Partial restoration of normal intestinal microbiota in morbidly obese women six months after bariatric surgery

Study population

Morbidly obese study subjects were recruited from patients undergoing bariatric surgery procedure as part of their normal treatment at the Hospital District of Southwest Finland (Dadson et al., 2017; Koffert et al., 2018). The surgical inclusion criteria were age between 18–60 years and BMI ≥40 kg/m² (or ≥35 kg/m² with an additional obesity-related comorbidity). In addition to standard exclusion criteria for bariatric surgery patients with T2D requiring insulin treatment were excluded. Healthy study subjects were recruited to study protocol by advertisements in local newspaper from January to December in 2011. The inclusion criteria for the control population were body-mass index 18–27 kg m², age 18–60 years, fasting plasma glucose less than 6.1 mM, and normal oral glucose tolerance test. Altogether 42 females were recruited for the study, including 27 obese subjects and 15 controls. For nine obese subjects, one of the time points was discarded due to the inability to PCR amplify the extracted DNA or low reproducibility of the phylogenetic microarray results (see below). Moreover, three obese subjects withdrew from the study for personal reasons, and further two subjects who had a successful microbiota profile only from a single time point were discarded. This yielded the final data set with 13 obese subjects who had successful microbiota profiles from both time points and 14 controls that were included in the analysis (Fig. 1). Of the 13 obese subjects, five had T2D and in eight had normal glycaemia. Only three study subjects with type 2 diabetes were on metformin medication. Written informed consents were obtained prior the studies. The protocol (NCT01373892/SLEEVEPET) was approved by the Ethics Committee of the Hospital District of Southwest Finland (ETMK 99/180/2010). The clinical study was performed at the Turku PET Centre, the University of Turku and Turku University Hospital (Turku, Finland) between January 2011 and October 2013 and fecal microbiota samples analyzed in 2014.

Study design

At screening visit, study subjects gave their signed approvement prior to their enrolment in to the study. The medical history was recorded, the physical examination, anthropometric measurements, blood and oral glucose tolerance tests were performed, and instructions for fecal sample collection were provided. On a separate study visit abdominal fat masses were assessed with MRI. Thereafter, obese subjects followed a very low-calorie diet (VLCD; 800kcal/day) for one month before the bariatric operation. Of the morbidly obese subjects, seven underwent sleeve gastrectomy and six Roux-en-Y gastric bypass (RYGB) surgery. All the morbidly obese study subjects received perioperative antibiotic Cefuroxime 3g intravenously. The post-operational phase was conducted six months after the surgery and all studies were repeated as in baseline. Food diaries were used to document dietary intake in healthy controls and before the VLCD and 6 months postoperatively in obese subject.

Fecal sample collection and microbiota profiling

Fecal samples for the controls were collected in a single time point and for the obese subjects both before the VLCD (one month before the surgery) and six months after the surgery. DNA from fecal samples was extracted with repeated bead-beating as described previously (Nylund et al., 2010) and used for phylogenetic profiling with the Human Intestinal Tract (HIT) Chip, a custom-made Agilent phylogenetic microarray (Rajilic-Stojanovic et al., 2009). The HITChip analysis allows phylotype-level profiling at a resolution of ∼0.1% relative abundance with >98% reproducibility (Claesson et al., 2009) based on >5,000 probes designed to target 16S rRNA gene sequences of over 1,000 intestinal bacterial phylotypes from 130 genus-like groups (defined by >90% 16S rRNA gene similarity). Further details on the sample processing, microarray, and hybridization protocols were reported previously (Rajilic-Stojanovic et al., 2009). The microarray hybridization data was extracted from scanned microarray images with the Agilent Feature Extraction software. The technical replicates were quantile-normalized with custom R scripts as previously described (Lahti et al., 2014), and reproducible hybridizations (Pearson r >  98%) were accepted for the analysis and stored in a custom MySQL database. Between-sample normalization was performed at probe level with the minimum–maximum method, and probe summarization at the phylotype-level (>98% 16S rRNA sequence similarity) was done with the frozen Robust Probabilistic Averaging (fRPA) method (Lahti et al., 2011; Lahti et al., 2013) and then to the genus (>90% 16S rRNA sequence similarity) and order-like 16S rRNA gene sequence groups by summing up phylotype abundances as described earlier (Lahti et al., 2014).The processing scripts are available at https://github.com/microbiome/HITChipDB. In this work, we focus primarily on genus-level variation.

Statistical analysis

The HITChip profiles were centered log-ratio transformation (CLR)-transformed before statistical testing to remove compositionality bias (Aitchison, 1986), and Euclidean distances for the CLR-transformed genus-level abundances were used as the beta diversity measure. For hierarchical clustering, we used complete linkage with Spearman correlation. The surgery effect on individual taxa between the matched obese subject time points 1–2 was quantified with a random effects linear model with subject as the random effect and time as the fixed effect (lmer R package). Unpaired t-test was used for two-group comparisons. Multi-group comparisons were performed with ANOVA and the Tukey post-hoc test. In order to reduce multiple testing, we removed from comparisons the rare taxa that were observed with <10% prevalence at >0.1% relative abundance. To further support the analysis, we verified the multi-group comparisons with non-parametric Kruskal-Wallis test and diagnostic plots. PERMANOVA (adonis in the vegan R package) was used for the group-wise community comparisons. The p-values were adjusted for multiple corrections with Benjamini–Hochberg method. The FDR<0.05 was considered statistically significant, and FDR<0.25 borderline significant. Spearman correlation of the probe-level HITChip signals were used for subject identification based on best match between the two timepoints. Microbiota richness and diversity were quantified from the probe-level HITChip data by observed probe count and Shannon index, respectively. Tools from the phyloseq (McMurdie & S, 2013) and microbiome (https://microbiome.github.io/tutorials/) R packages were used to support analysis and visualization.

