The Gut Bugs Trial
To understand the level of HGT in the gut microbiomes of FMT donors and recipients, we analysed metagenomic sequences obtained from a previously published double-blinded, randomised, placebo-controlled trial investigating the impact of FMT on adolescent obesity that was carried out in Auckland, New Zealand [18, 32]. Each recipient received capsules containing microbiota from four-sex-matched donors, with the same donors used throughout the trial [29]. There was no effect of FMT on weight loss at 6 weeks post-intervention, although improvements in a marker of abdominal obesity (android-to-gynoid fat) were observed [32]. Post-hoc analysis identified improvements in insulin sensitivity and glucose metabolism in a subset of participants with metabolic syndrome. Sustained shifts in both the structure and functional potential of recipient gut microbiomes in response to FMT were observed, with highly variable rates of strain engraftment [18].
Metagenome assembled genome (MAG) assembly
Metagenomic sequencing data were available for 381 trial samples, including 58 FMT donor samples collected from 9 donors over the 12 month stool donation period, and 323 samples belonging to 42 FMT and 45 placebo recipients collected at baseline, and 6-, 12-, 26-weeks post-intervention. An average of 96,604 contigs with a minimum length of 500 bp were assembled per sample (range 27,729 − 202,760). Clustering of genes with > 95% identity resulted in a gene catalogue containing 2.9 million genes from across all samples. Across all samples, we assembled 20,941 metagenome assembled genomes (MAGs), of which 4,189 (20.0%) were high quality (> 90% completeness and < 5% contamination). Most high quality MAGs were taxonomically classified as belonging to the phyla Firmicutes (68.2%), Bacteroidetes (21.8%) and Actinobacteria (4.2%), and the species Agathobacter rectale (4.2%), Agathobacter faecis (4.1%) and Faecalibacterium prausnitzii (3.9%). A total of 9,425,258 genes were present on high quality MAGs, representing 1,090,166 gene clusters, 79% of which were assigned a COG functional annotation.
FMT does not increase HGT compared with a placebo
We analysed the assembled metagenomic contigs for historic HGT events by identifying DNA segments assigned to a separate ancestral lineage compared to the surrounding DNA [23]. We identified that 0.15% of assembled contigs per sample (range 0.074–0.26%) harboured segments implicated in HGT. The mean prevalence of contigs without HGT was 58.64% (range 38.19–80.33%) (Supplementary Fig. 1, Supplementary Table 1).
We hypothesised that the acute competition associated with FMT would stress the recipient microbiome and lead to an increase in HGT events. However, it is also possible that metagenome samples with increased microbial species richness have an increased propensity for HGT events. Therefore, we normalised HGT event counts within each microbiome sample by species richness. Linear mixed models did not identify evidence of a treatment effect on normalised HGT events (LMM, b = 0.0076, 95%CI [-0.11, 0.12], p = 0.90). Similarly, there was no evidence of a significant sex (b = 0.11, 95%CI [-0.0036, 0.23], p = 0.060), or timepoint effect from baseline to week 6 (b = -0.0043, 95%CI [-0.16, 0.15], p = 0.96), or baseline to week 12 (b = 0.032, 95%CI [-0.13, 0.19], p = 0.71). By contrast, there was a significant longitudinal effect, in both FMT and placebo recipients, from baseline to week 26 (b = 0.37, 95%CI [0.21, 0.53], p < 0.001) (Fig. 1), which is likely due to drift over time.
A limitation of the aforementioned approach to detect HGT events is its inability to differentiate between HGT events that have occurred following FMT treatment and historical signatures of HGT present in bacterial genomes. We hypothesised that HGT due to FMT treatment, specifically, would increase following FMT. Therefore, we utilised a complementary method to quantify the number of FMT donor genes that transferred to FMT recipients, at 6-weeks post-intervention. Genes that were present on high quality MAGs (i.e. >90% completion and < 5% contamination) were clustered using a > 95% identity threshold. To identify putative HGT events, we performed a time-course analysis for each individual (FMT and placebo recipients). Gene clusters that were present within the participant samples at 6 weeks post-intervention, absent at baseline, and were also present in the respective donor samples were identified. Gene clusters were then classified as being horizontally transferred if their taxonomic identification differed between donor and recipient MAGs. Using this approach we identified 57,590 putative HGT events occurring post-intervention in 39 FMT recipients. Using the same criteria, we observed 111,273 putative HGT events that occurred in 44 placebo recipients. Adjusting for gene richness, there was no difference in the percentage of genes involved in HGT for the FMT and placebo groups, including when subset by sex (Wilcoxon rank sum test, overall, p = 0.56; males, p = 0.84; females, p = 0.63). This finding is consistent with our earlier observations and supports the conclusion that FMT does not increase the rate of HGT events above the background rates in the gut microbiome (Fig. 2a).
