Predicting and Manipulating Cardiac Drug Inactivation by the Human Gut Bacterium Eggerthella lenta

See allHide authors and affiliations

Science  19 Jul 2013:
Vol. 341, Issue 6143, pp. 295-298
DOI: 10.1126/science.1235872

Digoxin Dangers

A proportion of patients treated with digoxin, a cardiac glycoside used to treat heart function abnormalities, generate the inactive metabolite, dihydrodigoxin, resulting in poor efficacy. Haiser et al. (p. 295) examined a potential culprit responsible for this transformation—the actinobacterium, Eggerthella lenta—to probe the microbiota-digoxin interaction. Microbe growth was promoted by arginine, and differential expression analysis revealed a two-gene cardiac glycoside reductase (cgr) operon that was induced by digoxin in low arginine conditions. Not all strains of E. lenta could reduce digoxin and, when fecal samples from healthy people were tested, a spectrum of digoxin inactivation was detected. When the digoxin-reducing strain of E. lenta was given to germ-free mice that were fed a high-protein (that is, high-arginine) diet, digoxin levels stayed high in serum, and drug inactivation was suppressed.


Despite numerous examples of the effects of the human gastrointestinal microbiome on drug efficacy and toxicity, there is often an incomplete understanding of the underlying mechanisms. Here, we dissect the inactivation of the cardiac drug digoxin by the gut Actinobacterium Eggerthella lenta. Transcriptional profiling, comparative genomics, and culture-based assays revealed a cytochrome-encoding operon up-regulated by digoxin, inhibited by arginine, absent in nonmetabolizing E. lenta strains, and predictive of digoxin inactivation by the human gut microbiome. Pharmacokinetic studies using gnotobiotic mice revealed that dietary protein reduces the in vivo microbial metabolism of digoxin, with significant changes to drug concentration in the serum and urine. These results emphasize the importance of viewing pharmacology from the perspective of both our human and microbial genomes.

Humans are home to large and diverse microbial communities, the most abundant of which resides in the gastrointestinal tract. Recent studies have highlighted the clinical relevance of the biotransformations catalyzed by the human gut microbiome, including alterations to the bioavailability, activity, and toxicity of therapeutic drugs (1, 2). Although >40 drugs are metabolized by the gut microbiome, little is known about the underlying mechanisms. This knowledge is critical to enable the rational design of pharmaceutical or dietary interventions.

The inactivation of the cardiac drug digoxin provides a promising starting point for understanding microbial drug metabolism. Digoxin and other cardiac glycosides have been widely used for hundreds of years to treat heart failure and arrhythmias. Therapeutic effects are accomplished indirectly when inhibition of the Na+- and K+-dependent adenosine triphosphatase (Na+/K+ ATPase) in cardiac myocytes raises the intracellular Ca2+ concentration (3). Digoxin has a narrow therapeutic range (0.5 to 2.0 ng/ml) (3), and some patients excrete the inactive digoxin metabolite, dihydrodigoxin, in which the lactone ring is reduced (fig. S1A) (4). This modification disrupts ring planarity, which is thought to shift positioning within the binding pocket of the Na+/K+ ATPase, resulting in decreased target affinity (5). Coadministration of broad-spectrum antibiotics increases serum digoxin (4), and Eggerthella lenta reduces digoxin in vitro (6). Before this work, the molecular mechanism of digoxin reduction and the factors that alter microbial drug inactivation in vivo were unknown.

We confirmed that E. lenta DSM2243, the type strain, reduces digoxin in vitro (7) and that arginine inhibits this reaction (Fig. 1A). The growth of E. lenta DSM2243 was stimulated by arginine supplementation (Fig. 1A and fig. S2), indicative of using the arginine dihydrolase pathway for adenosine 5′-triphosphate (ATP) (8). Citrulline (an intermediate upstream of ATP production) stimulated growth, whereas ornithine (an end product) did not (figs. S2 and S3).

Fig. 1 Discovery of a bacterial operon induced by digoxin.

(A) Arginine stimulates the growth of E. lenta DSM2243 in vitro while blocking the reduction of digoxin. Maximum optical density (absorbance) at 600 nm (OD600) (solid line; values are means ± SEM; n = 3) and digoxin percentage reduction efficiency (dashed line; values are means; n = 2) after 48 hours of growth. (B) RNA-Seq profiles of the cgr operon are shown with and without digoxin during exponential growth in medium containing low or high arginine. The height is proportional to the natural log of the number of unambiguous sequencing reads mapped to each base. (C) cgr2 transcription as determined by QRT-PCR. Asterisks indicate statistical significance by Student’s t test (P < 0.05). Horizontal lines are means; n = 2 to 3. (D) Identification of two strains of E. lenta incapable of reducing digoxin. Values are means ± SEM; n = 3. ND, no reduction detected.

