Gene expression profiling reveals distinct features of various porcine adipose tissues

Background The excessive accumulation of body fat is a major risk factor to develop a variety of metabolic diseases. To investigate the systematic association between the differences in gene expression profiling and adipose deposition, we used pig as a model, and measured the gene expression profiling of six variant adipose tissues in male and females from three pig breeds which display distinct fat level. Results We identified various differential expressed genes among breeds, tissues and between sexes, and further used a clustering method to identify sets of functionally co-expression genes linked to different obesity-related phenotypes. Our results reveal that the subcutaneous adipose tissues mainly modulate metabolic indicators, nonetheless, the visceral adipose tissues as well as the intermuscular adipose tissue were mainly associated with the impaired inflammatory and immune response. Conclusions The present study provided the evidence of gene expression profiling that the subcutaneous adipose tissues are mainly affected the metabolism process, whereas the visceral and intermuscular adipose tissues should been term as the metabolic risk factors of obesity.


Background
Obesity is becoming a major global health hazard mainly because of the increased ingestion of high energy food and the decreased amounts of physical activity in populations. A gain in the adipose mass in the body is generally accompanied by increased risk of metabolism-related diseases. However, it is not only the amount of body fat mass, but also the distribution of adipose tissue that is now recognised as an important predictor of metabolic abnormalities.
Based on their anatomical location, adipose tissue can be divided into two major types: visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT). The VAT is a higher risk factor for obesity than SAT [1]. In particular, basal free fatty acid flux, lipolysis rate, and secretory protein expression are markedly higher in visceral adipocytes compared with subcutaneous adipocytes [2]. Our previous reports have provided epigenetic evidence that VAT is mainly associated with impaired inflammatory and immune responses and is a metabolic risk factor of obesity, while SAT is mainly associated with metabolic processes [3,4]. In addition, ethnic group and gender are important factors that affect adipose deposition [5,6].
Pig (Sus scrofa) has been recognised as an attractive biomedical model for human because its physiology and anatomy are remarkably similar to humans and pigs are easy to breed and handle [7]. Here, we present a comprehensive genome-wide comparison of gene expression profiling among six adipose tissues from different body sites in both sexes of three well-defined pig breeds displaying distinct adipose phenotypes. Using microarray technology, we identified various differentially expressed genes that may be related to the different adipose deposition phenotypes among the breeds, and between the sexes and anatomic locations. We also used a clustering method to identify sets of co-expressed genes that may be functionally linked to different obesity-related phenotypes. These results can be used to understand the different accumulation mechanisms of the distinct adipose tissue types.

Results and discussion
Obesity-related phenotypes Our previous report based on same individuals have demonstrated that the adipose volume were significant difference among adipose tissues and breeds, and between sexes [3]. In addition, the 24 kinds of reprehensive metabolism indicators in serum also revealed the same ranking from the leaner Landrace, the wild Tibetan and the fatty Rongchang pigs [3].

Identification of differentially expressed (DE) genes and relatedness of adipose samples
The number of DE genes (P < 0.05, three-way repeatedmeasures analysis of variance [ANOVA]) was lowest between male and female (1,225), higher among different adipose tissues (2,750), and highest among the different breeds (2,949), indicating significant biological differences between the groups in the latter two categories ( Figure 1A). To further characterise variability in the gene expression profiling, we performed a principal components analysis (PCA) based on the DE genes in each of the compared groups. The six adipose tissues clustered into two groups, VATs and SATs, which reflected their distinct functional and metabolic features ( Figure 1B). Interestingly, the intermuscular adipose (IAD) tissue that is deposited between muscle bundles, clustered with the VATs in terms of gene expression profiling, suggesting that IAD may be a potential new risk factor for obesity-related diseases, in agreement with our findings from a previous study on methylation [4].
Although the three pig breeds clustered into three groups, the Rongchang and Tibetan breeds were closer to each other than the Landrace breed ( Figure 1C). These results not only reflect the marked phenotypic differences in adipose deposition among the three pig breeds which have undergone different/opposite breeding directions, but also highlight the deep phylogenetic split between the European (Landrace) and Asian pigs (Rongchang and Tibetan) [3].
In addition, the male and female were also been roughly distinguish ( Figure 1D), which consisted with the lowest variation of gene expression profiling between sexes, which only have less than half of the amount of DE genes than among tissue types and breeds.

