A1 and overrepresentation analyses mapping significant genes to the Reac- NR4A2). In particular, NR4A2 was the most significant gene asso- tome and WikiPathways databases included in the Consensus- ciated with the STROOPI test (which evaluates inhibitory control), PathDB. We found an overrepresentation of pathways with while NR4A1 was among the top genes associated with trail making important role in the CNS associated with measurements of inhib- test part B (TMTB) (executive function). In SAT, NR4A1 and itory control, working memory, immediate memory, and attention NR4A2 were negatively associated with long-term memory (fig. (Fig. 1, D to F; fig. S2, C to F; and tables S13 to S17). Executive func- S3A), while the complement genes were particularly associated tion was associated with axon guidance, nervous system develop- with attention (fig. S3C). A gene from the NR4A (NR4A3) had ment, synaptic vesicle pathway, and signaling by receptor tyrosine again the strongest negative association with inhibitory control kinases (Fig. 1D). The neuronal system and the serotonin and dop- (table S35). Last, AMPH was the gene most significantly associated amine neurotransmitter release cycles were the most significant with digit span forward test (table S31) and the total digit span test pathways associated with immediate memory (Fig. 1E) and atten- (table S33). tion. The axon guidance and the nervous system development Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 2 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Fig. 1. Associations of VAT gene expression and cognitive domains in the discovery cohort. Volcano plots of differentially expressed genes in the VAT associated with (A) the STROOP Interference tests, (B) the trail making test part B (TMTB), and (C) California verbal learning test immediate recall (CVLT_IR) scores in discovery cohort (IRONMET, n = 17) identified by limma-voom analysis controlling for age, BMI, sex, and education years. The log2 fold change associated with a unit change in the cognitive test score and the log10 P values adjusted for Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 multiple testing (pFDR) are plotted for each gene. Differentially ex- pressed genes (pFDR < 0.05) are colored in red and green indicating down-regulation and up-regulation, respectively. (D) Manhattan-like plot of pathways significantly associated (q < 0.1) with the TMTB in the VAT identified from a pathway overrep- resentation analysis mapping sig- nificant genes to the Reactome and WikiPathways databases. (E) Dot plot of pathways significantly associated (q < 0.1) with the CVLT_IR in the VAT identified from a pathway overrep- resentation analysis mapping sig- nificant genes to the Reactome database. Dots are colored by the q value. (F) Manhattan-like plot of pathways significantly associated (q < 0.1) with the STROOP Interference in the VAT identified from a pathway overrepresentation analysis mapping significant genes to the Reactome and WikiPathways data- bases. In the Manhattan-like plots, the bubble size represents the ratio of input genes that are annotated in a pathway (GeneRatio). CREB, aden- osine 30 ,50 -monophosphate re- sponse element–binding protein; tRNA, transfer RNA; NCAM1, Neural Cell Adhesion Molecule 1; TP53, Tumor Protein 53; PKA, Protein Kinase CAMP-Activated Catalytic Subunit Alpha. Adipose tissue genes linked to cognitive function are the subset of genes included mostly pathways with key roles in syn- involved in neuronal system and synaptic formation aptic function (Fig. 3, B and C; fig. S4, A and B; and table S45) and To gain further insights into the coincident associations between inflammation (tryptophan metabolism and complement activa- both cohorts, we performed an enrichment analysis considering tion). Mapping these genes to the Search Tool for the Retrieval of only those genes significantly associated with at least one cognitive Interacting Proteins/Genes (STRING) database to construct a gene- test in all three tissues (VAT from discovery and VAT and SAT from gene interaction network, a cluster of highly interconnected genes validation cohorts), and a total of 188 coincident genes were found mainly involved in neuronal system and synapse formation and (Fig. 3A and table S44). The most overrepresented pathways from maintenance (26) (DLG4, DLGAP1, DLG2, ATP1B1, HOMER1, Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 3 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 Fig. 2. Longitudinal associations of VAT gene expression at baseline and the scores in different cognitive domains later in life. Volcano plots of differentially expressed genes in the VAT at baseline associated with (A) the STROOP color word tests (STROOPCW), (B) the TMTA, and (C) the Forward Digit Span scores 2 to 3 years later in the validation cohort (INTESTINE, n = 22) identified by limma-voom analysis controlling for age, BMI, sex, and education years. The log2 fold change associated with a unit change in the cognitive test score and the log10 P values adjusted for multiple testing (pFDR) are plotted for each gene. Differentially expressed genes (pFDR < 0.05) are colored in red and green indicating down-regulation and up-regulation, respectively. (D) Manhattan-like plot of pathways significantly associated (q < 0.1) with the TMTA in the VAT identified from a pathway overrepresentation analysis mapping significant genes to the Reactome and WikiPathways databases. (E) Dot plot of pathways significantly associated (q < 0.1) with the STROOPCW and (F) the Forward Digit Span in the VAT identified from a pathway overrepresentation analysis mapping significant genes to the Reactome database. The x axis in the dot plots and the bubble size in the Manhattan-like plots represent the ratio of input genes that are annotated in a pathway (GeneRatio). Dots are colored by the q value. NAD, nicotinamide adenine dinucleotide; TYROBP, Transmembrane Immune Signaling Adaptor TYROBP; SLIT, Slit Guidance Ligand; ROBO, Roundabout Guidance Receptor. Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 4 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Fig. 3. Pathway analysis of differentially ex- pressed genes in common among the VAT and SAT in the discovery and validation cohorts (n = 188). (A) Venn diagram representing the overlap of significant genes associated with at least one cog- nitive test in VAT of the discovery cohort, the VAT of the validation cohort, and the SAT of the validation cohort. (B) Dot plot of significantly overrepresented pathways (q < 0.1) mapping common differentially expressed genes (n = 188) to the WikiPathways and (C) Reactome databases. The x axis in the dot plots represents the ratio of input genes that are anno- tated in a pathway (GeneRatio). Dots are colored by the q value. (D) Gene-gene interaction network constructed using common differentially expressed genes via the STRING database. The network nodes Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 are genes, and the edges represent the predicted functional interactions. The thickness indicates the degree of confidence prediction of the interaction. Functional gene clusters are colored on the basis of the Markov cluster algorithm (MCL) with an inflation parameter of 1.4. Only connected nodes are shown. A highly connected functional cluster (in red) was detected comprising genes with important roles in the CNS. (E) Enrichment map of the interrelation of significant pathways identified using an active sub- network oriented approach. Each color displays a cluster of related pathways using a threshold for kappa statistics = 0.35. The size of the nodes corre- sponds to its −log10(pFDR). The thickness of the edges between nodes corresponds to the kappa statistic between the two nodes. IP3, inositol 1,4,5- trisphosphate; IP4, inositol 1,4,5,6-tetrakisphos- phate; PI, phosphatidylinositol; TNF, tumor necrosis factor; NMDA, N-methyl-D-aspartate; CNV, copy number variations; L1CAM, L1 Cell Adhesion Molecule. NLGN3, SNCA, BCHE, RIMS1, SLC18A2, ADRB2, EZR, PODXL, overrepresented Reactome pathways (pFDR < 0.05). To reduce and NCAM2) was disclosed (Fig. 3D). complexity derived from this large number of pathways, we per- To obtain a better picture of the associated pathways, we lever- formed a hierarchical clustering to collapse redundant pathways aged the interaction information from the gene-gene interaction and ease the identification of mechanisms relevant to cognition network to identify distinct active subnetwork and then performed (table S46). Figure 3E shows the functional network of the clustered an active subnetwork–oriented enrichment analysis on these sub- pathways. Of 53 clusters, the most significant one included path- networks. Using this approach, we identified a total of 120 ways related to the neuronal system and synaptic function Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 5 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E (RIMS1, SLC18A2, DLG2, DLG4, DLGAP1, HOMER1, NLGN3, DIAPH3, DLG4, RHOBTB1, SGO2, and SRF), and the release of PTPRD, BCHE, and KCNN3) (Fig. 4A). This cluster was strongly neurotransmitters (RIMS1 and SLC18A2) (Fig. 4B). connected with three other cluster involved in the synthesis and me- To gain better insights into potential mechanisms linking tabolism of phospholipids, particularly phosphatidylinositols adipose tissue to cognition, we next performed a multiblock (PIPs) (PIK3C2B, PIK3C2G, SYNJ1, and PLEKHA6), the signaling Partial Least Squares (PLS) analysis to integrate these 188 adipose by Rho guanosine triphosphatase (GTPases) (ARHGAP23, tissue genes with the plasma levels of metabolites measured by nuclear magnetic resonance (NMR), high-performance liquid Fig. 4. Selected clusters from active subnetwork oriented pathway analysis of common sig- nificant genes among the VAT and SAT in the discovery and validation cohorts (n = 188) and integration with circulating me- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 tabolites. (A) Dot plot of enrich- ment analysis results performed on active subnetworks grouped by se- lected clusters. The x axis repre- sents the fold enrichment defined as the ratio of the frequency of input genes annotated in a pathway to the frequency of all genes annotated to that pathway. The dot size indicates the number of differentially expressed genes in a given pathway. Dots are colored by the –log10(pFDR), with red indi- cating high significance. (B) Gene- concept network depicting signifi- cant genes involved in enriched pathways from selected clusters. The dot size of the pathways rep- resents the −log10(pFDR). Pathways with the same color correspond to the same cluster. (C) Correlation circle plot for the integration of the adipose tissue genes, metabolites (NMR, HPLC-ESI-MS/MS in positive and negative mode), and cytokines and neurotoxic proteins using a multiblock PLS model in canonical mode. Strongly positively associated variables or groups of variables are projected close to one another on the correlation circle (<0° angle). The variables or groups of variables strongly negatively associated are projected diametrically opposite (<180° angle) on the correlation circle. Variables not correlated are situated <90° from one another. Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 6 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E chromatography–electrospray ionization tandem mass spectrome- between the expression levels of genes involved in the metabolism try (HPLC-ESI-MS/MS) in positive and negative modes and the cir- of vitamins in the adipose tissue, particularly TCN2 and ABCC1, culating levels of cytokines and neurotoxic proteins [IL-6, IL-4, Tau, and the circulating levels of retinoic acid (Fig. 4C). Notably, α-synuclein, and amyloid-β42 (Aβ42)]. The expression levels of RBP1, which encodes for cellular retinol binding protein 1 involved adipose tissue genes involved in phospholipid metabolism, such in the transport of retinol, had a strong negative association with the as PIK3C2B, PLA2G6, or SYNJ1, had a strong association with plasma levels of retinoic acid. Together, these results suggest that the several circulating phospholipids and lysophospholipids (Fig. 4C). adipose tissue could modulate cognitive performance through the Although lipids are essential to brain function and, in particular, metabolism of vitamin A. phospholipids are crucial for synaptic plasticity, neurotransmission, From the identified cluster of genes in Fig. 4B, SLC18A2 and and memory (27), the brain can synthesize a limited number of RIMS1 (regulating synaptic exocytosis 1) were those with the lipids. However, plasma fatty acids can cross the Blood-Brain largest number of significant associations across the different cog- Barrier (BBB) as nonesterified free fatty acids or esterified in lyso- nitive domains in the three tissues analyzed. SLC18A2 encodes for phospholipids such as lysophosphatidylcholine (LPC) (28, 29). VMAT2 (vesicle monoamine transporter member 2), a protein that Notably, some of the genes from the phospholipid metabolism transports cytoplasmic monoamine neurotransmitters (dopamine, cluster codified for phospholipase A2 (PLA2G6), which catalyzes norepinephrine, histamine, and serotonin) into presynaptic vesicles Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 the hydrolysis of the sn-2 position of glycerophospholipids to and has been linked to the pathophysiology of different neuropsy- yield nonesterified fatty acids and lysophospholipids. Hence, these chiatric and neurological disorders (39). The Drosophila and mouse phospholipids could be mediating the effects of adipose tissue on RIMS1 orthologues have shown to be a central component in the cognition. regulation of synaptic vesicle exocytosis and the neurotransmitter We also identified three highly interconnected clusters of genes release (40). Mice lacking Rim1α show severely impaired learning related to inflammatory response that included the complement and memory (41), and RIMS1 has recently been identified as the cascade (C4A and C4B) and the IL-4, IL-10, IL-13 interleukin sig- highest-ranked candidate gene in autism spectrum disorder (42). naling (IL18 and TIMP1). Chronic low-grade inflammation is a hallmark of obesity, and the association between obesity and cogni- Morbid obesity modulates the direction of the associations tive decline has recently been shown to be mediated by inflamma- between adipose tissue genes and cognition tion (30, 31). In addition, a complement-dependent synapse We sought to further validate these results using an additional elimination has been recently identified as a mechanism for cohort. Hence, we measured the expression levels of EZR, memory loss (32). In line with our findings, NLRP3 from the UNC5B, NUDT2, and NR4A2 in the VAT from a third independent VAT has recently shown to impair cognition in mice by activating validation cohort (ADIPOBRAIN cohort) consisting of n = 40 pa- microglial IL-1 receptor (7), whereas IL-4 has shown to mediate the tients with and without obesity aged 40 to 64 years old (Fig. 5A and protective effects of beige fat on obesity-induced cognitive impair- table S1) who underwent the Rey-Osterrieth complex figure ment by restoring synaptic plasticity in the hippocampus (14). Con- (ROCF) test 4 to 10 years later. The ROCF is one of the top 10 sistent with recent findings that inhibition of Na,K-ATPase tests used by neuropsychologists for the evaluation of visuospatial signaling resulted in improved cognitive function (12), we also iden- constructional ability, visual memory, and executive function tified a cluster containing the Na+/K+ ATPase transporting subunit (43). Consistent with our findings in the discovery cohort (includ- beta 1 gene (ATP1B1). Notably, the Na,K-ATPase signaling pathway ing only patients with morbid obesity), the expression of EZR requires the activation of PIP 3-kinase, and we identified a cluster of (Fig. 5B) and UNC5B (Fig. 5C) at baseline was strongly negatively pathways involved in PIP metabolism strongly associated with cog- associated with the ROCF copy scores in subjects with morbid nition (Fig. 4B). Last, the genes involved in the inflammatory cluster obesity (BMI > 35 kg/m2). Conversely, the expression of these (C4A, C4B, IL18, and TIMP1) had a strong positive association with genes was positively associated with executive function in subjects the circulating levels of IL-6 and a strong negative association with with a BMI of <35 kg/m2 (Fig. 5, F and G). The expression of the IL-4 levels in plasma (Fig. 4C). Therefore, these adipokines NUDT2 was positively associated with the ROCF copy performance could be released by the adipose tissue and mediate its effects on in subjects with morbid obesity (Fig. 5D) but not in subjects with a cognition. Accumulating evidence indicates that peripheral inflam- BMI of <35 kg/m2 (Fig. 5H), whereas we did not find any significant mation contributes to increased neuroinflammation (33) and could association with the expression of NR4A2 (Fig. 5, E and I). be a potential causal mechanisms involved in the obesity-related cognitive impairment (34, 35). Blood expression levels of key genes linked to cognitive Other significant clusters included the nuclear receptor tran- performance scription (NR4A1, NR4A2, and NR2F1) and the metabolism of vi- To further validate these findings and provide noninvasive bio- tamins (ABCC1, TCN2), particularly vitamin A (RBP1 and GPC4). markers of cognitive function as an alternative to adipose tissue bi- Recent studies have shown that NR4A2 activation enhances long- opsies, we measured the expression of some of the most promising term memory in young mice and ameliorates age-related memory genes in the circulation of n = 816 individuals from the general pop- impairments in old mice (36). Several studies have also suggested ulation aged ≥50 years taking part in the Aging Imageomics study that water-soluble vitamins (C, E, and Bs) may affect cognitive per- (Fig. 5J and table S1). Initially, we validated those genes with the formance by reducing reactive oxygen species generation and largest number of associations across the different neuropsycholog- proinflammatory mediators such as NLRP3 inflammasome (37). ical tests in the discovery cohort (NUDT2, AMPH, UNC5B, and Retinoic acid, the active form of vitamin A, is essential for regulating OAT) and validation cohort 1 (EZR and NR4A2). In agreement synaptic plasticity in regions of the brain involved in memory and with our findings, the circulating levels of NUDT2 were negatively learning (38). Notably, we found a strong positive association associated with executive function (Fig. 5K), while the circulating Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 7 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Fig. 5. Associations of expression levels of selected genes with cog- nition in a validation cohort, the Aging Imageomics cohort, and D. melanogaster. (A) Main baseline characteristics of the validation cohort 3 (ADIPOBRAIN). Scatter plots of the partial Spearman’s rank cor- relations (adjusted for age, BMI, sex, and education years) between ROCF copy scores and the VAT expression levels of (B) EZR, (C) UNC5B, (D) NUDT2, and (E) NR4A2 in patients with a BMI of >35 kg/m2 from the ADIPOBRAIN cohort. (F to I) The same associations but in patients Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 with a BMI of <35 kg/m2. (J) Main baseline characteristics of the vali- dation cohort 2 (Imageomics). Violin plots of the score in several cognitive tests and the tertiles (T1, T2, and T3) of the circulating expression levels of selected genes: (K) normalized STROOP color word test–color word (SCWT CW) versus NUDT2; (L) symbol digit test (SDT) versus AMPH; (M) total paired recall (TFP) versus OAT; (N) normalized digit span test (DST) backward versus UNC5B; and (O) normalized DST backward versus EZR. The ranked residuals after con- trolling for age, BMI, gender, and education level are plotted. Overall significance was assessed using a Mann-Kendall trend test. (P) Scheme of the RNA interference via the UAS-GAL4 system. (Q) Courtship conditioning paradigm. (R to W) The graphs display courtship condition- ing paradigm results of fat body promoter line w; C7-GAL4; UAS-Dcr-2 crossed with RNAi lines targeting (R) unc-5, (S) Datp, (T) Moe, (U) Hr38, (V) Amph, (W) Oat, and their corre- sponding genetic background con- trols (controls 1 and 2). Boxplots represent the CI of naïve (N) and trained (T) males. Significance of courtship suppression upon training was assessed with Kruskal-Wallis nonparametrical test and post hoc Dunn’s multiple comparison test. LI statistical significance was deter- mined with the nonparametrical bootstrap analysis with 10,000 iter- ations. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. ns, not significant. Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 8 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E expression of AMPH was negatively associated with processing we found that in subjects with morbid obesity (Fig. 6A), the expres- speed (Fig. 5L) and memory (assessed by the total free recall; sion of SLC18A2 at baseline was strongly negatively associated with Ptrend = 0.02). Increased levels of OAT and UNC5B tended to be as- the ROCF copy performance 4 to 10 years later, whereas it was pos- sociated with delayed memory (Fig. 5M) and executive function itively associated in subjects with a BMI of <35 kg/m2 (Fig. 6B). (Fig. 5N). Last, we also found consistent negative associations To demonstrate a possible causal effect of adipose tissue between the circulating expression of EZR and executive function- SLC18A2 on cognition, we studied two preclinical models. In a ing (Fig. 5O) and the quintiles of NR4A2 with memory (Ptrend mice model, we used an AAV2/8-RSV-GFP-Adp-miRNASLC18A2 = 0.027). vector that expresses a miRNA to selectively down-regulate slc18a2 gene expression in adipose tissue (Fig. 6C). Eight-week-old mice Down-regulation of candidate genes in the fat body of were fed a normal chow diet (ND; n = 23) or a high-fat diet Drosophila melanogaster improves cognition (HFD; n = 25) for 8 weeks (Fig. 6D). Three days after the start of We next assessed the causal effect of the adipose tissue expression of the diet, each dietary group was divided into two groups and inject- these candidate genes (AMPH, UNC5B, NUDT2, OAT, NR4A2, ed intravenously via tail vein either with the AAV2/8-RSV-GFP- NR4A3, and EZR) on cognitive function using the model organism Adp-miRNASLC18A2 vector or with saline. The expression levels D. melanogaster (fruit flies). A main advantage of Drosophila is the of green fluorescent protein (GFP) contained in the vector as a re- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 possibility of manipulating gene expression in a tissue-specific porter gene was significantly increased in both inguinal and mesen- manner by means of the UAS-GAL4 system (Fig. 5P) (44). To eval- teric white adipose tissue (iWAT and mWAT) of mice from the uate Drosophila cognition, we used the courtship conditioning par- vector-injected groups compared to the saline groups (fig. S5, A adigm to measure associative learning and memory capabilities (45) and B), thereby indicating that the vector was able to reach the . Male courtship behavior is conditioned by the exposure to nonre- adipose tissue. We also found a significant decrease in the solute ceptive premated females resulting in a suppression of courtship carrier family 18 member A2 (slc18a2) protein levels in the after a learning process. Male courtship rates can be assessed at dif- mWAT of virus-injected mice fed an HFD compared to their re- ferent times after training periods with premated females. Learning spective controls (fig. S5, C and D). Four weeks after injection, we is assessed immediately after the training period, and short-term assessed short- and long-term memory and locomotor activity. As memory is typically assessed 1 hour after training. The courtship expected, mice fed an HFD gained more weight in parallel to food index (CI; the percentage of time spent on courtship during an 8- intake compared with those fed an ND (Fig. 6E and fig. S6, A and min interval) of trained and of socially naïve male is used to calcu- B). There is good evidence that obesity affects cognition in rodents late the learning index (LI). A higher LI is indicative of a better (35). Consistently, mice fed an HFD had lower short- and long-term learning or short-term memory (Fig. 5Q). memory than mice fed an ND (Fig. 6, F and G). Down-regulation of We down-regulated Amph, unc-5, Datp, Oat, Hr38, and Moe (the slc18a2 in the mice adipose tissue reversed the memory impairment well-conserved orthologues of AMPH, UNC5B, NUDT2, OAT, induced by an HFD (Fig. 6, F and G). We observed no differences in NR4A2/NR4A3, and EZR, respectively) specifically in the fat body the exploration time between groups (fig. S6, C and D). Although of Drosophila, which is the equivalent of the adipose tissue in we found a diet effect on locomotion, down-regulation of slc18a2 humans. The fat body promoter line w; C7-GAL4; UAS-Dcr-2 was had no substantial effect (fig. S6, E to H). crossed with at least one RNA interference (RNAi) line targeting To further assess the role of adipose tissue SLC18A mRNA on each gene of interest (table