E. lenta cultures were grown anaerobically in rich medium supplemented with low and high levels of arginine (0.25% and 1.25%, respectively) in the presence or absence of digoxin (10 μg/ml), and we performed RNA sequencing (RNA-Seq) on the resultant cellular biomass (figs. S4 to S6 and table S1). A two-gene operon was highly up-regulated after exposure to digoxin during exponential growth (>100-fold) (Fig. 1B and tables S2 and S3). These two genes, referred to here as the cardiac glycoside reductase (cgr) operon (gene labels: cgr1 and cgr2), encode proteins that are homologous to bacterial cytochromes and are therefore potentially capable of using digoxin as an alternative electron acceptor. Incubation of E. lenta with multiple cardiac glycosides and their reduced forms revealed that the cgr operon is broadly responsive to compounds with an α,β-unsaturated butyrolactone ring (figs. S7 to S9 and table S5).

Digoxin induction was increased in low-arginine conditions during both its exponential and stationary phases, relative to cultures exposed to high levels of arginine (fig. S10, A and B). cgr induction by digoxin and the growth phase–dependent effects exerted by arginine were confirmed on independent samples using quantitative reverse transcription polymerase chain reaction (QRT-PCR) (Fig. 1C, fig. S10C, and table S4). Unlike arginine, ornithine did not repress cgr2 expression (fig. S11). These results are consistent with the hypothesis that arginine represses cgr operon expression and thereby inhibits digoxin reduction.

Next, we tested three strains of E. lenta (DSM2243, FAA 1-3-56, and FAA 1-1-60) (9, 10) for digoxin reduction; the type strain was the sole strain capable of digoxin reduction in vitro (Fig. 1D). Comparative genomics revealed that the type strain was nearly indistinguishable from the other two strains when common marker genes were used (fig. S12). Reciprocal BLASTP comparisons of all protein-coding sequences of the three fully sequenced E. lenta strains revealed that the type strain shared 79.4% and 90.5% of its proteome with strains FAA 1-3-56 and FAA 1-1-60, respectively (fig. S12). The cgr operon was unique to the type strain (table S6); furthermore, the two nonreducing E. lenta strains were missing three genomic loci, which were also up-regulated by digoxin, and are predicted to encode membrane transporters for the uptake of small molecules and glycosides (fig. S13). Arginine did not significantly decrease the expression level of these transporters (fig. S14).

Strain-level variation provides an explanation for the difficulties in predicting dihydrodigoxin levels in cardiac patients by the presence or absence of E. lenta (6, 11). We used quantitative PCR (QPCR) to measure the abundance of the cgr operon relative to the E. lenta 16S ribosomal RNA (rRNA) gene (the “cgr ratio”) in microbial community DNA from 20 unrelated healthy people, along with ex vivo digoxin reduction assays. The results stratified our cohort into low reducers (12.82 ± 10.68% reduction; n = 6) and high reducers (96.25 ± 7.69% reduction; n = 14) (Fig. 2A). The cgr ratio was significantly increased for the high reducers (1.058 ± 0.562) when compared with low reducers (0.425 ± 0.582; P < 0.05, Student’s t test) (Fig. 2B and fig. S15). Linear regression of reduction efficiency with the cgr ratio revealed a significant correlation (R2 = 0.22, P < 0.05), whereas the abundance of E. lenta failed to predict the extent of reduction (R2 = 0.06, P = 0.30). The optimal cgr ratio cutoff (0.6) predicted digoxin reduction efficiency with a sensitivity of 86%, specificity of 83%, and precision of 92%.

Fig. 2 A microbial biomarker predicts the inactivation of digoxin.

(A) Liquid chromatography–mass spectrometry (LC-MS) was used to quantify digoxin reduction in the fecal microbiomes of 20 unrelated individuals. (B) The cgr ratio was significantly different between low and high reducers. Data represent QPCR with the cgr2 gene, and E. lenta–specific 16S rDNA primers (table S4). (C) Five low-reducing fecal microbial communities were incubated for 5 days in the presence or absence of E. lenta DSM2243 or FAA 1-3-56. LC-MS was used to quantify the completion of digoxin reduction. Supplementation with the nonreducing strain of E. lenta did not significantly affect digoxin reduction efficiency. (D) The cgr ratio was obtained for each of the low-reducing microbial communities after incubation. Outliers were identified using Grubbs’ test (P < 0.01) and removed. Values are means ± SEM. Points in (A) and (B) represent biological replicates. Asterisks indicate statistical significance by Student’s t test (*P < 0.05; ***P < 0.001; ****P < 0.0001).

Coculture of E. lenta with the fecal microbiome enhanced the efficiency of digoxin reduction. Each low-reducing fecal sample was incubated with the type (reducing) and FAA 1-3-56 (nonreducing) strains of E. lenta. The communities incubated with the type strain reduced more digoxin (95.39 ± 2.41%) than the type strain alone (68.91 ± 7.70%; P < 0.05, Mann-Whitney U test) (Fig. 2C). The cgr ratio was significantly elevated after coculture (Fig. 2D) and was tightly linked to reduction efficiency (R2 = 0.74, P < 0.0001). An explanation for the observed microbial synergy is that the fastidious growth of E. lenta is promoted by growth factors supplied by the gut microbiota, a phenomena that is known to affect the metabolism of environmental pollutants by soil microbial communities (12), along with competition for arginine that boosts digoxin reduction by E. lenta. Consistent with these hypotheses, the abundance of the E. lenta type strain was significantly increased in the presence of a complex microbial community (1.6e6 ± 4.8e5 versus 1.8e5 ± 8.4e3 in isolation; P < 0.05, Mann-Whitney test), and arginine supplementation suppressed the reduction of digoxin during coculture (fig. S16).