To understand the occurrence of treatment-specific HGT events, we quantified the number of transfer events for each horizontally transferred gene cluster (HTGC), across FMT- and placebo-specific HTGCs. FMT-specific HTGCs were defined as distinct HTGCs that were present in FMT recipients at 6 weeks post-intervention, and absent in the respective placebo group samples. Placebo-specific HTGCs were defined as the reverse. We observed that 42% (4,260/10,109) of HTGCs in female FMT recipients were FMT-specific and 63% (9,794/15,643) of HTGCs in female placebo recipients were placebo-specific. In males, 63% (3,321/5,258) of HTGCs in FMT recipients were FMT-specific and 69% (4,379/6,316) of HTGCs in placebo recipients were placebo-specific (Fig. 2c). Quantifying the number of individual transfer events for each FMT-specific and placebo-specific HTGC identified that FMT- and placebo-specific HTGCs were predominantly involved in single transfer events (FMT-specific mean 2.7 ± 3.2; placebo-specific mean 4.7 ± 5.4). Therefore, we did not observe any increase in the occurrence of FMT-specific HGT events, relative to the background rate of HGT observed in the placebo cohort (Fig. 2b).
Engrafted bacterial species horizontally transfer genes to recipient bacteria
The engraftment of microbial strains within the recipient microbiomes was observed to be donor-specific in the Gut Bugs FMT trial for adolescent obesity [18]. We hypothesised that the engraftment of novel donor strains into the recipient gut would promote HGT with other microorganisms within the recipient’s gut. To investigate this, we selectively focused on HGT events that occurred at 6 weeks post-intervention between donor-engrafting strains and distinct strains in the FMT recipient’s gut [18]. HGT events involving engrafted bacteria were identified for three female donors and four male donors. Engrafted strains contributing to HGT events in FMT recipients most commonly originated from donor DF16 in females and donor DM08 in males (Fig. 3), consistent with the higher levels of strain engraftment observed from these donors [18]. Between the male donor DM07 and recipient TM04, specifically, there were a high number (n = 161) of HGT events, with 158 of these being between Bacteroides uniformis and Bacteroides vulgatus.
We investigated if the HGT events associated with engrafted bacteria occurred through inter- or intra-phylum transfers. The majority of HGT events were facilitated by engrafted species in the Bacteroidetes phylum (159/166 events for females and 375/400 events for males) (Fig. 4a and Fig. 4c). We compared the distribution of engrafted phyla that contributed to HGT, with all engrafted phyla, to determine if the Bacteroidetes phylum was overrepresented amongst the engrafted donor bacteria (Fig. 4b). In the female cohort, Bacteroidetes were overrepresented (adj. p < 0.001, chi-squared test) and Firmicutes were underrepresented (adj. p < 0.001, chi-squared test) in the engrafted bacteria contributing to HGT. In the male cohort, Bacteroidetes were overrepresented (adj. p < 0.001, chi-squared test), while Firmicutes and Proteobacteria were underrepresented (both adj. p < 0.001, chi-squared test) in the engrafted bacteria contributing to HGT.
Engraftment-dependent HTGCs are maintained following FMT
We assessed the retention of engraftment-dependent HTGCs in FMT recipients. Gene abundance data was obtained for each engraftment-dependent HTGC in the FMT recipients, post-intervention. We hypothesised that HTGCs acquired by 6 weeks post-intervention would further proliferate within participants at week 12 and week 26. In total, there were 289 male and 139 female distinct engraftment-dependent HGT events detected in FMT recipients, 6 weeks following the intervention. Of these, 248 (85.8%) male and 134 (96.4%) female HTGCs were retained at week 12, while 232 (80.3%) and 135 (97.1%) female HTGCs were retained at week 26. We found no evidence of timepoint- or sex-specific effects on the relative abundance of engraftment-dependent HTGCs using linear mixed models to compare the relative abundance of HTGCs at 6-, 12- and 26 weeks post-intervention (Fig. 5). These data suggest that while the HTGCs did not proliferate, they were maintained within the host for up to 26 weeks following the intervention.
Functional annotation of engraftment-dependent HTGCs
We investigated the assigned COG functions for the engraftment-dependent HTGCs [33]. Across the female recipients, 76 (54.7%) of engraftment-dependent HTGCs detected at 6 weeks post-intervention were able to be assigned a functional classification (cellular processes and signalling, information storage and processing, or metabolism), while 37 (26.6%) were poorly characterised, and 26 (18.7%) had no classification. In the male cohort, 166/289 (57.4%) had a functional classification, 62/289 (21.5%) were poorly classified, and 61/289 (21.1%) had no classification (Fig. 6). Comparisons of the relative abundance of gene clusters with each functional classification at each timepoint identified no significant differences (PERMANOVA test females, p = 0.87; males, p = 0.92). The relative abundance of individual COG functional categories assigned to these HTGCs was also not significant (PERMANOVA test females, p = 0.92; males, p = 0.99) (Supplementary Fig. 2).