Gut microbiota and physiologic bowel ¹⁸F-FDG uptake

Participants were recruited from the Kangbuk Samsung Health Study, which is a cohort study of Korean men and women who undergo a comprehensive annual or biennial examination at Kangbuk Samsung Hospital Screening Centers in South Korea. Stool samples were collected from 1463 adult participants (men: 907, women: 556) between the ages of 23 and 78 who underwent a comprehensive health checkup between June 2014 and September 2014. Among them, 76 males and 5 females had completed PET images and only males were included in this study.

Participants who met any of the exclusion criteria were not enrolled in this study. We excluded one participant with less than 5000 sequences per sample and five participants with antibiotic use within 6 weeks prior to enrollment, regular use of statins, or probiotic use within 4 weeks prior to enrollment. To avoid bowel FDG uptake caused by Metformin, we excluded six participants who had diabetes, were taking a medication for diabetes, or had a fasting glucose level ≥ 126 mg/dL. We excluded a participant showing a hot spot at the left femur in an ¹⁸F-FDG PET/CT image, which is suspicious for a benign or malignant tumor. Finally, 63 participants (100% male) with a mean of 25,077 sequences per sample were included.

Medical history was collected through a self-administered questionnaire. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. After fasting overnight, blood samples were taken from the antecubital vein before the ¹⁸F-FDG PET/CT scan. Serum glucose, HbA1c, total cholesterol, triglycerides, uric acid, high-density lipoprotein, low-density lipoprotein, high-sensitivity C-reactive protein, free T4, free T3, and TSH were measured according to standard procedures.

Participants were asked to fast overnight before PET/CT scan. Blood glucose levels at the time of injection of ¹⁸F-FDG were lower than 160 mg/dl. PET/CT scans were performed on Discovery D600 (for 57 subjects) or Discovery STE scanners (for 6 subjects) (GE Medical Systems, Waukesha, WI, USA) with a tracer uptake time of 60 min. ¹⁸F-FDG was injected based on weight (0.1 mCi/kg on the D600 scanner, 10-13 mCi on the STE scanner). No intravenous contrast agent was given. For the Discovery D600 scanner, slice thickness was 3.75 mm, current was 40–120 mAs, and energy was 120 kVp. For the Discovery STE scanner, slice thickness was 3.3 mm, current was 40–200 mAs, and energy was 140 kVp. Following CT acquisition, a PET emission scan was acquired with an acquisition time of 2.5–3 min per bed in 3D mode from the proximal thigh to the skull base. CT data were used for attenuation correction. The images were reconstructed using a conventional iterative algorithm (OSEM). Using dedicated software (Advanced Workstation, GE Healthcare, Milwaukee, WI, USA), CT, PET, and fused PET/CT images were reviewed.

We assessed intestinal ¹⁸F-FDG uptake by visual analysis and quantitative analysis. For the visual analysis, we classified subjects into three groups. Group 1 included subjects with intestinal FDG uptake lower than liver FDG uptake. Group 2 included subjects with intestinal FDG uptake equal to or higher than liver FDG uptake; among group 2 subjects, subjects with relatively focal or intense FDG uptake in the intestine were reclassified as group 3 (Fig. 1). For the quantitative analysis, we measured maximum and mean standardized uptake values (SUV_max and SUV_mean) in each segment of the intestine using a three-dimensional volume of interest (VOI). The intestine was divided as follows: third portion of the duodenum, jejunum, ileal loop, ileocecal junction, ascending colon, transverse colon, descending colon, and sigmoid colon. The SUV_mean of each segment was measured with a margin threshold of 60% SUV_max [5]. After SUV measurement, we added all the segment values, which indicated the SUVs of the total bowel: TB SUV_max and TB SUV_mean. For the measurement of liver SUV_mean, we drew 3 cm VOIs in both lobes of the liver as previously reported [6]. The uptake ratio of TB SUV_max and TB SUV_mean to liver SUV_mean (TB to liver uptake ratio, TBR_max and TBR_mean) was calculated for each subject.

Fecal samples were immediately frozen after defecation at −20 °C and were placed at −70 °C within 24 h. Within 1 month, the stool specimens were vortexed to achieve a homogenous suspension and homogenized to isolate DNA using the PowerSoil® DNA Isolation Kit (MO BIO Laboratories, Carlsbad, CA) according to the manufacturer’s instructions.