S47). Learning capabilities were assessed cognition, we performed RNAi-mediated knockdown of Vmat immediately after training using the courtship conditioning assay. (the orthologue in Drosophila) using the w; C7-GAL4; UAS-Dcr-2 Down-regulation of unc-5, Datp, Hr38, and Moe lead to a significant promoter line for specific knockdown in the fat body (the equiva- decrease in learning capabilities compared to their corresponding lent of adipose tissue in humans). The courtship-conditioning par- genetic background controls in at least one of the RNAi lines adigm was performed, and short-term memory was assessed 1 hour tested (Fig. 5, R to U). Conversely, down-regulation of Amph lead after training. Males of Vmat-RNAi1were able to suppress court- to an improvement of learning capabilities for one RNAi line ship significantly better than the corresponding genetic background (Fig. 5V). Down-regulation of Oat in the fat body was lethal for control flies (Fig. 6H). Consequently, the LI was significantly in- one RNAi line or did not affect learning captivities when using a creased (Vmat-RNAi1: LI = 0.577, control 2: LI = 0.366, P = second RNAi line (Fig. 5W). Our findings demonstrate that specific 0.0064), indicating that Vmat-RNAi1 presented more short-term down-regulation of some of these genes in the fat body of Drosophi- memory (Fig. 6H). Next, we performed real-time polymerase la modulates learning, thereby suggesting a direct effect on cogni- chain reaction (RT-PCR) to prove that Vmat was specifically tion uniquely depending on its expression and/or function in knocked down in the fat body but not in the fly head (Fig. 6I), in- fat body. dicating that when decreasing Vmat expression exclusively in the fat body flies are able to learn better. To unravel a potential mechanism Neurotransmitter release cycle–associated genes SLC18A2 through which Vmat expression in the fat body could influence cog- and RIMS1 modulate cognition in mice adipose tissue or nition, we performed RT-PCR to evaluated the expression of ruta- Drosophila fat body baga (rut), dunce (dnc), amnesiac (amn), homer, CAMKll, and orb2 Among the genes associated with cognition in both VAT and SAT in Drosophila heads. These genes are part of the cyclic adenosine 50 - from the discovery and validation cohort 1, the cluster of genes in- monophosphate pathway, which have been demonstrated to influ- volved in the release of neurotransmitters (RIMS1 and SLC18A2) ence learning and short-term memory in the courtship-condition- (Fig. 4B) had the largest number of significant associations across ing assay (46). RNAi-mediated knockdown of Vmat in the fat body different cognitive domains. Therefore, we also validated these as- lead to a significant increase in memory related genes (rut, dnc, sociations in the third validation cohort (ADIPOBRAIN). Again, caMKll, and orb) in the fly head (Fig. 6J). To further validate our Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 9 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 Fig. 6. Associations of SLC18A2 with cognition in the validation cohort 3 and altered expression in preclinical models. Scatter plots of the partial Spearman’s correlations (adjusted for age, BMI, sex, and education years) between the SLC18A2 VAT expression and the ROCF copy scores in (A) morbid obese patients (BMI > 35 kg/ m2) and (B) patients with a BMI of <35 kg/m2 (ADIPOBRAIN cohort, n = 40). (C) Schematic representation of the AAV packaging plasmid used to down-regulate SLC18A2 gene expression. (D) Timeline of events and mice cognitive testing. Arrows indicate punctual events such as virus injection, tests, and euthanasia. (E) Weekly body weight (in grams). Means ± SEM; n = 12 normal diet + saline (ND-S), n = 11 normal diet + virus (ND-V), n = 13 high fat diet + saline (HFD-S), n = 12 high fat diet + virus (HFD-V). &&&P < 0.001 week effect; +++P < 0.001 diet effect, @@@P < 0.001 week × diet interaction, and ^^^P < 0.001 week × treatment interaction (three-way ANOVA). (F and G) Short-term (3 hours) and long-term (24 hours) memory using the novel object recognition test. Dots with the means ± SEM. **P < 0.01, ***P < 0.001 ND-S versus HFD-S, %P < 0.05, %%P < 0.01 ND-V versus HFD-S, and ##P < 0.01, ###P < 0.001 HFD-S vs HFD-V (two-way ANOVA). (H) Male short-term memory in the courtship-conditioning paradigm. Control 2 (w; C7-GAL4/+; UAS-Dcr-2/+) and Vmat-RNAi1 fat body–specific knockdown flies (w; C7-GAL4/+; UAS-Dcr-2/Vmat-RNAi1). (I) Relative gene expression of Vmat in fly heads and fat body and (J) rutabaga (rut), dunce (dnc), amnesiac (amn), homer, CAMKll, and orb2 in fly heads of Vmat-RNAi1 fat body–specific knockdown flies and its corresponding genetic background control. Means ± SEM. (t test: *P < 0.05, **P < 0.01, and ****P < 0.0001). Data are based on a minimum of five biological and two technical replicates. (C) and (D) were created with BioRender.com. results, we used an independent RNAi line against Vmat (Vmat- of RIMS1 was negatively associated with ROCF in subjects with RNAi2). When down-regulating Vmat in the fat body with severe obesity (Fig. 7A) but positively in subjects with a BMI of RNAi2, we observed a significant increase in short-term memory <35 kg/m2 (Fig. 7B). To prove a possible causal effect, we assessed (Vmat-RNAi2: LI = 0.650, control 1: LI = 0.437, P = 0.002) (fig. learning in Drosophila after down-regulation or overexpression of S7A) and a significant overexpression of dnc, CAMKll, and orb2 Rim (the well-conserved orthologue of RIMS1) in the fat body. A in the fly head (fig. S7B), confirming our previous data. significant decrease in CI after learning was observed in two inde- In the third validation cohort, VAT RIMS1 mRNA was also eval- pendent RNAi lines and corresponding genetic background con- uated. In line with the results obtained for SLC18A2, the expression trols, but we did not observe significant differences in the LI (fig. Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 10 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E Fig. 7. Associations of RIMS1 with cognition in the validation cohort 3 and Rim overexpression in Droso- phila fat body effects in learning and gene expression profiles. (A and B) Scatter plots of the partial Spearman’s rank correlations (adjust- ed for age, BMI, sex, and education years) between the VAT expression levels of RIMS1 and the ROCF copy scores in (A) morbid obese patients (BMI > 35 kg/m2) and (B) patients with a BMI of <35 kg/m2 from the ADIPO- BRAIN cohort (n = 40). (C) Rim over- expression in the Drosophila fat body and associations with learning. Diff- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 erences in CI between naïve and trained males were assessed with Kruskal-Wallis nonparametrical test and post hoc Dunn’s multiple com- parison test. LIs to assess either short- term memory (for Vmat) or learning (for Rim) were calculated from CIs as specified in Materials and Methods. Statistical significance was deter- mined with the nonparametrical bootstrap analysis with 10,000 repli- cates. ****P < 0.0001. (D) Rim relative expression assessed by quantitative RT-PCR (qRT-PCR) in fly brain or fat body in UAS-Rim fat body–specific overexpression flies and its corre- sponding genetic background control. Data are derived from a minimum of five biological and two technical replicates. (E) Dot plot of significantly (q < 0.1) overrepresented Reactome pathways, (F) GO biological processes (GO-BP), and (G) molecular functions (GO-MF) associated with the miss expressed genes in the fat body of UAS-Rim flies. (H and I) Gene- concept network depicting significant genes involved in enriched Reactome pathways [from (E)] and GO-GO-BP [from (F)]. (J) Overrepresentation analysis of pathways (q > 0.1) associ- ated with differentially expressed gene transcript in UAS-Rim fly heads based on the Reactome database. (K) Gene-concept network associated with significant genes involved in Re- actome pathways of (J). (L) Signifi- cantly overrepresented GO molecular functions of significant gene tran- scripts in fly heads (q > 0.1). HDL, high- density lipoprotein; HSF1, heat shock factor 1; IGF, insulin-like growth factor; IGFBP, insulin-like growth factor binding protein; NADP+, nicotinamide adenine dinucleotide phosphate; PCNA, proliferating cell nuclear antigen. Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 11 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E S7, C and D). However, when Rim was overexpressed (UAS-Rim) in phospholipid homeostasis has been found to be important for the the fat body, males were able to suppress courtship significantly regulation of dendrite morphogenesis (54), and lipid imbalance is better than the corresponding genetic background control flies used as a target to protect against Aβ42-induced cytotoxicity in Alz- (Fig. 7C), resulting in a significant increase in the LI (UAS-Rim: heimer’s disease (AD) fly model (55). LI = 0.539, control 3: LI = 0.237, P = 0.0001) (Fig. 7D). Our Overall, we have demonstrated that overexpression of Rim results indicate that flies that overexpress Rim exclusively in the uniquely in Drosophila fat body is sufficient to promote an incre- fat body are able to learn better. Next, we used RT-PCR to prove ment in learning and that the effect of Rim overexpression in the that Rim was uniquely targeted in the fat body but not in the fat body can mediate changes at the transcriptional level in the brain (Fig. 7D). Again, learning improvement was exclusively pro- adipose tissue itself and in peripheral tissues as the brain modulat- moted by Rim altered expression in the fat body and not due to a ing pathways associated with neuronal functioning, cellular oxida- “leakage” of the RNAi in the brain. tive balance and detoxification, and genes from the phospholipid To elucidate the potential mechanisms underlying the overex- metabolism. Together, these results suggest that altering Vmat or pression of Rim in learning, we performed an mRNA-seq in the Rim expression in the fat body mediates changes in gene expression fat body and fly heads of UAS-Rim flies that overexpressed Rim spe- in the head (brain) throughout a putative fat body-to-brain axis. cifically in the adipocytes. We identified the differentially expressed Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 genes between the flies overexpressing Rim and the corresponding genetic backgrounds in the fat body (table S48) and fly heads (table DISCUSSION S49). In the fat body, consistent with our results in humans, an en- While the brain regulates adipose tissue performance, here, we show richment analysis based on Reactome identified an overrepresenta- that the reverse is also plausible. We found that the expression of tion of pathways involved in vitamin metabolism and lipid genes linked to axon guidance, nervous system development, neu- metabolism, including the hydrolysis of LPC and the glutamate ronal system, tryptophan, and inflammation activity in human neurotransmitter release cycle (Fig. 7, E and H). It also identified adipose tissue were all linked to cognition in three different genes involved in other pathways involved in biological oxidations cohorts of subjects. and cytochrome P450 (Cyp4e1, Cyp4ac2, Cyp4ac3, Cyp4p1, and SLC18A2 expression in the brain has been linked to the patho- Cyp4p2). Notably, a gene ontology (GO) enrichment analysis iden- physiology of different neuropsychiatric and neurological disorders tified mainly biological processes participating in neurotransmitter (39). Similarly, decreased expression of Rim1α in the brain of mice regulation, in particular, the metabolism of biogenic amines (dop- led to impaired learning and memory (41) . For this reason, and amine, serotonin, and norepinephrine), which is strongly consistent given the associations of both SLC18A2 and RIMS mRNA in the with our findings in humans (Fig. 7, F and I). At the molecular func- adipose tissue with cognitive function, we selected these neuro- tion level, we identified several functions involved in monooxyge- transmitter release cycle–associated genes to test their possible nase and oxidoreductase activity (Fig. 7G). Notably, we also functional relevance. Knockdown of slc18A2 exclusively in identified several functions involved in the PIP transfer and phos- adipose tissue reversed the memory impairment induced by an pholipid transfer and transport. These results were replicated in the HFD in mice (Fig. 6, F and G). Down-regulation of Vmat in the fly heads. We found again an overrepresentation of pathways in- fat body of two different lines of Drosophila also led to improved volved in the metabolism of lipids (particularly phospholipids short-term