Four-way identification of DEGs
Fig. 1 shows a flowchart of the study procedure. After producing mRNA-seq data from the purified ACs and preACs of lean and obese individuals, we investigated two different questions: 1) which genes are significantly upregulated or downregulated under obese conditions in comparison with lean conditions, and 2) which alterations in gene expression detected for obese adipogenesis (i.e., the information extracted from the DEGs obtained by comparing ACs and preACs from obese individuals) are significantly different from those detected for lean adipogenesis (i.e., the information extracted from the DEGs obtained by comparing ACs and preACs from lean individuals). As a validation, we compared our conclusion with that of a previously published paper that provided a list of DEGs obtained by comparing gene expression between lean and obese ACs derived from lean and obese SAT samples; the original dataset produced by the microarray platform was downloaded from the GEO database (GSE80654). How similar that list of DEGs is to our results discussed later in the DISCUSSION section.
We decided to estimate four different classes of DEGs from four different types of mRNA-Seq data, i.e., ‘lean AC’ (L-AC), ‘obese AC’ (O-AC), ‘lean preAC’ (L-preAC), and ‘obese preAC’ (O-preAC) (see Materials and methods) (Additional file 1: Figure S1): ‘Class I: AC-DEGs’ from the comparison of expression between ‘L-AC’ and ‘O-AC’, ‘Class II: preAC-DEGs’ between ‘L-preAC’ and ‘O-preAC’, ‘Class III: Lean_Ag-DEGs’ between ‘L-preAC’ and ‘L-AC’, and ‘Class IV: Obese_Ag-DEGs’ between ‘O-preAC’ and ‘O-AC’. Various thresholds were tested to select DEGs (Additional file 2: Table S1), and DEGs were identified for the four classes mentioned above. Class I and II DEGs were investigated to answer question #1 described above, i.e., to determine the differences between obese ACs and lean ACs and between obese preACs and lean preACs. Class III and IV DEGs were chosen to answer question #2 described above, i.e., to determine how gene expression is altered during obese and lean AC adipogenesis.
Defining intermediate obesity samples in ‘Class I: AC-DEGs’
A total of 1,198 genes and 314 genes were classified as ‘Class I: AC-DEGs’ with P < 0.01 and Q < 0.05, respectively (Additional file 2: Table S1). ‘L-AC’ samples could not be differentiated from ‘O-AC’ samples by either of these DEGs in the analysis of unsupervised clustering processed with a heatmap; the samples in the middle of the heatmap did not show gene expression patterns pertinent to the ‘L-AC’ and ‘O-AC’ categories (Fig. 2A). Notably, gene ontology (GO) analysis showed that canonical obesity-related genes involved in inflammation and ECM were downregulated in obese ACs rather than in lean ACs (Fig. 2B). The ambiguity of sample classification by these DEGs was also confirmed in principal component analysis (PCA) (Fig. 2C). Thus, we subcategorized these ambiguous samples separately into the ‘intermediate (I-AC)’ group (#9, 24, 30, 35, 36, 40, 42, 51). The remaining two extreme samples were then named lean-extreme (‘Le-AC’) and obese-extreme (‘Oe-AC’), which were ultimately categorized as 8 ‘Le-AC’, 8 ‘I-AC’, and 7 ‘Oe-AC’, as indicated on top of the heatmap (Additional file 3: Table S2).
Subsequently, for these redefined three groups of samples, some obesity-related clinical information was investigated, including BMI, waist circumference (WC), fasting plasma glucose (FPG) level, C-peptide (C-pep), high-density lipoprotein (HDL) level, and low-density lipoprotein (LDL) level (Fig. 2D). Interestingly, ‘I-AC’ samples were special in that the BMIs of ‘I-AC’ were expectedly positioned between ‘Le-AC’ and ‘Oe-AC’; however, the WC and FPG level of the ‘I-AC’ samples were comparable to those of ‘Le-AC’. In addition, except for BMI, WC, and FPG, all the other clinical levels showed no significant difference among these three groups, although ‘I-AC’ was located between ‘Le-AC’ and ‘Oe-AC’. Notably, FPG levels seemed to be associated with WC rather than with BMI.
The existence of a third group, ‘I-AC’ samples, may not be surprising, considering the complexities of molecular etiologies causing obesity involved with various genetic and epigenetic alterations and the unclear association between obesity and obesity-related metabolic diseases.
Inflammatory genes are expressed at lower levels in obese ACs than in lean ACs
DEGs were re-estimated involving the ‘I-AC’ in three different comparison sets, between ‘Le-AC’ and ‘Oe-AC’ (named ‘LO-DEGs’), between ‘Le-AC’ and ‘I-AC’ (named ‘LI-DEGs’), and between ‘I-AC’ and ‘Oe-AC’ (named ‘IO-DEGs’), with various thresholds (Additional file 4: Table S3). To understand alterations in gene expression related to obesity, we applied different thresholds to produce similar numbers of DEGs for these three categories (indicated ‘*’ in Additional file 4: Table S3). Consequently, a total of 2,657 (Q < 0.01), 1,474 (P < 0.01), and 1,324 (Q < 0.05) DEGs were selected for ‘LO-DEGs’, ‘LI-DEGs’, and ‘IO-DEGs’, respectively; the heatmap of each group was constructed to visualize gene expression differences (Fig. 3A). Notably, the highest number of genes was allocated in ‘LO-DEGs’, despite the stringent threshold applied. Note that we focused on collecting similar numbers of DEGs rather than on determining a single criterion or a single most important gene (although all the thresholds were chosen in ranges considered statistically significant) to reveal trends in gene expressions between two different conditions.
