The liver–gut microbiota axis modulates hepatotoxicity of tacrine in the rat

ANIMAL STUDY DESIGN

All experiments were conducted in compliance with the GSK Policy on the Care, Welfare, and Treatment of Laboratory Animals, the guidelines of the Institutional Animal Care and Use Committee (protocol number 080342).

MAIN STUDY

Thirty-eight male Lister hooded (LH) rats (Harlan, UK) (weight, 237-373 g) were used. Of these, 26 rats received by oral gavage a single dose of 20 mg · kg⁻¹ tacrine, dissolved in 1% methylcellulose, at a volume of 10 mL · kg⁻¹. This dose was chosen to induce liver injury without causing death.6, 13, 16 The remaining 12 rats in the control group were administered an oral dose of 1% methylcellulose vehicle. The sample size was estimated by assuming that 50% of tacrine-dosed rats exhibit transaminitis and using the estimated standard deviation of ALT of 79 U · L⁻¹ from a pilot study. Plasma, urine, and feces samples were collected from each rat over a period of 176 hours. Livers (perfused with cold phosphate-buffered saline) were collected immediately at the end of the study, snap frozen, and stored at −80°C before analysis.

VALIDATION STUDY

Sixty-nine male LH rats (weight, 236-293 g) were divided into three treatment groups (sample size calculation is described in the Supporting Information). In group 1 (conventional treated), 23 rats were first given an oral gavage of vehicle used to prepare the β-glucuronidase enzymes as described previously.17 Five minutes later, 17 rats received a single oral dose of 20 mL · kg⁻¹ tacrine and six rats were orally administered vehicle (1% methylcellulose). Group 1 mirrored the main study and served as a control for the other treatment group. In group 2 (β-glucuronidase treated), all 23 rats were given a single oral dose of 168,400 U · mL⁻¹ of β-glucuronidase, at a dose of 0.6 mL · kg⁻¹, before oral gavage of either 20 mg · kg⁻¹ tacrine or 1% methylcellulose vehicle. To mitigate degradation, we calculated for a much higher dose of the β-glucuronidase enzyme (>1000-fold excess of β-glucuronidase required for conversion of the drug). In group 3 (antibiotics treated), all 23 rats were administered 50 mg/kg/d each of vancomycin and imipenem (given in drinking water) 3 days before similar treatment with tacrine. Plasma samples were prepared and collected for analysis.

CLINICAL CHEMISTRY AND TOXOCITY PHENOTYPING

Plasma aspartate aminotransferase (AST) and ALT were determined using a Cobas C111 analyzer (Roche Diagnostics, Basel, Switzerland). Based on LiverTox grading of drug-induced liver injury (https://livertox.nih.gov/Severity.html), tacrine-dosed rats with AST more than 3 times the mean AST of controls over the entire study duration were classified as responders to tacrine-induced transaminitis, while rats not meeting this criterion were classified as nonresponders (NonR). The responders were further arbitrarily classified as strong responders (StrR, AST elevation ≥3 measurements), moderate responders (ModR, AST elevation >2 measurements) or mild responders (MilR, AST elevation >1 measurement) based on the duration of transaminitis.

GC/TOFMS-BASED METABONOMICS

Urine and fecal metabonomic experiments were performed using the established in-house protocol as described by Chan et al.18 Detailed descriptions of the sample preparation, GC/TOFMS metabonomics and data analysis are provided in the Supporting Information.

PHARMACOKINETICS

Tacrine and hydroxytacrine metabolites (1-hydroxytacrine, 2-hydroxytacrine, and 4-hydroxytacrine [OH-THA]) were extracted from the rat plasma and fecal samples and their levels were determined via multiple reaction monitoring using liquid chromatography tandem mass spectrometry (Supporting Information). The plasma concentrations of tacrine and OH-THA metabolites of 26 tacrine-dosed rats (0-24 hours; main study) and 34 tacrine-dosed rats (0-32 hours; validation study) were determined. Pharmacokinetic parameters were determined using noncompartmental analysis (Phoenix WinNonlin 6.0, Tripos, L.P.). Pharmacokinetic parameters obtained included C_max, plasma clearance, and area under the plasma concentration time curve, with area under the curve (AUC)_0-last for time of last quantifiable concentration and AUC_0-∞ for AUC extrapolated to infinity. AUCs were calculated using linear up log down trapezoidal rule. Tacrine and OH-THA were quantitated using liquid chromatography tandem mass spectrometry in the feces of four StrR (T8, T10, T12, and T14) and six NonR rats (T1, T2, T11, T16, T18, and T27) collected over 0-8 and 8-24 hours. The amount of tacrine and OH-THA (nmol) was normalized against the dose (mg) administered to each LH rat. The percentage conjugation of OH-THA and tacrine were calculated using the equation below, where “Total” refers to the summation of free unconjugated and conjugated analyte and “Free” refers to the free unconjugated analyte.

