Obesity is a rising threat to public health with comorbidities including type 2 diabetes and hypertension.¹ To date, bariatric surgery is the most effective procedure to treat obesity and achieve weight loss.^1,2 The benefits of bariatric surgery, including improved outcomes from multiple comorbidities and reduced all-cause mortality, far outweigh the low short-term risks and rare complications.¹

The molecular and cellular mechanisms behind the effectiveness of bariatric surgery remain elusive.³ One hypothesis is that the bariatric procedure reprograms metabolic networks in adipose tissue. Subcutaneous adipose tissue (SAT) is a major subtype of human adipose with an important role in fat storage and has been implicated in obesity.^4,5 Proteome-wide profiling, validated by targeted experiments, showed substantially reduced fat stored in SAT after bariatric surgery.²

Transcriptome-wide studies also attempted to address the biological changes after bariatric surgery.^3-6 However, existing studies in SAT profiling suffer from small sample sizes with as few as eight subjects measured in multiple batches.⁴ The present study leverages publicly available transcriptomic datasets and conducts pooled profiling with rigorous statistics to investigate the effect of bariatric surgery on SAT gene expression.

Datasets access and processing

Summarized in Fig. 1A, Supplementary Table 1, five datasets (GSE29411, GSE42715, GSE65540, GSE66921, and GSE199063) were publicly accessed from Gene Expression Omnibus (www.ncbi. nlm.nih.gov/geo).^3-7 The study is Institutional Review Board exempt because all clinical information was de-identified, previously published, and publicly available. Further processing of gene expression data was performed in two phases. First, genes coded outside of chromosomes 1-22 and X, measured by control probes, without the Hugo/RefSeq annotation, or having flat-zero expression were excluded. Expression values of genes with duplicate names were mean-aggregated, standard-normalized, and subjected to a 0.5% least-variant filter. Second, all datasets were inner-joined followed by batch-effect correction (sva-ComBat3.38.0⁸), constrained within ±5.0, and standard-scaled. Non-SATs and one SAT with distinct global expression were excluded, leaving n=237 from 118 unique subjects. Pre-operative (PreOP) SATs were defined as being collected before or at the time of bariatric surgery, while post-operative (PostOP) SATs were collected 3 to 60 months after surgery.

Unsupervised analyses

The first two principal components (PCs) were computed on all 5,000 genes with R function ‘prcomp.’ PC regions were identified based on whether a sample is above or below the negative identity line. Unsupervised hierarchical clustering with Manhattan distance and Ward-D linkage was performed on 1,188 (23.8%) genes with inter-sample standard deviation exceeding 0.99 (‘pheatmap1.0.12’).

Differential gene expression

A robust linear mixed effect model (LIMMA3.46.0)⁹ was used to identify differentially expressed genes in PostOP SATs with explicit consideration of the 49.8% (118 of 237) samples from repeated measurements, as follows:

PostOP_yes=FixedEffect (gene)+RandomEffect (subject, withinsubject correlation coefficient)

The significantly over- and under-expressed genes, identified at Bonferroni P<0.05, were visualized with EnhancedVolcano1.8.0 and separately tested for Gene Ontology biological processes noredundant and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment using all 5,000 genes as the background set on the WebGestalt server (http://www.webgestalt.org).¹⁰

Statistical analysis

Two-sided Welch’s t and Fisher’s exact tests were used to evaluate differences and associations, respectively. All analysis code was executed in R4.0.3 and available at https://github.com/ydavidchen/bariatric-sat-profiling.

SAT global gene expression showed clearcut differences before and after bariatric surgery. The first two PCs could be delineated by a simple, linear decision boundary, and the regions created were significantly associated with PreOP/PostOP status (odds ratio [OR], 7.18; 95% confidence interval [CI], 3.93 to 13.42; Fisher’s exact P=1.91E-12) (Fig. 1B). Both PCs were significantly lower in Post- OP (both Welch t-test P<5E-7) (Fig. 1C). Clustering of the most variable genes segregated SAT samples into groups significantly associated with PreOP/PostOP status (OR, 3.65; 95% CI, 1.94 to 7.06; Fisher’s exact P=1.44E-5) (Fig. 1D). There was no detectable association between expression and dataset/batch (Fig. 1D, Supplementary Fig. 1).

A robust linear mixed effect model adjusting for repeated measurements identified 44 (0.9%) over- and 395 (7.9%) under-expressed genes in PostOP SATs (all Bonferroni-adjusted P<0.05) (Fig. 1E, Supplementary Table 2). Gene Ontology identified 18 significant biological processes associated with the under-expressed genes (all OR >2.1 and Bonferroni P<0.05), 14 (77.8%) playing a direct role in immunity and one (5.6%) in amino/proteo-glycan metabolism (A in Table 1). KEGG identified two (66.7%) immune-related and one (33.3%) amino/proteo-glycan metabolic pathways (all OR >2.6 and Bonferroni P<0.05) (B in Table 1). No significant ontology or pathway was associated with the overexpressed genes.

Obesity is a global health problem with many comorbidities. Bariatric surgery, the most effective obesity intervention to date, remains underexplored. Through a pooled re-analysis maximizing sample size and power, this study confirms the gene and pathwaylevel changes post-surgery despite the heterogeneity in measurement and study designs.

All unsupervised analyses support a distinct global transcriptomic landscape associated with bariatric surgery. Many genes, including those with a known role in obesity, showed differential expression. CIRBP involved in liver glucose metabolism¹¹ was the most overexpressed PostOP. SIK2, under-expressed in individuals with obesity and negatively associated with diabetes,¹² was the second. These findings could imply bariatric surgery’s efficacy through related metabolic processes.

Many downregulated genes, including receptor CD4, were immune- related and enriched for immune-related biological processes. Notably, “granulocyte activation” and “neutrophil-mediated immunity” showed the highest enrichment and significance (OR >3.6 and Bonferroni P<4E-14). These findings, alongside previous reports,^3,6 provide insights into how the bariatric procedure might yield beneficial outcomes through cellular-molecular modulation. Other downregulated processes were involved in intercellular signaling and could broadly contribute to proper immune function. Both Gene Ontology and KEGG also detected reduced amino/ proteo-glycan metabolism in PostOP SATs, consistent with prior knowledge.¹³

This study has limitations. First, potential confounders including obesity severity, the type of bariatric surgery, and the presence of other illnesses were not addressed due to limited data availability. Second, the biological findings of this study, especially the downregulated immune-related pathways, need elucidation partly because acquiring SATs for expression analysis requires a biopsy. Since molecular changes such as wound healing can be detected after solidtissue biopsy,¹⁴ future studies should consider evaluating less invasive procedures (e.g., blood sampling) that can detect bariatric surgery- associated changes with minimal bias. It will also be interesting to identify the specific cell types that contribute to such molecular changes, possibly through novel cell-type deconvolution methods, single-cell approaches, and integration of other molecular data types.

Supplementary materials can be found online at https://doi.org/10.7570/jomes22065.

The author declares no conflict of interest.

The author would like to thank the peer reviewers and the editorial team for their input.