A striking observation emerged from GO analysis. Specifically, a total of 1,874 genes of the 2,657 ‘LO-DEGs’ (70.5 %), i.e., genes largely assigned to the inflammatory response and cell adhesion, were expressed at significantly lower levels in ‘O-AC’ than in ‘L-AC’ (Fig. 3B, in the wide blue box). A similar result was observed in Fig. 2B, and the downregulation of these genes seems to be amplified when ‘I-AC’ samples are excluded from the DEG analysis. Moreover, ‘IO-DEGs’ revealed the same pattern as did ‘LO-DEGs’ (Fig. 3B, in the wide blue box). This observation is striking because these classes of genes are all canonical obesity-related genes that are known to be expressed at higher levels in obese AT [19,20].
By contrast, other DEGs involved in fat metabolism in ‘LO-DEGs’ and ‘IO-DEGs’ were consistent with the previous findings, i.e., upregulation under obese conditions. For instance, a total of 783 genes of the 2,657 ‘LO-DEGs’ (29.5 %), including LEP, CES1, and NQO1, that were expressed at higher levels in ‘Oe-AC’ than in ‘Le-AC’, were largely assigned to mitochondrial metabolism and the oxidation-reduction process (ROS) (Fig. 3B, in the wide red box).
‘LI-DEGs’ were also assigned to distinctive GO functions; RNA metabolism genes were expressed at lower levels in ‘I-AC’ than in ‘Le-AC’ (i.e., downregulated; narrow blue box in Fig. 3B), and genes involved in centrosome organization and protein phosphorylation were expressed at higher levels in ‘I-AC’ than in ‘Le-AC’ (i.e., upregulated; narrow red box in Fig. 3B), confirming that ‘I-AC’ is distinct and not comparable to ‘Le-AC’ or ‘Oe-AC’.
Gene set enrichment analysis (GSEA), i.e., a tool designed to see whether an a priori defined set of genes shows significant differences in expression between two different biological conditions, led to the same conclusion as did GO analysis. Specifically, genes belonging to inflammatory or angiogenesis functions were significantly upregulated in lean ACs rather than obese ACs, whereas genes belonging to cellular respiration functions were significantly upregulated in obese ACs (Additional file 5: Figure S2).
Detailed examination of expression alterations between lean and obese ACs
The regrouped ‘Le-AC’, ‘I-AC’, and ‘Oe-AC’ were assumed to reflect different degrees of obesity based on BMI, as shown in Fig. 2D. Thus, changes in gene expression can be examined further by investigating ‘LI-DEGs’ and ‘IO-DEGs’. For instance, if a gene that is upregulated both in ‘LI-DEGs (i.e., genes expressed at higher levels in ‘I-AC’ than in ‘Le-AC’) and ‘IO-DEGs (i.e., genes expressed at higher levels in ‘Oe-AC’ than in ‘I-AC’), it can be concluded that the gene is ‘progressively upregulated (named ‘progressive-up’) because it is upregulated in ‘Le’ compared with that in ‘I’ and upregulated again in ‘I’ compared with that in ‘Oe ’ (refer to Additional file 6: Figure S3). Similarly, if a gene is downregulated in both ‘LI-DEGs’ and ‘IO-DEGs’, the gene is defined as progressively downregulated (named ‘progressive-down’). Using this scheme, genes were allocated into eight different categories as shown in Fig. 4 and Additional file 6: Figure S3. As a result, only seven genes, including ACAA1, KLHL22, and AKR1C3, were identified as ‘progressive-up’, while a total of 66 genes were categorized as ‘progressive-down’. Notably, genes assigned to cell migration, cell adhesion, and angiogenesis were allocated to the ‘progressive-down’ category.
Cell cycle and metabolic process genes were upregulated in ‘I-AC’ compared with those in ‘L-AC’ and sustained their expression in ‘O-AC’, i.e., upregulated genes in the ‘LI-DEGs’ category but not in the ‘IO-DEGs’ category (named ‘initial-up’). By contrast, RNA processing, angiogenesis, and signal transduction genes were downregulated in ‘I-AC’ and sustained their expression in ‘O-AC’, i.e., downregulated genes in the ‘LI-DEGs’ category but not in the ‘IO-DEGs’ category (named ‘initial-down’) (Fig. 4). This result indicates that angiogenesis alteration and cell proliferation may start in the early stage of obesity. Genes involved in ROS metabolism and fatty acid biosynthetic process genes were upregulated in the later stage in obesity, i.e., genes not in the ‘LI-DEGs’ category but upregulated in the ‘IO-DEGs’ category (named ‘later-up’). By contrast, genes involved in inflammation, cell adhesion, and ECM organization were downregulated in a later stage in obesity, i.e., genes not in the ‘LI-DEGs’ category but downregulated in the ‘IO-DEGs’ category (named ‘later-down’) (Fig. 4). Notably, genes in the ‘later-down’ category showed that most canonical obesity genes were downregulated in in the later stage of obesity.