memory. Overexpressing Rim [the orthologue of regu- and the hydrolysis of LPC), vitamins, and biological oxidations lating synaptic membrane exocytosis 1 (Rims1)] exclusively in the (table S50). An analysis based only on up-regulated genes also iden- fat body also led to an increment in learning abilities in Drosophila. tified alterations in DNA metabolism (Fig. 7, J and K), while a GO The altered expression of other genes linked to cognition in humans analysis identified once more molecular functions related to mono- (AMPH, UNC5B, NUDT2, NR4A2/NR4A3, and EZR) in the fat oxygenase and oxidoreductase activity (Fig. 7L). body of Drosophila again modified cognition of the latter. Oxidative damage and reduction of antioxidants are associated In the search of more accessible biomarkers, we evaluated the with a decline in cognitive function due to neurodeterioration (47). expression of most of these genes in PBMCs. PBMC mRNA levels Vitamins can act as potent antioxidant molecules protecting of adipogenic genes decreased after weight loss in subjects with neurons from oxidative stress and improving overall neuronal func- obesity (56), while an HFD-induced cognitive disruption led to al- tioning (48). Reducing oxidative stress in Drosophila has been asso- terations in the PBMC transcriptome in rats (57). Here, we show ciated with an increase in lifespan (49) and improves age-related that the expression of genes in PBMCs that were most associated memory impairment (47). Similarly, monooxygenase and oxidore- with cognition in adipose tissue (NUDT2, AMPH, UNC5B, OAT, ductase enzymes are contributing to the cellular levels of oxygen re- EZR, and NR4A2) were also linked to cognitive traits in a cohort active species, and cytochrome P450 family members have been of 816 subjects. found to be up-regulated in the presence of oxidative stress The main limitation of the current study is that it is only focused having a detoxification and antioxidant function in Drosophila on morbid obese subjects (BMI > 35 mg/kg2); hence, our results (50). Notably, low-grade inflammation is also associated with cellu- may not be generalizable to other populations. Findings should lar oxidative imbalance and one of the physiopathological mecha- be confirmed in longitudinal, larger samples of subjects to general- nisms behind obesity-associated syndromes (51). Phospholipids are ize these results to the general population. also abundant molecules in Drosophila brain, where they play essen- Together, these findings suggest a systemic developmental tial roles in cognition, by both supplying energy for brain activity program that changes in parallel in different tissues and cells, and particip4ating in membrane trafficking and neurotransmission being reflected in neuronal networks. Current findings hint at po- (52). Similar to mammals, main source of phospholipids in Droso- tential therapeutic targets: Peripheral administration of miRNAs phila are lipid droplets inside fat body cells (53). In Drosophila, targeting the adipose tissue and influencing cognition seems far Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 12 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E safer than targeting the brain. Current observations also provide in- Neuropsychological assessment formation about useful biomarkers in predicting cognitive decline The primary end point in all these cohorts was the study of cognitive and in monitoring response to therapy using accessible sources. function. In these subjects, the following tests were collected (58, 59). Digit span tests MATERIALS AND METHODS The digit span is a subtest of the Wechsler Adult Intelligence Scale- Clinical cohorts III (61). It is based on numbers and includes the forward and the In discovery cohort (IRONMET, n = 17) (58, 59), from January 2016 backward digit span tests. In the forward digit span test, the exam- to October 2017, a cross-sectional case-control study was undertak- inee repeats a number sequence in the same order as presented. This en in the Endocrinology Department of Dr. Josep Trueta University constitutes a measure not only of working memory but also of at- Hospital (Girona, Spain). In 17 morbidly obese patients (BMI > 35 tention. In the backward digit span task, the examinee repeats in mg/kg2) aged 28 to 60 years old, adipose tissue–stranded RNA-seq reverse order series of digits that became gradually longer. This is was analyzed. In validation cohort 1 (INTESTINE, n = 22) (58), an executive task particularly dependent on working memory. A consecutively, 22 morbidly obese (BMI > 35 kg/m2) subjects with higher score reflects a better working memory. In a standardization different degrees of insulin action (measured using hyperinsuline- sample of 394 participants (aged 16 to 89 years), the reliability co- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 mic-euglycaemic clamp) were recruited at the Endocrinology efficient was very high, ranging from 0.94 to 0.97 (62). Service of the Dr. Josep Trueta University Hospital to validate Trail making test adipose tissue RNA-seq analysis. The TMT is composed by two parts: TMTA and TMTB. The TMTA In both cohorts, all subjects were of Caucasian origin and report- (greater focus on attention) consisted of a standardized page in ed a body weight stable for at least 3 months before the study. Sub- which numbers 1 to 25 are scattered within the circles, and partic- jects were studied in the postabsorptive state. Exclusion criteria were ipants were asked to connect the numbers in order as quickly as pos- previous type 2 diabetes mellitus, chronic inflammatory systemic sible. Before starting the test, a six-item practice test was diseases, and acute or chronic infections in the previous month; administered to the participants to make sure they understood severe disorders of eating behavior or major psychiatric anteced- both tasks. A maximum time of 300 s was allowed before suspend- ents; neurological diseases, history of trauma or injured brain, lan- ing the test. The direct scores of TMTA were the time in seconds guage disorders; and excessive alcohol intake (≥40 g of OH/day in taken to complete each task. In the same way, the TMTB (greater women or 80 g of OH/day in men). Body fat composition was esti- focus on executive function) consisted of an alternating sequence mated using bioelectrical impedance analysis (BC-418, Tanita Cor- of numbered circles and letters (63, 64). In both tests, shorter poration of America, Illinois, USA). times to completion indicate better performance. This protocol was revised, validated, and approved by the ethics California verbal learning test-II committee of the Dr. Josep Trueta University Hospital. The purpose The California verbal learning test-II (CVLT) is used to assess verbal of the study was explained to participants, and they signed written learning and memory (65). It consists of five learning tests in which informed consent before being enrolled in the study. a list of words (list A) is presented and the subject is asked, imme- In validation cohort 2 (Imageomics, n = 816) (58, 59), the Aging diately after each presentation, to recall as much words as possible. Imageomics study is an observational study including participants Then, an interference list (list B) is presented, and the subject is from two independent cohort studies (MESGI50 and MARK). De- asked to repeat the same task. CVLT immediate recall score is a tailed description of the cohorts can be found elsewhere (60). result of the first five tests and provides information about the learn- Briefly, the MESGI50 cohort included a population aged ≥50 ing process. In the short delay test, the patient is asked to recall list years old, while the MARK cohort included a random sample of pa- A, free (CVLT short-delayed free recall) or with semantic facilitation tients aged 35 to 74 years with intermediate cardiovascular risk. El- (CVLT short-delayed cued recall). After 20 min, in which nonverbal igibility criteria included ages of ≥50 years, dwelling in the tasks are carried on, the subjects are asked to repeat a list of words community, no history of infection during the past 15 days, no con- that the examinee has previously presented without semantic facil- traindications for magnetic resonance imaging, and consent to be itation (the CVLT long-delayed free recall) to assess long-term informed of potential incidental findings. The Aging Imageomics memory. A higher score reflects a better memory function. About study protocol was approved by the ethics committee of the Dr. 50 min are necessary to administrate this test, and its reliability Josep Trueta University Hospital. ranges from 0.78 to 0.94 (66). In validation cohort 3 (ADIPOBRAIN, n = 40), all the samples STROOP test (Golden’s version) and data of the participants included in this study were provided by The STROOP test was used to assess cognitive flexibility, selective the FATBANK (platform promoted by the CIBEROBN and coordi- attention, inhibition, and information-processing speed. This nated by the IDIBGI Biobank) and integrated in the Spanish Na- version consists of three different parts: (i) 100 words (color tional Biobank Network. A total of 40 patients, ages 40 to 64 names) are printed in black ink, and the subject is asked to read years, were included. From these, 26 subjects had morbid obesity them as fast as possible; (ii) 100 “XXX” are printed in color ink (BMI > 35 mg/kg2), aged between 40 and 63. Patients had no sys- (green, blue, and red), and the subject is asked to name as fast as temic disease other than obesity, and all were free of any infection in possible the ink color; and (iii) 100 color names (from the first the month before the study. Liver disease [specifically tumor disease page) were printed in color ink (from the second page), the color and Hepatitis C Virus (HCV) infection] and thyroid dysfunction name and the ink color do not match, and the subject is asked to were specifically excluded. Adipose tissue samples were obtained name the ink color (and not to read the color name). The subject from VAT depots during elective surgical procedures. is given 45 s for each task. After the 45 s, the last item completed is noted, obtaining three scores: one for each part of the test (“W,” “C,” Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 13 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E and “CW”). The interference (“I”) index was also obtained from the following the manufacturer’s protocol for dual indexing. Image subtraction CW – CW0, where CW0 = (C × W)/(C + W). Standard analysis, base calling, and quality scoring of the run were processed administration procedures were followed as indicated in the test using the manufacturer’s software Real Time Analysis (RTA v3.4.4) manual (67). and followed by generation of FASTQ sequence files. Memory binding test RNA-seq reads were mapped against human reference genome The memory binding test consists of 16 category cues two word lists (GRCh38) using STAR software version 2.5.3a (69) with ENCODE containing 16 word items in each list. Both lists are learned and re- parameters. Genes were quantified using RSEM version 1.3.0 (70) called by controlled learning and cued and paired recall. The with default parameters and using the annotation file from memory binding test produces four principal measures, the total GENCODE version 29. Only protein-coding genes that were ex- paired recall, total free recall, total delayed free recall, and total pressed >1 count per million (cpm) in at least 10 samples were con- delayed paired recall. sidered for the association of gene expression with clinical variables. Symbol digit test The symbol digit test is a subtest of the Wechsler Adult Intelligence Circulating gene expression analysis (validation cohort 2, Scale-III. It includes a coding key that shows nine abstract symbols, Aging Imageomics) each paired with a number, and below the key, series of symbols Total RNA was purified from blood using PAXgene Blood RNA Kit Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 were presented. The participants were asked to write down the cor- (QIAGEN, Gaithersburg, MD). RNA concentrations were assessed responding numbers associated with the abstract symbols as quickly with Nanodrop ND-1000 Spectrophotometer (Thermo Fisher Sci- as possible. This task is a measure of information processing speed. entific, Wilmington, DE). Total RNA was reverse-transcribed to The number of correct substitutions during a 90-s interval was cDNA using High Capacity cDNA Archive Kit (Applied Biosys- scored, and higher numbers indicate better performance. tems, Darmstadt, Germany). Gene expression was assessed by