Diet could also explain interindividual variations in digoxin reduction. In vitro growth of E. lenta showed that, although arginine stimulated cell growth, it decreased cgr operon expression and prevented the conversion of digoxin to dihydrodigoxin (Fig. 1, A and C, and fig. S10). These observations led us to hypothesize that increased consumption of dietary protein, and the corresponding increase in arginine, would inhibit the in vivo reduction of digoxin by E. lenta. Germ-free adult male Swiss-Webster mice were colonized with the type strain before being fed diets differing only in the amount of total protein (n = 5 mice/group) (tables S7 and S8 and fig. S17A). E. lenta colonized mice fed both diets (fig. S18A) and exhibited high levels of expression of the cgr operon (fig. S18B). Quantification of serum and urine digoxin (7) revealed significant increases in mice fed the high-protein diet, indicative of suppressed digoxin reduction by E. lenta (Fig. 3, A and B). These trends were also consistent with fecal analysis of samples from each group of mice 4 to 16 hours after digoxin administration (Fig. 3). We also confirmed that the high-protein diet significantly elevated the amino acids in the distal small intestine (7), which resulted in a fold increase of 1.71 ± 0.06 (P < 0.001, Wilcoxon test) (tables S9 and 10).

Fig. 3 Dietary protein blocks the inactivation of digoxin.

Serum (A) and urinary (B) digoxin levels from the type strain experiment. Fecal digoxin levels showed a consistent trend: the mean area under the curve was 6.226 ng digoxin per hour per ml in germ-free mice, 3.576 for mice fed the 0% protein diet, and 6.364 for mice fed the 20% protein diet. Serum (C) and urinary (D) digoxin levels from each group. Digoxin levels were quantified by enzyme-linked immunosorbent assay (ELISA) (7). Values are means ± SEM. Asterisks indicate statistical significance by Student’s t test (*P < 0.05; **P < 0.01). n = 4 to 5 mice per group. NS, not significant.

We controlled for the indirect effects of host diet and colonization that might alter digoxin pharmacokinetics irrespective of reduction by E. lenta. Germ-free mice were colonized with either the digoxin-reducing type strain or the nonreducing FAA 1-3-56 strain and subsequently fed the same two diets (fig. S17B). As seen before, we detected colonization with both strains, high–cgr operon expression, and elevated serum and urine digoxin on the high-protein diet for mice colonized with the type strain (Fig. 3, C and D, and fig. S18, C and D). Diet did not significantly affect the serum or urine digoxin levels of mice colonized with the nonreducing strain (Fig. 3, C and D). Serum digoxin was significantly lower in mice colonized with the type strain fed the 0% protein diet, relative to those colonized with the nonreducing strain (4.91 ± 1.56 ng/ml vs. 13.8 ± 1.25; P < 0.01, Student’s t test) (Fig. 3C). Together, these results suggest that the enhanced free amino acids available to E. lenta inhibited the activity of the cgr operon and increased the bioavailability of digoxin.

An expanded model of digoxin pharmacokinetics is now emerging: Colonization by distinct strains of E. lenta, microbial interactions, and host diet act together to influence drug levels (fig. S19). Follow-up studies in cardiac patients are necessary to determine whether rapid QPCR-based biomarker assessments of the gut microbiome can guide dosage regimes. It may also be possible to provide dietary guidelines or supplements that prevent microbial drug metabolism. More broadly, our results emphasize that a comprehensive view of pharmacology includes the structure and activity of our resident microbial communities and a deeper understanding of their interactions with each other, with their host habitat, and with the nutritional milieu of the gastrointestinal tract.

Supplementary Materials

Materials and Methods

Figs. S1 to S19

Tables S1 to S10

References (1325)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: B. Budnik and S. Trauger for liquid chromatography–mass spectrometry (LC-MS) analyses; V. Yeliseyev, A. Liou, and R. Carmody for mouse studies; C. Reardon and C. Daly for sequencing support; C. Maurice, L. David, R. Dutton, B. Wolfe, J. Button, M. Elliot, Y. Falanga, R. Losick, A. Murray, and B. Stern for helpful discussions. Mouse experiments were done with the generous support of the Harvard Digestive Diseases Center and the University of North Carolina gnotobiotic cores. This work was supported by grants from the NIH (P50 GM068763) and the Harvard Digestive Diseases Center (2P30DK034854-26). H.J.H. is supported by the Canadian Institutes of Health Research (MFE-112991). RNA-Seq data are deposited in the Gene Expression Omnibus (GEO) database (accession GSE43919).

Stay Connected to Science

Navigate This Article