DE genes between the VATs and SATs
The PCA classified six adipose tissues into two groups: the VAT group which included greater omentum adipose (GOM), mesenteric adipose (MAD), retroperitoneal adipose (RAD) and IAD; and the SAT group which included upper layer of backfat (ULB) and inner layer of backfat (ILB).
We performed a pairwise comparison analysis and identified 192 and 181 DE genes that were specifically upregulated in the SAT and VAT groups, respectively (Figure 2 and Table 1). The DE genes which were specifically upregulated in the VAT group were mainly enriched in the GO categories of the immune processes, including positive regulation of defence response (8 genes, P = 5.63 × 10 -4 ), positive regulation of immune system process (10 genes, P = 1.11 × 10 -4 ), adaptive immune response (6 genes, P = 3.54 × 10 -4 ), innate immune response (7 genes, P = 7.86 × 10 -4 ) and B cell mediated immunity (5 genes, P = 9.91 × 10 -4 ) ( Figure 3). Various genes known to be involved in the immune response were highly expressed in VATs compared with in SATs (P < 0.05). For example, C3 (complement 3), which was reported to play a central role in the activation of the complement system [8,9] was upregulated in the VATs. CLEC12B (C-type lectin domain family 12B) limits the activity of monocyte-derived immune cells after cell differentiation and possibly during inflammatory diseases [10] (Figure 4). The action of these genes which were highly expressed in VATs may contribute to the obesity-induced chronic inflammation in adipose tissue that precedes the development of insulin resistance and type II diabetes.

DE genes among the three pig breeds
The number of DE genes (P < 0.05, three-way repeatedmeasures ANOVA) was highest between the fatty Rongchang and leaner Landrace pigs (1,855), lower between the wild Tibetan and Landrace pigs (1,440), and lowest between the Tibetan and Rongchang pigs (868) ( Figure 2).
The highly expressed genes in the Landrace pigs were mainly involved in the GO and KEGG categories of skeletal system development processes and immune processes, including positive regulation of skeletal system development (7 genes, P = 2.22 × 10 -4 ), bone development (8 genes, P = 4.16 × 10 -2 ), positive regulation of immune system process (20 genes, P = 1.19 × 10 -3 ) and adaptive immune response (8 genes, P = 2.39 × 10 -2 ) ( Figure 5). The highly expressed genes in the Rongchang pigs were mainly involved in the processes of lipid metabolism and immune processes, such as lipid oxidation (8 genes, P = 1.40 × 10 -3 ), fatty acid oxidation (8 genes, P = 2.53 × 10 -3 ), fatty acid metabolic process (21 genes, P = 4.71 × 10 -4 ), activation of immune response (6 genes, P = 1.30 × 10 -2 ) and positive regulation of immune response (8 genes, P = 5.93 × 10 -3 ). This result reflects the phenotypic divergence of adipose deposition between these two pig breeds that have undergone artificial selection for different breeding targets. The Landrace breed is under strong artificial selection for more muscle production, while the Rongchang breed is selected for more adipose deposition.

Validation of gene expression changes by Q-PCR
The expression changes of nine genes that are wellknown to be related to adipose metabolism (ACADL, ADIPOQ, ADIPOR 1, CAV 3, MDH 1, MDH 2, ME 1, Insig 1 and SCAP) showed significant positive correlations between the Q-PCR results and their expressions determined by microarray (P < 0.05, Figure 8), which highlighted the high confidence of the results from the microarray approach.     Co-expressed gene sets linked to the obesity-related phenotypes Generally, quantitative characters (such as adipose deposition) in mammals are determined by a set of genes that cooperate in certain biological processes [11]. To further explore the relationships between the genes and obesity-related traits, we applied a clustering method to identify sets of functionally related genes linked to the phenotypic traits as reported previously [19].
We identified eight gene modules containing 947 genes for ULB and twelve gene modules containing 1,277 genes for ILB. In three typical VATs, we identified nine, nine and eleven gene modules containing 933, 1,058 and 1,409 genes for the MAD, GOM and RAD tissues, respectively. For the IAD, we identified eleven modules containing 1,423 genes ( Figure 9).
Next, we explored the correlations between the modules of co-expressed genes and the various obesity phenotypes. Three gene modules in the SATs (i.e. ULB and ILB) were significantly correlated with adipocyte volume and the concentration of plasma lactate dehydrogenase in serum (false discovery rate adjusted permutation, P < 0.05) ( Figure 10 and Table 2). The co-expressed genes within these modules were mainly involved in metabolic processes, such as fatty acid metabolic process (23 genes, P = 4.38 × 10 -5 ), fatty acid beta-oxidation (3 genes, P = 2.65 × 10 -3 ), coenzyme metabolic process (6 genes, P = 4.77 × 10 -5 ) and carbohydrate metabolic process (12 genes, P = 2.94 × 10 -3 ) (Table 3), in agreement with an earlier report that SATs contribute mainly to metabolic processes [20,21].
Similar to the findings for the high-risk VATs, the coexpressed genes in IAD were also found to be significantly correlated with immune-related TNF-α ( Figure 10 and Table 2), further suggesting that the IAD is an independent risk factor for metabolic diseases [25].