General cognition RT-PCR using the LightCycler 480 Real-Time PCR System A composite cognitive score was computed as the sum of the stan- (Roche Diagnostics SL, Barcelona, Spain), using SYBR green tech- dardized scores of each individual for each test included in the nology suitable for relative genetic expression quantification. Glyc- Aging Imageomics cohort. eraldehyde-3-phosphate dehydrogenase (GAPDH) was used as endogenous control. Adipose tissue collection and handling The commercially predesigned KiCqStart primers used were as Adipose tissue samples were obtained from SAT and VAT during follows: NR4A2, 50 -GACTATCAAATGAGTGGAGATG-30 elective surgical procedures (cholecystectomy, surgery of abdominal (forward) and 50 -GACCTGTATGCTAATCGAAG-30 (reverse); hernia, and gastric bypass surgery) (68). Both SAT and VAT OAT, 50 -CAGACCTGATATAGTCCTCC-30 (forward) and 50 - samples were collected from the abdomen, following standard CTTCTTCTA AAACCTCAAGGG-30 (reverse); amphiphysin procedures. Samples of adipose tissue were immediately (AMPH), 50 -CCTGACATAAAGAATCGCATC-30 (forward) and transported to the laboratory (5 to 10 min). The handling of 50 -CAAACACTTTCTGTGCTTTC-30 (reverse); nudix hydrolase 2 tissue was carried out under strictly aseptic conditions. Adipose (NUDT2), 50 -TTCTGCTTCGGTCCTTAG-30 (forward) and 50 - tissue samples were washed in phosphate-buffered saline, cut off AACTCAATTGCATTGTTGTC-30 (reverse); Unc-5 netrin receptor with forceps and scalpel into small pieces (100 mg), and immediate- B (UNC5B), 50 -AGGACAGTTACCACAACC-30 (forward) and 50 - ly flash-frozen in liquid nitrogen before stored at −80°C. GGGGATCTCCTGGTATTTG-30 (reverse); EZR, 50 - 0 TGACTTTGTGTTTTATGCCC-3 (forward) and 50 - VAT- and SAT-stranded RNA-seq (discovery IRONMET; CTCCTTTTCTT CTCTGTTTCC-3 (reverse); GAPDH, 50 - 0 validation cohort 1 INTESTINE) ACAGTTGCCATGTAGACC-30 (forward) and 50 -TTGAGCA VAT and SAT RNA purification was performed using RNeasy- CAGGGTACTTTA-30 (reverse). Tissue Mini-Kit (QIAGEN). Total RNA was quantified by Qubit RNA BR Assay kit (Thermo Fisher Scientific), and the integrity VAT expression of candidate genes by RT-PCR (validation was checked by using the RNA Kit (15NT) on 5300 Fragment An- cohort 3 ADIPOBRAIN) alyzer System (Agilent Technologies) (68). Total RNA was purified from VAT using RNeasy Lipid Tissue Mini The RNA-seq libraries were prepared with Illumina TruSeq Kit (QIAGEN, Izasa, SA). RNA concentrations were assessed with Stranded Total RNA Sample Preparation kit following the manufac- Nanodrop ND-1000 Spectrophotometer (Thermo Fisher Scientific, turer’s recommendations with some modifications. Briefly, in func- Wilmington, DE), and the integrity was checked by Agilent Bioana- tion of availability, 100 to 500 ng of total RNA was ribosomal RNA lyzer (Agilent Technologies, Palo Alto, CA). Total RNA was reverse- depleted using the RiboZero Magnetic Gold Kit and fragmented by transcribed to cDNA using High Capacity cDNA Archive Kit divalent cations. The strand specificity was achieved during the (Applied Biosystems, Darmstadt, Germany). Gene expression was second-strand synthesis performed in the presence of deoxyuridine assessed by RT-PCR using a LightCycler 480 Real-Time PCR triphosphate. The cDNA was adenylated and ligated to Illumina System (Roche Diagnostics SL, Barcelona, Spain), using TaqMan platform compatible IDT adaptors with unique dual indexes with technology suitable for relative genetic expression quantification. unique molecular identifiers (Integrated DNA Technologies) for Peptidylprolyl isomerase A (PPIA) and GAPDH were used as en- paired-end sequencing. The ligation products were enriched with dogenous control. The commercially available and prevalidated 15 PCR cycles, and the final library was validated on an Agilent TaqMan primer/probe sets used were as follows: SLC18A2 2100 Bioanalyzer with the DNA 7500 assay (Agilent Technologies). (Hs00996835_m1); RIMS1 (Hs01112189_m1), PPIA The libraries were sequenced on NovaSeq 6000 (Illumina) in a frac- (Hs99999904_m1), and GAPDH (4352934). The commercially tion of sequencing flow cell with a read length of 2 × 101 base pairs Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 14 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E predesigned KiCqStart primers used were as follows: NUDT2, EZR, 2.73 s with a Recycling Delay (RD) of 2 s, and the 90° pulse UNC5B, NR4A2, and PPIA. length was automatically calibrated for each sample at around 11.1 μs. For each sample, eight dummy scans were followed by Plasma HPLC-ESI-MS/MS metabolomics 256 scans and collected in 64,000 points over a spectral width of (discovery IRONMET) 20 ppm. TSP was used a general reference for NMR samples Metabolites were extracted from plasma samples with methanol because it does not introduce any additional signals apart from (containing phenylalanine-C13 as an internal standard) according the sharp methylsilyl resonance at 0 ppm. In addition, a high con- to previously described methods (58, 59, 71). Briefly, samples (30 μl) centration of TSP was used to release low–weight metabolites with of cold methanol were added to 10 μl of each sample, vortexed for 1 high affinity for serum proteins by binding competition with TSP. min, and incubated for one hour at −20°C. Samples were homoge- All 1H NMR spectra were recorded at 300 K on an Avance III 600 nized using FastPrep-24 (MP Biomedicals) and were incubated spectrometer (Bruker, Germany) operating at a proton frequency of overnight in a rocker at 4°C. Then, all samples were centrifuged 600.20 MHz using a 5 mm Broad Band Multinuclear (PABBO) gra- for 3 min at 12,000g, and the supernatant was recovered and filtered dient probe and automatic sample changer with a cooling rack with a 0.2 μm Eppendorf filter. Two microliters of the extracted at 4°C. sample were applied onto a reversed-phase column (Zorbax SB- Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 Aq 1.8 μm, 2.1 × 50 mm; Agilent Technologies) equipped with a Plasma cytokines and neurotoxic proteins precolumn (Zorbax-SB-C8 Rapid Resolution Cartridge 2.1 × 30 Commercially available enzyme-linked immunosorbent assay mm, 3.5 μm; Agilent Technologies) with a column temperature of (ELISA) kits were used to measure the plasma of IL-6 (RND- 60°C. The flow rate was 0.6 Ml/min. Solvent A was composed of HS600C, human IL-6 Quantikine HS ELISA kit, R&D Systems), water containing 0.2% acetic acid, and solvent B was composed of IL-4 (RND-HS400, human IL-4 Quantikine HS ELISA kit, R&D methanol containing 0.2% acetic acid. The gradient started at 2% B, Systems), α-synuclein (RND-DY1338-05, human α-synuclein increased to 98% B in 13 min, and held at 98% B for 6 min. Posttime DuoSet ELISA, R&D Systems), Aβ42 [RND-DAB142, human Aβ was established in 5 min. (amino acids 1 to 42) Quantikine ELISA kit, R&D Systems], and Data were collected in positive and negative electrospray modes tau (NBP2-62749, human tau ELISA kit, Novus Biologicals) follow- time of flight operated in full-scan mode at 50 to 3000 mass/charge ing the manufacturer’s protocol. ratio in an extended dynamic range (2 GHz), using N2 as the neb- ulizer gas (5 liters/min at 350°C). The capillary voltage was 3500 V D. melanogaster experiments with a scan rate of 1 scan/s. The ESI source used a separate nebulizer D. melanogaster stocks and maintenance for the continuous, low-level (10 liters/min) introduction of refer- Flies were raised on standard medium (cornmeal, sugar, and yeast) ence mass compounds 121.050873 and 922.009798, which were and cultured according to standard procedures at 28°C in a 12:12- used for continuous, online mass calibration. MassHunter Data hour light-dark cycle. Analysis software (Agilent Technologies, Barcelona, Spain) was Conditional RNAi lines targeting the genes of interest are sum- used to collect the results, and MassHunter Qualitative Analysis marized in table S47. v60000 (control 1) and v60100 (control 2) software (Agilent Technologies, Barcelona, Spain) was used to were obtained from Vienna Drosophila RNAi Centre (http:// obtain the molecular features of the samples, representing different, stockcenter.vdrc.at/control/main) and used as the corresponding comigrating ionic species of a given molecular entity using the Mo- genetic background controls for KK and GD RNAi line collections, lecular Feature Extractor algorithm (Agilent Technologies, Barcelo- respectively. The Rim overexpression line (UAS-Rim), stock na, Spain). We selected samples with a minimum of two ions. BL78050, referred to as UAS-Rim and the corresponding genetic Multiple charge states were forbidden. Compounds from different background control (BL36304; control 3) were obtained from the samples were aligned using a retention time window of 0.1% ± 0.25 Bloomington Drosophila Stock Centre (http://flystocks.bio. min and a mass window of 20.0 parts per million (ppm) ± 2.0 mDa. indiana.edu/). UAS-Rim was subsequently isomerized for six gener- We selected only those present in at least 50% of the samples of one ations into the (BL36304; control 3) genetic background. group and corrected for individual bias. RNAi expression in the fat body was induced by w; C7-GAL4; UAS-Dcr-2 driver line, which carries a fat tissue–specific promoter, Plasma 1H NMR metabolomics (discovery IRONMET) driving the expression of GAL4 specifically on Drosophila adiposi- Plasma samples were thawed at room temperature. For each sample, ties, was provided by M. Jindra (72). 400 μl of plasma were combined with 200 μl of phosphate buffer [9% Drosophila courtship conditioning (w/v) NaCl and 100% D2O] that contained 10 mM of 3-trimethyl- Drosophila courtship conditioning was used to assess Drosophila silyl-1-[2,2,3,3-2H4] (TSP) (58, 59). Samples were mixed with the learning and memory capabilities. Courtship conditioning was per- use of a vortex and centrifuged (10,000g) for 10 min. Then, a 550 formed as described previously (45). RNAi and overexpression lines μl aliquot was transferred into a 5 mm NMR tube before NMR anal- together with their corresponding genetic background controls ysis. 1H spectra of low–molecular weight metabolites were per- (table S47) were crossed with w; C7-GAL4; UAS-Dcr-2 driver line formed using a Carr-Purcell-Meiboom-Gill (CPMG) sequence and raised at 28°C to down-regulate or overexpress the correspond- (RD–90°–[t–180°– t]n–ACQ-FID) with spin-echo delay of 400 μs ing targeted genes in the adipose tissue. Males were collected at (for a total T2 filter of 210 ms) allowing an efficient attenuation eclosion. of the lipid NMR signals. The CPMG sequence generates spectra and kept in isolation until 7 days of age at 25°C. Males were ran- edited by T2 relaxation times, reducing broad resonances from domly assigned to either trained or naïve groups. Learning capabil- high–molecular weight compounds facilitating the observation of ities were assessed by pairing individual males with a single low–molecular weight metabolites. The total acquisition time was premated wild-type female for 2.5 hours during the training Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 15 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E period. Immediately after training, the males were tested to assess quality of the libraries were assessed using a High Sensitivity learning. Short-term memory was assessed by pairing individual DNA Bioanalyzer assay (Agilent Technologies). The libraries were males with a single premated wild-type female for 6 hours during sequenced on a NovaSeq 6000 (Illumina) in paired-end mode, with the training period, and, subsequently, the male was individually a read length of 2 × 51 base pairs, following the manufacturer’s pro- isolated for 1 hour before testing. Testing was performed by trans- tocol for dual indexing. Image analysis, base calling, and quality ferring male to a 1-cm-diameter chamber with in the presence of a scoring of the run were processed using the manufacturer’s soft- premated female and filmed for 8 min. The courtship behavior of ware, Real Time Analysis (RTA 3.4.4). RNA-seq reads were male flies toward females was quantified manually from videos. The mapped D. melanogaste reference genome (BDGP6.32) with mean CIs (defined as the percentage of time spent on courtship STAR/2.7.8a (ref1) with ENCODE parameters (69) with during an 8-min interval) of trained males and naïve males were ENCODE parameters. Gene quantification was performed with used to calculate the LI, which is defined as the per cent reduction RSEM/1.3.0 (70) with default parameters and using ensembl109 an- in mean courtship activity in trained males compared with naïve notation. Only protein-coding genes that were expressed >1 cpm in males; LI = (CInaïve − CItrained)/CInaïve. For all conditions, male- at least 10 samples were considered. female pairs were analyzed on a minimum of three different test days, and the data were pooled. Mice experiment Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 Nonparametrical statistical bootstrapping analysis was pre- Animals formed to assess differences in LI (learning and short-term Male wild-type C57BL/6J (Charles River, France) 8-week-old mice memory) between RNAi knockdown or overexpression lines and were housed individually in controlled laboratory conditions (21° ± their corresponding genetic background controls. The R script pro- 1°C, 55 ± 10%) with illumination at 12-hour cycles (lights on at 7:30 vided by Koemans et al. (45) was used to perform the bootstrapping a.m. and off at 7:30 p.m.). Animals were fed ad libitum for 8 weeks analysis. Briefly, CI values were randomly sampled with replace- either with an ND (~3.514 kcal/kg: 10% fat, 66% carbohydrate, and ment to generate 10,000 hypothetical LIs, which were used to deter- 24% protein; n = 23) or an HFD (~5.228 kcal/kg: 60% fat, 16% car- mine the 95% confidence interval of the difference between the bohydrate, and 24% protein; n = 25). ND and HFD were obtained index of the control and the index of the knockdown and the prob- from Altromin Spezialfutter GmbH & Co. (KG, Lage, Germany), ability that this difference is >0. t test or one-way analysis of variance references C1090-10, and C1090-60, respectively. Three days after (ANOVA) was performed to assess statistical differences in CI of the the start of the diet, weight-matched mice of each group (ND or compared conditions. HFD) were divided into two groups and injected intravenously mRNA extraction and cDNA synthesis via tail vein with 200 μl of either the AAV8-RSV-GFP-Adp-miR- Seven-day-old male were selected for quantitative RT-PCR (qRT- NASLC18A2 vector containing 1 × 1012 Genome Copies (GC) PCR) or RNA-seq. A total of 10 adult heads or 10 fat bodies were (Fig. 6C) in phosphate-buffered saline or with saline. Adeno-asso- collected per sample and transferred to RNAlater solution (Sigma- ciated virus (AAV) vector was produced by the Viral Production Aldrich). Total RNA was extracted using the Arcturus PicoPure Unit, UAB (www.viralvector.eu/). The AAV2/8-RSV-GFP-Adp- RNA Purification kit (Thermo Fisher Scientific). To avoid amplifi- miRNASLC18A2 vector expresses the GFP and an miRNA designed cation from genomic DNA, deoxyribonuclease treatment was per- to down-regulate SLC18A2 gene expression. It has a recombinant formed using the DNA-free Ambion kit, and RNA was reverse- AAV2/8 serotype that has been demonstrated to be effective at tar- transcribed into cDNA using the High-Capacity cDNA reverse geting adipose tissue (73). Adipose selectivity was enhanced by in- transcription (Applied Biosystems) according to the manufacturers’ corporating a proximal promoter and distal enhancer of the procedures. adiponectin gene into the AAV vector. To decrease vector expres- Quantitative real-time polymerase chain reaction sion in the liver, this AAV incorporates eight copies of the liver-spe- Quantitative PCRs were performed using the LightCycler 480 SYBR cific miR-122T into the 30 untranslated region of the expression Green Master (Roche) in a LightCycler 480 II machine (Roche). cassette (Fig. 6C) (73). Bodyweight and food intake were monitored Primer sequences and the reference genes RNApol2 and 1tub23cf weekly for the entire protocol. All experiments were performed are provided in table S51. For each condition, a minimum of five during the light phase. Mice were euthanized by decapitation, and biological replicates and two technical replicates were analyzed. Dif- mWAT was collected and frozen with dry ice. All procedures were ferential gene expression was calculated using the 2ΔΔCt method. conducted in strict accordance with the guidelines of the European The average threshold cycle (Ct) value for each sample was calculat- Communities Directive 86/609/EEC regulating animal experimen- ed and subtracted from the geometric mean Ct value of the reference tation and were approved by the local ethical committee (Comitè genes RNApol2 and 1tub23cf to calculate the ΔCt value. t test Ètic d’Experimentació Animal-Parc de Recerca Biomèdica de Bar- (GraphPad Prism version 5.00 for Windows) was used for calcula- celona, CEEA-PRBB, agreement no. 9687). tions of P values. Experimental procedure Behavioral tasks aimed to assess short- and long-term memory mRNA sequencing of fat body and fly heads (novel object recognition in V-maze), and locomotor activity (hor- The starting material for sequencing library preparation was total izontal and vertical activity assessment) was performed chronolog- RNA. The samples were quantified using the Qubit RNA BR ically as indicated in (Fig. 6D) using standardized tests. Assay kit (Thermo Fisher Scientific), and RNA integrity was esti- Novel object recognition test (V-MAZE) mated with the RNA 6000 Nano Bioanalyzer 2100 Assay (Agilent Object-recognition memory was assayed in a V-shaped maze (30 Technologies). The RNA-seq libraries were constructed using the cm long × 4.5 cm wide × 15 3 cm height each corridor) (74, 75). KAPA Stranded mRNA-Seq Illumina Platforms Kit (Roche), fol- On day 1 of the test, mice were habituated for 9 min to the maze. lowing the manufacturer’s recommendations. The size and Then, 24 hours later, mice were put back in the maze for 9 min, two Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 16 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E identical objects were presented, and the time that the mice spent Biotechnology, CA) at 4°C for 30 min. Cellular debris were elimi- exploring each object was recorded. The mice were again placed in nated by centrifugation at 13,000g for 15 min (4°C). Protein concen- the maze 3 hours (30 days after the AAV injection to assess short- tration was determined using the RC/DC Protein Assay (Bio-Rad term memory) or 24 hours (37 days after the AAV injection to assess Laboratories, Hercules, CA). For Western blot analysis, samples long-term memory) later for 9 min, with one of the familiar objects were resolved by SDS–polyacrylamide gel electrophoresis and trans- replaced with a novel object. The total time spent exploring each of ferred to a polyvinylidene difluoride membrane (Bio-Rad Laborato- the two objects (novel and familiar) was computed. A discrimina- ries, Hercules, CA). Membrane was exposed overnight at 4°C to tion index was calculated as the difference between the times spent primary antibody VMAT2/SLC18A2 (EB06558) at 1:250 dilution exploring the novel object divided by the total time exploring both (VWR International EuroLab, SL) diluted in 2% bovine serum objects. A higher discrimination index is considered to reflect albumin – 1× tris-buffered saline containing 0.1% Tween 20, follow- greater memory retention for the familiar object (76, 77). ing the recommendations of the manufacturer. After secondary an- Locomotion tibody incubation (anti-goat horseradish peroxidase), signal was On day 46 after AAV injection, mice underwent a 1-hour locomo- detected using enhanced chemiluminescence horseradish peroxi- tion test in individual locomotor activity boxes (10.8 × 20.3 × 18.6 dase substrate (Millipore) and analyzed with a luminescent image cm; Imetronic, Pessac, France) equipped with infrared sensors to analyzer ChemiDoc MP Imaging System (Bio-Rad Laboratories, Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 detect locomotor activity and an infrared plane to detect rearing Hercules, CA). TGX Stain-Free gels (#4568096, Bio-Rad Laborato- (76, 77). ries, Hercules, CA) were used as protein loading control. Statistical analysis Statistical analysis Differential expression gene analyses were performed on gene The number of animals (n) in each experimental condition is indi- counts using the “limma” R package (78). First, low expressed cated in figure legends. All graphs and statistical comparisons were genes were filtered, so that only genes with more than 10 reads in performed using GraphPad Prism software version 8.0 for Mac at least two samples were selected. RNA-seq data were then normal- (GraphPad Software, San Diego, CA, USA). To test whether the ized for RNA composition using the trimmed mean of M value as data were normally distributed, the Kolmogorov-Smirnov test was implemented in edgeR package (79). Normalized counts were then applied. Comparison of means between different groups was per- converted to log2 count per million with associated precision formed using two-way ANOVA, followed by Newman-Keuls mul- weights to account for variations in precision between different ob- tiple comparisons test as post hoc test. Three-way ANOVA was servations using the “voom” function with donor’s age, BMI, sex, performed when required to test the evolution over time. Results and education years as covariates. A robust linear regression are expressed as individual values with the means ± SEM, specified model adjusted for the previous covariates was then fitted to the in the figure legend. A P < 0.05 was used to determine statistical data using the “lmFit” function with the option method = significance. “robust,” to limit the influence of outlying samples. Last, an empir- RNA expression ical Bayes method was applied to borrow information between RNA purification was performed using RNeasy Lipid Tissue Mini genes with the “eBayes” function. P values were adjusted for multi- Kit (QIAGEN, Hilden, Germany), and concentrations were mea- ple comparisons using the pFDR. A cutoff for the pFDR of <0.05 sured using a Nanodrop ND-1000 Spectrophotometer (Thermo was used a threshold for statistical significance. The functional Fisher Scientific, Waltham, MA). Integrity was checked by roles of differentially expressed genes were characterized using over- Agilent Bioanalyzer (Agilent Technologies, Palo Alto, CA). Total representation analyses based on the Reactome and WikiPathways RNA was reversed transcribed to cDNA using High Capacity databases using ConsensusPathDB (80). Pathway significance was cDNA Archive Kit (Applied Biosystems, Darmstadt, Germany). assessed using a hypergeometric test, and a Storey procedure (q Gene expression was assessed by RT-PCR using a LightCycler 480 value) was applied for multiple testing corrections. Statistical signif- Real-Time PCR System (Roche Diagnostics SL, Barcelona, Spain), icance was set at q < 0.1. For differentially expressed genes simulta- using SYBR Green technology suitable for relative genetic expres- neously in the VAT and SAT of the discovery and validation cohorts, sion quantification. The RT-PCR reaction was performed in a data were further analyzed using functional gene-gene interaction final volume of 10 μl. The cycle program consisted of an initial de- networks mapping genes to the STRING database (which integrates naturing of 10 min at 95°C and then 40 cycles of 15 s denaturizing known and predicted protein/gene interactions) (81). Then, func- phase at 95°C and 1 min annealing and extension phase at 60°C. A tional local clusters in the interaction network were determined Ct value was obtained for each amplification curve and then a ΔCt using a Markov cluster algorithm (MCL) with an inflation param- was first calculated by subtracting the Ct value for endogenous eter = 1.4. Active interacting sources including text mining, exper- control from the Ct value for each sample. Fold changes compared iments, databases, coexpression, and cooccurrence and an with the endogenous control were then determined by calculating 2 interaction score of >0.4 were used to construct the interaction net- − ΔCt. β-Actin (Actb) was used as housekeeping. Actb and GFP works. In addition, we integrated the information provided from were analyzed by SYBR green technology using the following differential expression analysis, gene-gene interaction networks, primer sets: Actb, 50 -GATGTATGAAGGCTTTGGTC-30 and pathway overrepresentation analysis using the R package “path- (forward) and 50 -TGTGCACTTTTATTGGTCTC-30 (reverse); finder” with default parameters (82). First, significant genes were GFP, 50 -CAACAGCCACAACGTCTATATCATG-30 (forward) and mapped onto a STRING gene-gene interaction network. Then, 50 -ATGTTGTGGCGGATCTTGAAG-30 (reverse). active subnetworks of interconnected genes (including genes that Protein analysis are not significant themselves but connect significant genes) in Adipose tissue samples (100 mg) were homogenized using radioim- this gene-gene interaction network were identified. Last, separate munoprecipitation lysis buffer system (sc-24948, Santa Cruz pathway overrepresentation analyses based on the Reactome Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 17 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E databases were performed for each active subnetwork using the sig- 10. F. Cardoso, R. G. J. Klein Wolterink, C. Godinho-Silva, R. G. Domingues, H. Ribeiro, J. A. da Silva, I. Mahú, A. I. Domingos, H. Veiga-Fernandes, Neuro-mesenchymal units control ILC2 nificant genes in each of the active subnetworks and genes in the and obesity via a brain–adipose circuit. Nature 597, 410–414 (2021). Protein-protein Interaction Network (PIN) as the background 11. S. F. Morrison, K. Nakamura, Central mechanisms for thermoregulation. Annu. Rev. Physiol. genes. Pathway statistical significance was set at the pFDR of 81, 285–308 (2019). <0.05. Significantly enriched pathways were clustered via hierarchi- 12. K. Sodhi, R. Pratt, X. Wang, H. V. Lakhani, S. S. Pillai, M. Zehra, J. Wang, L. Grover, cal clustering. Genes in each pathway were used to calculate the B. Henderson, J. Denvir, J. Liu, S. Pierre, T. Nelson, J. I. Shapiro, Role of adipocyte Na,K- pairwise kappa statistics, a chance-corrected measure of cooccur- ATPase oxidant amplification loop in cognitive decline and neurodegeneration. iScience 24, 103262 (2021). rence between pathways. The distance matrix 1-kappa statistic was 13. D. H. Guo, M. Yamamoto, C. M. Hernandez, H. Khodadadi, B. Baban, A. M. Stranahan, used for agglomerative hierarchical clustering and a threshold of Visceral adipose NLRP3 impairs cognition in obesity via IL-1R1 on CX3CR1+ cells. J. Clin. 0.35 for the kappa statistics was used to define strong relation. Investig. 