We then examined the similarity between the expression profiles of individual genes to build the co-expression network underlying the described functional interactions. Among the genes annotated with significantly over-represented GO categories, 40 genes (12.5%, among which 24 genes were up-regulated and 16 genes down-regulated) were found to encode various structural components of the ECM or molecules involved in ECM remodeling and regulation (Additional data file 2 and supplementary online table 1).

Figure 4 depicts a bi-modular co-expression network relating genes annotated with significantly over-represented GO Biological Process categories in the obese WAT (Figure 2a). The first co-expression module (Figure 4, module 1) groups up-regulated genes associated with processes constituting the first functional interaction module (Figure 2b, module 1). This module includes representatives from all major classes of ECM components, namely structural proteins such as members of the collagen family, adherent proteins such as fibronectin and laminin family members, glycosaminoglycans and proteoglycans, and specialized glycoproteins such as integrins, as well as several enzymes involved in ECM remodeling (Additional data file 2 and supplementary online table 1). A sub-network grouping all ECM related genes, showing significant differential expression in obese WAT, is presented in Figure 5.

Among various ECM components, several genes coding for members of the integrin family were found to be significantly induced and co-expressed in obese WAT, occupying central positions in the first co-expression module (Figure 4, module 1). This module included integrins alpha V (ITGAV), referred to as the vitronectin receptor, and alpha M (ITGAM), as well as integrins beta 1 (ITGB1; also named fibronectin receptor or beta polypeptide), beta 2 (ITGB2) and beta 3 (ITGB5). These integrins displayed strong co-expression with other key components of the ECM (Figures 4 and 5; Additional data file 2 and supplementary online table 1), such as members of the collagen family, including the major type IV alpha collagen chain of basement membranes (COL4A1), and members of the fibril associated collagen (COL5A2 and COL12A1). They were also co-expressed with members of the glycosaminoglycan and proteoglycan family (syndecan binding protein (SDCBP), lumican (LUM)), known to play an important role in the initiation of inflammatory phenomena, as well as in the recruitment, rolling, and subsequent extravasation of lymphocytes 24, the laminin beta 1 (LAMB1), and with several proteases and other enzymes involved in ECM remodeling and cell-cell or cell-matrix interactions. Some of the genes coding for these enzymes were significantly induced in the obese WAT. Among them, metalloproteinases domain 12 (ADAM12) and domain 9 (ADAM9), which belong to the disintegrin family, are known to modulate the communication between the fibronectin-rich ECM and the actin cytoskeleton, and are also involved in the early stages of pre-adipocyte differentiation 25. Lysyl oxidase (LOX) is involved in cross-linking extracellular matrix proteins, while chondroitin sulfate GalNAcT-2 (GALNACT-2) plays a central role in the synthesis of some members of the glycosaminoglycan and proteoglycan family. Other ECM related genes were significantly under-expressed in WAT of obese subjects, such as metallopeptidases domain 17 (ADAM17) and domain 15 (ADAM15), or the collagen type I alpha 1 (COL1A1).

Interestingly, the first co-expression module (Figure 4, module 1) grouped not only genes related to ECM components, but also a number of genes coding for cytokines and surface markers secreted by immune cells possibly infiltrating WAT in obese subjects. A number of these genes showed significant co-expression with members of the integrin family and are known to be involved in the recruitment and activation of immune circulating cells, such as monocytes, lymphocytes or neutrophils. Among them were markers of the alternative pathway of macrophage activation, as the CC chemokine ligand 18 (CCL18) and the macrophage scavenger receptor (CD163), which showed strong co-expression with the integrin alpha V (ITGAV) and the macrophage receptor 1 (Mac-1) complex formed by integrins alpha M (ITGAM) and beta 2 (ITGB2). Available data demonstrate that the synthesis of CCL18 by alternatively activated macrophages is induced by Th2 cytokines, integrin beta 2 (ITGB2) and the scavenger receptor (CD163) 26. CCL18 is also known to be involved in the recruitment and activation of CD4+ and CD8+ T cells and, more remarkably, is credited with playing a central role in perpetuating fibrotic processes through its involvement in a positive feedback loop that links activated macrophages to fibroblasts 26. Moreover, expression of the Mac-1 complex is increased by conditions such as diabetes, being overweight and tissular hypoxia 2728, and plays an important role in the recruitment, adhesion, and activation of circulating monocytes and neutrophils, and in the phagocytosis of complement coated particles 2829. Co-expressed with Mac-1 components, the hypoxia-inducible factor 1 (HIF1A) is a well characterized transcription factor that performs an essential role in cellular responses to hypoxia. HIF1A is also involved in the regulation of macrophage migration, and modulates the metabolism of immune cells exposed to low oxygen tensions in hypoxic areas of inflamed tissues 30.

