Introduction

The human protein currently annotated as agmatinase (AGMAT) belongs to the arginase or ureohydrolase superfamily. Most members of this family catalyze the hydrolysis of a guanidine group to urea and the respective amines in metabolites such as arginine1, agmatine2 and guanidinobutyrate (GBA)3. In the active site of these enzymes, two metal ions, typically Mn2+, are coordinated by highly conserved Asp and His residues. The di-metal center positions a hydroxide ion for the nucleophilic attack at the guanidine moiety4. AGMAT was proposed to be an agmatinase catalyzing the hydrolysis of agmatine to putrescine and urea. However, the designation of the enzymatic activity relied on indirect evidence by complementation of an agmatinase-deficient yeast strain with the human candidate gene5 and a biochemical study that lacked demonstration of activity using a purified enzyme6. It has been noted before that AGMAT activity has never been demonstrated conclusively and it was speculated by others that so-far unidentified co-factors might be necessary for its activity7. For the recently discovered guanidine hydrolase, also a member of the ureohydrolase family, we showed that its activity depends on the chaperone-mediated insertion of Ni2+ rather than Mn2+ into the active site8. It is conceivable that isolated or recombinant AGMAT similarly lacks essential factors for its activity. However, as we show with the present work, purified recombinant AGMAT was active when the appropriate substrates were provided.

Arginine is by far the most abundant guanidine compound and to current knowledge the direct or indirect precursor of all other linear guanidine derivatives. Apart from its role as proteinogenic amino acid, arginine serves as the precursor for the generation of urea, nitric oxide, polyamines, and creatine. In creatine biosynthesis, the intermediate guanidinoacetate is formed from glycine and arginine by mitochondrial glycine amidinotransferase (GATM)9. Creatine, a bi-substituted guanidine, is used as transitory acceptor of high-energy phosphates in vertebrates with creatine kinases mediating the reversible transfer of a phosphate group between ATP and creatine10. In some lower animals, homologs of creatine kinases use alternative guanidine derivatives such as taurocyamine as phosphate acceptor11. The fundamental roles of the mentioned metabolites in human physiology depend on the presence of the guanidine moiety and many of the producing and degrading enzymes are well characterized.

Besides arginine, guanidinoacetate, and creatine, a range of other linear guanidine derivatives have been detected in the human brain12 and in body fluids such as urine, saliva and blood plasma13,14,15. However, the metabolism and potential physiological roles of guanidine derivatives such as agmatine, taurocyamine (TC), GBA and guanidinopropionate (GPA) are only poorly understood in humans. Agmatine was detected in the human brain16, and many studies have since linked agmatine to neuroprotection or nerve regeneration, and proposed its use as a potential therapeutic for neuropathies17. Agmatine blocks nicotinic receptors18 and binds to imidazoline and α2-adrenergic receptors16 and has been suggested to function as an endogenous neurotransmitter19.

Only two years after the discovery of agmatine in the vertebrate brain, an agmatinase activity was reported in the rat brain20. This activity is mediated by LIM and calponin homology domains-containing protein 1 (LIMCH1), also named agmatinase-like protein (ALP) which also depends on Mn2+ but is not a member of the ureohydrolase family21. These findings raised the question whether agmatine is metabolized in humans by LIMCH1 or by AGMAT. The AGMAT homologs in mouse, rat and related rodents lack some of the conserved residues for Mn2+ coordination and are therefore considered to be catalytically inactive22.

Far less is known about the functions of the additional guanidine derivatives TC, GBA and GPA that have been detected in humans. Generally, concentrations of all kinds of guanidine compounds were higher in patients with renal failure compared to healthy subjects14. GPA acted antihyperglycemic23 by the induction of GLUT4 expression through reduced ATP, creatine and phosphocreatine levels24. Guanidine derivatives were reported to have various effects on the central nervous system. For example, increased concentrations of a range of guanidines have been related to convulsions25,26. GPA has been reported to depolarize neurons27 and TC interacts with neuroreceptors28,29,30,31. These findings demonstrate the general interest in mechanisms that modulate the levels of guanidine derivatives in human physiology.

