Gut microbial metabolism of 5-ASA diminishes its clinical efficacy in inflammatory bowel disease

Abstract

For decades, variability in clinical efficacy of the widely used inflammatory bowel disease (IBD) drug 5-aminosalicylic acid (5-ASA) has been attributed, in part, to its acetylation and inactivation by gut microbes. Identification of the responsible microbes and enzyme(s), however, has proved elusive. To uncover the source of this metabolism, we developed a multi-omics workflow combining gut microbiome metagenomics, metatranscriptomics and metabolomics from the longitudinal IBDMDB cohort of 132 controls and patients with IBD. This associated 12 previously uncharacterized microbial acetyltransferases with 5-ASA inactivation, belonging to two protein superfamilies: thiolases and acyl-CoA N-acyltransferases. In vitro characterization of representatives from both families confirmed the ability of these enzymes to acetylate 5-ASA. A cross-sectional analysis within the discovery cohort and subsequent prospective validation within the independent SPARC IBD cohort (n = 208) found three of these microbial thiolases and one acyl-CoA N-acyltransferase to be epidemiologically associated with an increased risk of treatment failure among 5-ASA users. Together, these data address a longstanding challenge in IBD management, outline a method for the discovery of previously uncharacterized gut microbial activities and advance the possibility of microbiome-based personalized medicine.

Extended Figure 1.. Fecal 5-ASA levels according to self-reported 5-ASA use in the IBDMDB.

We determined 5-ASA use according to detection of drug levels in stool using LC-MS, defined as detection of fecal 5-ASA levels > 10e7. Concordance, determined according to statistical accuracy (Methods), between self-reported use of 5-ASA and detection of fecal 5-ASA was 80.3%. Users (light blue) had 5-ASA levels which were ~10,000x greater than non-users (navy blue). Each individual is represented by multiple points on this graph given that participants provided multiple samples across the year-long cohort. Boxplots show median and lower/upper quartiles; whiskers show inner fences.

Extended Figure 2.. Representative indirect effects of 5-ASA on the fecal metabolome include shifts in Vitamin B3 and its metabolism towards glycine conjugation and formation of nicotinuric acid.

A) We considered that 5-ASA may lead to highly specific markers of medication use mediated through indirect microbial pathways, such as promotion of growth of certain bacteria which, in turn, generate compounds. Here, we find that the abundance of nicotinic acid (NA) dropped as nicotinuric acid (NUA) dramatically increased; there was no consistent change in the third niacin metabolite detected on our platform, N¹-methyl-nicotinamide. While the precise role of 5-ASA in determining the fate of NA is unknown, for >70 years, anaerobic bacteria, including Clostridia species have been known to metabolize NA (24). Furthermore, medications, such as aspirin, are known to affect the balance of NA and NUA in the blood. Intriguingly, fecal NA levels have previously been detected at lower levels in IBD patients compared to healthy controls, but confounding by 5-ASA use was not explored. Given the role of anaerobic gut bacteria in metabolism of NA, it is conceivable that gut bacteria promoted by 5-ASA – or 5-ASA itself – shunts NA towards the glycine conjugation pathway without profound impact on amidation. This phenomenon was not seen in initiators of steroids or biological drugs in the cohort; NUA levels were undetectable in non-5-ASA users. B) These effects were highly specific to 5-ASA use and could not be attributed simply to suppression of inflammation: in a subset of 9 participants in the IBDMDB who started biologic drugs, there was no change in nicotinic acid levels, and nicotinuric acid levels were undetectable in non-users of 5-ASA. In both panels, boxplots show median and lower/upper quartiles; whiskers show inner fences.

Extended Figure 3.. Impact of 5-ASA on the fecal metabolome is influenced by the gut microbiome.