Variable V3 and V4 regions of the 16S rRNA were amplified with the universal primers F319 (5′ TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG) and R806 (5′–GGACTACHVGGGTWTCTAAT–3′) [7], with each primer modified to contain a unique 8-nt barcode index by combination with the NexteraR XT DNA Library Preparation kit (Illumina, San Diego, CA). PCR reactions contained 5 ng/uL of DNA template, 2× KAPA HiFi HotStart Ready Mix (KAPA Biosystems, Wilmington, MA), and 2 pmol of each primer. Reaction conditions consisted of an initial incubation at 95 °C for 3 min, followed by 25 cycles of 95 °C for 30 s, 55 °C for 30 s, and 72 °C for 30 s. Samples were subjected to a final extension incubation at 72 °C for 5 min. After PCR clean-up and index PCR, sequencing was performed on the Illumina MiSeq platform according to the manufacturer’s specifications [8, 9]. The 100 bp of overlapping paired-end reads were merged using PandaSeq (version 2.7). To analyze the 16S rRNA gene sequence, quality filtering, including determination of the sequence length (>300 bp), end trimming, and determination of the number of ambiguous bases and the minimum quality score were performed using Trimmomatic (version 0.32).

Chimeras were detected and removed using USERCH 6.1 within the QIIME package (version 1.9) [10]. To identify OTUs from the non-chimeric sequences, an open-reference OTU picking approach was performed using representative sequences with pre-assigned taxonomy from Greengenes (version 13_8) [11]. This analysis was performed in QIIME, with a 97% similarity threshold. In an open-reference OTU picking process, reads are clustered against a reference sequence collection, and any reads that do not hit the reference sequence collection are subsequently clustered de novo. The initial data set for all 1463 samples included 278,619 OTUs, and the sequencing depth ranged from 38 to 137,059 reads per sample (mean = 24,644, SD = 17,384). OTUs with <0.005% of the total number of sequences were discarded as recommended by Navas-Molina et al. [12] for downstream analysis, resulting in 1011 OTUs.

For cross-sectional analyses of baseline characteristics, differences were evaluated using the Kruskal-Wallis test and the Spearman correlation test. STATA version 13.0 (StataCorp LP, College Station, Texas, USA) was used for these analyses. All statistical analyses were performed in RStudio (version 0.98.983) using packages phyloseq 1.14.0 [13], DESeq2 1.10.1 [14] and ggplot2 2.1.0 [15]. We compared the composition and structure of the gut microbiota between group 1 vs. 2, group 1 vs. 3, group 2 vs. 3, group 1 vs. 2 and 3, and groups 1 and 2 vs. 3. Various indices (Richness: Chao1, ACE; richness and evenness: Shannon, InvSimpson) were obtained using the estimate_richness function of the phyloseq package with untrimmed datasets for meaningful results, as described by the phyloseq manual. Statistical tests for alpha diversity among groups of ¹⁸F-FDG uptake were performed using the t test and Wilcoxon rank-sum test with the mentioned indices. Beta diversity was obtained using the ordinate function of the phyloseq package. For beta diversity analysis, we normalized OTU matrices with a cumulative sum scaling (CSS) approach [16]. To identify factors that could drive groupings of similar communities, principal coordinate analysis (PCoA) was performed on the Bray-Curtis matrices generated from beta diversity calculation, and the resulting PCoA plots were visualized using the phyloseq. The Bray–Curtis index of similarity was calculated to investigate the similarity of microbiota composition between different samples.

To determine statistically significant differences between samples from groups 1, 2, and 3, taxa seen more than three times in at least 10% of the samples were removed from the analyses to protect against an OTU with a small mean and a trivially large coefficient of variation. The number of OTU-mapped reads was normalized across all samples using the variance-stabilized transformation method as implemented in DESeq2 using a generalized linear model (GLM). Statistical inference was conducted using the negative binomial GLM fitting and Wald significance tests. We adjusted for age and BMI by including them as covariates in the linear model. The resulting p values were corrected for multiple comparisons at each phylogenetic level using the Benjamini-Hochberg correction (FDR). Significant findings were reported for OTUs that had fold changes exceeding two in any group and significant statistical associations (FDR q < 0.05).

All the subjects had no history of cancer, diabetes, heart disease including coronary disease, stroke, or inflammatory bowel disease. Four of 63 subjects had a history of hypertension. Participant characteristics are summarized in Table 1. The mean age was 44.7 (range 39–58) years. Twenty-six subjects (41%) were in group 1, 20 (31%) were in group 2, and 17 (27%) were in group 3. There was no significant difference in age and body mass index (BMI) between groups. There were significant differences in HbA1c, C-reactive protein (CRP), and Free T4 between the groups. Serum glucose and lipid profiles showed no significant differences. Measured values for the quantitative analysis, TBR_max and TBR_mean, showed strong significant differences between groups sorted by visual analysis. In the Spearman correlation test, TBR_max and TBR_mean were evaluated with age, BMI and laboratory data. There was no significant correlation between them.

The 16S RNA gene sequencing produced 1,579,858 sequences across 63 samples. The average number of sequences was 22,331, 23,923, and 30,636 for groups 1, 2, and 3, respectively. The number of generated sequences was not affected by grouping for intestinal ¹⁸F -FDG uptake (p = 0.29).