130, 1961–1976 (2020). Partial Spearman’s correlation analysis controlling for age, BMI, 14. D. H. Guo, M. Yamamoto, C. M. Hernandez, H. Khodadadi, B. Baban, A. M. Stranahan, Beige sex, and education level was used to determine the correlation adipocytes mediate the neuroprotective and anti-inflammatory effects of subcutaneous fat in obese mice. Nat. Commun. 12, 4623 (2021). between circulating levels of genes and the neurocognitive tests 15. M. Kivipelto, F. Mangialasche, T. Ngandu, Lifestyle interventions to prevent cognitive im- scores. As it works by obtaining the residuals of the ranked variables pairment, dementia and Alzheimer disease. Nat. Rev. Neurol. 14, 653–666 (2018). after removing the effect of the ranked covariates, scatter and violin Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 16. S. Anazi, S. Maddirevula, E. Faqeih, H. Alsedairy, F. Alzahrani, H. E. Shamseldin, N. Patel, plots were generated with the ranked residuals of the model adjust- M. Hashem, N. Ibrahim, F. Abdulwahab, N. Ewida, H. S. Alsaif, H. Al Sharif, W. Alamoudi, ing for selected covariates. Nonparametric monotonic trends ac- A. Kentab, F. A. Bashiri, M. Alnaser, A. H. Alwadei, M. Alfadhel, W. Eyaid, A. Hashem, A. Al cording to the gene tertiles or quintiles for the selected cognitive Asmari, M. M. Saleh, A. Alsaman, K. A. Alhasan, M. Alsughayir, M. Al Shammari, A. Mahmoud, Z. N. Al-Hassnan, M. Al-Husain, R. Osama Khalil, N. Abd Elmeguid, A. Masri, R. Ali, T. Ben- tests were assessed by the Mann-Kendall trend test. Omran, P. Elfishway, A. Hashish, A. Ercan Sencicek, M. State, A. M. Alazami, M. A. Salih, N. Altassan, S. T. Arold, M. Abouelhoda, S. M. Wakil, D. Monies, R. Shaheen, F. S. Alkuraya, Clinical genomics expands the morbid genome of intellectual disability and offers a high Supplementary Materials diagnostic yield. Mol. Psychiatry 22, 615–624 (2017). This PDF file includes: 17. F. Diaz, S. Khosa, D. Niyazov, H. Lee, R. Person, M. M. Morrow, R. Signer, N. Dorrani, A. Zheng, Figs. S1 to S7 M. Herzog, R. Freundlich, J. B. Birath, Y. Cervantes-Manzo, J. A. Martinez-Agosto, C. Palmer, Legends for tables S1 to S51 S. F. Nelson, B. L. Fogel, S. K. Mishra, Novel NUDT2 variant causes intellectual disability and polyneuropathy. Ann. Clin. Transl. Neurol. 7, 2320–2325 (2020). Other Supplementary Material for this 18. K. Takei, V. I. Slepnev, V. Haucke, P. De Camilli, Functional partnership between amphi- manuscript includes the following: physin and dynamin in clathrin-mediated endocytosis. Nat. Cell Biol. 1, 33–39 (1999). Tables S1 to S51 19. S. Bergström, J. Remnestål, J. Yousef, J. Olofsson, I. Markaki, S. Carvalho, J. C. Corvol, K. Kultima, L. Kilander, M. Löwenmark, M. Ingelsson, K. Blennow, H. Zetterberg, B. Nellgård, F. Brosseron, M. T. Heneka, B. Bosch, R. Sanchez-Valle, A. Månberg, P. Svenningsson, P. Nilsson, Multi-cohort profiling reveals elevated CSF levels of brain-enriched proteins in REFERENCES AND NOTES Alzheimer’s disease. Ann. Clin. Transl. Neurol. 8, 1456–1470 (2021). 1. P. S. Minhas, A. Latif-Hernandez, M. R. McReynolds, A. S. Durairaj, Q. Wang, A. Rubin, 20. Y. Yu, T. Niccoli, Z. Ren, N. S. Woodling, B. Aleyakpo, G. Szabadkai, L. Partridge, PICALM A. U. Joshi, J. Q. He, E. Gauba, L. Liu, C. Wang, M. Linde, Y. Sugiura, P. K. Moon, R. Majeti, rescues glutamatergic neurotransmission, behavioural function and survival in a Droso- M. Suematsu, D. Mochly-Rosen, I. L. Weissman, F. M. Longo, J. D. Rabinowitz, phila model of Aβ42 toxicity. Hum. Mol. Genet. 29, 2420–2434 (2020). K. I. Andreasson, Restoring metabolism of myeloid cells reverses cognitive decline in 21. C. Sheng, U. Javed, M. Gibbs, C. Long, J. Yin, B. Qin, Q. Yuan, Experience-dependent ageing. Nature 590, 122–128 (2021). structural plasticity targets dynamic filopodia in regulating dendrite maturation and syn- 2. A. N. Suarez, T. M. Hsu, C. M. Liu, E. E. Noble, A. M. Cortella, E. M. Nakamoto, J. D. Hahn, G. De aptogenesis. Nat. Commun. 9, 3362 (2018). Lartigue, S. E. Kanoski, Gut vagal sensory signaling regulates hippocampus function 22. A. Ginguay, L. Cynober, E. Curis, I. Nicolis, Ornithine aminotransferase, an important glu- through multi-order pathways. Nat. Commun. 9, 2181 (2018). tamate-metabolizing enzyme at the crossroads of multiple metabolic pathways. Biology 3. H. Endle, G. Horta, B. Stutz, M. Muthuraman, I. Tegeder, Y. Schreiber, I. F. Snodgrass, (Basel) 6, 18 (2017). R. Gurke, Z.-W. Liu, M. Sestan-Pesa, K. Radyushkin, N. Streu, W. Fan, J. Baumgart, Y. Li, 23. M. Arnoriaga-Rodríguez, J. Mayneris-Perxachs, O. Contreras-Rodríguez, A. Burokas, F. Kloss, S. Groppa, N. Opel, U. Dannlowski, H. J. Grabe, F. Zipp, B. Rácz, T. L. Horvath, J. A. Ortega-Sanchez, G. Blasco, C. Coll, C. Biarnés, A. Castells-Nobau, J. Puig, J. Garre-Olmo, R. Nitsch, J. Vogt, AgRP neurons control feeding behaviour at cortical synapses via pe- R. Ramos, S. Pedraza, R. Brugada, J. C. Vilanova, J. Serena, J. Barretina, J. Gich, V. Pérez-Brocal, ripherally derived lysophospholipids. Nat. Metab. 4, 683–692 (2022). A. Moya, X. Fernández-Real, L. Ramio-Torrentà, R. Pamplona, J. Sol, M. Jové, W. Ricart, 4. Y. Qu, H. Y. Hu, Y. N. Ou, X. N. Shen, W. Xu, Z. T. Wang, Q. Dong, L. Tan, J. T. Yu, Association of M. Portero-Otin, R. Maldonado, J. M. Fernández-Real, Obesity-associated deficits in in- body mass index with risk of cognitive impairment and dementia: A systematic review and hibitory control are phenocopied to mice through gut microbiota changes in one-carbon meta-analysis of prospective studies. Neurosci. Biobehav. Rev. 115, 189–198 (2020). and aromatic amino acids metabolic pathways. Gut 70, 2283–2296 (2021). 5. X. Tang, W. Zhao, M. Lu, X. Zhang, P. Zhang, Z. Xin, R. Sun, W. Tian, M. A. Cardoso, J. Yang, 24. M. Arnoriaga-Rodríguez, J. Mayneris-Perxachs, A. Burokas, O. Contreras-Rodríguez, R. Simó, J. B. Zhou, C. D. A. Stehouwer, Relationship between central obesity and the in- G. Blasco, C. Coll, C. Biarnés, R. Miranda-Olivos, J. Latorre, J. M. Moreno-Navarrete, cidence of cognitive impairment and dementia from cohort studies involving 5,060,687 A. Castells-Nobau, M. Sabater, M. E. Palomo-Buitrago, J. Puig, S. Pedraza, J. Gich, V. Pérez- participants. Neurosci. Biobehav. Rev. 130, 301–313 (2021). Brocal, W. Ricart, A. Moya, X. Fernández-Real, L. Ramió-Torrentà, R. Pamplona, J. Sol, M. Jové, 6. A. K. Dahl, L. B. Hassing, E. I. Fransson, M. Gatz, C. A. Reynolds, N. L. Pedersen, Body mass M. Portero-Otin, R. Maldonado, J. M. Fernández-Real, Obesity impairs short-term and index across midlife and cognitive change in late life. Int. J. Obes. (Lond) 37, working memory through gut microbial metabolism of aromatic amino acids. Cell Metab. 296–302 (2013). 32, 548–560.e7 (2020). 7. H. Gardener, M. Caunca, C. Dong, Y. K. Cheung, T. Rundek, M. S. V. Elkind, C. B. Wright, 25. D. Carulli, F. de Winter, J. Verhaagen, Semaphorins in adult nervous system plasticity and R. L. Sacco, Obesity measures in relation to cognition in the Northern Manhattan study. disease. Front. Synaptic Neurosci. 13, 672891 (2021). J. Alzheimers Dis. 78, 1653–1660 (2020). 26. T. C. Südhof, Neuroligins and neurexins link synaptic function to cognitive disease. Nature 8. M. C. Farruggia, D. M. Small, Effects of adiposity and metabolic dysfunction on cognition: A 455, 903–911 (2008). review. Physiol. Behav. 208, 112578 (2019). 27. V. García-Morales, F. Montero, D. González-Forero, G. Rodríguez-Bey, L. Gómez-Pérez, 9. W. Zeng, R. M. Pirzgalska, M. M. A. Pereira, N. Kubasova, A. Barateiro, E. Seixas, Y. H. Lu, M. J. Medialdea-Wandossell, G. Domínguez-Vías, J. M. García-Verdugo, B. Moreno-López, A. Kozlova, H. Voss, G. G. Martins, J. M. Friedman, A. I. Domingos, Sympathetic neuro- Membrane-derived phospholipids control synaptic neurotransmission and plasticity. PLOS adipose connections mediate leptin-driven lipolysis. Cell 163, 84–94 (2015). Biol. 13, e1002153 (2015). 28. R. W. Mitchell, G. M. Hatch, Fatty acid transport into the brain: Of fatty acid fables and lipid tails. Prostaglandins Leukot. Essent. Fatty Acids 85, 293–302 (2011). Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 18 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E 29. L. N. Nguyen, D. Ma, G. Shui, P. Wong, A. Cazenave-Gassiot, X. Zhang, M. R. Wenk, 56. A. Costa, B. Reynés, J. Konieczna, M. Martín, M. Fiol, A. Palou, D. Romaguera, P. Oliver, Use of E. L. K. Goh, D. L. Silver, Mfsd2a is a transporter for the essential omega-3 fatty acid do- human PBMC to analyse the impact of obesity on lipid metabolism and metabolic status: A cosahexaenoic acid. Nature 509, 503–506 (2014). proof-of-concept pilot study. Sci. Rep. 11, 18329 (2021). 30. K. Bourassa, D. A. Sbarra, Body mass and cognitive decline are indirectly associated via 57. M. Cifre, A. Palou, P. Oliver, Cognitive impairment in metabolically-obese, normal-weight inflammation among aging adults. Brain Behav. Immun. 60, 63–70 (2017). rats: Identification of early biomarkers in peripheral blood mononuclear cells. Mol. Neu- 31. Y. Yang, G. S. Shields, Q. Wu, Y. Liu, H. Chen, C. Guo, The association between obesity and rodegener. 13, 14 (2018). lower working memory is mediated by inflammation: Findings from a nationally repre- 58. J. Mayneris-Perxachs, A. Castells-Nobau, M. Arnoriaga-Rodríguez, M. Martin, L. de la Vega- sentative dataset of U.S. adults. Brain Behav. Immun. 84, 173–179 (2020). Correa, C. Zapata, A. Burokas, G. Blasco, C. Coll, A. Escrichs, C. Biarnés, J. M. Moreno-Nav- 32. C. Wang, H. Yue, Z. Hu, Y. Shen, J. Ma, J. Li, X. D. Wang, L. Wang, B. Sun, P. Shi, L. Wang, Y. Gu, arrete, J. Puig, J. Garre-Olmo, R. Ramos, S. Pedraza, R. Brugada, J. C. Vilanova, J. Serena, Microglia mediate forgetting via complement-dependent synaptic elimination. Science J. Gich, L. Ramió-Torrentà, V. Pérez-Brocal, A. Moya, R. Pamplona, J. Sol, M. Jové, W. Ricart, 367, 688–694 (2020). M. Portero-Otin, G. Deco, R. Maldonado, J. M. Fernández-Real, Microbiota alterations in 33. R. Dantzer, J. C. O’Connor, G. G. Freund, R. W. Johnson, K. W. Kelley, From inflammation to proline metabolism impact depression. Cell Metab. 34, 681–701.e10 (2022). sickness and depression: When the immune system subjugates the brain. Nat. Rev. Neu- 59. J. Mayneris-Perxachs, A. Castells-Nobau, M. Arnoriaga-Rodríguez, J. Garre-Olmo, J. Puig, rosci. 9, 46–56 (2008). R. Ramos, F. Martínez-Hernández, A. Burokas, C. Coll, J. M. Moreno-Navarrete, C. Zapata- 34. A. A. Miller, S. J. Spencer, Obesity and neuroinflammation: A pathway to cognitive im- Tona, S. Pedraza, V. Pérez-Brocal, L. Ramió-Torrentà, W. Ricart, A. Moya, M. Martínez-García, pairment. Brain Behav. Immun. 42, 10–21 (2014). R. Maldonado, J. M. Fernández-Real, Caudovirales bacteriophages are associated with 35. O. Guillemot-Legris, G. G. Muccioli, Obesity-induced neuroinflammation: Beyond the hy- improved executive function and memory in flies, mice, and humans. Cell Host Microbe 30, 340–356.e8 (2022). Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 pothalamus. Trends Neurosci. 40, 237–253 (2017). 