A) Drug levels, followed by gut microbiome taxonomic profiles, independently explain variation in most 5-ASA-derived metabolite levels. These were analyzed systematically by linear models constructed for each metabolite with independent variables being: fecal 5-ASA levels; microbiome data; host data, including diet, disease type, age, other medications, and sex; and other/unexplained. We quantified variance explained (EV) by each model and partitioned EV by each term for each 5-ASA-shifted metabolite (shown along the x-axis, as a stacked proportional bar plot). As an example, N-acetyl 5-ASA had a moderate correlation with 5-ASA levels in stool (Spearman rho 0.50, p=4e-9), with 35% of variance explained by drug levels, 15% explained by the microbiome, and 7% explained by other host features. B) Associations between differentially abundant (DA) metabolites and DA species using HAllA (Methods). Block associations are numbered in descending order of significance (max FDR<0.25), with each numbered block corresponding to a group of co-occurring metabolites with a species. A white dot indicates marginal significance of a particular pair of features (p<0.05). C) Abundance of R. inulinivorans was inversely correlated with the metabolite peak 242.0458 (represented by Block 17 in (B)), which was 37,661-fold greater on average among users than non-users (FDR 0.001). Boxplots show median and lower/upper quartiles; whiskers show inner fences. D) F. prausnitzii was positively associated with the peak 373.1254 (represented by Block 20 in (B)), which was 73-fold greater among users than non-users (FDR 0.001). Boxplots show median and lower/upper quartiles; whiskers show inner fences.

Extended Data Figure 4:. Microbial species previously characterized to acetylate 5-ASA have low or absent representation in the gut microbiomes of patients with IBD.

A) Phylogenetic analysis shows the taxonomic contributions to the MTX (purple, n=301 species) and MGX (gold, n=322 species) data in the IBDMDB. Also shown are bacteria (green, n=20 species) previously found to acetylate 5-ASA in vitro, all of which belong to the Proteobacteria phylum. Notably, only six of these had detectable MGX and MTX levels in the HMP2, which are labeled and listed. B) The abundance of Proteobacteria is low across the vast majority of participants, relative to Firmicutes and Bacteroidetes species. C) The prevalence of the six species with potential for 5-ASA acetylation activity is low, except E. coli. Most importantly, none of these six bacterial species have detectable arylamine N-acetyltransferase enzymes at the metatranscriptomic level in the IBDMDB.

Extended Figure 5.. Human gut microbiome genes were prioritized for experimental characterization via a three-pronged multi-omics approach.

We took three-pronged approach to identifying 5-ASA metabolizing enzymes; each part was independent of the other. In Part 1, we used two arylamine N-acetyltransferase (NAT) sequences from Salmonella enterica serovar typhimurium LT2 (nhoA) (UniProt accession, Q00267) and Pseudomonas aeruginosa (UniProt accession, Q9HUY3), previously shown to metabolize 5-ASA, as well as 105 NAT or N-hydroxyarylamine O-acetyltransferase microbial protein sequences predicted to metabolize 5-ASA (Methods, Extended Table 4) as the query for a BLASTP search (Diamond v0.9.24.125) of the Human Microbiome Project (HMP) reference isolate genomes with an e-value of 10 down to the 25% identity. Given that there were no hits at the metatranscriptomic level, in Part 2, we then used differential abundance testing to identify significantly overexpressed acetyltransferases, fit with a per-feature linear mixed-effects model, which adjusted for DNA copy number, which allows for biological and technical zero values while also controlling for underlying DNA levels (q < 0.25). These two hits were expanded at 80% homology to find similar functional proteins. In Part 3 of our approach to identifying 5-ASA metabolizing enzymes, we compared metatranscriptomic abundance of individual gene families (present/absent) with the metabolomics readout of dichotomized N-acetyl 5-ASA (high/low) in each stool sample, and then derived measures of sensitivity (true positives / (true positives + false negatives)) and specificity (true negatives / (true negatives + false positives)) for each gene cluster to estimate how well a given cluster correlates with the drug metabolite. All of these were pooled to arrive at 12 candidates.

Extended Figure 6.. Heterologous expression of thiolase and acyl-CoA N-acyltransferase enzymes in E. coli BL21.

(A) Coomassie-stain gel shows purity of indicated enzymes after overexpression and cobalt purification. N=1 protein purification per enzyme; repeated independently n=2-4 for each enzyme. Abbreviations: FcTHL = a predicted thiolase (R6CZ24) from an uncultured Firmicutes; FpGNAT = a predicted acyl CoA N-acyltransferase (C7H1G6) from F. prausnitzii; StNAT = a known arylamine N-acetyltransferase from Salmonella enterica serovar typhimurium LT2. (B) Confirmation of in vitro acetylation of 5-ASA by known S. typhimurium enzyme using 1 mM of each substrate and 5 uM enzyme incubated for 1hr at RT. Data are presented as mean values +/− SEM.