Sample size for original study protocol in PET-imaging was determined by joint a priori power analysis based on longitudinal PET studies using 18F-FTHA (Karmi et al., 2010; Viljanen et al., 2009) which suggested that a sample size of 10+10 would be sufficient for establishing the predicted effects at P<0.05 with actual power exceeding 0.95. These study subjects were used for intestinal microbiota analysis.

Effect of bariatric surgery on glucose metabolism

Before surgery, the obese study group was insulin resistant and had increased systemic inflammation compared to healthy controls (Table 1). In addition, the obese study subjects had higher HbA1c and 2-hour glucose concentrations in OGTT compared to healthy controls (Table 1). Bariatric surgery resulted in marked weight loss (on average 23%) with improved insulin resistance (Table 1). Furthermore, diabetes remission occurred in 80% (4/5) of the T2D subjects after surgery.

Dietary intake preceding and after surgery

Mean caloric intake was lower in obese subjects compared to healthy controls but the difference was only marginally significant (1,304 ±  570 kcal vs. 1,625 ±  518, p = 0.07, respectively). The total calorie intake decreased in obese subject by 35% on average after bariatric surgery (p < 0.01). Nutrient intake was lower postoperatively, while carbohydrate-, protein- and fat-intake decreased on average by 27% (p < 0.01), 18% (p < 0.01) and 6% (p = 0.04) respectively.

Control subjects form two distinct groups

Unsupervised hierarchical clustering of the genus-level profiles indicated two distinct subgroups among the control subjects (Fig. S1). These included 4 normal-weight control subjects with mean BMI 24.9 kg/m² (group A) and 10 “lean” control subjects with a lower average BMI of 21.5 kg/m² (group B). Differences in body fat and energy consumption were not significant. The group A controls had on average higher levels of bacteria related to Allistipes. In contrast, the group B controls had on average higher levels of bacteria belonging to Ruminococcus obeum, Coprococcus eutactus and Clostridium nexile (p < 0.25).

Surgery type has no effect on microbiota composition

No differences were observed between the alternative surgery procedures (bypass n = 6; sleeve n = 7) in post-surgery samples in community composition quantified by the overall beta diversity or individual genus-like groups (Fig. S2A). These groups were combined in the subsequent microbiota analyses.

Microbial richness shifts towards controls in patients following surgery

Before surgery, the average richness was significantly lower in the obese subjects compared to the group A controls (p = 0.01) and borderline significant compared to the group B controls (p = 0.12) (Fig. 2). After the surgery, the difference to group A controls was borderline significant (p = 0.14), and difference to the group B controls was not significant (p = 0.87).

Microbiota community structure shows changes following bariatric surgery

Bariatric surgery was associated with a shift in the Bacteroidetes-Firmicutes ratio towards that of the control subjects (Fig. 2B). The overall differences in microbiota composition (beta diversity) between the obese and control subjects were not significant after the surgery. Notably, whereas the average abundance of Firmicutes decreased after surgery, specific members of the Firmicutes exhibited qualitatively different increasing shifts following the surgery (Fig. 3), indicating that the phylum-level comparison of the Bacteroidetes-Firmicutes ratio should be complemented by the analysis of specific members within these groups.

Genus-level shifts following the surgery

At the level of specific bacterial genera, bariatric surgery was associated with an increased abundance of specific members of bacteria belonging to the Bacteroidetes, Streptocooccus spp. (S. mitis and S. bovis.), as well as Sporobacter termitidis (Clostridium cluster IV) (Fig. 4), so that the abundances of these bacteria were shifting towards the group A controls. The observed shift towards these controls was evident also as a reduced abundance in specific members of the Clostridium cluster XIVa (including bacteria related to Ruminococcus obeum, Coprococcus eutactus, and other Clostridium groups) after surgery.

Individuality of microbiota composition

Microbial composition remained individual-specific and identifiable following surgery for most (11/14; 79%) subjects based on the closest match with the pre-surgery samples (see Methods). However, three subjects experienced so remarkable changes during the intervention that reliable subject identification based on microbiota composition was not possible after the surgery (best match was with a sample from another subject; Fig. S3). However, for two of these incorrectly identified subjects, the second best match was with the correct subject. For the third misidentified obese subject, his/her own pre-surgery sample was only the 9th best match, suggesting drastically altered microbiota composition or potential shift in community type following the surgery.

Effect of diabetes status

In addition, the microbiota characteristics of metformin-treated obese subjects (n = 3) and obese subjects with no metformin treatment (n = 10) were compared at the baseline time point. Diabetes status did not have a statistically significant effect on the overall microbiota composition as quantified by the standard permutation analysis of variance (PERMANOVA; p = 0.28). However, metformin was associated with a significant reduction in intestinal microbiota diversity (p = 0.02; Fig. 5). A similar non-significant trend is seen in microbiota richness (p = 0.21).