Between June 2009 and March 2010, 19 patients undergoing elective cardiac operations at Massachusetts General Hospital were enrolled in this study. From these patients, RNA samples were isolated from SQAT, PCAT, sqAds, and pcAds, of which 53 samples (11 SQAT, 11 PCAT, 15 sqAds, and 16 pcAds) passed quality standards for hybridization to Affymetrix U133A Plus 2.0 microarrays. Using nuclear and lipid stains, we determined that our isolated adipocytes contained 96.7% single-nuclei adipocytes (Fig. S1
As our primary aim was the identification of depot-specific transcription patterns, we first performed unbiased hierarchical clustering of whole tissue and isolated adipocytes. This showed clustering of samples which were processed in one specific month, suggesting a strong batch effect (Fig. S2
). To remove this effect, principal component analysis was performed. After correction for batch effects in processing (see Methods) clustering analysis revealed 3 distinct clusters. One cluster was composed solely of pcAds samples, a second cluster contained 10 samples of which 9 were pcAds, and a third cluster of 12 samples contained 9 pcAds samples ().
Experimental overview and clustering results.
We then used linear models to identify differentially expressed genes between populations PCAT and SQAT, and between pcAds and sqAds. Given the multiple hypothesis testing burden of microarray analyses, a nominal p-value of 0.05 almost certainly includes a large number of false positives. We thus restricted our analysis to those genes that were differentially at a false-discovery rate (FDR) of <0.25 (, Methods). In PCAT vs. SQAT, this approach identified 2,284 differentially expressed probesets (). In pcAds vs. sqAds, this analysis yielded 657 differentially expressed probesets.
To characterize the differentially expressed transcripts, we used unbiased gene enrichment analysis (GEA) to look for enriched Gene Ontology (GO) terms in the candidate probesets overexpressed in PCAT, SQAT pcAds and sqAds individually. Since multiple probesets map to the same genes, a list of unique candidate genes were generated from probeset lists. These candidate genes were ranked in order of greatest fold-change and analyzed for GO category enrichment as ordered lists using FuncAssociate 2.0. 
Our GEA revealed that the four tissue types were each enriched for multiple gene ontology categories ( and Table S1
). Because we analyzed both whole tissue and isolated adipocytes, we were also able to infer whether the differentially expressed genes arise from predominantly either the adipocytes themselves, or from the more heterogeneous adipose tissue mixtures.
Enriched Gene Ontology (GO) Categories in sqAds and pcAds identified by GEA.
In sqAds, our GEA revealed a striking enrichment of pattern specification genes (). Because the current GO classifications do not include all known homeobox genes, we manually referenced our differentially expressed genes against a comprehensive list of human homeobox genes 
. Compared to pcAds, sqAds demonstrated increased expression of 12 homeobox genes, including PAX3, HOXA10, HOXA9, and HOXB7 (). All of the homeobox genes that were relatively higher in sqAds were also increased in SQAT vs. PCAT (Table S2
). For example, PAX3 was 3.3-fold (p
0.00002) increased in sqAds vs. pcAds and 3.8-fold (p
0.0004) increased in SQAT vs. PCAT.
When we looked at homeobox expression in the pericardial depot, we found significantly increased expression of two homeobox genes, HOXA2 (2.53-fold, p
0.00012) and SATB1 (1.84-fold, p
0.015), in both pcAds and PCAT ( and Table S2
). In aggregate, the observed transcriptional patterns of these 14 homeobox genes constitute depot-specific signatures.
GEA also found that our pcAds were significantly enriched for immune response genes (p-adj<0.001, and Table S3
), with 26 immune response genes found among the top 2500 genes. A similar result was seen in PCAT (p-adj
0.005, Table S1
). These genes included 7 chemokines, such as chemokine (C-C motif) ligand 4 (CCL4) which was overexpressed 2.79-fold (p<0.0001) in pcAds and 2.9-fold (p
0.0009) in PCAT. To better visualize the molecular relationships between these inflammatory genes, we mapped the overexpressed genes to KEGG pathways (), revealing that the chemokines upregulated in pcAds belong to the tumor necrosis factor (TNF), CXC and CC chemokines, and IL-1 families. Since expression levels were not significantly lower in pcAds compared to PCAT, it is possible that pericardial adipocytes are a primary site for the synthesis of inflammatory mediators. We thus sought to determine, using qPCR, the expression of CCL4 in isolated pcAds and isolated pericardial stromal vascular fraction (SVF). We found 12.5-fold (p
0.005) higher expression of CCL4 in the pericardial SVF vs. the pcAds fraction. Altogether, these data demonstrate that pcAds express inflammatory mediators but not to the same degree as the SVF, which contains immune and inflammatory cells. In contrast to PCAT, our GEA revealed that SQAT was significantly enriched in many metabolism-related GO categories (Table S1
), including oxidoreductase activity (p-adj
0.002) and lipid metabolic processes (p-adj
0.008). KEGG pathway-mapping did not identify any recognized, adipose-tissue specific pathways.
Chemokines overexpressed in pcAds.
To validate the differential gene expression patterns identified by our microarray analysis, we performed qPCR analysis on patient-matched sqAds and pcAds samples (). For this analysis, we selected homeobox genes HOXA9 and HOXB7; and the adipocyte identity gene leptin. For all three genes, the expression trend matched our microarray results.
Quantitative PCR confirms differential expression of homeobox genes HOXA9 and HOXB7; and adipocyte-identity gene leptin (LEP) in patient-matched, isolated subcutaneous and pericardial adipocytes.
Lastly, we confirmed that a core set of adipocyte identity and function genes was not differentially expressed in pcAds vs. sqAds (Table S4
). The genes we selected included the well-studied adipocyte identity maintenance factors perilipin (PLIN1) and cell death-inducing DFFA-like effector c (CIDEC); the transcription factors Peroxisome proliferator-activated receptor gamma (PPARγ) and the CCAAT/enhancer binding proteins (CEBPA, CEBPB, CEBPD); the lipid catabolism enzymes lipoprotein lipase (LPL) and hormone-sensitive lipase (LIPE); the adipocyte anabolic enzymes fatty acid binding protein 4 (FABP4) and diacylglycerol O-acyltransferase 1 (DGAT1); and the adipokines leptin (LEP), adipsin (CFD), and adiponectin (ADIPOQ). We also considered the brown fat identity gene uncoupling protein 1 (UCP1) and found ≤0.50 fold change between the sample classes.