Extended Fig. 7.. Extended biochemical data for the 5-ASA metabolizing thiolase enzymes.

(A) In vitro competition assay with 1 mM of 5-ASA, 4-ASA, procainamide, hydralazine and isoniazide demonstrates relative specificity of the Firmicutes CAG:176 thiolase (FcTHL) for 5-ASA. N=3 biologically independent samples per enzyme/condition. (B) Shown is a representative Michaelis-Menten plot of n=1 thiolase enzyme preparation, conducted in technical triplicates at each concentration of 5-ASA; summary data is from n=5 biologically independent experiments each conducted in technical triplicate. (C) Live culture of Oscillibacter sp., strain KLE 1745 encoding another predicted thiolase gene (UniRef90 R6TIX3) was capable of acetylation of 5-ASA to N-acetyl 5-ASA. N=3 biologically independent samples per condition, representative data from n=2 independent experiments, unpaired, two-sided T-test, *** = p = 0.0015. Data are presented as mean values +/− SEM.

Extended Figure 8.. Comparison of canonical acetyltransferase reaction mechanisms for thiolase and arylamine NAT enzymes, now both shown to acetylate 5-ASA.

A) The thiolase two-step “ping pong” mechanism (Modis) begins after a cysteine is activated by a nearby histidine residue and then performs a nucleophilic attack on an acetyl CoA molecule to form a covalent acetyl-enzyme intermediate (shown in Fig 4d). In the second step, the substrate (classicaly a second acetyl CoA molecule) nucleophilically attacks the acetyl-enzyme intermediate to yield the final acetyoacetyl-CoA and enzyme. The second nucleophilic attack is activated by a second cysteine residue in the active site, which deprotonates the substrate. B) In a similar two-step “ping pong” mechanism in the arylamine NAT, a cysteine is also activated by a nearby histidine residue and then performs a nucleophilic attack on an acetyl CoA molecule to form a covalent acetyl-enzyme intermediate. In contrast, rather than abstracting a proton from another cysteine residue in the active site, the departing CoA molecule deprotonates the histidine residue, which allows an arylamine substrate to perform nucleophilic attack on the acetyl-enzyme intermediate to yield the final acetylated substrate and enzyme. C) We speculate that in some cases, an acetyl-thiolase enzyme intermediate is formed which may allow nucleophilic attack by an arylamine, such as 5-ASA.

Extended Figure 9.. Overview of the Study of a Prospective Adult Research Cohort with IBD (SPARC IBD) study.

A) Timeline shows the stool sampling scheme in relation to clinical assessments throughout the cohort. At study entry, all participants provided a single stool sample. The median time to event (use of steroids) was 229 days. Among a minority of participants (n=33) who voluntarily provided additional stool samples throughout the cohort (denoted by empty blue circles, Methods), median interval sampling time was 133 days. B). Flow diagram illustrating inclusion and exclusion criteria for this analysis. C) In a sensitivity analysis, we limited our analysis to a single (baseline) stool sample per participant. With diminished power, the SPARC IBD-specific estimate was no longer significant (OR 2.22, 95% CI 0.74-6.69). Accordingly, age and CD were not significantly associated with risk of steroids despite being established risk factors for 5-ASA treatment failure. Nevertheless, the pooled meta-analysis was not meaningfully different from the primary analysis (OR 2.78, 95% CI 1.19-6.50).”

Extended Figure 10:. Validity of 5-ASA and N-acetyl 5-ASA annotation by metabolomics methods.

Two separate standards for 5-ASA (SIGMA catalog PHR1060 and 18858) confirmed the identity of the 5-ASA peak in the IBDMDB (m/z: 154.0502, RT: 3.83 min) through retention time (panel A) and spectral matching (panel B). A standard for N-acetyl 5-ASA (Cayman Catalogue 27618) produced two peaks which matched two peaks in the IBDMDB (QI3818 and QI3816) (panel C). The later eluting peak (QI3816), was more abundant in the IBDMBD stool, correlated better with 5-ASA (Pearson’s correlation r=0.89 vs r=0.60), and had excellent MS/MS spectral matching (panel D). Further still, levels of QI3816 perfectly discriminate 5-ASA users from non-users (c-statistic 0.99).

