The sBF and dBF tissues, which are located in different anatomical locations (Figure 1a), exhibited significantly different adipocyte volumes (p = 0.043) (Figure 1b); this confirms the findings of a previous report in which the adipocytes from deep sites were smaller than those from superficial sites [10]. In addition, the sBF and dBF tissues had distinct fatty acid compositions for saturated fatty acids (p = 0.005) and monounsaturated fatty acids (p = 0.038) (Table 1), which suggests that their metabolic activities, such as the rate of deposition and mobilization, and also the rate of endogenous synthesis of fatty acids, differ between the different body sites [16]. These phenotypic differences between the sBF and dBF imply intrinsic differences in their molecular regulation.

We generated 42.80 gigabases (Gb) of methylated DNA immunoprecipitation sequencing (MeDIP-seq) data from six samples (~7.13 Gb per sample), of which 32.91 Gb (76.89%) of clean reads could be aligned on the pig reference genome. We used statistics to measure the methylation rate changes, and defined DMRs between the sBF and dBF tissues. The high correlation (average Pearson’s r = 0.995) between the number of DMRs (Figure 2a), the number of CpGs within the DMRs (Figure 2b), and the length of the DMRs (Figure 2c) implied that DMR detection in regions of different length and number of CpGs was non-biased. The number of DMRs varied considerably among 10 defined features of canonical gene structure: the distal (1254 DMRs), intermediate (877 DMRs), and proximal (1020 DMRs) regions of the promoter, the first exon (931 DMRs), the first intron (8518 DMRs), internal exons (6547 DMRs), internal introns (27,615 DMRs), the last exon (1429 DMRs), 2 kb downstream of the transcription end site (TES) (2356 DMRs), and intergenic regions (61,807 DMRs). We then looked at the percentage of CpGs within the DMRs for each genomic feature (Figure 2d). Intriguingly, DMRs occurred more frequently in exons (29.96%) than in promoters (% of CpGs within the DMRs, 7.22%), introns (9.95%), 2 kb downstream of the TES (9.97%), or intergenic regions (7.18%). A previous report indicated that 16% of all CpG islands in the human brain were methylated, whereas 98% of CpG islands associated with annotated promoters were unmethylated [17]. This suggests that in the human brain, DNA methylation may serve a broader role in intragenic regions than in promoters. DNA methylation in promoters suppresses gene expression, yet in the main body of the gene (particularly in coding exons), it is also important in regulating alternative promoters and preventing spurious transcription initiation [18].

We explored the correlation between the changes of DMR in promoters and the associated mRNAs expression levels. To obtain high-confidence gene expression data, we mapped 43,603 60-mer probes of a microarray to the pig reference genome; this resulted in 4983 (11.43%) probes uniquely mapped to exons of Ensembl genes. Multiple probes that mapped to the same or different exons of a specific gene were filtered out. Only 3074 probes (7.05%), uniquely representing 3074 genes, were used in subsequent analyses.

As shown in Figure 3, there was a negative correlation between mRNA expression and DMRs across the entire promoter (Pearson’s r = −0.205, p = 4.269 × 10⁻⁵). In detail, the correlation coefficient in the different promoter regions decreased from the proximal (Pearson’s r = −0.255, p = 0.001) to the intermediate (Pearson’s r = −0.189, p = 0.037) to the distal (Pearson’s r = −0.165, p = 0.016) region. This validates previous findings in human melanoma cells [19]. Transcriptional repression driven by promoter methylation is inversely related to the distance between the promoter sub-regions and the transcription start site.

To identify genes potentially responsible for functional and metabolic differences between the sBF and dBF, we performed functional enrichment analysis of Gene Ontology (GO) for genes with DMRs in their promoters using DAVID software [20]. We found that the top 10 significantly overrepresented categories of GO biological processes (GO-BP) were related to the two main classes of biological functions (Figure 4), i.e., ‘lipid metabolism’ (such as catabolic process, glycerolipid metabolic process, lipoprotein metabolic process, lipid transport, and regulation of appetite) and “regulation of immune-related cytokines” (such as cytokine production, immune response-regulating signal transduction, immune response signaling pathway, interleukin 1 production, and JAK-STAT cascade), which reflects the distinct roles of the sBF and dBF in energy homeostasis and adipokine-induced immune responses [21,22].

Notably, the most significantly overrepresented category was enriched for 13 genes related to “positive regulation of cytokine production”, of which the promoters of 11 (84.62%) genes were significantly hypermethylated in the sBF compared with the dBF (Figure 5). This suggests a higher mRNA expression level of these genes in the dBF compared with the sBF, which may result in higher cytokine levels secreted from the dBF than from the sBF. It is well established that adipocyte-derived adipokines can regulate systemic processes, particularly those involved in insulin sensitivity, immune responses, and inflammation [21,23]. These results are consistent with those of a previous report, which found that, compared with human superficial SAT, deep SAT appears to be a distinct adipose depot that supports an independent metabolic function, and may be associated with the risk of obesity-associated complications [8]. It is believed that the SATs can have direct and beneficial effects on the control of body weight and metabolism [24]. Visceral adipose tissues were negatively associated with peripheral insulin sensitivity, while thigh SAT, in contrast, was positively associated with peripheral insulin sensitivity [25]. Our findings of anatomical location-specific methylation patterns imply intrinsic functional and metabolic differences between the superficial and deep SATs. A recent study of the autologous transplantation of visceral adipose tissues to subcutaneous sites demonstrated the impact of local (residence) factors influencing the epigenetic memory of adipose depots in different body sites [26].