Considering that AGMAT had been annotated as an agmatinase despite an apparent lack of a demonstration of agmatine hydrolysis by AGMAT, we set out to investigate the reactions catalyzed by this enzyme. When we compared AGMAT to other members of the ureohydrolase family, we observed the highest sequence similarity to guanidinobutyrase (GbuA) from Pseudomonas aeruginosa3. We demonstrate here that the enzyme efficiently hydrolyzed linear guanidino acids such as TC and GBA, whereas GPA and arginine were turned over less efficiently and guanidinoacetate or compounds lacking acidic residues such as agmatine were not accepted as substrates. Modeling of the 3D-structure of AGMAT revealed that two naturally occurring variants of the human enzyme differ in a residue (R/G105) forming part of the substrate-binding pocket. We characterized both variants with regard to their activity towards a broad panel of guanidine-containing compounds and found slightly different substrate specificities. Apart from being present in human diets32,33 and potentially produced by the human microbiome34, we further show that the identified guanidino acid substrates were produced by the promiscuous action of human glycine amidinotransferase. Taken together, the discovery of a guanidino acid hydrolase activity revises some fundamental aspects of the metabolism of guanidine-containing compounds in humans.

Results

Human AGMAT efficiently hydrolyzes guanidino acids

Sequence comparison revealed that the mature form of AGMAT, obtained after removal of the predicted mitochondrial transit peptide (35 amino acids), shared 39.1% identical amino acids with the functionally characterized agmatinase (SpeB) from E. coli and 35.8% with human arginase (ARG1, Fig. 1A and Supplementary Fig. S1). However, two loops that were reported to mediate substrate specificity of SpeB and Arg135 show no conservation between AGMAT and SpeB or Arg1. Furthermore, we noted that guanidinobutyrase GbuA and guanidinopropionase GpuA from Pseudomonas aeruginosa3,36 share considerably higher degrees of similarity with 60.2% and 45.2% identical amino acids, respectively with AGMAT. Since the activity of human AGMAT has never been demonstrated conclusively, we tested whether guanidinobutyrate is hydrolyzed more efficiently compared to the currently assumed substrate agmatine. In accordance with the activity of the more similar enzyme, we observed that recombinant AGMAT catalyzed the hydrolysis of GBA whereas agmatine hydrolysis was not detected (Fig. 1B).

Figure 1
figure 1

Sequence comparison and catalytic activity of human AGMAT variant R105. (A) Amino acid identity of the mature form of AGMAT compared to guandidinobutyrase (GbuA) and guanidinopropionase (GpuA) from Pseudomonas aeruginosa, agmatinase from Escherichia coli (SpeB) and human arginase 1 (ARG1). The numbers of identical (blue) and different (red) residues of the respective enzymes in comparison to AGMAT were derived from a multiple sequence alignment (Supplementary Fig. S1). (B) Specific activity of recombinantly expressed and purified AGMAT in the presence of 10 mM GBA or agmatine (Agm). Column represents the average of technical triplicates and consistent results were obtained with independent preparations (Data from Fig. 2C). (n.d. = not detected; limit of detection ~ 0.02 nmol s-1 mg-1), error bars, s.d.

Full size image
Figure 2
figure 2

Structural model and substrate specificity of GDAH/AGMAT. (A) Structures of guanidine derivatives tested as substrates for GDAH. Boxed substrates were hydrolyzed by both variants of GDAH, whereas substrates in a dashed box were only hydrolyzed by either the R105 or the G105 variant. (B) Close-up of the substrate binding pocket of a GDAH model showing the position of R105 from the neighboring subunit in relation to the substrate taurocyamine (TC). A global view of the two subunits is provided in Supplementary Fig. S2. Details about the modelling and substrate docking are provided in the main text. (C) Specific activities of the two GDAH variants recombinantly expressed in E. coli with various substrates. Columns represent the average of technical triplicates and consistent results were obtained with independent preparations. (n.d. = not detected; limit of detection ~ 0.02 nmol s−1 mg−1), error bars, s.d. For the sensitivity of the colorimetric assay, see the urea calibration curve (Supplementary Fig. S3).

Full size image

The surprising discovery of guanidinobutyrate as substrate for the human agmatinase prompted us to test a variety of different substrates. We found that guanidino acids of a certain chain length such as guanidinobutyrate and taurocyamine are generally accepted as substrates for AGMAT (Fig. 2A). In order to gain structural insights into the substrate preference of AGMAT, we built a model of the 3D-structure of a monomer using the Robetta server37. Members of the arginase or ureohydrolase family commonly form hexamers, where a neighboring subunit contributes to the formation of the substrate-binding groove in another subunit leading to a shared active site. To properly assess the selectivity of the enzyme and the features of the AGMAT active site, we modeled the local quaternary structure of the enzyme based on the arrangement of subunits in the crystal structure of GbuA (PDB entry 3NIO)36 (Fig. 2B). For this, the catalytic residues of AGMAT and GbuA were superimposed to match two neighboring subunits as present in the biological assembly of GbuA. Ions, which are also not included in Robetta models, were placed according to their location relative to the active site residues in GbuA. The residues coordinating two Mn2+ in the active site as well as His201 and Glu320 essential for catalytic activity in human arginase38 are fully conserved between GbuA and AGMAT. In addition, eight out of 10 residues in close proximity to the substrate-binding pocket are conserved between AGMAT and GbuA (Supplementary Fig. S1). Due to this high similarity, GbuA is an excellent template to evaluate the active site features of AGMAT. Within the GbuA structure, it can be seen that the active site of a subunit is capped by residue R72 from a neighboring subunit, so that this residue could interact with the carboxylate group of the substrate guanidinobutyrate. Interestingly, in the equivalent position to R72 in AGMAT (position 105) two variants exist, with either arginine or glycine in this position. In the genomes covered by the NCBI short genetic variation database, the G105 variant is slightly more abundant (57.7%) in the sampled human population compared to the R105 variant (42.3%) (dbSNP, entry rs6429757). Both variants have been used independently in studies aiming to demonstrate agmatinase activity of AGMAT5,6. To assess the possible influence of having a large, positively charged R105 residue instead of a small glycine, R105 was modeled using the rotamer observed in GbuA and taurocyamine, the best substrate in initial experiments, was docked into place in the active site using Autodock Vina. This complete substrate-bound AGMAT model shows that R105 can indeed reach into the active site of the neighboring subunit and potentially interacts with the sulfonic group of taurocyamine (Fig. 2B).

To compare the enzymatic activity and substrate specificity of both AGMAT variants (R105 and G105), we expressed the mature forms of the mitochondrial enzyme lacking the predicted N-terminal transit peptide of 35 amino acids (MitoFates39) in E. coli and purified them by Ni-affinity chromatography (Supplementary Fig. S1). Consistent with previous studies, we observed no agmatine hydrolysis with both variants of purified AGMAT. In contrast, both variants efficiently hydrolized GBA (Fig. 2A,C). When we provided a range of similar guanidine derivatives that differ in chain length and additional substitutions at a substrate concentration of 10 mM, both AGMAT variants additionally hydrolyzed TC, hypotaurocyamine (HTC, a reduced form of TC), and guanidinopropanesulfonate (GPS). For variant R105, the specific activity was three times higher for TC (1.4 nmol s−1 mg−1) compared to the second-best substrate GBA (0.4 nmol s−1 mg−1), whereas in comparison variant G105 showed reduced activity towards TC and slightly enhanced activity for GBA (Fig. 2C). As described above, when TC is placed into the structural model, its sulfonate group is in proximity of the guanidine group of R105 (Fig. 2B). Variant G105 hydrolyzed arginine to a small extent, whereas variant R105 was inactive with arginine but hydrolyzed GPA instead (Fig. 2C). Urea release from agmatine or other guanidine derivatives lacking a negatively charged group (mercaptoethylguanidine, methylguanidine, guanidine) was not detected, except for guanidinoethanol that was a poor substrate for variant G105. Similarly, guanidinosuccinate, guanidinoacetate and creatine were not hydrolyzed by either variant of AGMAT at a substrate concentration of 10 mM. Taken together, AGMAT exhibits a rather broad substrate specificity towards linear guanidino acids with intermediate chain lengths, but is almost strictly dependent on a negatively charged head group. Variant R105 prefers shorter substrates compared to variant G105 that even showed weak activity with arginine. In light of our results, we propose to re-annotate the human enzyme formerly known as agmatinase (AGMAT) as guanidino acid hydrolase (GDAH) to better reflect its enzymatic activity.

Of the identified substrates of GDAH, only TC, GBA and GPA have been detected in human samples according to the Human Metabolome Database (HMDB)40. For this reason, we determined the concentration dependence of GDAH activity for these three compounds. The Michaelis constants (KM) for TC, GBA and GPA hydrolysis by GDAH were all > 50 mM for both, the GDAH variant R105 (Fig. 3A) and the variant G105 (Supplementary Fig. S1). The exact values could not be determined because no saturation was achieved within the concentration limits set by the solubility of the compounds. Throughout the concentration range tested, the specific activity for TC hydrolysis was approximately three times higher compared to GBA hydrolysis with variant R105, and GPA hydrolysis rates were substantially lower. GDAH was further characterized by using variant R105 with TC as best substrate. When GDAH was expressed in normal LB medium, its specific activity was low but the specific activity was increased four-fold when the protein was expressed in medium supplemented with 0.5 mM Mn2+ (Fig. 3B), consistent with a dependence on Mn2+ common to most members of the ureohydrolase family. The apparent pH optimum of GDAH was 10 and the apparent temperature optimum 67 °C (Fig. 3C,D). We determined the activation energy for TC hydrolysis to be 46 ± 1 kJ/mol (Fig. 3D).

Figure 3
figure 3

Kinetic characterization of GDAH variant R105. (A) The specific activity of GDAH variant R105 was determined for different substrate concentrations of TC, GBA and GPA. Means of triplicates were plotted against the concentration with error bars representing the SD. Curves were fitted with Michaelis–Menten kinetics. (B) Activity of purified GDAH expressed in medium supplemented with 0.5 mM Mn2+ (Mn) or without supplemental Mn2+ (w/o). Columns represent the average of technical triplicates and consistent results were obtained with independent preparations. Error bars, s.d. (C) For the determination of the pH optimum of GDAH, specific activity of GDAH is plotted against the pH. GDAH was incubated with 10 mM TC in pH ranges from 5.5 to 11.5 and the specific activity was determined. (D) The temperature dependency of GDAH was assessed by determining the specific activity at 13 °C to 80 °C with 10 mM TC as substrate. Means of triplicates are plotted against the temperature. Inlet shows the Arrhenius-plot for the determination of the activation energy. The linear range of the Arrhenius-plot was used for fitting (red line).

Full size image

GDAH substrates are not used as alternative phosphagens

Various guanidine derivatives ranging from arginine to creatine are used as phosphagens in different organisms, serving as cellular buffers of high energy phosphates. Phosphorylation of the guanidine group of TC yields phosphotaurocyamine, which is used as phosphagen in marine annelids and some protists41,42. In order to shed light on potential physiological roles of the identified GDAH substrates TC, GBA and GPA we tested whether these guanidine derivatives could serve as so far unrecognized phosphagens in humans by getting phosphorylated by one of the human creatine kinase isoforms. For this purpose, we overexpressed and purified the four charaterized human creatine kinase isoenzymes43,44. The creatine kinases were incubated with γ-32P-ATP as phosphate donor and creatine, TC, GBA, or GPA as substrates. Analysis of the reaction products by thin-layer chromatography revealed that all four creatine kinases accepted exclusively creatine as substrate (Fig. 4A). As a control we overexpressed and purified the TC kinase from Arenicola brasiliensis41. With this enzyme we observed a phosphorylation product specifically with TC as substrate under the same assay conditions utilized for the creatine kinases (Supplementary Fig. S1). These results indicate that the physiological function of GDAH is not concerned with the modulation of the levels of potential alternative phosphagens.

Figure 4
figure 4

Formation of guanidino acids by GATM and their potential as phosphagens. (A) Radiograph of the phosphorylation reaction of four different human creatine kinase isoforms. Recombinant human creatine kinases CKB, CKM, CKMT1 and CKMT2 were incubated with γ-32P-ATP and 10 mM of the respective substrate (creatine (crea), GBA, GPA, TC and no substrate (-)). Reaction products were separated by TLC with Silica 60 as stationary phase. A phosphorylation product was only observed for creatine (blue arrow) for all creatine kinase isoforms. (B) Reactions of human GATM with its main substrate glycine in comparison to GABA, β-alanine and taurine forming the canonical GATM product guanidinoacetic acid (GAA) or the guanidino acids GBA, GPA and TC, respectively. Columns represent the average of technical triplicates and consistent results were obtained with independent preparations. Error bars, s.d.

Full size image

GATM synthesizes GDAH substrates

GBA, GPA and TC have been detected in human fluids but an endogenous source has so far not been identified. However, all three compounds have been shown previously to be generated in side reactions of mitochondrial glycine amidinotransferase in rats from the respective amino homologs γ-aminobutyric acid (GABA), β-alanine and taurine45. In order to clarify whether the identified substrates for GDAH could be produced by an endogenous activity, we overexpressed and purified human mitochondrial glycine amidino transferase (GATM) lacking the mitochondrial transit peptide (uniprot entry: P50440) and performed activity assays with the potential substrates. Similar to the previously characterized enzyme from rat, human GATM had the highest activity with glycine, but also GABA, β-alanine and taurine were accepted as substrates with decreasing preference (Fig. 4B). Both GATM and GDAH are mitochondrial enzymes especially expressed in kidney and liver (https://www.ncbi.nlm.nih.gov/gene/2628; https://www.ncbi.nlm.nih.gov/gene/79814).

Discussion

20 years ago, a human gene from the ureohydrolase family has been annotated as agmatinase. However, we demonstrate that GDAH/AGMAT hydrolyzes guanidino acids rather than agmatine (Fig. 2C). Human GDAH has a high degree of sequence similarity with the guanidinobutyrase GbuA of Pseudomonas aeroguinosa and its three-dimensional structure can be predicted with high confidence based on the GbuA structure (PDB 3NIO). Two enzyme variants exist in humans in regard to residue 105 that likely participates in the binding of the substrates. In 57.7% of human sequences arginine is replaced by glycine at this position. Our modeling based on the P. aeroguinosa guanidinobutyrase and subsequent docking of TC suggests that R105 can extend into a neighboring subunit’s active site, positioning R105 to directly contact the negatively charged head group of the substrate. Although we observed residual arginine hydrolysis by GDAH variant G105, agmatine was not hydrolyzed by both tested GDAH variants. Instead, both GDAH variants hydrolyzed TC and GBA efficiently. Variant R105 exhibited the highest activity with TC, whereas variant G105 turned over GBA and TC to the same extent. Both GDAH variants also hydrolyzed GPA but for the G105 variant, this activity was only detectable at very high substrate concentrations far above the concentrations detected in human samples that are in the low µM range (HMDB)40. Still, we cannot exclude that concentrations might be higher in cells or certain compartments such as the mitochondria, where both GDAH and GATM are predicted to be localized (NCBI Gene database). Overall, variant G105 preferred longer substrates compared to variant R105 and even accepted arginine and guanidinoethanol as poor substrates. Both KM and reaction rate for all substrates are representative of only low to intermediate enzymatic activity. However, kinetic constants in this range do not allow to draw conclusions about a potential physiological relevance. For example, we recently reported that despite its relatively high KM the guanidine hydrolase GdmH allowed Synechocystis to utilize guanidine as sole nitrogen source8. The analysis of further substrates that are not known to occur in humans indicated that the activity of GDAH strictly depends on a negatively charged functionality opposite to the guanidine group. The first report of an enzymatic activity of this enzyme allowed a more thorough characterization: GDAH shares common features such as rather high KM values, Mn2+-dependency, and high apparent temperature and pH optima with other members of the ureohydrolase superfamily46.

We show that GDAH substrates GBA, GPA and TC were not accepted as substrates of human creatine kinases but could be formed by the action of human GATM. GATM promiscuously transferred the amidino group of arginine to glycine, GABA, β-alanine and taurine yielding the respective guanidine compounds GBA, GPA and TC as it has been demonstrated for some mammalian homologs before45,47. Expression data and subcellular localization suggest that GDAH and GATM co-localize in the mitochondria of kidney, liver and brain. Even if taurine is a minor substrate of GATM, TC could be formed in substantial amounts via this reaction as taurine is produced in the liver and reaches concentrations of 10–50 mM48. In addition to the activity of GATM, the identified guanidino acids could be produced by so-far unknown metabolic processes. GBA could be an intermediate of agmatine/arginine catabolism via guanidinobutyraldehyde49. However, the activity of diamine oxidase towards agmatine to produce guanidinobutyraldehyde is controversial50. Alternatively, GBA, GPA and TC could originate from direct uptake from the diet32,33 or from gut microbial activities34.

TC has been used since the late 1970’s to manipulate taurine levels in different organs or tissues since it is a competitive inhibitor of the taurine transporter TAUT51,52. Additionally, TC has been reported as an antagonist of GABAA and glycine receptors28,29,30,31. GBA has been detected in the brain, liver and kidney of vertebrates and insects and elevated levels of GBA and TC in the brain of rabbits have been related to convulsions25,26. Given these effects of the identified GDAH substrates, a possible function of GDAH could be the hydrolysis of potentially deleterious concentrations of guanidino acids as a means of detoxification. This hypothesis would also agree with the observed broad substrate specificity of GDAH. GDAH expression has been reported to be reduced in diabetic patients with breast cancer or clear cell type of renal cell carcinoma53,54. GDAH variant R105 of GDAH has been linked to type 1 diabetes55. On the other hand, GDAH was reported to promote lung adenocarcinoma tumorigenesis and the progression of colorectal cancer by inducing chronic inflammation56,57.

The potential involvement of GDAH/AGMAT in agmatine metabolism has received increasing attention from neuroscientists starting in 1994 when agmatine had been detected in the human brain16. Agmatine is discussed to have beneficial effects on disorders of the central nervous system like depression, nerve regeneration and epilepsy among others and is generally believed to be neuroprotective7,17. GDAH is highly expressed in kidney, liver and hippocampal interneurons58,59,60. The presented findings implicate that GDAH is unlikely to contribute directly to the modulation of endogenous agmatine concentrations. Further evidence for a physiological role of GDAH as guanidino acid hydrolase comes from two recent observations: Concentrations of GBA and GPA have been found to negatively correlate with the expression level of GDAH in a genome-wide study of genes affecting metabolite concentrations in saliva15. In addition, a further study identified an association of altered concentrations of GPA with genetic variations in the chromosomal locus of GDAH61. In conclusion, the findings presented in this study revise some fundamental aspects of the metabolism of guanidine derivatives in humans and pose further questions regarding the functional or pathological roles of GDAH and its substrates in human physiology.

Methods

Bacterial cultivation, cloning, protein overexpression and purification

E. coli BL21(λDE3) gold (Invitrogen) or E. coli SoluBL21 (Genlantis) were grown in LB with 50 µg/ml kanamycin when transformed. Genes of interest (human GDAH, GATM, CKM, CKB, CKMT1, and CKMT2, as well as TCK from Arenicola brasiliensis41) were codon optimized for E. coli and synthesized by GeneArt (LifeTechnologies). Mitochondrial target sequences were predicted with MitoFates39 and excluded to obtain mature enzymes. DNA constructs were directly cloned into a pET24 derivative by Gibson assembly to obtain enzymes with an N-terminal 6xHis-tag followed by a TEV cleavage site. For recombinant protein expression, E. coli BL21(λDE3) gold (Invitrogen) or E. coli SoluBL21 (Genlantis), in case of GDAH and CKMT1, were transformed with the expression constructs, grown in LB supplemented with 0.5 mM MnCl2, if not stated otherwise, at 37 °C to an OD600 of 0.6, transferred to 18 °C and induced over night with 1 mM IPTG. The cells were harvested by centrifugation, resuspended in enzyme buffer (50 mM Tris–HCl pH 8, 100 mM NaCl) supplemented with 1 × EDTA-free cOmplete protease inhibitor (Roche) and lysed by ultrasonication (Branson). After centrifugation at 12,000 g for 20 min at 4 °C, the soluble protein fraction was incubated with His-bind NiNTA resin (Qiagen, Hilden, Germany) for 30 min at 4 °C. The resin was loaded into gravity flow columns and washed sequentially with enzyme buffer supplemented with 20 mM or 50 mM imidazole. The His-tagged enzymes were eluted in enzyme buffer supplemented with 500 mM imidazole and desalted into enzyme buffer by passage through PD50 columns (GE lifesciences).