Figure 1:. Identification of microbial 5-ASA inactivating enzymes from IBD microbiome population multi-omics.

As part of the HMP2 IBDMDB (timeline, upper left), 132 participants with and without IBD were followed for 1 year, each completing multiple dietary and medication questionnaires, and each providing stool every two weeks and blood samples approximately quarterly. After excluding participants without IBD or without metabolomics data, we identified 45 verified users of 5-ASA in the cohort and 34 non-users. Among 5-ASA users, we found 13 individuals who started or resumed using the drug during the cohort follow-up. Stool from >1,000 samples was then profiled through metagenomics, metatranscriptomics, and/or metabolomics; blood was analyzed by exome sequencing which was ultimately leveraged to determine human NAT2 acetylation phenotypes (“fast” vs. “slow”) for our clinical exploration. In the analysis phase, we first studied the impact of 5-ASA on the fecal metabolome and then identified gut bacterial enzymes involved in inactivating 5-ASA to N-acetyl 5-ASA. Finally, we related these enzymes to risk of disease relapse.

Figure 2:. 5-ASA directly impacts the fecal metabolome and undergoes biotransformation by the microbiome.

(A) The IBD fecal metabolome segregates by 5-ASA status (PERMANOVA R2=6.8%, p<0.001) more than by UC or CD diagnosis (R²=2.2%), suggesting a substantial role of medication in modulating the fecal biochemical environment of IBD patients (95% bivariate normal confidence ellipses shown, Methods). (B) Initiation of 5-ASA among a subset of participants (n=13) reveals 2,306 (total n=81,868, 2.8%) altered metabolomic features when comparing profiles pre- and post- 5-ASA administration (paired two-sided Wilcoxon, FDR q < 0.25), collected an average of 13.0 (± 8.7) weeks apart. Only 17 were assigned Human Metabolome Database (HMDB) identifiers, including the known 5-ASA metabolite, N-acetyl 5-ASA, as well as potential off-target effects - including shifts in vitamin B3 metabolism and bacterial products implicated in oxidative stress (23, 24). **, q<0.05; *, q<0.25; NS, not significant. Boxplots show median and lower/upper quartiles; whiskers show inner fences. (C). Examining the remaining 2,293 unannotated metabolomic features, we identified two promising candidates in the IBDMDB and independently profiled PRISM datasets using mass differences and retention time matching as likely N-propionyl 5-ASA and N-butyryl 5-ASA that also discriminated 5-ASA users from non-users (c-statistic >0.95) that were not initially annotated by the HMDB (Methods). Boxplots show median and lower/upper quartiles; whiskers show inner fences.

Figure 3:. Microbial genes metatranscriptomically implicated in generation of N-acetyl 5-ASA cluster into thiolase and acyl CoA N-acyltransferase superfamilies.

