In a large Swedish population-based cohort, we report associations between specific gut species and functional GMMs and commonly used medications/medication-classes and polypharmacy. We replicate earlier associations between use of PPIs, metformin, and SSRIs with specific gut species, and describe new associations with use of metformin, antidepressants/SSRIs, NSAIDs and NSAID-subtypes, ARBs, calcium channel blockers, beta-blocking agents, nasal preparations, inhaled or oral steroids and polypharmacy. Regarding metformin, we replicate earlier findings of increased lactate consumption I and arginine degradation II in users of this drug, and present new observations of elevated methionine degradation and anaerobic fatty acid oxidation in users of antidiabetic medications. We replicate previously described negative associations between Shannon index and use of PPI, paracetamol, metformin, and metoprolol and report new negative associations for inhalants for obstructive airways, oral corticosteroids, and levothyroxine. Further, among individuals with polypharmacy, we report differences in the overall gut microbial composition and decreased richness and Shannon index, replicating previous findings5, and describe both new and replicated associations between polypharmacy and specific gut species, of which some did not associate with any other specific medications. We further report numerous associations with deviations in the functional potential of the gut microbiome among those with polypharmacy, most of which were solely associated with polypharmacy and not with use of any individual medication.Earlier studies of associations between medication use and gut microbiota using shotgun metagenomics in large cohorts have been scarce5,12,24. MOS cohort represents the general population and both registry data on prescription medicines and self-reported medication use in the past week, including over-the-counter medications, were utilized. As not all medication-classes were available from the medication register due to the original design of MOS, the self-reported data added valuable complementary information, also regarding ongoing concomitant medication-use.Lower gut microbiota alpha diversity, commonly measured as richness (number of species) and richness and evenness (Shannon index), has been observed in numerous diseases11,25 as well as among users of specific medications5,11,17,24,26,27. We observed lower gut microbiota richness amongst users of almost 40% of the investigated medications, and 70% of these additionally associated with lower Shannon index. In line with Jackson et al.11, we found lower Shannon index in users of metformin, paracetamol, and PPI. Contrary to our results, Nagata et al.5 observed a higher Shannon index in PPI users, and our associations did not withstand the sensitivity-analyses, which showed substantial reductions in the correlation coefficients and thus the association might rather reflect previous antibiotic use or occurrence of polypharmacy. The associations of paracetamol and metformin with Shannon index did not withstand sensitivity analysis excluding polypharmacy users, which could reflect either decreased power as these medications were amongst the most common among those with polypharmacy, or an impact of polypharmacy on Shannon index, or both. In our study, lower richness and Shannon index were also seen among users of beta-blockers, replicating earlier result in the Estonian microbiome cohort24. Further, we observed lower richness and Shannon index in users of oral corticosteroids, and medications for obstructive airways, and decreased Shannon index amongst users of levothyroxine, which to our knowledge have not been reported before. The strengths of these correlations were generally preserved in the three sensitivity analyses. Overall, the further adjustment for fiber intake and physical activity were not observed to have any major impact on the associations between medications or polypharmacy and Shannon index.Our study provides further convincing evidence, put forward in several earlier studies, for a strong connection between PPI use and gut microbiota. Concordant with earlier observational studies, a higher relative abundance of species of oral origin, like those of Streptococcus and Rothia genera, was characteristic for the gut microbiota of PPI users in our study5,12,16,17,26. By repeated sampling, Nagata et al.5 demonstrated increased relative abundance of Streptococcus in the gut microbiota in individuals after starting with PPI, and a corresponding decrease after termination. This has been suggested to be mediated by the increased pH in the gut due to PPI use, which can promote colonization of certain bacteria at more distal parts of the digestive tract10,26,28.Most gut species with elevated abundance among PPI-users are normal habitants of the human oral flora. For 10 of 16 such species in the gut, we identified oral counterparts in MODS, and the relative abundance of all these correlated with the corresponding oral species, which is in line with the view that PPI use may promote transferal of certain oral species to the gut.Both observational and interventional studies have demonstrated specific changes in the gut microbiota of metformin users and a direct anti-diabetic effect mediated by gut bacteria has been demonstrated by fecal transplantation from metformin treated donors to germ-free mice14. Both improved glucose control and negative side effects of metformin have previously been attributed to gut microbiota mediated mechanisms12,13,29. Results from our study provide support for metformin-related changes in the gut microbiota. We observed a higher relative abundance of two Escherichia