PROFILING OF HEPATIC mRNA EXPRESSION

Total RNA was isolated from approximately 10 mg of tissue from each liver sample using RNeasy Mini Kit, according to the manufacturer's protocol. The hepatic expression of a panel of 17 candidate genes potentially involved in tacrine's disposition based on literature review were profiled using quantitative polymerase chain reaction, and their respective primer sequences are reported in Supporting Table 1. Details of the mRNA profiling experiments are described in the Supporting Information.

MICROBIAL DNA ISOLATION FROM RAT FECES

Microbial DNA was isolated from 84 LH rat fecal samples collected at –24-0, 0-8, 8-24, and 72-96 hours (Supporting Table 2) using the QIAamp DNA Stool Minikit by following a modified Stool Pathogen Detection protocol to maximize DNA extraction. Isolated DNA was checked for purity spectrophotometrically to assess 260/280 and 260/230 values. Concentration of double-stranded DNA (dsDNA) was quantified using Quant-iT PicoGreen kit.

16S rRNA GENE SEQUENCING

Polymerase chain reaction amplification of the V3-V4 region was performed, using the primer pair as described by Klindworth et al.19 The samples were prepared following the protocol in Nextera XT DNA Sample Preparation kit. Paired-end 2 × 250 bp sequencing was performed on the Illumina MiSeq using a 500-cycle MiSeq Reagent Standard Kit (version 2). Details of sample preparation and data processing are provided in the Supporting Information.

METAGENOMIC LIBRARY PREPARATION AND SEQUENCING

DNA libraries were prepared and paired-end sequencing (2 × 76 bp reads) was performed on the Illumina HiSeq2000 platform as detailed in the Supporting Information.

METAGENOMIC ASSEMBLY, OPEN READING FRAME (ORF) PREDICTION AND ANNOTATION

Low-quality bases (base quality <3) were trimmed and read pairs with short reads (<60 bp) were filtered out. De novo assembly was performed on the trimmed reads using SOAPdenovo20 (version 1.05) with parameters “-K 41 -M 3 -R -d -D” and open reading frame (ORF) prediction was performed using Prodigal21 (parameter: metagenomic mode). Genes were annotated based on matches to the KEGG database using RapSearch22 (default parameters).

ORFs annotated as β-glucuronidase were extracted from all metagenome assemblies and merged into a single reference. Rat genomic reads were identified and filtered out by mapping to the rat genome (rn6) using BWA-MEM with default parameters (bwa-0.7.9a). Filtered reads (microbiome reads) were then mapped using BWA-MEM to β-glucuronidase reference sequences to compute relative abundance. For computing the abundance of the uidA gene, mapping was performed with the uidA gene sequence as reference. Differential abundance testing (StrR versus NonR group) was performed using the webtool EDDA (version 1.4.0) (http://edda.gis.a-star.edu.sg/; upper-quartile normalization, cuffdiff mode).23

Metaphlan24 (default parameters) was used to profile community abundance at the genus level. Differential abundance testing (StrR versus NonR) was performed with the webtool EDDA (version 1.4.0) (upper-quartile normalization, cuffdiff mode)23 using the mapped clade-specific (genus) reads produced by metaphlan. An false discovery rate (FDR) cutoff (< 0.05) was used to filter for significant differentially abundant genera (StrR versus NonR). Bacteria genera that are present in less than 1% (defined by metaphlan) in all analyzed samples were filtered out (StrR versus NonR) as we regard their contributions to the microbiome as minimal. The raw sequencing reads in this study have been deposited in National Center for Biotechnology Information's Sequence Read Archive. The project number is PRJNA268906 and the Sequence Read Archive project number is SRP051029.

DATA AND STATISTICAL ANALYSIS

Details of data and statistical analysis are presented in the Supporting Information.