Conclusions
The present study describes a genome-wide analysis of gene expression profiling among various adipose depots between the sexes of three well-defined pig breeds displaying distinct adipose phenotypes. We present evidence that the SATs mainly affect metabolic processes, while the VATs and the IAD are associated mainly with immune and inflammation responses that are seen as the metabolic risk factors of obesity. This study also supports pig as an ideal model for studying human obesity.

Tissue collection
Three females and three males at 210-days-old for each of the leaner Landrace pigs (a Western breed), the wild Tibetan pigs (a feral, indigenous Chinese pig that has not undergone artificial selection) and the fatty Rongchang pigs (a Chinese breed) were used in this study as previously described [3]. Animals were humanely sacrificed, in compliance with experimental animals established by the Ministry of Agriculture of China. Three VATs (i.e. GOM, MAD and RAD), two SATs (i.e. ULB and ILB) as well as IAD were rapidly separated from each carcass, immediately frozen in liquid nitrogen, and stored at −80°C until RNA extraction. For more information, please refer to Li et al. [3].

Measurements of the obesity-related phenotypes
Measurements of the pig body density, the concentrations of 24 serum-circulating indicators of metabolism and the adipocyte volume are from our previous report based on same individuals. For more information, please refer to Li et al. [3].

RNA isolation and microarray analysis
The total RNA (10 μg) of the 108 tissue samples was extracted with TRIzol (Invitrogen) and further purified using an Rneasy column (Qiagen). Hybridization, washing, and scanning were done according to standard Agilent protocols. The data analysis was performed with MultiExperiment Viewer (MeV) [24]. Array data have been uploaded to NCBI's Gene Expression Omnibus

Identification of differentially expressed genes
We used three-way repeated-measures ANOVA (n = 3 per breed per sex per tissue) to identified differentially expressed (DE) mRNA among six adipose tissues between sexes of three pig breeds (P < 0.05, corrected with adjusted Bonferroni method, FDR <0.05, 1,000 permutations), which were further analyzed using DAVID to examine whether these DE genes were enriched for specific functional catalogs (i.e. KEGG pathways and Gene Ontology-Biological Processes and -Molecular Functions) (P < 0.05, corrected with adjusted Bonferroni method) [26].

Identification of co-expressed gene modules
To identify sets of functionally related genes in six adipose tissues, linked to the phenotypic traits, we applied a clustering method as described previously [4,19] with some modifications. Spearman rank correlation coefficients were determined between all possible gene-pairs across 18 samples for each adipose tissue. The strongest correlated gene-pair was selected, and grouped together in a set that was assigned the average expression value of the two genes that constitute this set. After addition of this newly created set to the dataset, the two individual genes were removed from the data and the strongest correlation in the dataset was again selected. This resulted in either the expansion of a set already created or in the creation of a new set. We kept repeating this as an iterative process until the most significantly correlated pair was coefficient of correlation r < 0.80. Only the sets containing 50 or more genes are kept for further analysis. The gene sets -reflecting the average expression value of the genes constituting that set -were correlated with obesity phenotypic traits using a Spearman rank correlation coefficient.

Quantitative PCR
Quantitative PCR (Q-PCR) was used to confirm the expression pattern observed in the microarray results. Q-PCR was performed on a CF96 Real-Time PCR Detection System (Bio-Rad) using SYBR® Green Real-time PCR Master Mix (TaKaRa, China). The PCR primer sequences are shown in Table 4. Porcine ACTB, TBP and TOP2B were simultaneously used as endogenous control genes. The 2 -ΔΔCt method was used to determine the relative mRNA abundance for the surveyed samples.