Adipose tissue coregulates cognitive function

Abstract

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 performance in different cognitive domains. These genes were mostly involved in synaptic function, phosphatidylinositol 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 misexpression of candidate gene ortholog in the Drosophila fat body significantly altered flies memory and learning. 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 function in humans, providing unprecedented diagnostic/therapeutic targets in AT.

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 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 TMTB 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 CVLT_IR in the VAT identified from a pathway overrepresentation analysis mapping significant 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 databases. In the Manhattan-like plots, the bubble size represents the ratio of input genes that are annotated in a pathway (GeneRatio). CREB, adenosine 3′,5′-monophosphate response element–binding protein; tRNA, transfer RNA; NCAM1, Neural Cell Adhesion Molecule 1; TP53, Tumor Protein 53; PKA, Protein Kinase CAMP-Activated Catalytic Subunit Alpha.

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.

Fig. 3.. Pathway analysis of differentially expressed 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 cognitive 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 annotated 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 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 subnetwork oriented approach. Each color displays a cluster of related pathways using a threshold for kappa statistics = 0.35. The size of the nodes corresponds 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-tetrakisphosphate; PI, phosphatidylinositol; TNF, tumor necrosis factor; NMDA, N-methyl-d-aspartate; CNV, copy number variations; L1CAM, L1 Cell Adhesion Molecule.

Fig. 4.. Selected clusters from active subnetwork oriented pathway analysis of common significant genes among the VAT and SAT in the discovery and validation cohorts (n = 188) and integration with circulating metabolites.

(A) Dot plot of enrichment analysis results performed on active subnetworks grouped by selected clusters. The x axis represents 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 indicating high significance. (B) Gene-concept network depicting significant genes involved in enriched pathways from selected clusters. The dot size of the pathways represents 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.

Fig. 5.. Associations of expression levels of selected genes with cognition 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 correlations (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 with a BMI of <35 kg/m². (J) Main baseline characteristics of the validation 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 controlling 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 conditioning 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 corresponding genetic background controls (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 determined with the nonparametrical bootstrap analysis with 10,000 iterations. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. ns, not significant.

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.

Fig. 7.. Associations of RIMS1 with cognition in the validation cohort 3 and Rim overexpression in Drosophila fat body effects in learning and gene expression profiles.

(A and B) Scatter plots of the partial Spearman’s rank correlations (adjusted 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 ADIPOBRAIN cohort (n = 40). (C) Rim overexpression in the Drosophila fat body and associations with learning. Differences in CI between naïve and trained males were assessed with Kruskal-Wallis nonparametrical test and post hoc Dunn’s multiple comparison 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 determined with the nonparametrical bootstrap analysis with 10,000 replicates. ****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 corresponding 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) associated 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 Reactome pathways of (J). (L) Significantly overrepresented GO molecular functions of significant gene transcripts 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.