TACRINE HEPATOXICITY CORRELATED WITH PHARMACOKINETICS BUT NOT RAT GENETICS

We administered an acute oral gavage of tacrine at 20 mg · kg⁻¹ body weight to 26 LH rats, while 12 control rats received 1% methylcellulose vehicle (Fig. 1a). Clinical chemistry measurement of plasma ALT and AST revealed variable susceptibility to tacrine-induced transaminitis (Fig. 1b; Supporting Fig. S1a,b). Based on the severity of transaminitis, 10 tacrine-dosed rats with elevation of AST >3× mean of controls were classified as StrR (elevations over ≥3 time points; n = 4), ModR (elevations over 2 time points; n = 3), or MilR (elevations over 1 time point; n = 3) while 16 rats with AST <3× mean of controls were classified as NonR to reflect the graded responses. For the responders, the development of transaminitis began at 0-8 hours, peaked at 8-24 hours, and recovered from 24 hours post-administration of tacrine (Fig. 1b). Good correlation observed between the plasma ALT and AST markers at the peak of transaminitis (r² > 0.8), and necrosis and fatty changes in the midzonal and pericentral regions of the liver lobules in LH rat liver sections confirmed the occurrences of hepatocellular injury after tacrine administration (Supporting Fig. S1c-g). The differential susceptibility to tacrine-induced transaminitis in LH rats corroborated its clinical prevalence7, 10, 25 and validated the suitability of LH rats for further investigation.

Because several studies have indicated a link between tacrine-induced hepatotoxicity and its pharmacokinetics,13, 25 we next sought to validate such an association. Large interindividual variation in the plasma concentration versus time profiles of tacrine was apparent despite controlled preclinical experimental conditions (Fig. 1c). The first maximum plasma concentration (C_max) and the initial absorption kinetic were found to be similar among the tacrine-dosed rats, thus ruling out variation in intestinal drug absorption (Fig. 1c). Interestingly, several rats exhibited a “double-C_max” profile that was most prominent in the StrR (Fig. 1c). In addition, the StrR were found to have the highest C_max of tacrine (C_max = 638 ± 184 ng · mL⁻¹), largest area under the curve (AUC_0-∞) (AUC_0-∞ = 5494 ± 1714 h · ng · mL⁻¹), and lowest clearance (1272 ± 337 mL · h⁻¹) compared with all other toxicity phenotypes (Fig. 1d). Taken together, the higher systemic exposure to tacrine in the StrR was linked to their double-C_max profile, which is in turn suggestive of secondary intestinal drug absorption. Our results suggest that tacrine is a direct hepatotoxicant and corroborated previous findings.13-16

Enterohepatic recycling is a common mechanism yielding double-C_max pharmacokinetic profiles.26, 27 We hypothesize that the glucuronidated metabolites of tacrine are excreted into the intestine via active biliary secretion, where it is subsequently deconjugated by gut bacteria and reabsorbed intestinally as the parent drug. The augmented enterohepatic recycling of tacrine among the StrR increases its systemic exposure and enhances their susceptibility to transaminitis. Consequently, one would expect differential fecal excretion patterns of glucuronidated metabolites. To test our hypothesis, we compared the levels of free and conjugated tacrine and hydroxytacrine metabolites (OH-THA) in StrR and NonR. No significant differences were observed in the excretion of OH-THA between the StrR and NonR (Fig. 1e). However, the mean amount of total tacrine (free tacrine and tacrine-N-glucuronides) excreted by the StrR over 24 hours was significantly higher than that for the NonR (127.4 ± 69.9 nmol · mg⁻¹ dose vs 40.4 ± 17.8 nmol · mg⁻¹ dose; P < 0.05 [Mann-Whitney U test]) (Fig. 1e). Tacrine-N-glucuronide is a phase 2 conjugated metabolite of tacrine generated in the liver. 58.8% ± 10.0% of the total tacrine were excreted as tacrine-N-glucuronide in the feces of StrR, whereas this proportion was significantly lower (15.6% ± 3.4%) for the NonR (P < 0.05 [Mann-Whitney U test]) (Fig. 1f). In the intestines, tacrine-N-glucuronide might undergo deglucuronidation catalyzed by microbial β-glucuronidase to yield free tacrine for completion of enterohepatic recycling. In summary, vastly different fecal excretion patterns were observed between the StrR and NonR that was consistent with their variable pharmacokinetics.

The liver is a major organ involved in drug disposition and detoxification processes. We next profiled the mRNA expression levels of disposition-related genes in rat livers to determine the potential role of host factors in defining the toxicological outcome. No significant difference was found in the mRNA expression levels of the panel of 17 genes (uptake and efflux transporters, phase 1 and 2 metabolizing enzymes, biliary excretion transporter, and nuclear receptors) relevant to the pharmacokinetics of tacrine between the tacrine-dosed rats (n = 26) and controls (n = 12). The expression were also generally similar between each toxicity phenotypes (n = 10) and the NonR (n = 16) (Fig. 2). This suggests that beyond the host's hepatic gene expression, there may be other factors underlying tacrine-induced transaminitis.