To the same group of pro-inflammatory molecules belong also interleukin (IL)1 receptor type I (IL1R1), which modulates many cytokine induced immune and inflammatory responses, and IL15 (IL15), which regulates T and natural killer cell activation and proliferation 3132. Both of them were strongly co-expressed with the Mac-1 complex and with C-type lectin domain family 4 member A (CLEC4A), known to play an important role in mediating the immune and inflammatory responses, especially in neutrophils 33.

Several molecules demonstrated strong co-expression with IL1R1, among which are the CD53 (CD53) and CD9 (CD9) markers, known to complex with integrins, and annexin I (ANXA1), credited with a potential anti-inflammatory activity, all of them performing important homeostatic roles by modulating innate immunity 343536. In the same spectrum, integrin alpha V (ITGAV) displayed strong co-expression with CD163, a well known macrophage-specific marker mediating an anti-inflammatory pathway that includes IL10, and whose synthesis was shown to be well correlated with local and systemic inflammatory phenomena 3738. Also, the heat shock protein 8 (HSPA8), a surface marker for the undifferentiated cellular state expressed on the surface of human embryonic stem cells 39, performs an important role in the repair processes following harmful tissular assaults (for example, hemorrhage or local ischemia) 40, and was found to be significantly co-expressed with integrin alpha V, annexin I and other ECM components.

A panel of the genes clustered in module 1 of the co-expression network (Figure 4) displayed significant positive correlations between their expression levels in WAT of obese and non-obese subjects and the BMI of these subjects (Figure 6 and Table 2). Among the genes showing the strongest association with the BMI were cathepsin S (CTSS), involved in the degradation of several components of the extracellular matrix 19, lymphocyte cytosolic protein 2 (LCP2) and CD247 (CD247), both related to T cell development and activation, as well as the hypoxia-inducible factor 1 (HIF1A).

The second co-expression module (Figure 4, module 2) grouped several genes encoding proteins involved in lipolysis pathways, which were down-regulated in the obese WAT, including hormone-sensitive lipase (LIPE), perilipin (PLIN), and monoglyceride lipase (MGLL). The insulin receptor (INSR) and antilipolytic adenosine A1 receptor (ADORA1) were also located in this module, together with a number of genes encoding mitochondrial enzymes, including NADH dehydrogenase 1 alpha subcomplex (NDUFA1) and cytochrome c oxidase assembly homolog (COX17). The NDUFA1 gene encodes a component of respiratory chain complex I that transfers electrons from NADH to ubiquinone, while COX17 might contribute in the mitochondrial terminal complex to the functioning of cytochrome c oxidase, which catalyzes electron transfer from the reduced cytochrome c to oxygen. Several genes of module 2 (Figure 4; online supplementary data 20) are involved in the synthesis, transport and oxidation of a variety of fatty acids. Among them, some genes are known to code for proteins intervening in the initial step (acyl-coenzyme A dehydrogenase (ACADS)), and the processing (3-hydroxyacyl-CoA dehydrogenase type II (HADH), 3,2 trans-enoyl-CoA isomerase (DC1)) and the termination (acyl-CoA thioesterase 4 (ACOT4)) of the mitochondrial fatty acid β-oxidation pathway. The β-oxidation of long-chain fatty acids usually implicates the sequential action of carnitine palmitoyltransferase I and carnitine palmitoyltransferase II together with a carnitine-acylcarnitine translocase. The expression levels of two members of the carnitine/choline acetyltransferase family (CPT1A and CPT1B) involved in this rate limiting step across the mitochondrial inner membrane were decreased as well as that of the CRAT gene, which catalyzes the reversible transfer of acyl groups from an acyl-CoA thioester to carnitine and regulates the ratio of acylCoA/CoA in the mitochondrial compartments. Interestingly, module 2 also gathered several genes involved in the induction of apoptosis, such as the death-associated protein (DAP), the death-associated protein kinase 2 (DAPK2), and the serine/threonine kinase 17a (STK17A), a member of the DAP kinase-related apoptosis-inducing protein kinase family, as well as the apoptosis-inducing factor (SIVA1), TNFRSF1A-associated via death domain (TRADD) and programmed cell death 5 (PDCD5), some being strongly co-expressed with mitochondrial enzymes described above. Protein kinase C epsilon (PRKCE), involved in several intracellular signaling pathways and particularly in apoptosis, was linked to DAPK2, CRAT, and ACADS in this module. Other down-regulated genes encode components of cytoplasmic or mitochondrial ribosomal subunits, which are part of ribosomal proteins, and several eukaryotic translation elongation factors implicated in protein synthesis.

