Projection of Gut Microbiome Pre- and Post-Bariatric Surgery To Predict Surgery Outcome

Abstract

Bariatric surgery is often the preferred method to resolve obesity and diabetes, with ∼800,000 cases worldwide yearly and high outcome variability. The ability to predict the long-term body mass index (BMI) change following surgery has important implications for individuals and the health care system in general. Given the tight connection between eating habits, sugar consumption, BMI, and the gut microbiome, we tested whether the microbiome before any treatment is associated with different treatment outcomes, as well as other intakes (high-density lipoproteins [HDL], triglycerides, etc.). A projection of the gut microbiome composition of obese (sampled before and after bariatric surgery) and lean patients into principal components was performed, and the relation between this projection and surgery outcome was studied. The projection revealed three different microbiome profiles belonging to lean, obese, and obese individuals who underwent bariatric surgery, with the postsurgery microbiome more different from the lean microbiome than the obese microbiome. The same projection allowed for a prediction of BMI loss following bariatric surgery, using only the presurgery microbiome. The microbial changes following surgery were an increase in the relative abundance of Proteobacteria and Fusobacteria and a decrease in Firmicutes. The gut microbiome can be decomposed into main components depicting the patient's development and predicting in advance the outcome. Those may be translated into the better clinical management of obese individuals planning to undergo metabolic surgery. IMPORTANCE BMI and diabetes can affect the gut microbiome composition. Bariatric surgery has large variabilities in the outcome. The microbiome was previously shown to be a good predictor for multiple diseases. We analyzed here the gut microbiome before and after bariatric surgery and showed the following. (i) The microbiome before surgery can be used to predict surgery outcomes. (ii) The postsurgery microbiome drifts further away from the lean microbiome than the microbiome of the presurgery obese patients. These results can lead to a microbiome-based presurgery decision whether to perform surgery.

Keywords: bariatric surgery; machine learning; obesity.

FIG 1

Distribution of age, gender, height, weight, BMI, A1C, glucose and triglyceride levels, and HDL and LDL levels at the samples taken in each group and time point (H is lean. All other samples are obese before [A and B] or after surgery [C, D, E, 1 year {1y}, 1.5y]. The time point codes are detailed in the bottom right part of the figure (W, week; M, month; Y, year).

FIG 2

Outline of analysis (from top left to top right and then from bottom right to bottom left). First, the Fastq sequences are quality controlled. The good quality sequences are translated to features using QIIME2. To homogenize the description level, the feature levels belonging to the same genus in a given sample are averaged to the genus level. (Bottom) The sample distribution is heavy-tailed. It is thus log transformed with a minimal value (0.1) added to each feature level to avoid log of zero values. The results are then z scored by removing the average and dividing by the standard deviation of each sample. The dimension of the z-scores is further reduced using PCA. The first eight PCs explain approximately 50% of the total variance (bottom left panel).

FIG 3

(A) Spearman correlation between BMI of samples and the eight highest variance PCs. ***, **, and * represent significance levels of 0.001, 0.01, and 0.05, respectively (in this and all following figures). PC2, PC4, and PC5 are the most correlated with BMI. (B) Projection of the significant PCs on the different stages. One can see in all PCs a clear difference between the H and obese states. Following surgery, the projection is farther away from the H state than before. Note that we do not have microbiome samples from the latest time points. TP, time point. (C to E) Receiver operating characteristic (ROC) curve of linear SVM classification using the projections on the first 8 PCs of the H versus O and within the O group before versus after surgery (C and D) and the resulting weights (E).

FIG 4

(A to D) Bacterial composition of PC1, PC2, PC4, and PC5. (E and F) Weights of linear SVM classifier for H versus O and before versus after surgery. Only the top 15% of features are presented (based on their absolute weights).

FIG 5

(A and B) Correlation of age, gender (represented as 0 for male and 1 for female) and BMI with the PC. (C and D) ROC curves of future BMI change based only on the PC at point A. The binary predicted change is above or below the median change. The ROC curves for above and below median change in BMI, using a LASSO regression and a LOO validation. C is for 6 months and D is for 18 months postsurgery. The first P value is the ROC curve P value compared with AUC of 1,000 scrambling of the predicted values. The second P is the Spearman correlation P value between predicted and actual change. (E) Average regression weights over all LOO learning sessions. The only nonzero coefficients are the fifth PC for 18M and the first PC for the 6M. (B to F) Average weights of coefficients in the LASSO regression for plots C and D over all LOO predictions. In all plots, * and ** represent significance levels of P < 0.05 and P < 0.01, respectively.