Transcriptome Changes and Metabolic Outcomes After Bariatric Surgery in Adults With Obesity and Type 2 Diabetes

Figure 1.. Flow of patients during the study period. CONSORT diagram of 53 patients who signed informed consent and were enrolled. Twenty-nine patients completed all procedures, and 26 completed follow-up at 1 year.

Figure 2.. Graphical summary of the study design. Twenty-six patients with obesity and type 2 diabetes who underwent bariatric surgery were studied. Plasma and peripheral blood mononuclear cells before surgery and at 2 and 12 months after surgery were used to determine fasting blood sugar, glycated hemoglobin A1c (HbA1c), C-peptide, and transcriptome. Body mass index, fasting blood sugar, HbA1c, and C-peptide before and after surgery are shown as violin plots. The responders vs nonresponders phenotypes were based on HbA1c and antidiabetic medications 12 months after surgery. The machine-learning method applied to a pool of 20,183 genes at baseline after classification into responders and nonresponders was used to identify genes that can predict the metabolic outcome preoperatively.

Figure 3.. Transcriptomic changes in adults with obesity in complete (r) and incomplete (NR) diabetes remission after bariatric surgery. A, Heat map of unsupervised hierarchical clustering (only on rows, ie, genes) showing the differentially expressed genes (DEGs) in R and NR separately across different time points (T0, T1, and T2). Each column represents the gene expression profile of a patient. Horizontal color bars at the top indicate the study time points T0 (baseline, before surgery), T1 (2 months after surgery), and T2 (12 months after surgery) for each group responders (R) or nonresponders (NR). The color scale is the normalized expression values from microarray data. B, Principal component analysis of gene expression data (containing approximately 20,000 genes) in R and NR. The first two principal components (explaining ∼75% of variance) were able to separate R and NR samples. C, The Venn diagram shows the number of different and shared DEGs in R (n = 16) and NR (n = 10) in T1 and T2 after bariatric surgery.

Figure 4.. Differential metabolic and signaling pathways between complete and incomplete diabetes remission after bariatric surgery. Comparison analysis of canonical pathways based on the Z score in responders (R) and nonresponders (NR) at 2 months (T1) and 12 months (T2) after bariatric surgery relative to before surgery (T0). The yellow and blue colors indicate the pathways that are predicted to be activated (Z score > 1) or inhibited (Z score < −1), respectively.

Figure 5.. Hippo signaling pathway with prediction of overlaid molecular activity in A, responders, and B, nonresponders after bariatric surgery. A, Diagram of the canonical Hippo signaling pathway showing 6 genes upregulated (red) and 12 genes that did not reach statistical significance (gray) 12 months after bariatric surgery (T2) in responders (R), and B, 13 genes upregulated (red) and 5 genes that did not reach statistical significance (gray) in nonresponders (NR), along with predictions of biological functions. Cell proliferation (colored orange) is predicted to increase both in R and NR. The inset square shows that the STK4 gene is upregulated in T1 (3) and T2 (4) in NR but not in R (1 and 2). Pathway and molecular activity prediction analyses were generated using QIAGEN Ingenuity Pathway Analysis (IPA, QIAGEN, www.qiagen.com/ingenuity).

Figure 6.. Diagram of the SIRT1 signaling pathway with predicted molecular activity overlaid on A, responders, and B, nonresponders after bariatric surgery. A, Diagram of the canonical SIRT1 signaling pathway showing 23 genes upregulated (red) and 1 downregulated (green) and 48 genes that did not reach statistical significance (gray) 12 months after bariatric surgery (T2) in responders (R). B, In nonresponders (NR), 50 genes were upregulated and 22 genes did not reach statistical significance, along with predictions of biological function. Cell proliferation, inflammation, neurogenesis, and elevated high-density lipoprotein (HDL) cholesterol (colored orange) are predicted to increase, while apoptosis, oxidative stress, hypertension, Alzheimer disease, and renal anemia (colored blue) are predicted to decrease in A, responders, and B, the inverse in NR. The inset square shows that SIRT1 is upregulated in T1 (1) and T2 (2) in R but not in NR (3 and 4). Pathway and molecular activity prediction analyses were generated using QIAGEN Ingenuity Pathway Analysis (IPA, QIAGEN, www.qiagen.com/ingenuity).

Figure 7.. Modeling of baseline transcriptome using machine learning. A, A predictive model to classify responders (R) and nonresponders (NR) using baseline gene expression data. Two different selection methods were applied in parallel using machine learning and computation of differentially expressed genes in R vs NR. Data training and testing performance were performed using the Python module “Scikit-learn.” The final gene set was selected based on the consensus of these methods. B, Heat map shows normalized gene expression profiles for the 10 most significant genes selected by the machine-learning approach (left) and for 10 randomly sampled genes (right) from a pool of 20,183 genes in 26 patients before bariatric surgery. A blue horizontal line separates responders (R) from nonresponders (NR).