In contrast with the genes comprising the co-expression module 1, the expression profiles of the majority of the genes comprising module 2 demonstrated significant negative correlations with BMI (Figure 7 and Table 2). Among them, some of the strongest negative correlations were observed for the insulin receptor (INSR), molecules of the adipocyte lipolytic pathway (LIPE, PLIN), some mitochondrial components (CRAT, ACADS, NDUFA1, COX17), and some members of apoptotic pathways (DAPK2, SIVA1, DAP). Also, the expression profiles of numerous components of cytoplasmic or mitochondrial ribosomal subunits showed significant negative correlations with the BMI (RPL28, RPS12, RPL35, RPS2, and RPS21 among others).

Since at the functional level the processes related to immune, inflammatory and stress responses, as well as to cell adhesion and signaling (Figures 2b and 3b, module 1), displayed an opposite regulation pattern to that of the metabolic functions (Figures 2b and 3b, module 2), we examined the links that may connect these two functional modules at the gene level, and searched for which genes could play a mediating role by linking the ECM to intracellular pathways. As shown in Figure 4, some ECM related genes were co-expressed with a set of inflammatory genes (module 1), while showing a significant inverse expression pattern to that of genes belonging to the metabolic module (module 2). Among them, integrin alpha V (ITGAV), CD163 and CCL18, two markers of the alternative pathway of macrophage activation, heat shock protein 8 (HSPA8), and contactin associated protein 1 (CNTNAP1), involved in the activation of intracellular signaling pathways, were strongly related to several genes encoding enzymes of the lipolytic pathway, including hormone-sensitive lipase (LIPE) and perilipin (PLIN), phosphatidic acid phosphatase type 2B (PAP2B), a member of the lipid phosphate phosphatases family, and to genes related to apoptosis, such as death-associated protein kinase 2 (DAPK2) and non-metastatic cells 3 protein (NME3).

We have shown previously that weight loss is associated with improvement in the inflammatory profile, together with regression of macrophage infiltration in WAT 11. To better characterize the association between adipose mass variation, local inflammatory phenomena and ECM remodeling, we further examined the functional profile of the transcriptomic signature of the obese WAT after a significant weight loss induced by bariatric surgery. Ten cDNA microarray experiments were performed from subcutaneous WAT biopsies carried out in morbidly obese subjects (BMI 47.65 ± 4.4 kg/m², range 42.5-57 kg/m²), before and three months after undergoing a laparoscopic gastric bypass 41. The detailed clinical and biochemical parameters of these subjects were presented elsewhere 13, and are provided as online supplementary data 20. The analysis of differential gene expression with the SAM procedure 21, performed on the cDNA measurements with signals recovered in at least 80% of the microarray experiments, detected 1,744 up- and 1,627 down-regulated genes, corresponding to a 5% FDR. Functional analysis of these genes identified 2,687 genes (1,390 up- and 1,297 down-regulated) annotated with GO categories, and 868 genes (450 up- and 418 down-regulated) annotated with KEGG categories.