36. S. Chatterjee, E. N. Walsh, A. L. Yan, K. P. Giese, S. Safe, T. Abel, Pharmacological activation of 60. J. Puig, C. Biarnes, S. Pedraza, J. C. Vilanova, R. Pamplona, J. M. Fernández-Real, R. Brugada, Nr4a rescues age-associated memory decline. Neurobiol. Aging 85, 140–144 (2020). R. Ramos, G. Coll-de-Tuero, L. Calvo-Perxas, J. Serena, L. Ramió-Torrentà, J. Gich, L. Gallart, M. Portero-Otin, A. Alberich-Bayarri, A. Jimenez-Pastor, E. Camacho-Ramos, J. Mayneris- 37. H. Shah, F. Dehghani, M. Ramezan, R. B. Gannaban, Z. F. Haque, F. Rahimi, S. Abbasi, Perxachs, V. Pineda, R. Font, A. Prats-Puig, M.-L. Gacto, G. Deco, A. Escrichs, B. Clotet, A. C. Shin, Revisiting the role of vitamins and minerals in Alzheimer’s disease. Antioxidants R. Paredes, E. Negredo, B. Triaire, M. Rodríguez, A. Heredia-Escámez, R. Coronado, W. de (Basel) 12, 415 (2023). Graaf, V. Prevost, A. Mitulescu, P. Daunis-I-Estadella, S. Thió-Henestrosa, F. Miralles, V. Ribas- 38. M. U. Wołoszynowska-Fraser, A. Kouchmeshky, P. McCaffery, Vitamin A and retinoic acid in Ripoll, M. Puig-Domingo, M. Essig, C. R. Figley, T. D. Figley, B. Albensi, A. Ashraf, cognition and cognitive disease. Annu. Rev. Nutr. 40, 247–272 (2020). J. H. C. Reiber, G. Schifitto, U. Md Nasir, C. Leiva-Salinas, M. Wintermark, K. Nael, J. Vilalta- 39. J. Ng, A. Papandreou, S. J. Heales, M. A. Kurian, Monoamine neurotransmitter disorders - Franch, J. Barretina, J. Garre-Olmo, The aging imageomics study: Rationale, design and Clinical advances and future perspectives. Nat. Rev. Neurol. 11, 567–584 (2015). baseline characteristics of the study population. Mech. Ageing Dev. 189, 111257 (2020). 40. P. S. Kaeser, L. Deng, M. Fan, T. C. Südhof, RIM genes differentially contribute to organizing 61. D. Wechsler, WAIS-IV. Escala de inteligencia de Wechsler para adultos-IV, in Manual técnico presynaptic release sites. Proc. Natl. Acad. Sci. U.S.A. 109, 11830–11835 (2012). y de interpretación (NCS Pearson, Madrid, Inc. Edici., 2012). 41. C. M. Powell, S. Schoch, L. Monteggia, M. Barrot, M. F. Matos, N. Feldmann, T. C. Südhof, 62. E. Strauss, E. M. S. Sherman, O. Spreen, A Compendium of Neuropsychological Tests: Ad- E. J. Nestler, The presynaptic active zone protein RIM1α is critical for normal learning and ministration, Norms, and Commentary (Oxford Univ. Press, 3rd Oxfo., 2006). memory. Neuron 42, 143–153 (2004). 63. J. D. Corrigan, N. S. Hinkeldey, Relationships between parts A and B of the trail making test. 42. N. Krumm, T. N. Turner, C. Baker, L. Vives, K. Mohajeri, K. Witherspoon, A. Raja, B. P. Coe, J. Clin. Psychol. 43, 402–409 (1987). H. A. Stessman, Z. X. He, S. M. Leal, R. Bernier, E. E. Eichler, Excess of rare, inherited trun- 64. M. D. Lezak, Neuropsychological assessment in behavioral toxicology - Developing cating mutations in autism. Nat. Genet. 47, 582–588 (2015). techniques and interpretative issues. Scand. J. Work Environ. Health 10 (suppl. 1), 43. M. S. Shin, S. Y. Park, S. R. Park, S. H. Seol, J. S. Kwon, Clinical and empirical applications of 25–29 (1984). the Rey-Osterrieth complex figure test. Nat. Protoc. 1, 892–899 (2006). 65. D. C. Delis, J. H. Kramer, E. Kaplan, B. A. Ober, Manual for the California Verbal Learning Test, 44. A. H. Brand, N. Perrimon, Targeted gene expression as a means of altering cell fates and (CVLT-II) (The Psychological Corporation, 2000). generating dominant phenotypes. Development 118, 401–415 (1993). 66. A. M. Paolo, A. I. Tröster, J. J. Ryan, Test-retest stability of the California verbal learning test 45. T. S. Koemans, C. Oppitz, R. A. T. Donders, H. Van Bokhoven, A. Schenck, K. Keleman, in older persons. Neuropsychology 11, 613–616 (1997). J. M. Kramer, Drosophila courtship conditioning as a measure of learning and memory. 67. C. Golden, A manual for the clinical and experimental use of the Stroop color and word test, J. Vis. Exp. 2017, 55808 (2017). in Faculty Books and Book Chapters. Stoelting Company, 751–758 (1978). 46. N. Raun, S. Jones, J. M. Kramer, Conditioned courtship suppression in Drosophila mela- 68. J. Latorre, J. Mayneris-Perxachs, N. Oliveras-Cañellas, F. Ortega, F. Comas, J. M. Fernández- nogaster. J. Neurogenet. 35, 154–167 (2021). Real, J. M. Moreno-Navarrete, Adipose tissue cysteine dioxygenase type 1 is associated 47. M. Haddadi, S. R. Jahromi, B. K. C. Sagar, R. K. Patil, T. Shivanandappa, S. R. Ramesh, Brain with an anti-inflammatory profile, impacting on systemic metabolic traits. EBioMedicine aging, memory impairment and oxidative stress: A study in Drosophila melanogaster. 85, 104302 (2022). Behav. Brain Res. 259, 60–69 (2014). 69. A. Dobin, C. A. Davis, F. Schlesinger, J. Drenkow, C. Zaleski, S. Jha, P. Batut, M. Chaisson, 48. D. Wang, L. Qian, H. Xiong, J. Liu, W. S. Neckameyer, S. Oldham, K. Xia, J. Wang, R. Bodmer, T. R. Gingeras, STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013). Z. Zhang, Antioxidants protect PINK1-dependent dopaminergic neurons in Drosophila. 70. B. Li, C. N. Dewey, RSEM: Accurate transcript quantification from RNA-seq data with or Proc. Natl. Acad. Sci. U.S.A. 103, 13520–13525 (2006). without a reference genome. BMC Bioinformatics 12, 323 (2011). 49. S. Bahadorani, P. Bahadorani, J. P. Phillips, A. J. Hilliker, The effects of vitamin supple- 71. W. R. Wikoff, G. Pendyala, G. Siuzdak, H. S. Fox, Metabolomic analysis of the cerebrospinal mentation on Drosophila life span under normoxia and under oxidative stress. J. Gerontol. fluid reveals changes in phospholipase expression in the CNS of SIV-infected macaques. A Biol. Sci. Med. Sci. 63, 35–42 (2008). J. Clin. Investig. 118, 2661–2669 (2008). 50. W. G. Roeterdink, K. E. Strecker, C. C. Hayden, M. H. M. Janssen, D. W. Chandler, Imaging the 72. J. Rynes, C. D. Donohoe, P. Frommolt, S. Brodesser, M. Jindra, M. Uhlirova, Activating rotationally state-selected NO(A,n) product from the predissociation of the A state of the transcription factor 3 regulates immune and metabolic homeostasis. Mol. Cell. Biol. 32, NO-Ar van der Waals cluster. J. Chem. Phys. 130, 134305 (2009). 3949–3962 (2012). 51. T. Kawai, M. V. Autieri, R. Scalia, Adipose tissue inflammation and metabolic dysfunction in 73. S. M. O’neill, C. Hinkle, S. J. Chen, A. Sandhu, R. Hovhannisyan, S. Stephan, W. R. Lagor, obesity. Am. J. Physiol. Cell Physiol. 320, C375–C391 (2021). R. S. Ahima, J. C. Johnston, M. P. Reilly, Targeting adipose tissue via systemic gene therapy. 52. Z. Liu, X. Huang, Lipid metabolism in Drosophila: Development and disease. Acta Biochim. Gene Ther. 21, 653–661 (2014). Biophys. Sin (Shanghai) 45, 44–50 (2013). 74. E. Puighermanal, G. Marsicano, A. Busquets-Garcia, B. Lutz, R. Maldonado, A. Ozaita, 53. R. P. Kühnlein, The contribution of the Drosophila model to lipid droplet research. Prog. Cannabinoid modulation of hippocampal long-term memory is mediated by mTOR sig- Lipid Res. 50, 348–356 (2011). naling. Nat. Neurosci. 12, 1152–1158 (2009). 54. S. Meltzer, J. A. Bagley, G. L. Perez, C. E. O’Brien, L. DeVault, Y. Guo, L. Y. Jan, Y. N. Jan, 75. A. Busquets-Garcia, E. Puighermanal, A. Pastor, R. De La Torre, R. Maldonado, A. Ozaita, Phospholipid homeostasis regulates dendrite morphogenesis in Drosophila sensory Differential role of anandamide and 2-arachidonoylglycerol in memory and anxiety-like neurons. Cell Rep. 21, 859–866 (2017). responses. Biol. Psychiatry 70, 479–486 (2011). 55. J. Muffat, D. W. Walker, S. Benzer, Human ApoD, an apolipoprotein up-regulated in neu- 76. M. Gomis-González, L. Galera-López, M. Ten-Blanco, A. Busquets-Garcia, T. Cox, rodegenerative diseases, extends lifespan and increases stress resistance in Drosophila. R. Maldonado, A. Ozaita, Protein kinase C-gamma knockout mice show impaired Proc. Natl. Acad. Sci. U.S.A. 105, 7088–7093 (2008). Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 19 of 20 S C I E N C E A D VA N C E S | R E S E A R C H A R T I C L E hippocampal short-term memory while preserved long-term memory. Mol. Neurobiol. 58, available as a service to the research community but charges for commercial use (for 617–630 (2021). permission requests contact the AECOM at biotech@einstein.yu.edu); the Spanish and Catalan 77. A. Escudero-Lara, D. Cabañero, R. Maldonado, Kappa opioid receptor modulation of en- adaptations used were provided by the Barcelona Beta Brain Research Center and the Pasqual dometriosis pain in mice. Neuropharmacology 195, 108677 (2021). Maragall Foundation with the AECOM’s permission. For further information about these 78. M. E. Ritchie, B. Phipson, D. Wu, Y. Hu, C. W. Law, W. Shi, G. K. Smyth, Limma powers diff- versions, contact Nina Gramunt at ngramunt@pmaragall.org. M.A.-R. is funded by Instituto de erential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. Salud Carlos III, Río Hortega (CM19/00190). N.O.-C. is funded by Instituto de Salud Carlos lll, PFIS 43, e47 (2015). (FI19/00293). A.C.-N. is funded by Instituto de Salud Carlos lll, Sara Borrell (CD20/00051). IDIBGI 79. M. D. Robinson, D. J. McCarthy, G. K. Smyth, edgeR: A Bioconductor package for differential is a CERCA center from the “CERCA Programme/Generalitat de Catalunya.” Author expression analysis of digital gene expression data. Bioinformatics 26, 139–140 (2010). contributions: J.M.-P. researched the data, performed the statistical analysis, and wrote the manuscript. N.O.-C., M.A.-R., and J.L.-L. researched the data. A.C.-N. and L.d.l.V.-C. performed and 80. A. Kamburov, U. Stelzl, H. Lehrach, R. Herwig, The ConsensusPathDB interaction database: wrote the Drosophila experiments. V.C.-A., E.M.-G., and R.M. performed and wrote the mouse 2013 update. Nucleic Acids Res. 41, D793–D800 (2013). experiment. A.M.-A. performed the neuropsychological examination. J.M.M.-N., J.P., J.G.-O., R.R., 81. D. Szklarczyk, A. L. Gable, D. Lyon, A. Junge, S. Wyder, J. Huerta-Cepas, M. Simonovic, and F.V. contributed to the discussion and reviewed the manuscript. J.M.F.-R. performed the N. T. Doncheva, J. H. Morris, P. Bork, L. J. Jensen, C. von Mering, STRING v11: Protein-protein statistical analysis and wrote the manuscript. J.M.-P. and J.M.F.-R. carried out the conception association networks with increased coverage, supporting functional discovery in and coordination of the study. All authors participated in final approval of the version to be genome-wide experimental datasets. Nucleic Acids Res. 47, D607–D613 (2019). published. J.M.F.-R. is the guarantor of this work and, hence, had full access to all the data in the 82. E. Ulgen, O. Ozisik, O. U. Sezerman, PathfindR: An R package for comprehensive identifi- study and takes responsibility for the integrity of the data. Competing interests: The authors cation of enriched pathways in omics data through active subnetworks. Front. Genet. 10, declare that they have no competing interests. Data and materials availability: All data Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 858 (2019). needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Acknowledgments Funding: This study was partially funded by Instituto de Salud Carlos III (ISCIII, Madrid, Spain) Submitted 21 December 2022 through the project PI18/01022 and PI21/01361 to J.M.F.-R. and the projects PI20/01090 Accepted 10 July 2023 (cofunded by the European Union under the European Regional Development Fund “A way to Published 11 August 2023 make Europe”) and CP18/00009 (cofunded by the European Union under the European Social 10.1126/sciadv.adg4017 Fund “Investing in your future”) to J.M.-P. and Generalitat de Catalunya 2021 SGR 01263. Albert Einstein College of Medicine owns the copyright for the Memory Binding Test and makes it Oliveras-Cañellas et al., Sci. Adv. 9, eadg4017 (2023) 11 August 2023 20 of 20 Adipose tissue coregulates cognitive function Nria Oliveras-Caellas, Anna Castells-Nobau, Lisset de la Vega-Correa, Jessica Latorre-Luque, Anna Motger-Albert, Maria Arnoriaga-Rodriguez, Josep Garre-Olmo, Cristina Zapata-Tona, Cludia Coll-Martnez, Llus Rami-Torrent, Jos Maria Moreno-Navarrete, Josep Puig, Francesc Villarroya, Rafel Ramos, Vernica Casad-Anguera, Elena Martn-Garca, Rafael Maldonado, Jordi Mayneris-Perxachs, and Jos Manuel Fernndez-Real Sci. Adv., 9 (32), eadg4017. Downloaded from https://www.science.org at Universitat Autnoma de Barcelona on August 29, 2023 DOI: 10.1126/sciadv.adg4017 View the article online https://www.science.org/doi/10.1126/sciadv.adg4017 Permissions https://www.science.org/help/reprints-and-permissions Use of this article is subject to the Terms of service Science Advances (ISSN ) is published by the American Association for the Advancement of Science. 1200 New York Avenue NW, Washington, DC 20005. The title Science Advances is a registered trademark of AAAS. Copyright © 2023 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC).