Sequence similarity network analysis of drug- and dye-modifying azoreductase enzymes found in the human gut microbiome

Abstract

Drug metabolism by human gut microbes is often exemplified by azo bond reduction in the anticolitic prodrug sulfasalazine. Azoreductase activity is often found in incubations with cell cultures or ex vivo gut microbiome samples and contributes to the xenobiotic metabolism of drugs and food additives. Applying metagenomic studies to personalized medicine requires knowledge of the genes responsible for sulfasalazine and other drug metabolism, and candidate genes and proteins for drug modifications are understudied. A representative gut-abundant azoreductase from Anaerotignum lactatifermentan DSM 14214 efficiently reduces sulfasalazine and another drug, phenazopyridine, but could not reduce all azo-bonded drugs in this class. We used enzyme kinetics to characterize this enzyme for its NADH-dependent reduction of these drugs and food additives and performed computational docking to provide the groundwork for understanding substrate specificity in this family. We performed an analysis of the Flavodoxin-like fold InterPro family (IPR003680) by computing a sequence similarity network to classify distinct subgroups of the family and then performed chemically-guided functional profiling to identify proteins that are abundant in the NIH Human Microbiome Project dataset. This strategy aims to reduce the number of unique azoreductases needed to characterize one protein family in the diverse set of potential drug- and dye-modifying activities found in the human gut microbiome.

Graphical Abstract

Introduction

The trillions of non-human cells comprising the gut microbiome are known to play a central role in human metabolism, including modifying xenobiotics like food additives and pharmaceuticals. (1–3) Drug metabolism by gut microbes has been shown to result in inactivation, activation, and toxification of drugs, leading to varied drug responses in individuals because of the diversity in interindividual microbiomes that are suspected to encode 150 times the genes as the human host. (1, 4, 5) Recent efforts have attempted to capture the diversity of drug metabolism by directly testing gut microbiomes in ex vivo drug screens or incubating drugs with purified bacterial strains. (6, 7) However, strain-specific metabolism is often not enough to be predictive of drug modification; determining and quantifying the specific gene responsible (or similar genes) for digoxin reduction or diltiazem deacetylation was required to correlate with drug metabolism but does not correlate with the abundance of the organism. (7, 8) Current methods do not account for the contribution of gut microbes in drug metabolism, which invariably leads to unforeseen side effects and poorer therapeutic outcomes.

Among the first drugs found to be modified by gut microbes was prontosil, as this prodrug requires reduction of the azo bond (-N=N-) to form the active compound, sulfanilamide. (9) Other FDA-approved drugs such as sulfasalazine, olsalazine, balsalazide, and phenazopyridine contain azo bonds and can be modified by azoreductases (AzoRs) in gut bacteria, most commonly by NAD(P)H- and flavin-dependent oxidoreductases (Figure 1). Studies have found that the ability to reduce sulfasalazine is found in nearly all individuals, although individuals can show varying rates of reduction and that different azo-bonded drugs have differing reduction rates. (6, 10) Other work has shown that significant sulfasalazine reduction can be found in 62 of 76 common strains of gut bacteria, but identifying the genes responsible for this reduction and to what extent it may be an enzymatic process remains elusive. (7, 11–13)

Figure 1.

AzoR-catalyzed reduction of sulfasalazine.

AzoRs from gut bacteria have been characterized for their activity on azo dyes found in food and textiles (14–16) and, in some cases, for azo drug reduction. (12, 17) Efforts to characterize the protein families of AzoRs have shown the diversity in characterized enzymes and the relative ubiquity but low abundance of AzoRs in the human gut microbiome. (11, 12, 18) However, we are still in the early stages of describing the diverse functionality of these enzymes and cannot yet predict substrate specificity. AzoRs are from various protein families, named for this non-native reduction of azo bonds, but can also reduce quinones and nitroaromatic compounds. (14, 18–22) A recent review article (18) classified AzoRs in four clades: clade I contains the NADPH-dependent FMN reductase protein family (PF03358/IPR005025), predominately found in Pseudomonadota (formerly Proteobacteria); both clade II and III contain the Flavodoxin-like fold protein family (PF02525/IPR003680), NAD(P)H-dependent FMN-reductases predominately found in Bacillota and Bacteriodota (formerly Firmicutes and Bacteroidetes, respectively); and clade IV contains the NmrA-like family (PF05368/IPR008030), NADH-dependent and flavin-independent enzymes found in most gut bacteria. (11) Furthermore, the reduction of azo dyes may occur due to nonenzymatic reduction by hydrogen sulfide, the end product of microbial metabolism of cysteine, sulfate, and other sulfur-containing compounds. (13, 23) A full understanding of drug and food dye will require understanding the relative contributions of these different and unique sources.

This study aimed to identify a Clostridial AzoR from the most well-characterized protein family (PF02525/IPR003680) that could reduce azo-bonded drugs. This was undiscovered when we began our investigations, although a recent report provided positive identification of a Clostridial AzoR capable of reducing sulfasalazine. (12) We characterized the AzoR from Anaerotignum lactatifermentan DSM 14214 (AlAzoR, formerly Clostridium lactatifermentan) for its ability to reduce a variety of dyes and drugs and have identified the first AzoR capable of reducing the analgesic phenazopyridine. (24) We then sought to determine how this enzyme compares with others in the family and the abundance levels in human gut microbiomes. To begin this effort, we created a sequence similarity network (SSN) of the Flavodoxin-like fold protein family (PF02525/IPR003680), the family of AzoRs with the most characterized enzymes and only known AzoRs with azo-drug activity. (12, 17, 25, 26) After identifying isofunctional groups of AzoRs (i.e., grouped proteins based on similar activities as reported in the literature), we determined the gut-abundant members in this family using chemically-guided functional profiling (CGFP).(27, 28) AlAzoR is found in the NIH Human Microbiome Project Dataset HMP-1, although not ubiquitously across individuals. We compared a methyl red-bound structure of an AzoR from Pseudomonas aeruginosa and an AlphaFold model of AlAzoR to determine candidate amino acids in the active site that interact with docked substrates to see how computational binding energies related to our experimentally derived values. This analysis provides the groundwork for continued study into this family, as understanding the structure-function relationship in this family can inform one aspect of the diverse azo bond metabolism found in the human gut and wastewater treatment efforts. (11, 12)

Results

AlAzoR, a Clostridial AzoR, was characterized as it is from an accessible organism and has a unique active site from those reported in the literature. The synthetic codon-optimized gene for AlAzoR, cloned into an E. coli expression vector, showed high yields of recombinant expression (>100 mg per liter of culture) and azo drug-modifying activity (vide infra). We examined the specificity of AlAzoR for both NADH and NADPH by testing with a commonly used AzoR substrate, ethyl red. (29, 30) AlAzoR will reduce ethyl red using either cofactor but has a ~10-fold increase in specificity and ~100-fold increase in catalytic rate using NADH over NADPH (Table 1, Figure 2a). Recombinantly produced and purified AlAzoR has a yellow color, and UV-Vis spectroscopy shows the characteristic absorbance profile of a flavin (Figure 2b, λ_max = 376 nm, 456 nm). Heating the enzyme until it releases the non-covalently bound flavin causes a shift in the absorbance profile (λ_max = 367 nm, 443 nm) and was verified to be FMN by mass spectrometry.

Table 1.

NAD(P)H specificity of AlAzoR

Substrate	kcat (S−1)	kcat/KM (M−1S−1)	KM (μM)
NADH	-------	(2.16 ± 0.05) × 103	>400
NADPH	0.04 ± 0.01	(2.0 ± 0.5) × 102	170 ± 60

Figure 2.

Cofactor determination. A. AlAzoR kinetics with ethyl red and varying NADH (blue circles) or NADPH (yellow triangles). The insert displays the data with a smaller y-axis maximum value to view the NADPH data and curve. B. UV-Vis spectra of recombinantly produced AlAzoR (forest green) and after protein denaturation (lime green). The kinetic experiments were conducted in NaPO₄ buffer (pH 7.2) at 25°C.

Using a library of azo-bonded drugs and food dyes, we found several potential substrates of AlAzoR (Figure 3, the full list of compounds tested is found in Table S1). The enzyme appears to act as an NADH oxidase, as seen in other azoreductases, (31) so kinetic studies of AlAzoR were performed by monitoring the changes in the absorbance of the substrates. This competing NADH oxidase activity is not likely to affect any in vivo considerations of enzyme activity, as AlAzoR natively functions in an anaerobic environment. AlAzoR has the highest efficiency with methyl red and ethyl red, pH indicators with azo bonds, followed by the reduction of azo-bonded drugs sulfasalazine and phenazopyridine (Table 2, Figure S1). The absorbance spectrum of sulfasalazine overlaps with NADH (λ_max of 360 nm and 340 nm, respectively), so these kinetics were performed using lower amounts of NADH and monitored at a wavelength off the peak absorbance (at 383 nm). We repeated the kinetic constant determination with ethyl red using the same lowered amount of NADH to allow for a direct comparison between these two substrates. We also determined the kinetic constants for the AlAzoR-dependent reduction of indigo carmine, an indigo derivative that lacks an azo bond.

Figure 3.

Chemical structures of AlAzoR substrates.

Table 2.

Catalytic constants of AlAzoR with different substrates.

Substrate	kcat (s−1)	kcat/KM (M−1s−1)	KM (μM)
Methyl Reda	0.49 ± 0.03	(2.8 ± 0.4) × 104	17 ± 3
Indigo Carminea	1.1 ± 0.1	(3.4 ± 0.2) × 103	340 ± 40
Phenazopyridinea	0.138 ± 0.008	(6.0 ± 0.7) × 103	23 ± 3
Ethyl Reda	0.78 ± 0.02	(2.6 ± 0.2) × 104	30. ± 2
Ethyl Redb	0.113 ± 0.004	(1.9 ± 0.2) × 104	6.1 ± 0.8
Sulfasalazineb	0.050 ± 0.002	(6.1 ± 0.9) × 103	8 ± 1

^aDetermined with constant 400 μM NADH

^bDetermined with constant 40 μM NADH

We tested for product formation after incubating various substrates with NADH and AlAzoR using LC-ESI mass spectrometry. We found the expected mass ions for the reduced products of the azo drugs olsalazine, balsalazide, and sulfasalazine; the non-azo food dyes brilliant blue and erythrosine; and other azo compounds 2-(4-hydroxyphenylazo)-benzoic acid, methyl red, and ethyl red (Table S3). We sought to further support substrate reduction by detecting changes in their ¹H NMR spectra before and after incubation with AlAzoR. After overnight incubation with methyl red, phenazopyridine, and sulfasalazine (Figs, S2a, S2b, and S2c) the solutions became colorless with clear shifts in the ¹H peaks. Although colorless in the NMR tubes, the methyl red incubation sample would turn black upon mixing in air while the phenazopyridine incubation sample would turn yellow-green. Indigo carmine was not soluble enough to observe in buffer directly by NMR spectroscopy. However, incubation of AlAzoR with indigo carmine formed a lighter blue solution, and the 1H NMR spectrum was determined after extraction.

AzoR enzymes are suspected to be quinone reductases in vivo, so the kinetic constants of some quinones tested in the literature were pursued.(19) AlAzoR demonstrated no observable activity with the simple 1,4-benzoquinone, but 5-hydroxy-2-methylnaphthalene-1,4-dione (HMND) and phenol blue are rapidly reduced by AlAzoR. Kinetic constants could not be determined, as they do not follow Michaelis-Menten kinetics and would decrease in rate as the quinone concentration increased (Figure S3). Instead, we compared the relative activity of this enzyme at a single substrate concentration (Table 3). The lowered activity of the enzyme with increasing amounts of quinone is likely due to covalent inactivation by these electrophilic substrates, as seen in other enzymes.(20)

Table 3.

Relative Activity of AlAzoR on various substrates.

Substrate	Relative Activity
Phenol Blue	100 ± 4 %a
HMND	26 ± 6 %
Methyl Red	2.14 ± 0.07 %
Ethyl Red	1.98 ± 0.17 %
Indigo Carmine	0.40 ± 0.02 %
Phenazopyridine	0.43 ± 0.02 %
Sulfasalazine	0.23 ± 0.06 %

^aSpecific activity was determined at 20 μM substrate concentration. Phenol blue had the highest specific activity of 15.2 ± 0.5 s⁻¹ (product turnover per enzyme)

To determine if AlAzoR is gut abundant and determine the relative diversity of this protein to its AzoR family, we first constructed a sequence similarity network (SSN). An SSN displays a flexible representation of the all-by-all sequence comparison of a protein family, allowing for determining isofunctional clusters of proteins (i.e., orthologs). (25, 28) SSNs contain nodes representing protein sequence(s) and edges that measure sequence similarity if above a user-defined score. The SSN for the Flavodoxin-like fold protein family (PF02525/IPR003680, InterPro version 94.0) was generated using the web-based Enzyme Function Initiative Enzyme Similarity Tool (http://efi.igb.illinois.edu/efi-est/). The 61,954 accession IDs found in these families in the UniProt Knowledgebase were downloaded, and the scores from the sequence comparisons of all proteins to one another were calculated. We compared the substrate specificities in the literature to determine the proper threshold score to identify orthologous AzoRs. (17, 18, 29, 30, 32–42) We mapped these characterized AzoRs to our SSN and changed the threshold score so AzoRs with similar substrate specificities would remain in the same cluster. Using an alignment score of 80 (pairwise sequence identity ≥ 65%, Figure S4) resulted in 52,101 unique protein sequences forming 8,112 clusters. As these clusters are expected to group together orthologous proteins, the hypothesis is that any study of one enzyme in the cluster should apply to the other orthologs.

To prioritize AzoRs found in the human gut, we used the web-based EFI Chemically Guided Functional Profiling tool (https://efi.igb.illinois.edu/efi-cgfp/) to generate markers and determine the abundance of genes that can encode these protein family members in the stool samples from healthy donors provided in the NIH Human Microbiome Project-1 (HMP-1) dataset. (27, 28) Sequences are grouped if they share greater than 85% sequence identity, and consensus sequences and unique motifs are determined for each grouping. The motif markers are then quantified in the NIH HMP-1 dataset, and the abundance is mapped to our SSN. The resulting sequences of this family that are found encoded in the NIH HMP-1 dataset are displayed in Figure 4a. The nodes in the SSN are binned and represented by one node if sequences share 90% identity or greater. The relative abundance of the sequences found in each cluster is then normalized to an average genome size and shown as a heatmap for the individual gut metagenomes in Figure 4b. Together with the SSN, the expectation is that we can prioritize gut abundant AzoRs from different clusters to test for their differing substrate specificities (the UniProt IDs and relative abundance can be found in the Supplemental File). Some clusters in this gut-abundant SSN have representative members that have been characterized previously: PaMdaB from Pseudomonas aeruginosa (17), the AzoR from Escherichia coli (EcAzoR) (43), and the AzoR from Clostridium sp. Marseille-P7770 (CmAzoR) (12). The AlAzoR characterized herein is abundant in the HMP-1 stool metagenomes.

Figure 4.

SSN and gut-abundant clusters of the Flavodoxin-like fold protein family (PF02525/IPR003680, InterPro: 94.0) in the HMP-1 dataset. A. An SSN of the Flavodoxin-like fold proteins from the UniRef90 dataset, only containing sequences of any abundance in the HMP-1 stool metagenomes. Proteins are colored by cluster. The “singleton” nodes at the bottom are colored if the UniRef90 dataset has more than one sequence represented in the UniRef90 dataset. The SSN is displayed using an alignment score of 80 (approximate pairwise sequence identity ≥ 65%) and is visualized with Cytoscape version 3.10.1. B. A heatmap displaying the relative abundance of proteins in HMP-1 stool metagenomes. The individual stool metagenomes are represented on the x-axis, and each cluster from the SSN is on the y-axis. The legend to the right indicates relative abundance through “gene copies per microbial genome.”

Examining the active site pockets of the Flavodoxin-like fold protein family and comparing their activities should aid our efforts to predict the specificity of these enzymes. To find potential amino acid interactions, we aligned the AlphaFold-predicted structure of AlAzoR with the FMN- and methyl red-bound structure of the azoreductase from Pseudomonas aeruginosa (PaAzoR1, PDB code: 2V9C).(35) Potential amino acid contacts between AlAzoR and azo compounds were determined by docking substrates in the methyl red binding site using AutoDock. Docking methyl red into PaAzoR1 results in a similar binding mode to that found in the crystal structure, and a similar pose is found when methyl red is docked into AlAzoR (Figure 5). AlAzoR can also make polar interactions given Arg12 and Asp172. In both structures, the docked substrates show unique interactions between the active sites and only in some cases show a binding mode productive to catalysis as determined by the distance between the N5-atom of FMN and a reducible carbon atom of the substrate (Figure S5). The experimentally determined K_M values, the binding energies from AutoDock, the calculated binding constants from the binding energies (assuming 298 K), and the distance between the reacting atoms of the FMN and docked substrate are given in Table 4.

Figure 5.

Comparison of docking methyl red into PaAzoR1 (PDB 2V9C) and an AlAzoR model. Coloring follows the CPK coloring scheme. A) The determined structure of PaAzoR (cyan) has methyl red (yellow) and FMN (white) in its active site. Docking methyl red (pink) results in a similar binding mode, making hydrophobic contacts with FMN and active site residues. B) An AlphaFold model of AlAzoR (slate) with docked methyl red (salmon) in the active site. The AlAzoR model was created by aligning the monomers with PaAzoR and the bound FMN.

Table 4.

Comparison of experimental K_M and computational K_d values

	PaAzoR1a				AlAzoR
Substrate	BE (kcal/mol)	Calc Kd (μM)	KM (μM)	N5 dist (Å)	BE (kcal/mol)	Calc Kd (μM)	KM (μM)	N5 dist (Å)
methyl red	−7.525	4.98	76	4.1	−8.054	2.11	17	3.7
sulfasalazine	−9.110	0.38	69	3.5	−9.299	0.28	8	3.9
balsalazide	−8.811	0.62	124	3.5	−8.916	0.52	N.A.	3.9
olsalazine	−8.515	1.00	104	3.4	−9.066	0.41	N.A.	3.8
phenazopyridine	−7.800	3.19	N.D.	3.5	−7.036	11.02	23	4.3
amaranth	−10.611	0.03	N.A.	8.6	−10.965	0.02	N.A.	8.6
orange II	−9.144	0.36	N.A.	3.5	−8.984	0.47	N.A.	3.9
phenol blue	−7.081	10.25	N.D.	3.6	−7.055	10.68	N.A.	4
plumbagin	−8.478	1.06	N.D.	3.7	−7.610	4.34	N.A.	3.9

^a- K_M values are provided in ref 40

N.D. – not determined in ref 40

N.A. – not applicable

Discussion

The well-behaved AlAzoR can efficiently reduce the azo drugs sulfasalazine and phenazopyridine, along with the food dye indigo carmine. AlAzoR can use both NADH and NADPH to reduce the bound FMN cofactor but shows a clear preference for NADH for maximal activity, although we do not see saturation by NADH. Comparing the relative k_cat/K_M values shows that the pH indicators methyl red and ethyl red are the most efficiently utilized substrates tested with AlAzoR (although previous reports suggest that members of this family are functionally quinone oxidoreductases in vivo, and our initial kinetic data supports this hypothesis). (19) Although a full determination of kinetic parameters for the quinone substrates was not possible, potentially due to inhibition or inactivation of AlAzoR, this finding presents a future area of study to find how AzoRs in this family may show more resistance to this inhibition. AlAzoR represents another gut-abundant AlAzoR that has been shown to reduce sulfasalazine, but AlAzoR is, to our knowledge, the first gut-abundant protein characterized in vitro to reduce this Phenazopyridine. Phenazopyridine is known to be reduced by gut microbes, where in one study, 62 of 76 common strains of gut bacteria showed the reductive metabolism of phenazopyridine; it is important to provide the identification of the protein and corresponding genes to be predictive in metagenomic sequencing. (7) One metabolite of phenazopyridine reduction, 2,3,6-triaminopyridine, is a toxic compound known to cause methemoglobinemia and has been reported to form unstable solutions that turn green and yellow, as witnessed in our NMR experiments. (44, 45) The black color observed after mixing the methyl red reaction is likely due to reactions with molecular oxygen and one product, N,N-dimethyl-1,4-phenylenediamine. These nonenzymatic reactions likely explain why our LC-MS experiments did not observe the products of some of our substrates, so we sought to use ¹H NMR to show structural and electronic changes for proof of substrate modification.

The reduction of indigo carmine has been previously reported for another AzoR and demonstrates the diverse substrates that “azoreductases” can reduce. (42) It is suggested that the azo bonds are not directly reduced, but substrates like sulfasalazine are reduced in a hydrazone tautomer, so this reactivity is consistent with previous observations (Figure 6). (19, 37) The proposed reduction mechanism requires a hydride transfer from N5 of FMNH₂ (or FMNH⁻) to the ortho position of the corresponding ketone or imine generated from a hydroxyl or amino group, respectively. Reduction of the hydrazone forms a hydrazine intermediate, which again tautomerizes to a quinone imine, breaking the N-N single bond. The resulting quinone imine is hypothesized to be reduced faster than the azo-bonded parent substrate. (19) The reduction of indigo carmine likely follows a similar mechanism.

Figure 6. AzoR-catalyzed reduction of azo-bonded substrates.

A. The tautomerization of sulfasalazine converts the azo bond to a hydrazone, forming a quinone imine. B. AzoR reduction of the quinone imine by bound FMNH₂ results in the formation of a hydrazine intermediate and the oxidized FMN.

To date, azoreductases have only recently been studied in the context of enzyme families by creating hidden Markov models of the AzoR clades (vide supra) or searching for proteins >25% identical to known AzoRs containing critical amino acids. (11, 12, 18) Here, we provide an SSN of the 61,106 proteins in the Flavodoxin-like fold protein family, containing the only known enzymes to reduce the prodrug sulfasalazine. (12, 17) This SSN was separated into isofunctional clusters with the expectation that single proteins can be chosen to represent the cluster, and we can prioritize unstudied clusters in future research projects (Figure 5a). We then mapped our SSN to the NIH HMP-1 dataset to further prioritize UniProtKB sequences if they are encoded in gut metagenomes (Figure 5b). This analysis has limitations: we require that 1) the sequences determined in one limited metagenome dataset have gene sequences that encode proteins found in the UniProtKB, 2) microbial genomes are fully and evenly sequenced, and 3) that the abundance of a gene in a metagenome indicates the abundance of the encoded protein in a microbial community. However, this SSN should not be used to determine the abundance of every AzoR in the gut microbiome but to highlight the number of different gut-derived AzoR orthologs that define the diverse activities in the protein family. Our approach provides a method to prioritize AzoRs found ubiquitously across these metagenome samples or in high abundance in some, as each has unique clinical implications. The SSN shows several groups of uncharacterized proteins, with the few characterized members representing AzoRs found in high abundance in a fraction of the individual metagenomes. Several clusters appear ubiquitously found in individual metagenomes but have yet to be studied. The heatmap indicates that some individuals may not have the common, ubiquitous AzoRs but are instead replaced with other sequences that presumably have unique substrate specificity.

Our resulting SSN is broader than the recent SSN provided in Simpson et. al. (12) with a total of 528 unique AzoRs compared to 192 AzoRs. The authors use the Integrative HMP-2 dataset, focusing on the Irritable Bowel Disease Multi-Omic Database (IBDMDB) and the Integrated Gene Catalog (IGC), with three times the number of protein-coding sequences compared to the HMP-1 dataset used to select our AzoRs encoded in individual human metagenomes. This dataset does not focus on IBD-associated metagenomes, but on those from healthy individuals, as AzoRs can reduce molecules outside the context of this disorder. Although some differences in the relative numbers of AzoRs may be due to the different sources and sizes of the reference metagenomes, our expanded dataset is likely due to identifying members of the Flavodoxin-like fold protein family as defined by the Hidden Markov Models generated by InterPro (in family IPR003690) and Pfam (in family PF02525). Simpson et. al. defined AzoRs through the conservation of five specific residues: Ser16, Pro95, Lys/Arg105, Asp/Glu109, and Gly142 (using the Escherichia coli AzoR UniProt ID P41407 numbering). This strategy results in a definition of AzoR that is too strict, missing proteins with azoreductase activity. For example, PaMdaB from Pseudomonas aeruginosa ATCC 15692 (UniProt ID Q9I0Q6) is a member of IPR003690 and PF02525 and is capable of reducing methyl red, sulfasalazine, and other azo drugs, but does not contain an equivalent to Ser16 or Gly142 (instead an Asn at each position).(17) Although PaMdaB is not a representative in the gut metagenome, this protein is part of a cluster containing such AzoRs (e.g. AzoR from Citrobacter sp. JGM124, UniProt ID A0A941Z646, with 65.3% ID to PaMdaB, containing Asp16 and Asn 142). AzrB from Bacillus sp. B29 (UniProt ID C0STY0) is also a member of IPR003690/PF02525, capable of reducing methyl red and ethyl red, but does not contain an equivalent to Lys/Arg105 or Asp/Glu109 (instead an Ile and Ser, respectively).(29)

After identifying likely amino acids in AlAzoR that can contribute to specificity, we wanted to compare the conservation of these residues in similar enzymes to AlAzoR and to the larger protein family. Two AzoR enzymes that have been studied extensively include PaAzoR1 from P. aeruginosa PAO1 (UniProt ID Q9I5F3) and EcAzoR from E. coli (UniProt ID P41407).(21, 22, 33, 35, 37, 46–48) Using our SSN as a guide, we constructed WebLogos of the isofunctional proteins in each cluster and found the conserved amino acids in each position of their active site (Figure 7, Figures S6–8). (49) In a structure with a bound FMN and methyl red, PaAzoR1 forms hydrophobic interactions with methyl red using residues Val 56, Phe 60, Leu 93, and Phe 173, along with hydrogen bonds from Asn 99 and the backbone atoms of Gly147. (35, 46) AlAzoR has Arg 58, Leu 62, Cys 120, Phe 157, Asp 92, and Gly 136 in these positions, indicating a change in the active site to contain more charged residues. AlAzoR has three Arg residues in the active site, providing more opportunities to form ionic interactions with azo drugs and dyes, generally bearing negative charges. The predicted structure of AlAzoR determined by AlphaFold has a more open active site than PaAzoR due to a five-residue deletion, pulling back a loop-turn helix in PaAzoR and opening the active site (Figure S9). In PaAzoR, this loop-turn helix contains Tyr 131, which forms hydrophobic interactions with its substrates. This hydrophobic interaction is unique to PaAzoR as this insertion is missing in AlAzoR and EcAzoR.

Figure 7.

The most common active site amino acids in the clusters containing AlAzoR, PaAzoR1, and EcAzoR from the SSN in Figure 2. Amino acids are given in their one-letter code, with the numbering corresponding to AlAzoR.

We sought to determine if docking of substrates into the determined PaAzoR1 structure and the AlphaFold model of AlAzoR would allow us to be predictive in relative activity by these enzymes. We converted the binding energy to a calculated K_d value by assuming the free energy change of binding at 298K was the same as the computationally determined binding energy. We then compared this to the experimentally determined K_M values (Table 4). Although this conversion of the binding energy does not directly relate to K_d values, and K_M is limited by but not equal to K_d, the computational results are generally consistent with the experimental K_M values. Both AlAzoR and PaAzoR cannot utilize amaranth as a substrate, and the docking results suggest that amaranth can bind well to both enzymes but in an unproductive mode for catalysis. Several substrates are predicted to bind in catalytically viable modes and with binding energies comparable to verified substrates but display no observable activity with our AlAzoR. These results suggest that docking studies of AzoRs may predict relative activities but not provide evidence if a substrate can be reduced by these enzymes.

There is a solved structure of EcAzoR with a bound FMN (PDB 1V4B), although there are no structures with a bound substrate. The corresponding active site residues of EcAzoR closely match those previously described for AlAzoR, although other residues in the active site are different (Figure 7). These three AzoRs all use methyl red and sulfasalazine as a substrate.(17) EcAzoR can uniquely reduce the food dye, amaranth. A recent article hypothesized that amaranth is reduced by an AzoR from Clostridium sp. Marseille-P7770 (CmAzoR) because of potential ionic interactions with Arg 12 and Arg 17.(12) AlAzoR shares these residues but does not have observable activity with amaranth. Given the lack of tested AzoRs, substrate-bound crystal structures, and mutagenesis data, the basis of AzoR specificity is still not well determined.

AlAzoR, EcAzoR, and PaAzoR have the highest k_cat/K_M values for azo-dye methyl red and ethyl red, continuing to show the highest activities for AzoRs, and may be a useful substrate for direct comparisons of enzyme efficiencies.(35) AlAzoR and EcAzoR show similar efficiencies (1.9 × 10⁴ M⁻¹ s⁻¹ and 3.5 × 10⁴ M⁻¹ s⁻¹, respectively), while being 100-fold more efficient than the recently characterized CmAzoR. We have developed an assay to determine the kinetic constants using sulfasalazine, so although we cannot compare these parameters, a comparison of the specific activity of EcAzoR for sulfasalazine is 7-fold greater than CmAzoR and 100-fold greater than AlAzoR under similar conditions.(12) There is a wide range in reported K_M values for NADH between the enzymes (31.6 μM for EcAzoR, 464 μM for PaAzoR, and 1500 μM for AlAzoR) and for NADPH (1100 μM for PaAzoR and 170 μM for AlAzoR), but k_cat/K_M values suggest that PaAzoR prefers NADPH and AlAzoR prefers NADH.(12, 33, 35)

A comprehensive study on the structure-function relationship of this enzyme family only begins to provide predictive power for azo bond and food dye reduction in the human gut. This protein family is one of three that has been identified to reduce azo bonds, and hydrogen sulfide can nonenzymatically reduce azo food dyes like Red 40 and the azo drug sulfasalazine when provided free FMN. (13, 23) Organisms such as E. coli may degrade these food additives and pharmaceutical compounds through both enzymatic and nonenzymatic methods, and a holistic approach may be needed to understand and control azo bond metabolism. Some structures, such as phenazopyridine, resist hydrogen sulfide reduction and may be privileged molecules that can only be enzymatically reduced. (23) The next step is to determine these enzymatic and nonenzymatic reduction rates and the potential overlap between these AzoR enzyme families and reduction by hydrogen sulfide.

We have analyzed a gut-abundant AzoR capable of reducing azo drugs, azo food dyes, and the non-azobonded food dye, indigo carmine. We then made a network of gut-abundant azoreductase enzymes to find the diverse activities of the protein family known to reduce azo drugs. We hope this can scaffold future efforts to explore the family and focus on clinically significant enzymes responsible for drug and food dye modifications. The reduction of food dyes has clinical implications, as it has been demonstrated that the activity of azoreductases on food dyes prevents the inhibition of intestinal drug transporters.(50) Although we have found some differences between characterized azoreductases and linked these activities to their active site residues, more work is still necessary to predict enzyme activity. Substrate-bound structures and kinetics on a wider pool of substrates, coupled with mutagenesis, should help define these enzymes’ specificities and rationalize the differences in the molecules these enzymes can reduce. Recent results find that sulfasalazine is a substrate for most of the members tested in this family, but we need to extend this work to other azo drugs and a wider scope of substrates. Although in the early stages, we hope this SSN and kinetic analysis lays the groundwork for predicting drug and food dye metabolism by metagenomic sequencing, leading to better health outcomes for patients with ulcerative colitis, Chron’s disease, arthritis, urinary tract infections, and other disorders.

Experimental procedures

General products and instruments.

Enzyme kinetics and UV-Visible absorption profiles were determined using a Cary 3500 spectrophotometer (Agilent Technologies). Cell suspensions and lysate were centrifuged using a 5910 Ri (Eppendorf) or a Microcentrifuge 24 (USA Scientific).

Protein Expression.

A codon-optimized gene of UniProt A0A1M6LNR3_9FIRM from Anaerotignum lactatifermentan DSM 14214, AlAzoR, cloned into pET28b was purchased from Twist Bioscience (the gene sequence can be found in Table S3). The plasmid was transformed into competent autolysis E. coli strain XJb (DE3) according to the manufacturer’s protocol (Zymo Research). A pET15b plasmid containing the thermostable phosphite dehydrogenase gene, 17x-PTDH, was obtained from Addgene (plasmid #166786).

Liquid cultures of transformed cells were prepared and grown overnight at 37°C in 5 mL LB culture with the appropriate antibiotics. To overexpress the protein, 500 microliters of the overnight culture were added to 50 mL sterile ZYM-5052 media (or 10 mL of overnight culture into 1000 mL media) supplemented with 0.3 mM arabinose and 60 μg/mL kanamycin when expressing AlAzoR and 100 μg/mL ampicillin for PTDH.(51) Cultures were grown at 37°C overnight and centrifuged at 4,000 × g at 4°C for 15 minutes. Cells were resuspended in 5 milliliters of Ni-NTA wash buffer (0.5 M NaCl, 50 mM sodium phosphate, 50 mM imidazole, pH 8.0) per mass of cells and stored at −20°C until purified.

Protein Purification.

Following autoinduction frozen cells (approximately 16.5 g from 2 liters of media, resuspended in 82.5 mL Ni-NTA wash buffer) were thawed at 37°C. When fully thawed, lysozyme solution was added to 0.5 mg/mL, and underwent a second freeze-thaw cycle from −80°C to 37°C. Cells were then spun down at 15,000 × g and 4°C for 45 minutes, and the clarified supernatant was added to 15 mL of HisPur^™ Ni-NTA Resin (Thermo Fisher Scientific) equilibrated with Ni-NTA wash buffer in 50 mL tubes. The resin and supernatant were placed on an end-over-end rotator for 15 minutes at 4 °C and spun down at 500 × g for 2 minutes. The supernatant was removed, 40 mL Ni-NTA wash buffer was added and the resin and wash buffer were put back on the end-over-end rotator for 15 minutes at 4°C. The resin and wash buffer were then transferred to a disposable column. The column was then repeatedly washed with Ni-NTA wash buffer until no trace protein was detected in the flow through by a Bradford Assay (Bio-Rad). Bound protein was eluted using Ni-NTA elution buffer (500 mM NaCl, 50 mM sodium phosphate, 250 mM imidazole, pH 8.0). Purification was verified by SDS-PAGE using a Mini-PROTEAN TGX gel (Bio-Rad) (Figure S8). Collected protein samples were dialyzed overnight at 4°C in 50 mM sodium phosphate buffer at pH 7.2.

Cofactor Determination.

The specific flavin cofactor for AlAzoR was determined by taking a spectral scan of a 3 mL-sample of AlAzoR in 25 mM ammonium bicarbonate buffer at pH 7.8 (A₂₈₀ = 1.391). The protein was denatured through heating from 20°C to 90°C with a ramp rate of 5 °C per minute. Another spectral scan was taken of the now denatured sample, and the extinction coefficients are listed in Table 4 to determine the concentration and flavin cofactor was present. A sample of this heated protein was spun in a 1.5 mL microcentrifuge tube at 15,000 × g for 10 minutes to pellet the protein. The supernatant was then spun through a Pierce Concentrator polyethersulfone columns (10K molecular weight cut-off, Thermo Scientific) for 10 minutes at 15,000 × g, and the flowthrough was submitted for mass spectrometry analysis.

Table 4.

Flavin Cofactor Extinction Coefficients

Wavelength (nm)	FMN ε (M−1cm−1)	FAD ε (M−1cm−1)
280	16,200	16,000
375	9,000	7,200
445	10,700	8,900
450	10,600	9,000

Enzyme Kinetics.

The activity of AlAzoR with its various substrates was measured through a kinetic assay that was conducted in 3 mL of 50 mM MOPS buffer at pH 7.2 over the course of one minute at 22°C. The assay contained a constant amount of either 40 or 400 μM NADH and 3.2–20 μL of substrate solutions prepared in DMSO. Concentrated substrate stock solutions were prepared and diluted 10-fold, to extend the tested substrate range while minimizing error. Upon addition of AlAzoR (in 50 mM sodium phosphate pH 7.2 buffer), the change in absorbance of substrate over time was measured in triplicate at each substrate concentration. The rates were converted to the change in concentration per second per concentration of the enzyme, and fitted to a nonlinear curve using Wolfram Mathematica (Mathematica, Version 13.2). This curve allowed for kinetic constants to be estimated using the modified Michaelis-Menten kinetic equation(52):

$$\mathrm{v}=\frac{k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_{\mathrm{M}}\left[\mathrm{S}\right]}{1+\left(k_{\mathrm{c}\mathrm{a}\mathrm{t}}/K_{\mathrm{M}}\left[\mathrm{S}\right]\right)/k_{\mathrm{c}\mathrm{a}\mathrm{t}}}=\frac{k_{\mathrm{S}\mathrm{P}}\left[\mathrm{S}\right]}{1+\left(k_{\mathrm{S}\mathrm{P}}\left[\mathrm{S}\right]\right)/k_{\mathrm{c}\mathrm{a}\mathrm{t}}}$$

Determining the Extinction Coefficient of Enzyme.

The concentration of AlAzoR was determined using the absorbance at 445 of the boiled sample of AlAzoR with the ${\epsilon }_{445}$ of FMN to determine the concentration of FMN. The concentration of FMN was then converted to the expected absorbance of FMN at 280 nm using the ${\epsilon }_{280}$ of FMN. The calculated FMN absorbance at 280 nm was then subtracted from the absorbance measured of the boiled sample of AlAzoR at 280 nm to calculate absorbance at 280 nm due to the protein. Using the theoretical protein ${\epsilon }_{280}$ of $24410{\mathrm{M}}^{-1}{\mathrm{c}\mathrm{m}}^{-1}$ determined by the ProtParam webtool (https://web.expasy.org/protparam/), the concentration of the protein in the boiled AlAzoR sample was determined. The relative concentration of FMN and protein can determine the percentage of holoprotein that could be active in the reduction reaction. This was then converted into an extinction coefficient for the AlAzoR using the equation below involving the ${\epsilon }_{280}$ of FMN ($\mathrm{16,200}{\mathrm{M}}^{-1}{\mathrm{c}\mathrm{m}}^{-1}$), the theoretical ${\epsilon }_{280}$ of AlAzoR, and the percent of protein loaded with FMN.

$${\mathrm{A}}_{280}={\mathrm{C}}_{\text{protein}}\left(16200{\mathrm{M}}^{-1}{\mathrm{c}\mathrm{m}}^{-1}(\right.\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{o}\left.\mathrm{\%})+24410{\mathrm{M}}^{-1}{\mathrm{c}\mathrm{m}}^{-1}\right)$$

AlAzoR kinetics was measured using a batch of enzyme that was 53.7% holo protein, resulting in an effective extinction coefficient of 31,000 M⁻¹cm⁻¹. This new extinction coefficient can be used with a DS-11FX+ Spectrophotometer (DeNovix) to determine the concentration of the AlAzoR that is loaded with flavin in a given sample. The concentration of AlAzoR was kept low to keep visible light absorbance between 400–500 nm below 0.02 AU.

Ethyl Red.

The activity of AlAzoR with ethyl red was measured in an assay containing 40 or 400 μM NADH and varied concentrations of ethyl red, ranging from 4.8 μM to 180 μM by using either a 4.5 mM or 45 mM ethyl red stock in DMSO. After the addition of 0.552 μM AlAzoR, the change in absorbance at 453 nm (λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (23,300 M⁻¹cm⁻¹).

Methyl Red.

The activity of AlAzoR with methyl red was measured in an assay containing 400 μM NADH and varied concentrations of methyl red, ranging from 2.4 μM to 150 μM by using either a 2.25 mM or 22.5 mM methyl red stock in DMSO. After the addition of 0.400 μM AlAzoR, the change in absorbance at 438 nm (λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (17,300 M⁻¹cm⁻¹).

Indigo Carmine.

The activity of AlAzoR with indigo carmine was measured in an assay containing 400 μM NADH and varied concentrations of indigo carmine, ranging from 24.2 μM-241.9 μM by using either a 3.75 mM or 37.5 mM indigo carmine stock in DMSO. After the addition of 0.243 μM AlAzoR, the change in absorbance at 612 nm (λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (11,400 M⁻¹cm⁻¹).

Sulfasalazine.

The activity of AlAzoR with sulfasalazine was measured in an assay containing 40 μM NADH and varied concentrations of sulfasalazine, ranging from 3.04 μM to 190 μM by using either a 2.85 mM or 28.5 mM sulfasalazine stock in DMSO. After the addition of 0.243 μM AlAzoR, the change in absorbance at 383 nm (off λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (15,800 M⁻¹cm⁻¹).

Phenazopyridine:

The activity of AlAzoR with phenazopyridine was measured in an assay containing 400 μM NADH and varied concentrations of phenazopyridine, ranging from 1.92 μM to 120 μM by using either a 1.8 mM or 18 mM phenazopyridine stock in DMSO. After the addition of 0.644 μM AlAzoR, the change in absorbance at 428 nm (λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (22,300 M⁻¹cm⁻¹).

NADH:

The activity of AlAzoR with NADH and NADPH was measured in an assay containing 120 μM ethyl red and varied concentrations of NADH or NADPH, ranging from 6.4 μM to 400 μM by using either a 6 mM or 60 mM stock in 50 mM MOPS buffer (pH 7.2). After the addition of 0.400 μM AlAzoR, the change in absorbance at 453 nm (λ-max) was measured over time and converted to changes in concentration using the experimentally determined extinction coefficient (23,300 M⁻¹cm⁻¹).

Phenol blue:

The activity of AlAzoR with phenol blue was measured in an assay containing 400 μM NADH and varied concentrations of phenol blue by using a 1.875 mM or 18.75 mM stock in DMSO. After the addition of 0.644 μM AlAzoR, the change in absorbance at 340 nm (due to changes in NADH concentration) was measured over time and converted to changes in concentration using the extinction coefficient 6,220 M⁻¹cm⁻¹. These rates were corrected for the background rate of absorbance change at 340 nm in the presence of NADH but in the absence of phenol blue.

5-hydroxy-2-methylnaphthalene-1,4-dione (HMND):

The specific activity of AlAzoR with HMND was measured in an assay containing 400 μM NADH and varied concentrations of HMND by using a 1.875 mM or 18.75 mM stock in DMSO. After the addition of 0.644 μM AlAzoR, the change in absorbance at 340 nm (due to changes in NADH concentration) was measured over time and converted to changes in concentration using the extinction coefficient 6,220 M⁻¹cm⁻¹. These rates were corrected for the background rate of absorbance change at 340 nm in the presence of NADH but in the absence of HMND.

NMR Spectroscopy.

For sulfasalazine, phenazopyridine, and methyl red, 1.0 mL reaction mixtures were made containing 0.1 mM NADH, 10 mM azo-substrate, 50 mM phosphite, 100 nM AlAzoR (from a 69.03 μM stock in sodium phosphate buffer), 3.5 μM phosphite dehydrogenase (from a 8.84 μM 17x-PTDH stock in sodium phosphate buffer) in 5% d₆-DMSO in 50 mM sodium phosphate buffer (pH 7.2). The reaction mixtures were incubated overnight and analyzed by ¹H NMR (Bruker Ascend 400 MHz) with solvent suppression at the water peak.

Due to difficulties with solubility, the indigo carmine reaction was conducted using the following assay conditions: 0.21 μM NADH, 0.74 μM of phosphite dehydrogenase (from a 8.84 μM 17x-PTDH stock in sodium phosphate buffer), and 5.74 μM of AlAzoR (from a 69.03 μM stock in sodium phosphate buffer) in 4 mL of 50 mM sodium phosphate buffer (pH 7.2). The reaction ran overnight at room temperature in a covered beaker. An extraction was performed with 3 increments of 2 mL of ethyl acetate in a separatory funnel, using sodium sulfate as a drying agent. Solvent was removed, and the residue was dissolved in 750 μL of d₆-DMSO.

Mass Spectrometry.

AlAzoR activity against 27 different substrates was tested through LC-ESI-MS (Table S1). Each reaction contained 0.721 μM AlAzoR, 400 mM NADH, and 182 μM substrate in 50 mM Ammonium bicarbonate buffer (pH 7.8) After an overnight incubation at ambient temperature, the samples were spun down in Pierce Concentrator PES columns (10K MWCO) (Thermo Scientific) for 5 minutes at 15,000 × g. The flow through of the column was then transferred into autosampler vials to be analyzed by LC-ESI mass spectrometry (Brown University). Resulting ions are reported in Table S2.

Sequence Similarity Network and Chemically-Guided Functional Profiling.

The EFI-EST web tool (https://efi.igb.illinois.edu/efi-est/) was used to generate an SSN using the “Familes” (Option B) tab, using the Flavodoxin_fold protein family IPR003680 and the sequences released in the UniProt: 2023–05 / InterPro: 97.0 dataset. Fragments were filtered using a minimum length cutoff of 165 amino acids. The UniRef90 cluster IDs (containing UniProt IDs that are 90% identical or more to a reference sequence) were used to convert the 63,672 proteins into 27,716 metanodes (representing one sequence or more). These nodes were segregated into isofunctional clusters using an alignment score threshold of 80 (corresponding to ~65% amino acid identity) following the substrate preferences identified in Suzuki, the references therein, and others found in the literature.(17, 18, 28–30, 32–42) This alignment score is similar in value to the negative base ten logarithm of the “e-value” obtained by basic local alignments, but provides a database size-independent value. This SSN was then sent to the EFI-CGFP web tool (https://efi.igb.illinois.edu/efi-cgfp/) for chemically-guided functional profiling (CGFP) using the default methods of the tool.(27, 28) This tool creates markers using the ShortBRED algorithm.(53) Briefly, proteins that are at least 85% identical are aligned, a consensus sequence is created, and markers are created from these sequences after ensuring they are unique. These markers are compared to the NIH HMP-1 dataset and are used to calculate the metagenome abundance. The abundance of the protein-coding genes is reported as the median abundance if multiple markers are identified, and metagenome abundances are normalized by average genome size.(54) The proteins identified in the HMP-1 metagenomes are at least 85% identical to the proteins identified in the SSN provided in Figure 5a.

Computational Modeling.

In PyMOL, an AlphaFold model of AlAzoR (AF-A0A1M6LNR3-F1) was aligned to the Pseudomonas aeruginosa azoreductase PaAzoR1 bound to methyl red (PDB: 2V9C) to create a dimeric model of AlAzoR bound to FMN and methyl red. Potential substrates were docked in the methyl red binding site of the PaAzoR1 structure and the predicted AlAzoR model using Autodock on Nanome.(55) In each case, the crystallographic methyl red was removed, and substrates were docked into the methyl red binding site.

• First identification of an azoreductase-dependent reduction of phenazopyridine

• The most ubiquitous gut-derived azoreductases are unstudied

• Computational docking provides some support of azoreductase substrate activity

Abbreviations and nomenclature

AzoR

Azoreductase

SSN

Sequence Similarity Network

PDB

Protein Database

ESI

Electrospray Ionization

NIH

National Institute of Health

CGFP

Chemically Guided Functional Profiling

EFI

Enzyme Function Initiative

EST

Enzyme Similarity Tool

AlAzoR

Anaerotignum lactatifermentans AzoR

HMP-1

Human Microbiome Project-1

ESI

Electron Spray Ionization

PTDH

Phosphite Dehydrogenase

PaAzoR

Pseudomonas Aruginosa AzoR

CPK

Corey–Pauling–Koltun

EcAzoR

Escherichia coli AzoR

NTA

nitrilotriacetic acid