species, E. marmotae and E. coli, among users of metformin, of which the association with E. coli has repeatedly been described in previous studies10,13,14,29. E. marmotae is a recently described species, phenotypically indistinguishable from E. coli, that in most earlier studies likely has been misidentified as E. coli30. Consistent with previous studies, we also observed nominally decreased abundance of I. bartlettii in metformin users13,14 and support for a causal connection was recently provided by a randomized trial where initiation of metformin in treatment of naïve adults resulted in decreased abundance of I. bartlettii29. Our study additionally identified R. timonensis in lower abundance among metformin users. Like E. marmotae, R. timonensis is a newly isolated species from the human gut31, and has not been captured in earlier data-base dependent gut microbiota studies. However, this finding is supported by another recent large metagenomic study from Sweden where metformin use was based on plasma metabolome data20.A major feature of the gut microbiome is the metabolic functional potential which we investigated utilizing GMMs, representing metabolic modules that are assigned based on known functions of bacterial genes22. Apart from the numerous GMM associations with polypharmacy, discussed later, associations with specific medications were restricted to antidiabetic medications including metformin, and ASA, of which the latter did not associate with any specific species in our study. In users of metformin, lactate consumption I was elevated, which is in line with associations of metformin with E. coli and E. marmotae, that harbor these functions, in our study. Metformin was further associated with arginine degradation II harbored by E. marmotae. In line with our results, Wu et al.14 observed that initiation of metformin treatment led to upregulation of bacterial amino acid metabolism, including arginine degradation. They further found increased fecal concentrations of lactate in the metformin-treated group, as compared to the placebo group, after 4 months of treatment. Here, we found an increase in lactate consumption I, and it is possible that the increased lactate levels in feces may benefit the expansion of species that can utilize lactate. Further evidence concordant with our results is by Nagata et al.5 reporting all genes related to lactate consumption I and arginine degradation II elevated in Japanese metformin users.Gut microbiota interacts bilaterally with the central nervous system via the gut-brain axis32 and has been associated with depression33,34. Previous association studies of antidepressants and gut microbiota have provided diverging results. We solely found negative associations between antidepressants and gut species, which is in line with a recent study showing anti-microbial activity of several antidepressants in vitro on specific habitants of normal gut microbiota, including Bifidobacterium bifidum35, which we found in decreased abundance among SSRI users. All other SSRI associated species in our study belong to Clostridia, and two of them belong to Clostridiaceae family, which associated with SSRI use in an earlier study11.Changes in the gut microbiota as a consequence of use of subclasses of NSAIDs have been described in numerous animal studies, while large human metagenomic population studies are lacking36. Of our study participants, 17% used NSAIDs and this was associated with increased abundance of Saccharomyces cerevisiae, a new finding that needs confirmation in other studies. S. cerevisiae is a common yeast in fermented food and a frequent habitant of the gut flora37 and has been attributed potential positive health outcomes such as protection of the gut mucosa and stimulation of the immune system38. Anti-saccharomyces antibodies have been described as predictors of inflammatory bowel disease progress39. One of two subspecies of Anaerostipes hadrus (HG3A.0003) was found in higher abundance among users of propionic acid derivative types of NSAIDs, while this species did not show any association with use of acetic acid derivative types of NSAIDs nor with the subordinate substance diclofenac. Previously, NSAID-type specific gut microbiota associations have been reported in a smaller study utilizing 16S sequencing40, while no associations with NSAID use was found in a larger metagenomic study that did not differentiate between different types of NSAIDs12. Further gut metagenomic studies and functional analyses are needed to elucidate the interconnectedness between the different NSAIDs, the gut microbiota and clinical outcomes.In rodents, the antihypertensive effect of ARBs has been connected to beneficial changes in the gut41,42. In our study we found increased relative abundance of L. vaginalis in users of both ARBs and calcium channel blockers, which is in line with increase of lactobacilli in the gut of mice treated with these medications41,43, while these associations in humans are new. However, blood pressure lowering effect of probiotics containing Lactobacillus spp. has been shown in human studies and animal models44. Another new finding of our study was the robust association between use of beta-blockers and increased abundance of one of 60 Oscillospiraceae species in our study (HG3A.1161).Polypharmacy has been associated with various unfavorable health outcomes7. Cumulative effects of several medications can be assumed to affect the intestinal environment45, and confounding by polypharmacy in gut microbiota studies needs to be considered5,10,12. Like in a previous report12, polypharmacy associated with differences in the overall gut microbial composition in our study. Further, among