Maintaining sufficient bacterial richness and diversity is important for providing gut microbiota with functional redundancy, adaptability, and thus systematic robustness against environmental challenges [17]. Using a total of 1,579,858 qualified sequences, we estimated bacterial richness and evenness. The Chao1 and ACE indices for alpha diversity were used to estimate the number of species (richness) in the microbiome, with correction for subsampling and metrics that aim to measure diversity by accounting for evenness or homogeneity using Shannon and InvSimpson. We compared the indices between group 1 vs. 2, group 1 vs. 3, group 2 vs. 3, group 1 vs. 2 and 3, and groups 1 and 2 vs. 3. There was no significant difference in ¹⁸F-FDG uptake in all indices of alpha diversity (Additional file 1: Figure S1 and Table S1) between groups.

The PERMANOVA analysis showed a significant (p = 0.047, pseudo-F = 1.57) difference in Bray-Curtis distances between groups 1 and 2, although the comparison of the bacterial communities by principal coordinate analysis (PCoA) plot with the Bray-Curtis distance could not easily distinguish between the two groups (Additional file 1: Figure S2).

We next examined whether differences in the abundances of specific bacterial taxa were associated with intestinal ¹⁸F -FDG uptake. For detailed taxonomic analyses, individual sequences were classified and assigned to 11 known phyla. Firmicutes and Bacteroidetes were the two most dominant phyla and they comprised an average of 93.04% of total classifiable sequences in all groups (group 1: 92.77%, group 2: 93.95%, and group 3: 92.37%) (Fig. 2), as previously reported in many studies [18]. Comparison of the mean abundances between groups by the Student’s t test showed that the phylum Fusobacteria was more abundant in group 3 than in group 1 (F = 14.44, p < 0.0001). Since the abundance data were not normally distributed and contained a large fraction of zero values, we employed a negative binomial GLM using DESeq, which was used as the main statistical test for comparison throughout this study. The tests showed that Fusobacteria were significantly associated with differences between groups 1 and 3 [log₂ fold change (log2FC) =4.50, q = 9.32 × 10⁻³] (Additional file 1: Table S2-2).

At the genus level, among 88 genera identified, Bacteroides spp., Prevotella spp., unclassified Ruminococcaceae, unclassified Lachnospiraceae, and Ruminococcaceae spp. were the top five most abundant genera identified (Fig. 3). The five genera comprised 56.52, 57.37, and 54.42% of groups 1, 2, and 3, respectively. The unclassified Leuconostocaceae and Mitsuokella spp. were significantly more abundant in group 2 than in group 1 after controlling for age and BMI (log2FC = 5.29, 4.46; q = 4.02 × 10⁻⁷, 2.85 × 10⁻³, respectively) (Table 2 and Fig. 4a). In contrast, the abundance of Klebsiella spp. was lower in group 2 than in group 1 (log2FC = −3.00, q = 1.81 × 10⁻²) (Fig. 4b). Mitsuokella spp. were also more abundant in group 3 than in group 1 (log2FC = 4.84, q = 3.85 × 10⁻²). We found that group 3 had more abundant Fusobacterium spp. than group 1 (log2FC = 5.74, q = 3.85 × 10⁻²), as well as the phylum Fusobacteria (Additional file 1: Table S2-2). However, we could not find the significant differences of taxa between group 2 and group 3 in the phylum and genus level (Additional file 1: Table S2-3 and Table S3). Moreover, there was no statistically significant difference in the unclassified Leuconostocaceae and Fusobacterium between group 2 and group 3 although there seemed to be differences in the Fig. 4.

To compare subjects showing diffuse FDG activity and relatively intense FDG activity in the intestine, we combined groups 1 and 2 and compared them with group 3. To compare subjects with lower FDG activity and higher activity than the liver, we combined groups 2 and 3 and compared them with group 1. The unclassified Clostridiales were more abundant in the group 1 + 2 than group 3 (log2FC = −1.11, q = 1.26 × 10⁻²) (Fig. 4c). The abundance of unclassified Clostridiales was shown the negative trend by groups in Fig. 4c and the taxon was also more abundant in group 2 than group 3 although it was not statistical significant (p = 0.003, q = 0.18) in (Additional file 1: Table S3). Comparison between groups 1 and 2 + 3 yielded results similar to those obtained from the comparison of groups 1 and 2.

We also tested whether TBR_max and TBR_mean were associated with changes in gut microbiota. The TBR_max and TBR_mean showed similar results (Table 3). At the phylum level, there was no significant association between TBR and the abundance of individual phyla. At the genus level, abundance of the unclassified Enterobacteriaceae was negatively associated with TBR_max and TBR_mean (log2FC = −0.35, −0.47; q = 3.56 × 10⁻³, 4.38 × 10⁻³, respectively) (Fig. 5).