MICROBIAL METABOLITES OF TACRINE-INDUCED HEPATOTOXICITY

Metabonomics has been demonstrated to provide mechanistic insights in toxicology and reveal information about the crosstalk between the host and gut microbiota in systems biology.4, 28, 29 We performed GC/TOFMS-based metabonomics to screen for transaminitis-related fecal and urinary marker metabolites. Unsupervised principal component analysis revealed similar predose urinary and fecal metabolomes among all control and tacrine-dosed rats (Supporting Fig. S2a,b). A careful analyses of the StrR revealed a metabolic trajectory that followed the progression and resolution of transaminitis over −24-96 hours (Fig. 3a,b). At 0-8 hours, both urinary and fecal metabolomes of tacrine-dosed rats were found to cluster distinctly from the vehicle-dosed controls, indicative of metabolic changes after tacrine administration (Supporting Fig. S2c,d). At the peak of transaminitis (8-24 hours), the StrR were found to form a distinct cluster from the rest of the rats (Supporting Fig. S2e,f).

To screen for transaminitis-related metabolites, we compared the metabolomes of the StrR with the NonR during the peak of transaminitis (8-24 hours). Our analysis uncovered 43 unique urinary marker metabolites and 62 unique fecal metabolites associated with tacrine-induced transaminitis (variable importance in the projection > 1 and/or P < 0.05, |r| > 0.6) (Fig. 3c,d; Supporting Tables S3 and S4). One toxicity hallmark of tacrine is mitochondrial dysfunction.14, 15 We observed an overall depletion in citrate acid cycle and oxidative phosphorylation intermediates such as citrate, cis-aconitate, 2-ketoglutarate, succinate, fumarate, and malate in the urine of StrR associated with tacrine-induced transaminitis (Fig. 3e). This finding is consistent with other earlier mechanistic studies that demonstrated the impact of tacrine on the hepatic mitochondria. The perturbation of these key pathways may hence contribute toward the decline in ATP production by tacrine as observed by Berson et al.14

Notably, 56% and 45% of the respective urinary and fecal marker metabolites are of cometabolic origin,2, 3, 28-31 suggesting the implication of gut microbiota in tacrine-induced transaminitis. We observed an elevated abundance of fecal bile acids such chenodeoxycholic acid and lithocholic acid in StrR. Chenodeoxycholic acid is formed from cholesterol in the liver and biliary excreted into the intestine. Lithocholic acid is a secondary bile acid formed in the intestines by bacterial 7-α dehydroxylation of chenodeoxycholic acid32 and it is a well-established hepatotoxicant when present at high levels.33, 34 Members of the Bacteroides, Clostridium, and Eubacterium genera are known to possess 7-α dehydroxylation activity.35, 36 The distinction in the levels of these bile acids suggested differential regulation of bile acid metabolism or an augmented enterohepatic recycling in the StrR (Supporting Fig. S3). Elevated levels of metabolites such as phenylethylamine, p-hydroxyphenylacetic acid and indoleacetic acid were detected in the feces of StrR compared to NonR. They are major colonic fermentation products of phenylalanine, tyrosine and tryptophan and several Clostridium and Bacteroides species were reported to be responsible for the metabolism of these aromatic amino acids.37 Several long chain fatty acids such as oleic acid, stearic acid and linoleic acid were found in higher abundance in feces of StrR. These fatty acids possess either bactericidal or bacteriostatic activities.38-40 Gram-negative bacteria (e.g., Escherichia coli) generally have higher resistance to the toxic effects of these long chain fatty acids than gram-positive bacteria.41, 42 The elevations of these fatty acids in rats that are highly susceptible to tacrine-induced transaminitis suggest structural changes in the microbial community within the gut of these rats. Collectively, our metabonomics findings suggest a distinctive gut microbiota associated with tacrine-induced transaminitis and provide the impetus for a deeper understanding of the specific bacteria associated with this phenomenon.

GREATER DEGLUCURONIDATION POTENTIAL IN HEPATOTOXICTY

To further characterize changes in the gut microbiome that are associated with tacrine-induced transaminitis, we used 16S rRNA sequencing to profile rat gut microbiota composition before and after tacrine dosing (Supporting Table S5). A high proportion of gut microbiome was determined to be common between LH rats and humans (Supporting Fig. S4). Our data showed increased levels of Bacteroides and Enterobacteriaceae (both known to exhibit β-glucuronidase activity43, 44) in the StrR group compared with the NonR group during peak transaminitis (8-24 hours) (Fig. 4a,b; Supporting Fig. S5). Nevertheless, at baseline levels (before dosing), their abundance was similar for both StrR and NonR (Fig. 4a,b), suggesting that Bacteroides and Enterobacteriaceae may have a preferential growth pattern associated with tacrine administration. In contrast, Lactobacillus, a genus reported to inhibit bacterial β-glucuronidase activity45 and associated with hepatoprotection46 was found at a lower level in the StrR versus NonR groups (Fig. 4c; Supporting Fig. S5).