(A) MTX abundance distributions for six putative 5-ASA-acetylating gene families in 5-ASA users vs non-users. Multivariate linear mixed effects models adjusted for DNA copy number (27) identified two significantly overexpressed gene clusters with putative acetyltransferase function in 5-ASA users vs. non-users: 1) a GNAT family N-acetyltransferase (UniRef90 ID: C7H1G6) and 2) an acetyl-CoA acetyltransferase (UniRef90 ID: R6TIX3) (FDR q 0.24 and 0.14, respectively). Searching for any additional sequences with at least 80% full-length sequence similarity yielded three additional hits, all from the first acetyl-CoA acetyltransferase, each nominally enriched in 5-ASA users compared to non-users. (B) Specificity and sensitivity for microbial transcripts with respect to presence/absence of fecal N-acetyl 5-ASA across samples from 5-ASA users identifies seven additional putative 5-ASA acetylating gene families. In our second criteria, we estimated sensitivity and specificity for how each metatranscriptomic gene cluster (presence/absence) detected dichotomized N-acetyl 5-ASA (high/low). Using a 50% cutoff (hashed line) for each test characteristic revealed an additional 7 putative acetyltransferase gene clusters. Shown outside of these bounds are results from the first criteria, highlighting high degrees of specificity, but lower sensitivity. (C) MTX relative abundances for three of these MBX-based candidates identified in the second criteria also correlated with fecal N-acetyl 5-ASA levels (R2 and p values inset). Error bands represent 95% confidence intervals. (D) Pooled metatranscriptomic families from parts A and B proved to cluster into two protein superfamilies – thiolase and acyl-CoA N-acyltransferases – which are carried primarily by Bacteroides and Firmicutes phylum members, respectively. (E). Multiple sequence alignment of these twelve candidate enzymes shows highly conserved sequences among the thiolase enzymes (green) but greater diversity among acyltransferases (light blue). (F) In genomes from isolate strains, the only acyl-CoA N-acyltransferase gene carried by a Firmicutes member was in a subset of F. prausnitzii strains. When hierarchically clustered according to a prior schema (repurposed with permission), these appeared to originate from only one of the clade’s major phylogroups, suggestive of acquisition via a horizontal transfer event.

Figure 4:. Heterologous expression and purification of a gut microbial acetyltransferase confirms 5-ASA acetylation activity in vitro.

(A) A predicted thiolase (R6CZ24) from an uncultured Firmicutes (FcTHL) and a predicted acyl CoA N-acyltransferase (C7H1G6) from F. prausnitzii (FpGNAT) convert 5-ASA to N-acetyl-5-ASA in the presence of acetyl-CoA. Reactions were carried out for 6 hr at 37°C with 1 mM of each substrate and 50 μM enzyme. N=3 biologically independent samples per enzyme/condition. (B) An in vitro pooled acyl-CoA competition assay with 1 mM of each acyl-CoA species, confirms varying length acyl groups can be donated to 5-ASA, as detected in our analysis in patients with IBD (Fig 2c). N=3 biologically independent samples per enzyme/condition. (****=p<0.0001, *=p= 0.02, one-way ANOVA followed by Tukey’s multiple comparisons test.) (C) Tetrameric crystal structure of FcTHL, shown as a ribbon diagram with one dimer in purple and orange and the other dimer in green and blue. Two tightly interacted dimers form a tetramer through an L-domain. D). Alignment of an acetylated monomer from the FcTHL with an acetylated biosynthetic thiolase monomer (1DM3) in complex with acetyl CoA from Zoogloea ramigera (ZrTHL) highlights good agreement and reveals an overlapping Cys-His-Cys catalytic triad poised to acetylate a substrate. (E) Using a similar method of protein structure superimposition, the crystallized Salmonella enterica typhimurium NAT (PDB ID: 1E2T, SeNAT, known to acetylate 5-ASA) and the Firmicutes thiolase have very different overall structures, yet both enzymes’ active sites contain cysteine and histidine residues in similar positions and perform similar reactions (Extended Fig11a-b). Where applicable, data are presented as mean values +/− SEM.

Figure 5.. Gut microbial 5-ASA-inactivating acetyltransferases are associated with greater risk of treatment failure in 5-ASA users.

(A) Bacterial acetyltransferase genes are associated with an increased risk of steroid use. Red squares with labels represent odds ratios, with 95% confidence intervals represented by gray bars. N=609 independent stool samples collected from 39 5-ASA users over the year-long HMP2 IBDMDB. (B) Metagenomic carriage of 3-4 acetyltransferases (identified in panel A) compared to those with 0-2 have an increased risk of steroid use in the IBDMDB. In an independent, prospective validation cohort, SPARC IBD (n=250 independent stool samples collected from 5-ASA users), baseline carriage of the same 3-4 acetyltransferases confers a similarly increased risk of future steroid use. Red squares with labels represent odds ratios. Pooled analysis (random effects meta-analysis) demonstrates consistent effect size across the two cohorts (pink diamond represents odds ratio) Studies are inversely weighted according to their variance. (C) Gene prevalence varies across participants with IBD but does not vary by 5-ASA status (GLMM, p≥0.07), arguing for effect modification of the effects of 5-ASA on prevention of disease relapse.