those with polypharmacy all three alpha-diversity measures were decreased, which is in line with some5 but not all previous studies12. Explanations to this could be differences in the study populations and the metagenomic methods. The large population size of our study, high resolution of the metagenomic method together with de novo identification of microbial species allowed us to identify a higher number of species than in most earlier studies, affecting the diversity measurements and making between-studies comparison difficult. Our results yet need to be interpreted with caution, as the number of individuals with polypharmacy in MOS was limited, their medication use differed widely and the differences in diversity measures could also reflect underlying disease states.In our study, polypharmacy associated with six species of which three, Enterococcus faecalis, Bacteroides uniformis and one Eubacterialis species (HG3A.0137), were not associated with any specific medication use. E. faecalis, is a common cause of antibiotic resistant hospital-acquired infection. Our finding is concordant with Nagata et al.5 report of a gradual increase of this pathobiont by the number of medications taken. Of the other polypharmacy associated species, R. mucilaginosa additionally associated with PPI use, E. coli with metformin/antidiabetics and L. vaginalis with use of PPIs, ARBs, and calcium-channel blockers. R. mucilaginosa, which is a normal habitant of the oral microbiota46, is an opportunistic pathogen that has been associated to various diseases. This bacterium has recently been described to harbor a catecholate-siderophore which produces iron-scavenging enterobactin47, similar to E. coli that gains survival advantage via production of enterobactin48. Further, R. mucilaginosa may increase virulence of other pathogenic species via a siderophore-sharing mechanism to supply them with iron. Interestingly, synergism between E. coli and B. uniformis has been suggested, where E. coli utilizes d-galactose generated by B. uniformis as a source of carbon for its own growth49. In the model adjusted for age, sex, and BMI but not for Shannon index, considerably more associations between polypharmacy and species were seen, predominantly negative ones. Taken together, these findings might indicate that polypharmacy suppresses a broad range of species to a certain degree, and thereby paves the way for certain pathobionts to thrive.Most of the observed associations with GMMs were observed for polypharmacy, with the strongest association noted for dissimilatory nitrate reduction. This path converts nitrate via nitrite to ammonium and represents a common detoxification process in facultative anaerobic bacteria like E. coli. Altogether 13 GMMs associated with polypharmacy, and the five species that positively associated with polypharmacy in our study contained between 3 and 11 of the associated GMMs each, with E. coli harboring 11 of them. Concordant with our results, a Japanese study5 reported positive associations with the number of medications taken and genes related to pyruvate dehydrogenase complex, proline degradation, lactate consumption I and NADH:ferredoxin oxidoreductase and ribose degradation all of which were among GMMs that associated with polypharmacy in our study. Nagata et al., also reported a positive association between polypharmacy usage and fatty acid degradation, which is in line with our findings of higher anaerobic fatty acid beta-oxidation in polypharmacy users.Our study has several important strengths. To our knowledge, this is the largest study of European general population cohort to report connections between medication use and gut microbiota using high resolution metagenomic sequencing5,12,24. Another strength is the high-quality medication data, collected from both questionnaire and medication-register. Further, the data of MOS cohort allows adjustments for known potential confounders such as diet and physical activity. Also, we performed analyses for the medications at different ATC-levels, which enabled identification of associations at both medication-class and substance level. Utilizing these strengths, we replicated several findings of previous studies, but also identified numerous new associations to be validated in future studies.Our study also has some limitations. First, the study population is Swedish, and although the results align with earlier results in European studies, they may not be generalizable to other populations. Second, the cross-sectional design prohibits any conclusions about causality. Third, we could not address medication dosages, intermittent or continuous use, nor the potential impact of specific co-administrated medications on the observed associations. Fourth, due to the lack of data we could not account for stool consistency as a potential confounder in our analyses. Fifth, adjusting for Shannon index aiming to reduce false positive findings due to the collinearity of the relative abundances of species and Shannon index, might concomitantly have increased bias and decreased power if Shannon index acted as a collider or mediator in some cases. Therefore, we also report all analyses unadjusted for Shannon index. Sixth, adjustments for fiber intake and physical activity could only be performed in the subset of 1475 participants which decreased the power, however, the estimates of most associations remained similar to those in the whole study cohort, indicating that confounding by these factors was limited. Finally, the associations between medication use and the gut microbiota can obviously be confounded by the indications for the treatments, or by other factors that co-vary with the medication.