Nine healthy female Landrace pigs at 210 days old were used in this study. After sacrifice, the sBF and dBF tissues near the last third or fourth rib were rapidly separated from each carcass, immediately frozen in liquid nitrogen, and stored at −80 °C until RNA and DNA extraction.

All adipose samples were fixed in 10% neutral buffered formalin solution, embedded in paraffin, sliced at a thickness of 6 μm using an RM2135 rotary microtome (Leica, Berlin, Germany) and stained with hematoxylin and eosin. The mean diameter of an adipocyte cell was calculated as the geometric average of the maximum and minimum diameter, and 100 cells were measured for each sample in randomly selected fields. The mean adipocyte volume (V) was obtained according to the following Formula 1:

where D_i is the mean diameter, and f_i denotes number of cells with the mean diameter D_i.

We determined the fatty acid composition of all adipose samples as previously described [27]. The fatty acid methyl esters (FAMEs) were quantified using GC-14C gas chromatography (Shimadzu, Kyoto, Japan). Response factors were determined by analyzing a standard solution of the relevant pure FAME. Individual compounds were identified by comparing their retention times with those of standards (Sigma-Aldrich, MO, USA). FAMEs were identified by comparison with standards previously run independently or together with samples.

We randomly selected three pigs as biological replicates. MeDIP DNA libraries were prepared following a previously described protocol [28]. In brief, DNA (5 μg) was sonicated to approximately 100 to 500-bp fragments with a Bioruptor sonicator (Diagenode, NJ, USA). Then, libraries were constructed using a Paired-End DNA Sample Prep kit (Illumina, CA, USA) following the manufacturer’s instructions. Adaptor-ligated DNA was immunoprecipitated by a monoclonal anti-methylcytidine antibody (Diagenode, NJ, USA). The enriched methylated fragments and 10% input DNA were purified on DNA Clean & Concentrator-5 columns (Zymo, CA, USA) following the manufacturer’s instructions. Enriched fragments were amplified by adaptor-mediated PCR. Each MeDIP library was subjected to paired-end sequencing with 50-bp read-lengths using an Illumina HiSeq 2000 Sequencing System. The MeDIP-seq data have been deposited in the NCBI Gene Expression Omnibus under the GEO Series accession number GSE30344.

After filtering the low-quality reads, the MeDIP-seq data were aligned to the UCSC pig reference genome (Sscrofa9.2) using SOAP2 (version 2.21) [29]. The genomic regions differentially enriched in methylated CpGs between the sBF and dBF were identified using our newly developed method. First, the normality and equal variances of read depth at each CpG across different sample groups were tested using Bartlett’s test (passed for p > 0.05, failed for p < 0.05). Second, the parametric Student’s paired t-test (if the data passed the Bartlett’s test) or the non-parametric Wilcoxon rank-sum test (if the data failed the Bartlett’s test) were used to select the highly variable CpGs (p < 0.01) as the seed site of a candidate DMR. Third, the 3′ downstream adjacent CpGs were singly incorporated with this seed CpG. To highlight the CpG-enriched regions, we allowed up to 200-bp separation between two adjacent CpGs, with which ~88.8% of all CpGs in the pig genome were covered (Supplementary Figure S1). The average read depth of a defined multi-CpG region was repeatedly subjected to a new round of tests. The resulting P values for DMRs were corrected using the Benjamini-Hochberg method (false discovery rate < 0.01; 1,000 permutations). If five or more CpGs in a genomic region demonstrated a significantly different (p < 0.01) read depth between samples, this region was taken to be a DMR.

We identified the genomic locations of the promoters, exons, and introns of the 21,533 Ensembl genes, together with the 13,626 intergenic regions, by referring to the UCSC pig reference genome (Sscrofa9.2). Each promoter of 2700-bp length was divided into three regions as previously described [19]: proximal (−200 to +500 bp), intermediate (−200 to −1000 bp), and distal (−1000 to −2200 bp).

Total RNA (10 μg) from three sBF and three dBF samples, which corresponded to the samples used for MeDIP-seq sequencing, was extracted with TRIzol (Invitrogen, CA, USA). The labeling procedure (Cy-3 dye only) was carried out using an RNA Fluorescent Linear Amplification Kit (Agilent Technologies, Santa Clara, CA, USA). The fragmented target was applied to a Pig Gene Expression Oligo Microarray (version 2; Agilent Technologies, Santa Clara, CA, USA, 2011). Data analysis was performed with MultiExperiment Viewer (MeV) [30]. The microarray data have been deposited in the NCBI Gene Expression Omnibus under the GEO Series accession number GSE30343.