Modeling

A comparative model of AGMAT was produced using the Robetta server37. This produced a model containing just one subunit. Therefore, to reconstitute the oligomeric state, the AGMAT model was aligned to chain A of 3NIO using the catalytic site residues (H129, D152, H154, D156 and D243 of 3NIO). A second copy of the AGMAT model was then structurally aligned to chain B of 3NIO. The active site manganese ions were modeled by using the positions of the ions in 3NIO relative to the active site residues. In the produced model R105 was modeled to have the rotameric conformation of R72 in PaGbuA. The AGMAT models were prepared for docking by removing the hydrogens from original Robetta models and re-adding and reparametrizing them using Chimera UCSF Dock Prep. Hydrogens and charges were calculated using ANTECHAMBER62. The AMBER ff14SB forcefield63 was applied for standard residues and Gasteiger for other residues64. Taurocyamine (TC) parameters (ZINC000004095756) were downloaded from ZINC15; https://zinc.docking.org/ and docked into the active site of AGMAT models using Autodock Vina 1.1.265. A search box large enough to include the entirety of residue 105 was used. Nine binding modes were tested, with an exhaustiveness of eight and a maximum energy difference of 3 kcal/mol. A representative TC docking result where TC occupied the active site was chosen to assess the potential distance between the TC sulfate group and R105 from the neighboring subunit.

Ureohydrolase assay

For the ureohydrolase activity assays, 4–8 µg of the purified enzyme was incubated at 30 °C with 10 mM substrate in a buffer containing 133 mM Tris–HCl, pH 8, 1 mM MnCl2 and 1 mM Na-maleate for 30 to 60 min. The reaction was terminated by the addition of 87% (v/v) acetic acid. For the kinetic analyses, substrate concentrations, buffer and pH, or temperature, were modified as indicated in the respective figures. Urea released by ureohydrolase activity was determined colorimetrically as described in reference66. In brief, 50 µl of enzyme assay were mixed with 100 µl of color reagent (62 mM butanedionmonoxime, 3.6 mM thiosemicarbazide) and 150 µl acid reagent (120 µM FeCl3 and 10 mM phosphoric acid in 20% (w/v) sulphuric acid) and incubated for 10 min at 96 °C. Samples with known concentrations of urea were treated in the same way and were used to establish a calibration curve. The absorption at 520 nm was measured in a 96-well plate (TECAN Spark Reader). Non-linear fitting of the data to Michaelis Menten kinetics {ν = (Vmax [S])/(KM + [S])} was performed with GraphPad Prism6.

Glycine amidinotransferase assay

3.4 µg glycine amidinotransferase was incubated in a buffer containing 80 mM KCl, 20 mM NaCl, 2 mM MgCl2, 80 mM Tris (pH 8) with 10 mM arginine and 10 mM of the indicated substrate. The ornithine concentration was measured as described67,68. In brief, 100 µl sample were mixed with equal volumes of both ninhydrin solution (25 mg ml-1 ninhydrin in 60% (v/v) acetic acid and 13.8% (w/v) phosphoric acid) and glacial acetic acid. The reaction mixture was incubated for 15 min at 96 °C. Reactions were chilled and the absorption at 520 nm was measured in a 96-well plate (TECAN Spark Reader).

Creatine kinase assay

5 µg of the respective creatine or TC kinase isoform was incubated with 10 mM creatine, GBA,GPA or TC and 1 mM ATP and 1 µCi γ-32P-ATP in 100 µl buffer containing 80 mM KCl, 20 mM NaCl, 2 mM MgCl2, 80 mM Tris (pH 8) for 1 h at 30 °C. The reaction mixture was separated by TLC with silica gel 60 (Merck) as stationary phase and isopropanol: 25% NH3: H2O (3:1:1) as mobile phase. Radiographs were recorded with a Typhoon FLA 7000 laser scanner (GE Healthcare).

Non-commercial guanidine compounds

Synthesis and characterization of commercially unavailable guanidino compounds is described in the supplementary information.