Figures 8a, 9a and 10a illustrate the biological themes characterizing the transcriptomic signature of the obese WAT three months after gastric surgery, as indicated by significantly over-represented categories from GO Cellular Component (Figure 8a) and GO Biological Process ontologies (Figure 9a), and from KEGG (Figure 10a). This analysis shows a diametrical shift in the functional profile of the obese WAT associated with weight loss. Indeed, the majority of the genes up-regulated in WAT after gastric bypass were associated with structural themes (GO Cellular Component) related to the intracellular domain and organelles ('protein complex', 'cytoplasm', 'mitochondrion', 'endoplasmic reticulum', 'lysosome', 'actin cytoskeleton', 'cytosolic part'), while the down-regulated genes (Figure 8a) were mostly associated with cellular membrane and extracellular space specific themes ('integral to membrane', 'plasma membrane', 'extracellular region', 'extracellular matrix part'). The cellular processes (GO Biological Process) associated with the WAT up-regulated genes (Figure 9a) were related primarily to carbohydrate and protein metabolisms, including ubiquitin-dependent protein catabolism ('cellular protein metabolism', 'carbohydrate metabolism', 'ubiquitin-dependent protein catabolism'), to energy metabolism ('oxidative phosphorylation') and to transcriptional, translational and transport processes ('RNA processing', 'tRNA metabolism', 'translation', 'protein transport'). In contrast, down-regulated genes were mainly associated with processes related to cell adhesion and signaling (Figure 9a), notably via G-protein coupled receptor proteins ('signal transduction', 'cell adhesion', 'G-protein coupled receptor protein signaling pathway', 'cell surface receptor linked signal transduction'), as well as to the immune response and apoptosis ('immune response', 'apoptosis'). Finally, the KEGG pathways involving WAT genes up-regulated after weight loss (Figure 10a) were related to energy and nucleotides metabolisms ('oxidative phosphorylation', 'purine metabolism'), as well as to the degradation of some key ECM constituents, namely the glycosaminoglycans ('glycan structures - degradation', 'glycosaminoglycan degradation'). In accordance with GO annotations, the down-regulated KEGG pathways were related mostly to signaling processes and immune and inflammatory responses (Figure 10a), including complement and coagulation cascades and signaling of T and B cell receptors ('MAPK signaling pathway', 'Wnt signaling pathway', 'Complement and coagulation cascades', 'T cell receptor signaling pathway', 'B cell receptor signaling pathway', 'mTOR signaling pathway', and so on).

The quantification of the transcriptomic interactions relating biological themes associated with various structures, processes or regulatory pathways identified a very distinct interaction pattern from that observed in the previous condition. Figures 8, 9 and 10 illustrate a very dense interaction pattern relating up- and down-regulated processes in a strongly interconnected network. Figure 9c depicts the two most representative functional interaction modules (GO Biological Process) in this condition; this illustrates the strong interactions that connect the up-regulated themes composing the first functional module, mostly related to carbohydrate, energy and protein metabolism, with the down-regulated themes grouped in the second interaction module and related essentially to immune and inflammatory responses, signaling, cellular proliferation and apoptotic processes.

Co-expression networks underlying these functional modules (see the online supplementary data 20) confirmed the dense interaction pattern associating genes related to the ECM and inflammatory and metabolic processes. A number of ECM components showed opposite expression patterns to those noted in the previous condition, some being induced by weight loss while others were down-regulated (online supplementary Table 2 20). Among others, several genes coding for structural proteins were significantly down-regulated after weight loss (online supplementary Table 2 20), including members of the integrin family, such as integrin alpha V (ITGAV), integrin beta 4 (ITGB4), and integrin beta 6 (ITGB6). Enzymes involved in the degradation of glycosaminoglycans and proteoglycans were also significantly up-regulated after weight loss, as shown by the induction of the related KEGG pathways (online supplementary data 20). In addition, some metallopeptidases implicated in the degradation of other ECM components were equally induced, such as the matrix metallopeptidase 2 (MMP2), concomitantly with several metallopeptidase inhibitors from the tissue inhibitor of metalloproteinase family (TIMP1, TIMP2). Finally, a remarkable number of genes related to mitochondrial enzymes involved in the oxidative phosphorylation pathway (Figure 10; online supplementary data 20) were significantly up-regulated after weight loss, including genes coding for NADH dehydrogenases (NDUFA3, NDUFA5, NDUFA6, NDUFA9, NDUFA11, NDUFA4L2, NDUFB7, NDUFB11, NDUFS2, NDUFS8), ATP synthases (ATP5G1, ATP5G2, ATP5H, ATP5I, ATP5O, ATP6AP1) and cytochrome c-1 (CYC1).

Functional analysis using KEGG annotations showed that the pathway of NK cell mediated cytotoxicity was significantly enriched in genes up-regulated in obese WAT (Figure 3a), while the T cell receptor signaling pathway was enriched in genes down-regulated after gastric bypass (Figure 10a). Immunostaining for T lymphocytes and NK cells using CD3 and NKp46 antibodies confirmed the presence of these cells in the adipose tissue of morbidly obese subjects (Figure 11a-d), although at low abundance. Macrophages, demonstrating cytoplasmic extensions, and lymphocytes were detected by electron microscopy in the vicinity of adipocytes and near vessel walls (Figure 11e-g).