Article pubs.acs.org/jpr
Defining the Human Adipose Tissue Proteome To Reveal Metabolic Alterations in Obesity Adil Mardinoglu,†,# Caroline Kampf,‡,# Anna Asplund,‡ Linn Fagerberg,§ Björn M. Hallström,§ Karolina Edlund,‡ Matthias Blüher,∥ Fredrik Pontén,‡ Mathias Uhlen,§,⊥ and Jens Nielsen*,†,§ †
Department of Chemical and Biological Engineering, Chalmers University of Technology, 412 96 Gothenburg, Sweden Department of Immunology, Genetics and Pathology, Uppsala University, 751 05 Uppsala, Sweden § Science for Life Laboratory, KTHRoyal Institute of Technology, 100 44 Stockholm, Sweden ∥ Faculty of Medicine, University of Leipzig, 04103 Leipzig, Germany ⊥ Division of Proteomics, KTHRoyal Institute of Technology, 106 91 Stockholm, Sweden ‡
S Supporting Information *
ABSTRACT: White adipose tissue (WAT) has a major role in the progression of obesity. Here, we combined data from RNA-Seq and antibody-based immunohistochemistry to describe the normal physiology of human WAT obtained from three female subjects and explored WAT-specific genes by comparing WAT to 26 other major human tissues. Using the protein evidence in WAT, we validated the content of a genome-scale metabolic model for adipocytes. We employed this high-quality model for the analysis of subcutaneous adipose tissue (SAT) gene expression data obtained from subjects included in the Swedish Obese Subjects Sib Pair study to reveal molecular differences between lean and obese individuals. We integrated SAT gene expression and plasma metabolomics data, investigated the contribution of the metabolic differences in the mitochondria of SAT to the occurrence of obesity, and eventually identified cytosolic branched-chain amino acid (BCAA) transaminase 1 as a potential target that can be used for drug development. We observed decreased glutaminolysis and alterations in the BCAAs metabolism in SAT of obese subjects compared to lean subjects. We also provided mechanistic explanations for the changes in the plasma level of BCAAs, glutamate, pyruvate, and α-ketoglutarate in obese subjects. Finally, we validated a subset of our model-based predictions in 20 SAT samples obtained from 10 lean and 10 obese male and female subjects. KEYWORDS: white adipose tissue, adipocytes, transcriptome, proteome, genome-scale metabolic modeling
■
INTRODUCTION Obesity has become a global pandemic, and it is considered as one of the greatest threats to human health.1 More than 68% of the adults living in the United States2 and 52% of the adults living in the European Union3 are characterized as obese or overweight. Obesity is also associated with one or more comorbidities, including type 2 diabetes (T2D), cardiovascular disease, and increased risk of cancer.4 The occurrence of the obesity and related systemic disorders is linked with a multitude of risk factors, including genetic components, physical inactivity, diet, and gut microbiome.5 White adipose tissue (WAT) has a major role in the progression of obesity and its adverse outcomes.6 In order to gain an increased understanding of its role in the occurrence of obesity and to develop adequate and efficient treatments strategies, metabolic differences in WAT between lean and obese states should be elucidated in a systematic way. The primary function of WAT is to store energy in the form of triacylglycerols (TAGs) in lipid droplets.7 It also regulates energy intake, energy expenditure, and carbohydrate and lipid metabolism by secreting hormones and cytokines, e.g., © XXXX American Chemical Society
adiponectin, estradiol, interleukins, leptin, plasminogen activator inhibitor-1, and tumor necrosis factor-α.8 The detailed physiological role of WAT has been revealed in lean and obese states through the generation of proteomics9 and transcriptomics10 data. WAT contains a variety of cell types, including adipocytes, pre-adipocytes, immune cells, and endothelial cells, but it is composed mostly of adipocytes. Hence, we recently evaluated the presence/absence of proteins in adipocytes using antibodies generated in the Human Protein Atlas (HPA, http://www.proteinatlas.org/)11 and studied the adipocyte biology at the genome-wide level.12 Obesity is driven or accompanied by metabolic reprogramming of adipocytes,13 which may be studied at the systems level in an unprecedented fashion.14 A large number of coordinated reactions, metabolites, and enzymes in adipocytes can be Special Issue: Proteomics of Human Diseases: Pathogenesis, Diagnosis, Prognosis, and Treatment Received: June 15, 2014
A
dx.doi.org/10.1021/pr500586e | J. Proteome Res. XXXX, XXX, XXX−XXX
Journal of Proteome Research
Article
antibody-based scores in HPA evidence (Supporting Information).
efficiently explored in a holistic manner, and the complexity of metabolism can be captured by the reconstruction of genomescale metabolic models (GEMs).15 These integrative models are widely used to understand how complex phenotypes arise from individual molecules and their interactions in a given cell type.15b,16 In order to have an increased elucidation of the WAT physiology, we first generated RNA-Seq data for WAT obtained from three healthy female individuals. Second, we combined transcriptomics and immunohistochemistry-based proteomics data to characterize and classify the proteins in WAT and explored the human WAT-specific protein-coding genes by comparing the expression of the proteins to 26 other human major tissues. We identified the localization of the proteins within the WAT and characterized the biological processes carried out by these genes. Next, we validated the content of the GEM for adipocytes iAdipocytes1809 on the basis of the WAT RNA-Seq data and updated the content of the model on the basis of the latest proteomics data in HPA version 12.0 and Human Metabolic Reaction database (HMR).17 Finally, we employed the GEM for adipocytes for the analysis of subcutaneous adipose tissue (SAT) gene expression data obtained from subjects included in the Swedish Obese Subjects (SOS) Sib Pair study. In this analysis, we focused on revealing the mitochondrial metabolic differences of SAT obtained from lean and obese individuals and on identification of the potential targets for drug development that aim to revert the disrupted metabolism in obese subjects. Following this analysis, we validated a subset of our findings in independent SAT samples obtained from 10 lean and 10 obese male and female subjects by quantitative reverse transcription PCR (RT-PCR). Our study demonstrated the value of integrating transcriptomics and proteomics data to characterize the human WAT proteome and the usage of high-throughput data to validate the GEM for adipocytes, which enables elucidating the genetic etiology of obesity.
■
Analysis of SAT Transcriptomics Data in SOS Sib Pair Study
Raw gene expression data were downloaded from the GEO public repository under the accession number GSE27916, and male and female data were analyzed independently. Gene expression data were analyzed using quantile normalization with the Piano R package.19 Differential expression analysis was carried out, and p-values were corrected for multiple testing by the Benjamini Hochberg method to calculate adjusted p-values. In the SAT, 7443 probe sets in microarray chip were differentially expressed (adj. P-value 1). A total of 18% of the genes presented a mixed expression and were not classified into any of the above-mentioned categories. The remaining 35% of the analyzed genes were not detected in WAT (FPKM < 1) (Figure 2A). We calculated the relative mRNA pools for each of the categories, and 87% of the mRNA molecules in the WAT accounted for house-keeping genes (Figure 2B). Notably, only
were observed (Spearman correlation coefficients 0.97, 0.98, and 0.96) (Figure 1A). This confirmed a high degree of technical and biological reproducibility and demonstrated minor individual genetic differences between the adipose tissue of healthy subjects. Similar transcriptomics analyses have been performed for another 26 different human tissues using RNASeq,21 and here we investigated the level of correlations between gene expression in the WAT and the other tissues investigated. From this comparison, we detected the lowest correlation between WAT and testis (Spearman correlation 0.71) and the highest correlation between WAT and endometrium (Spearman correlation 0.91) (Figure 1B). We evaluated the presence of each protein-coding gene in WAT on the basis of the mRNA expression value, and the total number of detected genes varied between 12,897 and 13,043 in each sample. In our study, we did not interpret low-abundance transcripts (FPKM < 1), and the chosen threshold is approximately equivalent to one copy per cell.22 On the basis of the average FPKM values from all biological replicates, we identified 13,064 (65%) of all putative protein coding genes (20,050) as being expressed in WAT. Classification of the Genes in WAT
Transcriptomics analyses of WAT and the other 26 major human tissues (Supporting Information, Dataset 1) allowed us to classify all 20,050 putative protein-coding genes into different categories on the basis of the FPKM values (Figure 2A). Three genes (ADIPOQ, LEP, and TUSC5) were highly C
dx.doi.org/10.1021/pr500586e | J. Proteome Res. XXXX, XXX, XXX−XXX
Journal of Proteome Research
Article
Figure 2. Classification of the genes in human WAT. (Top) Pie charts showing the distribution of (A) all 20,050 genes into six different categories based on the transcript abundance as well as the number of detected tissues and (B) the percentages of expressed mRNA molecules, i.e., the sum of all FPKM values for each of the categories. (Bottom) FPKM values across all 27 tissues for (C) the 30 most abundant genes in WAT, (D) 3 highly WAT-enriched genes, and (E) 29 moderately WAT-enriched genes.
enriched WAT genes in connection to expression in other tissue types are presented. Functional characterization of the WAT-specific genes was revealed through the Database for Annotation, Visualization and Integrated Discovery (DAVID),23 and gene ontology biological process (GO BP) terms (Hypergeometric test, Pvalues