Comparison of gene expression profiles between L-preACs and O-preACs
We obtained a total of 213 ‘Class II: preAC-DEGs’ estimated from 3 ‘L-preACs’ and 10 ‘O-preACs’ (Additional file 2: Table S1, Additional file 3: Table S2), as shown in a heatmap (Fig. 5A). The expression differences were somewhat vague between L-preACs and O-preACs (Fig. 5A). Unlike ‘Class I: AC-DEGs’ in Fig. 2, unsupervised clustering and PCA analysis produced a distinctive classification between ‘L-preAC’ and ‘O-preAC’ (Additional file 7: Figure S4), mainly due to the smaller sample size.
Remarkably, GO analysis of ‘Class II: preAC-DEGs’ revealed the opposite direction for some gene expression levels compared to those observed for ‘Class I: AC-DEGs’. Inflammatory response and chemotaxis genes, i.e., genes that were downregulated in AC comparisons, were upregulated in obese samples compared with lean samples (Fig. 5B). Interestingly, a similar inverse profile between AC and preAC has been reported for miRNAs , implicating crosstalk between AC and preAC during AC differentiation. However, cell adhesion genes were significantly downregulated in obese samples, consistent with the direction found in AC comparisons.
Both lean and obese adipogenesis are required for the enhancement of inflammatory genes
Comparing expression between preACs and ACs was assumed to reveal the changes in gene expression during the process of AC differentiation from preACs (i.e., during adipogenesis). Under this assumption, we investigated how obese AC adipogenesis is differentiated from lean AC adipogenesis by obtaining Class III and Class IV DEGs; ‘Class III: Lean_Ag-DEGs’ were estimated by comparing ‘L-preAC’ and ‘Le-AC’, and ‘Class IV: Obese_Ag-DEGs’ were obtained by comparing ‘O-preAC’ and ‘Oe-AC’ (Additional file 1: Figure S1 and Additional file 2: Table S1). For the two classes, ‘upregulation’ or ‘downregulation’ was determined based on the expression between ACs and preACs, i.e., genes that were expressed at higher levels in ACs than in preACs were upregulated genes, and genes that were expressed at lower levels in ACs than in preACs were downregulated genes.
We found that alterations in gene expression between preACs and ACs measured in the lean condition (i.e., ‘Lean_Ag-DEGs’) was significantly positively correlated with alterations in gene expression between preACs and ACs measured in the obese condition (i.e., ‘Obese_Ag-DEGs’) (g=0.96 and p < 2.2e-16, Pearson correlation) (Fig. 6A), indicating that AC differentiation from preACs to ACs involves similar gene expression alterations for both the lean and obese conditions. The degree of alterations in gene expression between preACs and ACs was extremely large, regardless of whether lean or obese samples were assessed; a total of 8,448 genes and 10,234 genes were identified as ‘Class III: Lean_Ag-DEGs’ and ‘Class IV: Obese_Ag-DEGs’, respectively at Q < 0.01. We categorized these DEGs into four subcategories by considering the log2 fold change (log2FC) in gene expression along with the Q < 0.01 threshold (Additional file 8: Figure S5). Subsequently, for each of these categories, Class III and Class IV intersected, which led to three subcategories of DEGs, i.e., ‘lean AC adipogenesis-specific (LS)’, ‘obese AC adipogenesis-specific (OS)’, and ‘commonly altered for both (CA)’. Interestingly, most inflammatory genes, such as leukocyte migration, cell chemotaxis, and complement activation genes, were allocated in the highest upregulation category (Q < 0.01 and |log2FC| > 4) for both lean and obese AC adipogenesis (‘CA’) (Additional file 8: Figure S5B and Additional file 9: Table S4). Surprisingly, the magnitude of upregulation of these inflammatory genes in ‘LS’ was significantly higher than that in ‘OS’ (Additional file 8: Figure S5B), indicating that both lean and obese adipogenesis are coupled with increased expression of inflammatory genes and that lean adipogenesis, rather than obese adipogenesis, requires stronger upregulation of these inflammatory genes. Consistently, these inflammatory genes showed greater upregulation in ‘Lean_Ag-DEGs’ than in ‘Obese_Ag-DEGs’ (Fig. 6A). Particularly, we confirmed that the extent of upregulation of three genes including IL-6, IL-1b, and TNF-a, i.e., the three most studied inflammatory genes related to obesity, was significantly greater in ‘Lean_Ag-DEGs’ than in ‘Obese_Ag-DEGs’ (p < 0.01) (Fig. 6B). Certainly, these results explain why the expression levels of these genes in obese ACs were lower than those in lean ACs (Fig. 1)