To study the differences in the functional composition of the microbiomes at the gene level, we employed shotgun sequencing of fecal samples collected during peak transaminitis (Supporting Table S6). We first used these data to confirm differences in microbial composition observed using 16S profiling, validating the higher abundance of β-glucuronidase producing bacterial genera in the StrR versus the NonR group (Fig. 4d). At the gene level, we first used the abundance of the uidA gene (encoding an extensively characterized β-glucuronidase) as a proxy and found the gene to be 6.1-fold higher in the StrR than in the NonR group (P < 8.41 × 10⁻¹²). Furthermore, we performed an unbiased analysis of all β-glucuronidases genes through a metagenome assembly, gene prediction, and mapping approach, and this also revealed their increased abundance within the microbiome in the StrR versus NonR groups (approximately 9% higher; P < 2.2 ×10⁻¹⁶). Taken together, our results demonstrate an association between gut microbial composition, β-glucuronidase abundance, and tacrine-induced transaminitis.

ORAL β-GLUCURONIDASE AND ANTIBIOTICS MODULATE SUSCEPTIBILITY

To validate the influence of intestinal β-glucuronidase activities on the susceptibility of LH rats to tacrine-induced transaminitis, we administered an oral gavage tacrine (20 mg · kg⁻¹) to LH rats pretreated with a single oral dose of 168,400 units · mL⁻¹ of β-glucuronidase 5 minutes before the administration of tacrine (n = 17) and compared this group with a conventional tacrine-dosed group without β-glucuronidase pretreatment (n = 17) (Fig. 5a). 41% of the conventional LH rats were classified as responders based on AST elevation > 3× average AST of the vehicle-dosed controls. Coadministration of oral β-glucuronidase significantly manipulated and augmented the susceptibility to tacrine-induced transaminitis from 41% to 76% (Supporting Fig. 6a–c) (P < 0.05; one-tailed Fisher's exact test). To confirm the strong correlation of tacrine systemic exposure with susceptibility to tacrine-induced transaminitis, in vivo pharmacokinetics studies of tacrine was further performed. β-glucuronidase treated tacrine-dosed rats were found to have a significantly higher systemic exposure to tacrine than the conventional tacrine-dosed rats (AUC_0-last of β-glucuronidase treated rats: 4135 ± 2055 h · ng · mL⁻¹ versus AUC_0-last of conventional rats: 2446 ± 1921 h · ng · mL⁻¹; P < 0.05 [Mann-Whitney U test]). C_max of tacrine was also found to be higher in the former group compared with the latter group (C_max of β-glucuronidase–treated rats: 608 ± 363 ng · mL⁻¹ versus C_max of conventional rats: 343 ± 316 ng · mL⁻¹; P < 0.05 [Mann-Whitney U test]). The difference in pharmacokinetic profiles between the two treatment groups corroborated with the higher susceptibility to tacrine-induced transaminitis in the LH rats supplemented with β-glucuronidase.

We next investigated the differences in the pharmacokinetic profiles between the toxicity phenotypes derived from the two treatment groups. For both groups of rats, the responders were found to have a significantly higher C_max and exposure (AUC_0-last) of tacrine compared with the NonR (Fig. 5b,c). In addition, these two pharmacokinetic parameters were comparable among the responders from both groups of rats. C_max and AUC_0-last of OH-THA metabolites were generally similar between the responders and NonR in both groups of rats except for a slightly higher C_max in the responders of conventional rats compared with their NonR (Fig. 5d,e). Because pretreatment of tacrine-dosed LH rats with β-glucuronidase did not alter the pharmacokinetics parameters among the responders from both groups, our findings suggest the kinetics of enterohepatic recycling of tacrine is saturable. Whereas the pretreatment of rats with vancomycin and imipenem did not influence the susceptibility of LH rats to tacrine-induced transaminitis based on AST measurement (47% versus 41%), the susceptibility was significantly decreased from 53% to 35% based on ALT measurement (Supporting Fig. 6a) (P < 0.05; one-tailed Fisher's exact test). In terms of pharmacokinetics profiles for the antibiotics-treated rats, similar observations were made where classification of responder phenotype was consistent with pharmacokinetics data.